Datasets:
Hendrik-a
/
arxiv

Dataset card Viewer Files Community
1
abstract_text
stringlengths
56
26.7k
article_id
stringlengths
9
15
article_text
stringlengths
27.2k
74.7k
we present in this paper multicolor surface and aperture photometry in the and bands for a sample of 34 lenticular galaxies from the ugc catalogue . from surface photometric analysis , we obtain radial profiles of surface brightness , colors , ellipticity , position angle and the fourier coefficients which describe the departure of isophotal shapes from purely elliptical form and find the presence of dust lanes , patches and ring like structure in several galaxies in the sample . we obtain total integrated magnitudes and colors and find that these are in good agreement with the values from the rc3 catalogue . isophotal colors are correlated with each other , following the sequence expected for early - type galaxies . the color gradients in lenticulars are more negative than the corresponding gradients in ellipticals . there is a good correlation between and color gradients , and the mean gradient in the and colors are , , magnitude per dex in radius respectively .
astro-ph0411026
hubble ( 1936 ) introduced lenticular ( s0 ) galaxies as a morphological transition class between ellipticals and early - type spirals . the lenticular galaxies have disks with luminosity ranging from ten to hundred percent of the bulge luminosity , but without any conspicuous spiral arms . from their appearance , and also their stellar content , they seem to be more like ellipticals rather than spirals , and have often been misclassified due to this fact . it has been suggested ( van den bergh 1994 ) that there could be different , but overlapping , sub - populations amongst the lenticulars . there are several scenarios possible for the formation of these galaxies . they could be of primordial origin , or could have been formed by the stripping of gas from spirals , which changes the morphology ( abadi , moore & bower 1999 ) , or through the mergers of unequal - mass spirals ( bekki 1998 ) . there could be a significant difference in the ages of disks and bulges , in which case they should have different stellar population , leading to different colors . bothun & gregg ( 1990 ) found that bulges and disks of lenticulars were well separated in the @xmath9 vs @xmath10 diagram which they interpreted to mean that the disks are younger than bulges by 3 - 5 gyr . but peletier & balcells ( 1996 ) found that the color differences between the disks and bulges are much smaller than those found by bothun & gregg ( 1990 ) , and inferred that the disks are only slightly younger than bulges , with the difference in age being in the range 0 - 3 gyr . in a study of 12 highly inclined lenticulars , michard & poulain ( 2000 ) found that disks were often redder than bulges , which indicates that there could be large concentration of dust in the disks . a detailed multiband study of the morphology of representative samples of lenticulars , with possible separation of bulge and disk components , and comparison of their properties with those of ellipticals , and with the bulges and disks of spirals will be very important in addressing these and other possibilities . we have observed a sample of 34 lenticular galaxies in the optical and infrared bands , and present in this paper their surface and aperture photometry , together with color profiles and mean color gradients . we will in a subsequent publication present a decomposition of the surface brightness distribution of the galaxies into bulge and disk components using the decomposition technique developed by wadadekar , robbason & kembhavi ( 1999 ) . using the results of the decomposition , it should be possible to see where the lenticulars are situated relative to the fundamental plane of ellipticals ( see jorgensen , franx & kjaergaard 1996 and references therein ) and the photometric plane for ellipticals and the bulges of spiral galaxies ( khosroshahi , wadadekar & kembhavi 2000a and 2000b ) . with multiband data , it might be possible to determine in this manner the bulge - to - disk luminosity ratios of these galaxies in different bands , as well as the distribution of different colors as a function of distance from the center in the bulge and disk components separately , thus facilitating the comparison with other galaxy types . our data set will be very useful for detailed comparison with other galaxy types , as well as in constraining stellar population models for the lenticulars . the present paper deals with the details of the sample selection , observations , data reduction and analysis . this paper is organized as follows : the selection criteria for the sample , details of observations and the technique used in data reduction are described in @xmath112 . the analysis and results from the surface photometry and aperture photometry are discussed in @xmath113 . the distribution of color gradients are discussed in @xmath114 and a summary of the results is presented in @xmath115 . an appendix contains comments on individual galaxies . ccccccc ugc 00080 & ngc 0016 & @xmath123.0 & 1.8x1.0 & 0.010194 & 12.50 & @xmath1220.93 + ugc 00491 & ngc 0252 & @xmath121.0 & 1.5x1.1 & 0.016645 & 13.40 & @xmath1221.65 + ugc 00859 & ngc 0473 & 0.0 & 1.7x1.1 & 0.007118 & 13.20 & @xmath1219.82 + ugc 00926 & ngc 0499 & @xmath122.5 & 1.6x1.3 & 0.014673 & 13.00 & @xmath1221.55 + ugc 01250 & ngc 0670 & @xmath122.0 & 2.0x1.0 & 0.012352 & 13.10 & @xmath1220.76 + ugc 01823 & ngc 0890 & @xmath123.0 & 2.5x1.7 & 0.013323 & 12.50 & @xmath1222.31 + ugc 01964 & ngc 0940 & @xmath122.0 & 1.2x1.0 & 0.017319 & 13.40 & @xmath1221.64 + ugc 02039 & ngc 0969 & @xmath122.0 & 1.7x1.6 & 0.015054 & 13.50 & @xmath1221.51 + ugc 02187 & ngc 1040 & @xmath122.0 & 1.7x0.8 & 0.016054 & 14.00 & @xmath1221.06 + ugc 02322 & ngc 1106 & @xmath121.0 & 1.8x1.8 & 0.014467 & 13.70 & @xmath1221.39 + ugc 03087 & & @xmath122.0 & 0.8x0.6 & 0.033010 & 14.20 & @xmath1222.28 + ugc 03178 & ngc 1671 & @xmath122.0 & 1.1x0.9 & 0.021270 & 13.90 & @xmath1221.63 + ugc 03452 & ngc 2208 & @xmath122.0 & 1.7x1.0 & 0.018763 & 14.00 & @xmath1221.26 + ugc 03536 & & @xmath122.0 & 1.1x0.6 & 0.015641 & 14.40 & @xmath1220.86 + ugc 03567 & & @xmath122.0 & 1.0x0.8 & 0.020164 & 14.20 & @xmath1221.26 + ugc 03642 & & @xmath122.0 & 1.5x1.1 & 0.015004 & 13.50 & @xmath1221.47 + ugc 03683 & & @xmath122.0 & 2.0x1.3 & 0.019076 & 14.10 & @xmath1221.53 + ugc 03699 & ngc 2332 & @xmath122.0 & 1.5x1.0 & 0.019467 & 14.00 & @xmath1221.52 + ugc 03792 & & 0.0 & 1.8x1.3 & 0.020608 & 14.00 & @xmath1221.46 + ugc 03824 & & @xmath122.0 & 0.9x0.8 & 0.017866 & 14.40 & @xmath1220.76 + ugc 04347 & ngc 2563 & @xmath122.0 & 2.1x1.5 & 0.014944 & 13.70 & @xmath1221.06 + ugc 04767 & & @xmath122.0 & 1.3x1.1 & 0.024127 & 14.00 & @xmath1221.80 + ugc 04901 & ngc 2804 & @xmath122.0 & 2.2x2.0 & 0.028099 & 14.00 & @xmath1222.23 + ugc 05292 & ngc 3032 & @xmath122.0 & 2.0x1.8 & 0.005114 & 13.00 & @xmath1219.43 + ugc 06013 & & @xmath122.0 & 0.9x0.8 & 0.021898 & 13.90 & @xmath1221.69 + ugc 06389 & ngc 3648 & @xmath122.0 & 1.3x0.8 & 0.006631 & 13.50 & @xmath1219.50 + ugc 06899 & ngc 3971 & @xmath122.0 & 1.4x1.2 & 0.022489 & 13.90 & @xmath1221.75 + ugc 07142 & ngc 4143 & @xmath122.0 & 2.3x1.4 & 0.003616 & 12.00 & @xmath1220.03 + ugc 07473 & ngc 4350 & @xmath122.0 & 3.0x1.4 & 0.004140 & 11.50 & @xmath1220.48 + ugc 07880 & ngc 4638 & @xmath123.0 & 2.2x1.4 & 0.003883 & 12.20 & @xmath1219.64 + ugc 07933 & ngc 4673 & @xmath125.0 & 1.0x0.9 & 0.022856 & 13.70 & @xmath1221.99 + ugc 08675 & ngc 5273 & @xmath122.0 & 2.8x2.5 & 0.003616 & 12.70 & @xmath1219.24 + ugc 09200 & ngc 5580 & @xmath122.0 & 1.8x1.8 & 0.010814 & 13.60 & @xmath1220.46 + ugc 09592 & ngc 5784 & @xmath122.0 & 1.9x1.8 & 0.017912 & 13.70 & @xmath1221.78 + ugc 11178 & ngc 6599 & @xmath122.0 & 1.3x1.2 & 0.010120 & 13.70 & @xmath1220.30 + ugc 11356 & ngc 6703 & @xmath122.5 & 2.5x2.3 & 0.008209 & 12.40 & @xmath1221.14 + ugc 11781 & & @xmath122.0 & 1.4x1.1 & 0.015501 & 13.70 & @xmath1221.74 + ugc 11972 & ngc 7248 & @xmath122.5 & 1.7x0.9 & 0.014627 & 13.60 & @xmath1221.22 + ugc 12443 & ngc 7539 & @xmath122.0 & 1.5x1.2 & 0.020174 & 13.70 & @xmath1221.88 + ugc 12655 & & @xmath122.0 & 1.4x0.9 & 0.017269 & 14.00 & @xmath1221.24 our sample consists of 40 bright and medium sized galaxies , classified as lenticulars , from the uppsala general catalogue of galaxies ( ugc ) . the ugc is essentially complete to a limiting major - axis diameter of one @xmath13 and/or to a limiting apparent magnitude of 14.5 on the blue prints of the palomar observatory sky survey ( poss ) . coverage is limited to the sky north of declination @xmath14 . for our sample we selected galaxies with apparent blue magnitude brighter than @xmath15 , diameter @xmath16 and declination in the range @xmath17 . the sample galaxies are listed in table 1 . our sample , though not complete , is representative of lenticular galaxies in the field with the above properties . our optical @xmath18-band observations were carried out with the _ observatorio astrofsico guillermo haro _ 2.1-m telescope at cananea , mexico . we used a tektronics ccd of @xmath19 pixel format at the @xmath20 cassegrain focus of the telescope with @xmath21 pixel binning , resulting in a scale of @xmath22 pixel@xmath23 and a field of view of @xmath24 . typically two exposures each of ten , five and five minutes in @xmath25 , and @xmath26 bands respectively were taken for each galaxy . exposure times were reduced if the central pixels were close to saturation . twilight sky exposures were taken for flat - fielding purposes . several bias frames were obtained at the start and end of each night . all the observations were carried out in three runs in december 2001 , february 2002 and october 2002 . a total of 34 galaxies were observed in the optical bands . for the galaxies ugc1964 , 3178 , 4901 , 6013 and 7933 from the observed set , the quality of the images was not good enough for photometric analysis . table 2 contains a detailed log of the observations . we obtained images of all 40 sample galaxies in the near infrared @xmath1 band with the _ observatorio astronomico nacional _ 2.1-m telescope at san pedro martir , mexico . the camila instrument ( cruz - gonzalez 1994 ) , which hosts a nicmos 3 detector of 256@xmath27256 pixel format , was used in the imaging mode with the focal reducer configuration @xmath28 in all our observations . this results in a spatial resolution of @xmath29 pixel@xmath23 and a total field of view of @xmath30 . each @xmath1 observation consisted of a sequence of object and sky exposures , with the integration time of an individual exposure limited by the sky counts ( or in some cases the nucleus ) , which was kept well below the non - linear regime of the detector . a typical @xmath1 image sequence consisted of 10 exposures , six on the object and four on the sky . the net exposure times were typically of 10 minutes . a series of twilight and night - sky images were taken for flat - fielding purposes . the @xmath1 observations were carried out in four runs in december 2000 , march 2001 , october 2001 and march 2002 . the sky conditions for both optical and near infrared observations were generally photometric and the seeing _ fwhm _ was in the range 1@xmath312@xmath31 on different nights . the average sky brightness was 21.14 , 20.72 , 20.18 and 12.27 magnitude arcsec@xmath32 in the @xmath33 and @xmath1 bands , respectively . the sky brightness in the @xmath1 band also includes the background emitted by the warm optics . ccccccccccc ugc 00080 & 04 - 10 - 2002 & 10 - 10 - 2001 & 1200 & 360 & 540 & 750 & 2.24 & 2.07 & 2.03 & 1.56 + ugc 00491 & 05 - 10 - 2002 & 12 - 10 - 2001 & 1200 & 600 & 600 & 750 & 1.85 & 2.18 & 2.18 & 1.52 + ugc 00859 & 14 - 12 - 2001 & 09 - 10 - 2001 & 1200 & 600 & 600 & 900 & 2.41 & 2.81 & 2.64 & 1.76 + ugc 00926 & 05 - 10 - 2002 & 10 - 10 - 2001 & 1200 & 360 & 540 & 750 & 1.80 & 1.66 & 1.78 & 1.49 + ugc 01250 & 04 - 10 - 2002 & 10 - 10 - 2001 & 1200 & 600 & 600 & 900 & 2.31 & 2.10 & 2.01 & 1.48 + ugc 01823 & 14 - 12 - 2001 & 09 - 10 - 2001 & 1200 & 600 & 600 & 750 & 2.43 & 2.78 & 2.67 & 1.61 + ugc 01964 & 14 - 12 - 2001 & 09 - 10 - 2001 & 1200 & 600 & 600 & 900 & 2.10 & 2.09 & 2.18 & 1.60 + ugc 02039 & 05 - 10 - 2002 & 12 - 10 - 2001 & 1200 & 600 & 600 & 600 & 1.91 & 2.11 & 2.01 & 1.64 + ugc 02187 & ... & 10 - 10 - 2001 & .... & ... & ... & 900 & ... & ... & ... & 1.47 + ugc 02322 & ... & 09 - 10 - 2001 & .... & ... & ... & 780 & ... & ... & ... & 1.49 + ugc 03087 & 08 - 02 - 2002 & 24 - 03 - 2002 & 1200 & 540 & 540 & 480 & 2.16 & 1.98 & 2.14 & 2.06 + ugc 03178 & 09 - 02 - 2002 & 23 - 03 - 2002 & 1200 & 600 & 600 & 320 & 2.02 & 1.87 & 1.87 & 1.75 + ugc 03452 & 14 - 12 - 2001 & 16 - 12 - 2000 & 1200 & 600 & 600 & 600 & 2.55 & 2.76 & 2.63 & 1.58 + ugc 03536 & 08 - 02 - 2002 & 26 - 03 - 2002 & 1200 & 540 & 540 & 105 & 2.14 & 2.01 & 2.05 & 1.54 + ugc 03567 & 07 - 02 - 2002 & 28 - 02 - 2002 & 1200 & 600 & 600 & 750 & 2.17 & 1.97 & 2.04 & 1.92 + ugc 03642 & 09 - 02 - 2002 & 26 - 02 - 2002 & 1200 & 720 & 720 & 750 & 1.86 & 2.04 & 1.90 & 1.48 + ugc 03683 & 08 - 02 - 2002 & 23 - 03 - 2002 & 1200 & 600 & 600 & 750 & 2.26 & 2.26 & 1.97 & 2.07 + ugc 03699 & 05 - 10 - 2002 & 12 - 10 - 2001 & 600 & 360 & 540 & 750 & 2.58 & 2.58 & 2.59 & 1.89 + ugc 03792 & 14 - 12 - 2001 & 10 - 10 - 2001 & 1200 & 600 & 600 & 750 & 2.28 & 2.69 & 2.43 & 1.85 + ugc 03824 & 07 - 02 - 2002 & 26 - 03 - 2002 & 1800 & 600 & 600 & 105 & 2.13 & 2.13 & 2.24 & 1.66 + ugc 04347 & 14 - 12 - 2001 & 08 - 03 - 2001 & 1200 & 600 & 600 & 540 & 2.51 & 2.71 & 2.55 & 1.88 + ugc 04767 & 08 - 02 - 2002 & 24 - 03 - 2002 & 1200 & 600 & 600 & 600 & 2.42 & 2.07 & 2.08 & 1.92 + ugc 04901 & 09 - 02 - 2002 & 26 - 03 - 2002 & .... & 600 & 600 & 105 & ... & 1.75 & 1.85 & 1.45 + ugc 05292 & ... & 07 - 06 - 2001 & .... & ... & ... & 900 & ... & ... & ... & 1.67 + ugc 06013 & 07 - 02 - 2002 & 08 - 02 - 2001 & .... & 600 & 600 & 840 & ... & 2.50 & 2.07 & 1.64 + ugc 06389 & 08 - 02 - 2002 & 23 - 03 - 2002 & 1200 & 540 & 540 & 700 & 1.92 & 2.05 & 2.03 & 1.89 + ugc 06899 & 08 - 02 - 2002 & 23 - 03 - 2002 & 1200 & 600 & 600 & 840 & 1.96 & 1.97 & 1.94 & 1.27 + ugc 07142 & ... & 23 - 03 - 2002 & .... & ... & ... & 450 & ... & ... & ... & 2.07 + ugc 07473 & 09 - 02 - 2002 & 22 - 03 - 2002 & 900 & 450 & 270 & 260 & 2.12 & 2.13 & 2.08 & 1.43 + ugc 07880 & 09 - 02 - 2002 & 25 - 03 - 2002 & 1200 & 270 & 600 & 500 & 2.37 & 2.64 & 2.41 & 1.79 + ugc 07933 & 07 - 02 - 2002 & 08 - 03 - 2002 & .... & 540 & 540 & 840 & ... & 2.47 & 2.26 & 1.79 + ugc 08675 & 08 - 02 - 2002 & 22 - 03 - 2002 & 1200 & 600 & 540 & 310 & 2.04 & 1.91 & 1.82 & 1.63 + ugc 09200 & ... & 07 - 03 - 2001 & .... & ... & ... & 840 & ... & ... & ... & 1.50 + ugc 09592 & ... & 24 - 03 - 2002 & .... & ... & .... & 600 & ... & ... & ... & 1.92 + ugc 11178 & 04 - 10 - 2002 & 22 - 03 - 2002 & 900 & 600 & 600 & 240 & 2.15 & 2.06 & 2.20 & 1.39 + ugc 11356 & 05 - 10 - 2002 & 23 - 03 - 2002 & 1200 & 240 & 360 & 420 & 1.91 & 1.91 & 1.91 & 1.80 + ugc 11781 & 04 - 10 - 2002 & 09 - 10 - 2001 & 900 & 600 & 600 & 540 & 1.99 & 1.94 & 2.07 & 1.83 + ugc 11972 & 05 - 10 - 2002 & 10 - 10 - 2001 & 900 & 540 & 360 & 900 & 2.07 & 1.79 & 1.78 & 1.50 + ugc 12443 & 14 - 12 - 2001 & 12 - 10 - 2001 & 1200 & 600 & 600 & 900 & 2.87 & 2.89 & 2.92 & 1.56 + ugc 12655 & 04 - 10 - 2002 & 09 - 10 - 2001 & 1200 & 600 & 600 & 120 & 2.24 & 1.98 & 2.12 & 1.59 the basic data reduction for both the optical and near infrared @xmath1 images involved subtraction of the bias and sky frames , division by flat field frames , registration of the images to a common co - ordinate system and then stacking all the images of a given galaxy in each filter . night to night variations of the optical bias frames were negligible , and hence bias frames of an entire run were stacked together using the median algorithm to form a master bias frame , which was then subtracted from all the other frames . preparation of the optical flat fields followed the conventional technique , wherein bias subtracted flats were stacked and the resultant frame was normalized to its mean value to form a master flat in each filter . bias subtracted images of the program galaxies were divided by the normalized flat field in the corresponding filter . the optical images suffered from a stray light problem that resulted in a gradient in the sky background , which roughly ran through one of the diagonals of the ccd chip . the gradient was found to be stable throughout each run and the mean counts scaled linearly with exposure time . after several experiments , we found that the best way to get rid of the gradient was to subtract a mean blank sky image from the data images . for this purpose , special blank fields were observed in each filter with exposure times matching the typical exposure times of the object frames . in the december 2001 run , three blank sky frames were used in each filter to prepare a median gradient image , while in the february 2002 run , six frames were used . the adopted procedure eliminated any gradient from the sky background . for the @xmath1 images , a bias frame taken immediately before an object exposure was subtracted as part of the data acquisition . a master @xmath1 flat field for each night of observing was prepared as follows . the night - sky flats were first stacked and then subtracted from stacked twilight flats . the frames obtained in this fashion for each run were then combined and normalized to the mean value of the resultant frame to form a master @xmath1 flat . the sky frames of each sequence of observations were combined and the resultant image was subtracted from each of the object frames to get a sky - subtracted image . flat fielding was done by dividing the sky subtracted images of the object by the normalized master flat . the resulting images were aligned to a common co - ordinate system using common stars in the frames and then combined using the median operation . only good images ( as defined in the camila manual see cruz - gonzalez et al . 1994 ) were used in the @xmath1 combination . the resulting combined @xmath1 images were aligned to corresponding images from the digitized sky survey ( dss ) . as a final step of the reduction procedure , the mutually aligned optical images were aligned to the @xmath1 image coordinate system . the transformed star positions in the images agreed to within @xmath34 as judged from the coordinates of common stars . all image reductions were carried out using the image reduction and analysis facility ( iraf ) and the space telescope science data analysis system ( stsdas ) . the iraf external package _ color _ was used to make color composite images , by combining images taken in three different bands , @xmath35 and @xmath26 or @xmath36 and @xmath1 . gray - scale representation of these color composite images are shown in fig . 1 , where north is up and east is to the left . dipper asterism stars in the m67 field were observed to enable accurate photometric calibration of our optical observations . the stars in this field span a wide color range ( @xmath37 ) , which includes the range of colors of the program galaxies , and hence are suitable for obtaining the transformation coefficients to the cousins @xmath18 system defined by bessell ( 1990 ) . the transformation equations are @xmath38 where @xmath35 and @xmath26 are standard magnitudes , @xmath39 and @xmath40 are extinction corrected instrumental magnitudes , @xmath41 and @xmath42 are the zero points and @xmath43 and @xmath44 the color coefficients in bands @xmath35 and @xmath26 respectively . typical extinction coefficients for the observatory ( 0.20 , 0.11 and 0.07 for @xmath35 and @xmath26 bands respectively ) were used . considering that the objects and the standard stars were observed as close to the meridian as possible , and in none of the cases the airmass exceeded 1.3 , the error introduced due to possible variation in the extinction coefficients is less than 0.02 magnitude . coefficients @xmath45 and @xmath46 were obtained by using the @xmath18 standard magnitudes of chevalier & ilovaisky ( 1991 ) . resulting values of @xmath45 and @xmath46 are given in table 3 . the relatively large color coefficient on the @xmath26-band calibration is due to the non - standard nature of the filter used in the observations . the stability of the @xmath45 on a given night was checked using at least two other standard fields from the landolt selected areas ( landolt 1992 ) . the standard fields observed during our runs are sa110@xmath12232 , pg2336 + 004 , rubin149 and pg1323@xmath12086 . overall , the zero points within a single night agreed to within 0.02 magnitudes . night - to - night variation of the zero point was also within 0.02 magnitudes . the detector and filter system combination that we used for the near infrared observations is identical to that used in the observations of standards by hunt et al . ( 1998 ) , and hence the color coefficients are expected to be negligibly small . we verified this by observing fields as17 and as36 , which contain stars spanning a wide range of colors . we observed at least 2 standard fields each night , each field containing more than one star and some fields such as as17 containing 5 stars . the @xmath1-band zeropoints was obtained for each night as the mean of zero points from different stars and the average value is @xmath47 = @xmath48 . for reddening corrections due to galactic extinction we adopt the values given by schlegal ( 1998 ) for each filter in the optical range . for near infrared observations we use the extinction law given by rieke ( 1985 ) . when galaxies at different redshift are seen through a fixed passband , the collected light comes from different wavelength ranges in the rest frame of the respective galaxies . @xmath49-correction is used to transform the rest frame magnitudes of each object to the passband of the filter . we have used an approximate form of the @xmath49-correction , applicable for small @xmath50 , from persson ( 1979 ) , frei ( 1994 ) and fukugita ( 1995 ) . for the four filters we have , @xmath51 , @xmath52 , @xmath53 and @xmath54 respectively , where @xmath50 is the redshift of the galaxy . we have listed the combination of galactic extinction and @xmath49-correction for individual galaxies in table 4 . ccc february 2002 & & + + @xmath55 & [email protected] & @[email protected] + @xmath57 & [email protected] & [email protected] + @xmath26 & [email protected] & @[email protected] + + + october 2002 & & + + @xmath55 & [email protected] & @[email protected] + @xmath57 & [email protected] & [email protected] + @xmath26 & [email protected] & @[email protected] an accurate estimation of the sky background is a crucial step in surface photometric analysis , as even small uncertainties in determining the sky can lead to significant errors in the estimation of the galaxy surface brightness and color profiles , especially where the surface brightness becomes comparable to or less than the sky brightness . in those cases where the galaxy image is small enough that the frame contains portions of the sky unaffected by the galaxy , it is possible to estimate the sky using the `` boxes method '' ( see peletier 1990a ) , where the background is estimated from a series of boxes chosen avoiding the sample galaxy , foreground stars and any other contaminating objects which may be present . we have used 20 boxes of size @xmath58 pixels near the corners of the ccd frame , and have adopted the mean of the median count in these boxes as the sky background . the rms dispersion over the mean of 20 boxes is found to be 0.6% of the mean in the @xmath35 , and @xmath26 bands and 0.2% of the mean in the @xmath1 band . when a galaxy is so large that the whole frame is affected by it , use of the boxes method leads to overestimation of the background . in such cases , the background can be determined by fitting a power - law of the form i(r ) = sky + i(0 ) r@xmath59 to the outer parts of the surface brightness profile . following jorgensen ( 1992 ) and goudfrooij ( 1994 ) , we fit the function to the region of the profile with @xmath60 typically , separately for @xmath61 and 3 , and use the mean of the respective best - fit constant values as the background . the error in the sky background estimation is taken to be half the difference of the sky values obtained for the two fits . we have used both the techniques for all galaxies in our sample , and find that in most cases the two methods agree within @xmath62 magnitude ( i.e. 1@xmath63 difference ) , because of the small size of the galaxies relative to the ccd frame . in the case of the galaxies ugc4347 , 7473 , 7880 , 8675 and ugc11356 , which cover most of the frame , the power - law fit provides a background which is @xmath64 magnitude ( i.e. 10@xmath63 difference ) fainter than the boxes method , and we adopt the former as the background estimation . in the @xmath1 band images , the background is already subtracted from the many individual frames for each galaxy during the pre - processing stage ( see @xmath65 ) . however , to check for any residual background which may be present in processed frames used in the analysis , we have again applied the two techniques for the background estimation . we find that the two estimates agree very well ( @xmath66 magnitude , except in two cases where the agreement is within @xmath67 magnitude ) . ccccccccc ugc 00080 & [email protected] & [email protected] & [email protected] & [email protected] & 0.25 & 0.06 & 0.10 & 0.20 + ugc 00491 & [email protected] & [email protected] & [email protected] & [email protected] & 0.32 & 0.07 & 0.13 & 0.27 + ugc 00859 & [email protected] & [email protected] & [email protected] & [email protected] & 0.41 & 0.09 & 0.16 & 0.30 + ugc 00926 & [email protected] & [email protected] & [email protected] & [email protected] & 0.36 & 0.08 & 0.14 & 0.29 + ugc 01250 & [email protected] & [email protected] & [email protected] & [email protected] & 0.36 & 0.08 & 0.14 & 0.29 + ugc 01823 & [email protected] & [email protected] & [email protected] & [email protected] & 0.39 & 0.09 & 0.15 & 0.31 + ugc 02039 & [email protected] & [email protected] & [email protected] & [email protected] & 0.57 & 0.13 & 0.22 & 0.44 + ugc 03087 & [email protected] & [email protected] & [email protected] & [email protected] & 0.42 & 0.10 & 0.17 & 0.35 + ugc 03452 & [email protected] & [email protected] & [email protected] & [email protected] & 0.74 & 0.17 & 0.29 & 0.56 + ugc 03536 & [email protected] & [email protected] & [email protected] & [email protected] & 0.54 & 0.12 & 0.21 & 0.42 + ugc 03567 & [email protected] & [email protected] & [email protected] & [email protected] & 0.44 & 0.10 & 0.18 & 0.36 + ugc 03642 & [email protected] & [email protected] & [email protected] & [email protected] & 0.26 & 0.06 & 0.11 & 0.23 + ugc 03683 & [email protected] & [email protected] & [email protected] & [email protected] & 0.48 & 0.11 & 0.19 & 0.39 + ugc 03699 & [email protected] & [email protected] & [email protected] & [email protected] & 0.45 & 0.10 & 0.18 & 0.37 + ugc 03792 & [email protected] & [email protected] & [email protected] & [email protected] & 0.38 & 0.09 & 0.15 & 0.32 + ugc 03824 & [email protected] & [email protected] & [email protected] & [email protected] & 0.31 & 0.07 & 0.13 & 0.27 + ugc 04347 & [email protected] & [email protected] & [email protected] & [email protected] & 0.25 & 0.06 & 0.10 & 0.22 + ugc 04767 & [email protected] & [email protected] & [email protected] & [email protected] & 0.19 & 0.04 & 0.08 & 0.21 + ugc 06389 & [email protected] & [email protected] & [email protected] & [email protected] & 0.11 & 0.02 & 0.04 & 0.10 + ugc 06899 & [email protected] & [email protected] & [email protected] & [email protected] & 0.17 & 0.04 & 0.08 & 0.19 + ugc 07473 & [email protected] & [email protected] & [email protected] & [email protected] & 0.13 & 0.03 & 0.05 & 0.10 + ugc 07880 & [email protected] & [email protected] & [email protected] & [email protected] & 0.12 & 0.03 & 0.05 & 0.09 + ugc 08675 & [email protected] & [email protected] & [email protected] & [email protected] & 0.05 & 0.01 & 0.02 & 0.05 + ugc 11178 & [email protected] & [email protected] & [email protected] & [email protected] & 0.70 & 0.16 & 0.27 & 0.51 + ugc 11356 & [email protected] & [email protected] & [email protected] & [email protected] & 0.41 & 0.09 & 0.16 & 0.31 + ugc 11781 & [email protected] & [email protected] & [email protected] & [email protected] & 1.12 & 0.26 & 0.43 & 0.82 + ugc 11972 & [email protected] & [email protected] & [email protected] & [email protected] & 0.76 & 0.17 & 0.29 & 0.56 + ugc 12443 & [email protected] & [email protected] & [email protected] & [email protected] & 0.52 & 0.12 & 0.21 & 0.42 + ugc 12655 & [email protected] & [email protected] & [email protected] & [email protected] & 0.32 & 0.07 & 0.13 & 0.27 we have fitted ellipses to the isophotes in our @xmath68 and @xmath1 band images , using the task ellipse in the stsdas package available in iraf . the fitting algorithm used is described in detail by jedrzejewski ( 1987 ) , and uses the intensity distribution along a trial ellipse , which can be expressed as a fourier series @xmath69 where @xmath70 is the ellipse eccentric anomaly , @xmath71 is the mean intensity along the ellipse and @xmath72 , @xmath73 are harmonic amplitudes . the fitting was started a few arcseconds from the center of the galaxy image to minimize the effect of seeing , and stopped at the isophotes where the mean count becomes comparable to three times the error in the sky background . stars , bad pixels etc . were identified in the first round of the fitting procedure and masked in the next run . all the parameters including the center of the ellipse were allowed to vary during the fitting . variation in ellipse center was found to be small ( @xmath74 ) , which provides a check on the accuracy of the sky subtraction . we repeated the fitting process with different modes of sampling and different starting major - axis lengths to check the stability of the extracted parameters . the fitting procedure provides the mean intensity ( surface brightness ) , the ellipticity and the position angle of the major - axis of the best fit ellipse , as a function of the semi - major axis length . the @xmath55 band surface brightness profile , and the @xmath75 and @xmath5 color profiles are shown for all the galaxies in fig . the profiles of the ellipticity , position angle and the fourier coefficients @xmath76 respectively in the @xmath77 and @xmath1 bands for one galaxy ( ugc7880 ) are shown in fig . tables and plots of surface brightness and color profiles as well as the profiles of the ellipticity , position angle and the fourier coefficients @xmath78 respectively in the @xmath77 and @xmath1 bands for all the galaxies in the sample are available at _ http://iucaa.ernet.in/@xmath79sudhan / s0.html_. some description of the profiles for individual galaxies is provided in the appendix . color measurements of galaxies involving frames with different seeing full width at half maximum ( _ fwhm _ ) can lead to errors at small radii , and in particular within the seeing disk which has to be approached for measuring nuclear colors . these errors have been discussed by franx ( 1989 ) , peletier ( 1990a ) and goudfrooij ( 1994 ) , who have derived a cutoff radius for color profiles and discarded colors measured inside this limit , which is usually @xmath80 . idiart ( 2002 ) used a method of equalization of the _ fwhms _ to reduce error due to different seeing _ fwhm _ but this method is feasible only if the frames involved in color measurement are observed in quick succession . for our observations the seeing _ fwhm _ was nearly the same for the optical @xmath35 and @xmath26 bands , but significantly better in @xmath1 band as indicated by table 2 . we have therefore degraded the @xmath1 images to the mean _ fwhm _ of the optical band images for each galaxy . the @xmath5 color profiles shown in fig . 2 were obtained using the degraded @xmath1 band images . cccccc ugc 00080 & 3.40 & 28.05 & @[email protected] & @[email protected] & @[email protected] + ugc 00491 & 3.40 & 51.00 & @[email protected] & @[email protected] & @[email protected] + ugc 00859 & .... & ..... & ..... & ..... & ..... + ugc 00926 & 3.40 & 34.00 & @[email protected] & @[email protected] & @[email protected] + ugc 01250 & .... & .... & ..... & ..... & @[email protected] + ugc 01823 & 3.40 & 34.00 & @[email protected] & @[email protected] & @[email protected] + ugc 02039 & 3.40 & 09.35 & @[email protected] & @[email protected] & @[email protected] + ugc 03087 & .... & .... & ..... & ..... & ..... + ugc 03452 & 3.40 & 09.35 & @[email protected] & @[email protected] & @[email protected] + ugc 03536 & 3.40 & 09.35 & @[email protected] & @[email protected] & @[email protected] + ugc 03567 & 3.40 & 09.35 & @[email protected] & @[email protected] & @[email protected] + ugc 03642 & 3.40 & 13.60 & @[email protected] & @[email protected] & @[email protected] + ugc 03683 & 3.40 & 13.60 & @[email protected] & @[email protected] & @[email protected] + ugc 03699 & 3.40 & 17.00 & @[email protected] & @[email protected] & @[email protected] + ugc 03792 & 3.40 & 17.00 & @[email protected] & @[email protected] & @[email protected] + ugc 03824 & 3.40 & 21.25 & @[email protected] & @[email protected] & @[email protected] + ugc 04347 & 3.40 & 22.95 & @[email protected] & @[email protected] & @[email protected] + ugc 04767 & 3.40 & 18.70 & @[email protected] & @[email protected] & [email protected] + ugc 06389 & 3.40 & 22.95 & @[email protected] & @[email protected] & @[email protected] + ugc 06899 & 3.40 & 17.85 & @[email protected] & @[email protected] & @[email protected] + ugc 07473 & 3.40 & 17.00 & @[email protected] & @[email protected] & @[email protected] + ugc 07880 & 3.40 & 16.15 & @[email protected] & @[email protected] & @[email protected] + ugc 08675 & 5.95 & 53.55 & @[email protected] & @[email protected] & @[email protected] + ugc 11178 & 3.40 & 17.85 & @[email protected] & @[email protected] & @[email protected] + ugc 11356 & 3.40 & 21.25 & @[email protected] & @[email protected] & @[email protected] + ugc 11781 & 3.40 & 08.50 & @[email protected] & @[email protected] & @[email protected] + ugc 11972 & 3.40 & 19.55 & @[email protected] & @[email protected] & @[email protected] + ugc 12443 & 3.40 & 14.45 & @[email protected] & @[email protected] & @[email protected] + ugc 12655 & 3.40 & 14.45 & @[email protected] & @[email protected] & @[email protected] lccc our sample & @xmath6 ( @xmath81 ) & @xmath7 ( @xmath81 ) & @xmath8 ( @xmath81 ) + + ellipticals & @xmath82 ( @xmath83 ) & @xmath84 ( @xmath85 ) & @xmath86 ( @xmath83 ) + ( peletier 1990a & b ) & & & + + bulges of disk galaxies & ... & @xmath87 ( @xmath85 ) & ... + ( peletier & balcells 1997 ) & & & in this work we obtain the total magnitude of each galaxy in all the bands by integrating the light within the isophote corresponding to @xmath88 = 25 magarcsec@xmath32 in each band . to obtain the @xmath75 and @xmath5 colors , however , we integrate out only to the radius at which the error in @xmath88 is 0.1 magarcsec@xmath32 ( the radius where the dotted horizontal line intersects the @xmath89 profile in fig . 2 ) . the colors obtained in this manner are consistent with those obtained using total magnitude as defined above , within the estimated errors . however , the errors on integrated colors obtained in this manner are substantially lower ( typically 0.01 vs 0.04 in @xmath2 , 0.01 vs 0.04 in @xmath3 , and 0.05 vs 0.12 in @xmath5 colors ) . we also obtain total @xmath1 magnitudes measured to @xmath90 = 20 magarcsec@xmath32 in order to make comparison with 2mass photometry . the measured total @xmath55 magnitudes and @xmath75 and @xmath5 colors along with the estimated errors for the 29 galaxies in the sample for which we have the necessary data are shown in table 4 . in fig . 4 ( left panel ) we have compared our photometry with that in the literature . we have plotted ( from bottom to top ) the difference between ( i - ii ) our @xmath55 magnitude and @xmath2 color and those of rc3 , ( iii ) our @xmath55 magnitude and that of ugc and ( iv ) our @xmath1 magnitude measured inside the isophote @xmath91 = 20 magarcsec@xmath32 and that of 2mass , all against our @xmath55 magnitude . agreement between our @xmath55 magnitude with that of rc3 is within the expected errors for all except two galaxies . our total @xmath2 colors are systematically around 0.04 magnitude bluer than those of rc3 . on the other hand , if we compare the @xmath2 color obtained by integrating only upto @xmath89 = 21 magarcsec@xmath32 ( _ bulge color _ ) , the agreement between the colors is excellent . the agreement between our @xmath1 magnitude is also reasonably good . however it is important to note that 2mass magnitudes correspond to @xmath92 = 20 magarcsec@xmath32 and considerable flux could be outside this radius our total @xmath1 magnitudes ( measured inside the radius corresponding to @xmath93 = 25 magarcsec@xmath32 ) are systematically brighter by @xmath790.4 magnitudes as compared to those of 2mass . hence @xmath1 magnitudes in the 2mass catalog underestimate the total flux . to check the internal consistency of our photometry we have plotted color - color diagrams as shown in fig . 4 ( right panel ) . in general , there is a good correlation between these independent color measurements and except for two galaxies ugc 3087 and ugc 11178 , the estimated colors follow the sequence expected for early - type galaxies . among these , ugc 3087 is clearly outside the 3@xmath63 error estimation in all the three colors . this galaxy is known to harbor an active galactic nucleus , which clearly affects the color profiles in fig . hence its colors are expected to depart significantly from the general correlation . ugc 11178 , though is within the 2@xmath63 error bar , is curious because it stands out in the plots , which are completely independent measurements . further photometry would be required to understand the nature of this object . we exclude these two galaxies from further analysis . we have used optical and @xmath1 band data to study the spatial distribution of dust . we have obtained extinction map in magnitude @xmath94 = @xmath95 , where @xmath96 is the observed image in same band , and @xmath97 is the smooth version of this image , obtained using the best fit ellipse parameters . extinction map particularly helping in identifying non - axisymmetric structure . color maps too are of help in identifying features with wavelength dependent intensity . we have noticed from the color and extinction maps that the lenticular galaxies ugc3178 , ugc3536 , ugc3792 , ugc4347 , ugc7473 and ugc7880 show clear evidence of the presence of dust in the form of lanes or patches . in all these cases there are differences in the ellipticity and position angle profiles , as well as in the higher order fourier coefficients profiles , in the different bands . this indicates that such differences could be used as indicators of the possible presence of dust , even when the dust is not otherwise discernible ( see peletier 1990a , goudfrooij 1994 ) . the lenticular galaxy ugc3792 has a very prominent dust lane along the major axis ( see the appendix for details ) . we will present elsewhere a detailed study of the dust in our sample of galaxies . [ fig : xxx ] ( continued ) fig . [ fig : xxx ] ( continued ) fig . [ fig : xxx ] ( continued ) fig . [ fig : xxx ] ( continued ) it is the usual practice to parameterize the color gradient over the galaxy as the slope @xmath98 , which is evaluated by making a straight line fit between an inner and an outer radius to the color profiles . the inner cutoff limit ( r1 ) is taken to be @xmath791.5 times the seeing fwhm ( peletier 1990a ) , so that seeing effects are lessened . the outer limit ( r2 ) is taken to be the point along the major axis at which the error on the mean surface brightness of the fitted ellipse reaches @xmath99 . it is seen from fig . 2 that plots of color against the logarithm of the semi - major axis for most of our sample galaxies , have a reasonably linear form between the inner and outer cutoff . there are some cases ( ugc859 , 1250 and 3087 ) for which the color profiles are non - linear between r1 and r2 because of the presence of a strong ring ( ugc859 , 1250 ) or an agn ( ugc3087 ) , which makes reliable estimation of the color gradient is difficult . the @xmath75 and @xmath5 color gradients for our galaxies are given in table 5 . the colors of the galaxies in our sample become bluer outwards , in keeping with the trend observed in ellipticals . the only exception to this is the positive gradient in the @xmath5 color of the galaxy ugc4767 . after excluding the galaxy ugc3792 ( which has a prominent dust lane ) , the mean logarithmic gradients in the @xmath75 and @xmath5 colors are @xmath6 , @xmath7 , @xmath8 magnitude per dex in radius respectively . a comparison of our mean color gradients with corresponding values from the literature for ellipticals and bulges of early - type spirals is shown in table 6 . the color gradients for lenticulars are more negative than the gradients for ellipticals , while the @xmath3 gradient for lenticulars is less negative than the corresponding gradient for bulges of early - type spirals . our steeper color gradients could imply that the metallicity gradients are stronger in lenticulars and/or that there is more recent star formation in the outer parts of lenticulars as compared to those of ellipticals . bothun & gregg ( 1990 ) reported significant differences in the colors of bulges and disks in lenticulars which they interpret as being due to age difference , and not metallicity difference , between the bulge and disk components . the large rms error in the mean color gradient of our sample could be due to the different colors of the bulge and disk components and varying proportion of these in different lenticulars . correlation between different color gradients have been noted by various authors ( peletier 1990a & b and idiart 2002 ) and have been used to derive information about the physical processes in galaxies . such correlations are displayed in fig . 5 for our sample . a clear correlation is seen between @xmath2 and @xmath3 gradients , with correlation coefficient 0.90 at a significance level better than 99.99 percent . correlation between @xmath2 and @xmath3 gradient has been noticed for a sample of ellipticals by peletier ( 1990 a & b ) . from their data we obtain a correlation coefficient of 0.97 at a significance level better than 99.99 percent . the plot of @xmath5 against @xmath2 gradients shows large scatter , and we do not see any correlation of the type reported by peletier ( 1990b ) , but it should be noted that they used only 13 galaxies to derive the correlation . it is necessary to investigate how the separate bulge and disk color gradients affect the total color gradients and hence the correlation between them . we will address in the future these issues using a bulge - disk decomposition technique which will allow us to study the color distribution in these components separately . in this paper we have presented detailed multicolor surface photometry performed with a ccd in the @xmath100 bands and a nicmos detector in the @xmath1 band , for a sample of 34 lenticular galaxies from the ugc catalogue . the galaxies were chosen in an unbiased fashion from a subset of ugc lenticulars as explained in section 2 . we have obtained total integrated magnitudes and colors for all the galaxies using elliptical annuli from surface photometry , and find that these are in good agreement with values from the rc3 catalogue . using isophotal analysis we have obtained radial profiles of the surface brightness , ellipticity , position angle , and higher order fourier coefficients in all the bands . the profiles in the different bands are consistent with each other , and any differences can be attributed to the presence of dust and other features which produce wavelength dependent effects . we have used the surface brightness profiles to obtain color profiles , and logarithmic color gradients , and find that the gradients are negative , indicating that the colors of lenticulars become redder towards the center , as is the case with elliptical galaxies . we have shown that there is good correlation between @xmath2 and @xmath3 color gradients for lenticulars . numerical profiles of all parameters that we have obtained from the isophotal analysis along with the color images are available at _ http://www.iucaa.ernet.in/@xmath79sudhan / s0.html_. our intention in obtaining the multiband data on lenticulars has been to study in detail these galaxies as a class , and to compare their properties with those of ellipticals and early type spirals . an important aspect of this study will be the decomposition of the lenticulars into bulge and disk components and the comparison of these separately with the bulges and disks in other types of galaxies where these components occur with varying degrees of prominence . we will use the results of the decomposition in a multiband study of the fundamental and photometric planes for lenticulars . we will also make a detailed study of the distribution of dust in lenticulars , particularly in those galaxies from our sample where there are prominent dust features and where we have multiband data at optical as well as near - infrared wavelengths . the staff at oagh and spm are greatfully acknowledged for their help during the observations . sb and skp thank iucaa for hospitality and the use of facilities without which this work could not have been done and for providing funds for observations . akk and sb thank inaoe for hospitality provided during there visits . we also thank the anonymous referee for several useful comments , which helped to improve the original manuscript . this research has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . in this appendix , we discuss structural properties of each galaxy , whenever isophotes in one or more bands depart from a smooth ellipse . the discussions are focused on any disagreement in the morphological classification in ugc and rc3 catalogs , presence of dust lane , change of ellipse - fit parameters ( b4 , pa or ellipticity ) , existence of any ring , and any evidence of nuclear activity . published information on any of these issues , if available , is noted . in total 12 galaxies ( ugc926 , 1823 , 3452 , 3567 , 3683 , 3699 , 3824 , 6013 , 6389 , 6899 , 11972 and 12443 ) are smooth lenticular galaxies without any identifiable feature in the direct images , the surface brightness profiles , color or extinction maps . comments for the remaining galaxies follow . * ugc80 :* ugc80 is a barred lenticular galaxy , having a very faint interacting companion . the faint bar is almost perpendicular to the galaxy major axis , and is clearly evident in the isophotal profile . the @xmath101 profile is indicative of a disky shape over the entire major - axis range explored . the ellipticity and position angle profiles show abrupt change at @xmath102 , where there is also a kink in the brightness profile . the @xmath3 color map of the galaxy shows a red feature near the center . * ugc491 :* ugc491 is the brightest galaxy in a small group with ngc0258 and ngc0260 . this galaxy is classified as s0 in the ugc catalogue and sa(r ) in the rc3 catalogue . in the optical images it indeed shows a ring , which is also evident in the @xmath3 and @xmath103 color maps . ellipticity and position angle profiles show a change at @xmath104 , which may be due to the presence of the ring . * ugc859 :* ugc859 shows an internal ring , but it is classified as s0 in the ugc catalogue . the @xmath3 and @xmath103 color maps of the galaxy show the ring as an inhomogeneous structure . a low surface brightness disk with some luminous spiral structure exists outside the ring . ugc 859 has been detected by iras and has flux densities of @xmath105 at @xmath106 and @xmath107 at @xmath108 ( knapp 1989 ) . the @xmath101 coefficient is disky towards the outer region . * ugc1250 :* while this galaxy is classified as s0 , the @xmath3 color map of the galaxy show a very clear , but distorted ring , just around the bulge . @xmath109 $ ] observations by pogge & eskridge ( 1993 ) show copious hii regions across the disk . the @xmath101 profile is disky at all major - axis lengths , with the coefficient being larger in the inner region of the galaxy . * ugc2039 :* ugc2039 is paired with ugc 2049 . there is a rise in the ellipticity from 0.1 to 0.5 towards the outer region . the @xmath101 coefficient is disky in the outer region , which suggests the presence of a significant disk . the galaxy has been detected by iras at 60 and @xmath108 . * ugc3087 :* previous studies of ugc3087 have mainly concentrated on its nuclear activity . it is a strong radio source ( 3c120 ) and the optical spectrum suggests a seyfert 1 nucleus ( tadhunter 1993 ) . it has a faint optical jet in the same apparent direction as the radio jet . the optical jet is clearly visible in our images and distorts isophotal profiles in the inner region . * ugc3178 :* a dust feature is clearly visible in the @xmath55 band extinction map of this galaxy , and is also detectable from the isophotal analysis of @xmath55 and @xmath1 band images . the @xmath101 coefficient , position angle and ellipticity are different in @xmath55 and @xmath1 , which may be due to dust absorption in @xmath55 . our @xmath57 and @xmath26 band images suffer from poor s / n . * ugc3536 :* this galaxy has a significant dusty disk and the @xmath101 coefficient is disky . the disk is also evident in the @xmath2 and @xmath3 color maps . there is a constant rise in ellipticity for semi - major axis length @xmath110 , beyond which there is a decrease , which could be because of the disk . * ugc3792 :* this galaxy is classified as s0 in the ugc but as sa0/a in the rc3 catalogue , and has no previous reported photometric study . the galaxy has a prominent dust lane along the major axis which greatly affects the various profiles . this dust lane is most prominent in the @xmath55 band . the @xmath101 coefficient is positive in the outer region . there is a large color gradient in @xmath75 and @xmath5 , presumably because of the dust . a detailed study of this galaxy in the @xmath111 bands will be presented in a forthcoming paper . * ugc4347 :* our @xmath2 and @xmath103 color and extinction maps of this galaxy reveal a large dust patch near the center . the @xmath101 profile is disky throughout the observed region . forbes & thomson ( 1992 ) have detected possible shells in this galaxy . the extinction images in all bands shows faint structures , which may be responsible for the non - zero values the of @xmath101 coefficient . * ugc4767 :* the position angle and ellipticity profile for this galaxy are different in the @xmath112 and @xmath1 bands for semi - major axis lengths @xmath113 , which could be because of a dust patch near the center of the galaxy . the various color maps also indicate the presence of the dust patch . * ugc4901 :* our @xmath55 band image of this galaxy has poor s / n . the color maps in @xmath114 and @xmath5 show no features in the galaxy . the @xmath101 coefficient is positive towards the outer region , suggesting the presence of a faint disk . * ugc7473 :* this well studied edge - on lenticular galaxy forms a pair with ngc 4340 . the surface brightness profile of this galaxy clearly indicates the presence of bulge and disk components . the mg@xmath115 line strength profiles along the major and minor axis of the galaxy differ dramatically and convincingly indicate the presence bulge and disk components ( fisher , franx & illingworth 1996 ) . michard & marchal ( 1993 ) described ugc7473 as having a disk fully embedded in a spheroidal halo . a disk of rapidly rotating gas is present within the inner @xmath116 ; this is decoupled from the stellar component ( fisher 1997 ) . a concentration of dust in the disk has been proposed by michard & poulain ( 2000 ) . our @xmath3 and @xmath103 color and extinction maps also reveal a clear inclined disk . the ellipticity rises from 0.2 to 0.61 over the observed region and the @xmath101 coefficient becomes significantly positive beyond @xmath117 , which reflects the presence of a strong disk . * ugc7880 :* this galaxy forms a pair with ngc 4635 , with a separation of at @xmath118 . the ellipticity changes from 0.1 to 0.5 upto @xmath119 , and decreases beyond that . the @xmath101 coefficient is positive upto the point at which the ellipticity begins to decrease . this may be due to the presence of an inner disk , the presence of which is also indicated by the @xmath3 and @xmath103 color and extinction maps . * ugc7933 :* this galaxy is classified as s0 in the ugc , but as e1 - 2 in the rc3 catalogue . it forms a non - interacting pair with ngc4670 at @xmath120 separation . the @xmath55 band image has poor s / n . the @xmath101 coefficient is positive towards the outer region . the color and extinction maps do not indicate any structure . * ugc8675 :* this is a little studied s0 galaxy hosting a seyfert 1.5 nucleus , and has been classified as a sa0 galaxy in the rc3 catalogue . the isophotes of the galaxy are nearly circular and at @xmath121 the ellipticity becomes 0.15 which is consistent with the values obtained by de robertis , hayhoe & yee ( 1998 ) and pierre ( 2000 ) . a dust absorption pattern is seen near the nucleus in the @xmath3 and @xmath103 color and extinction maps . pierre ( 2000 ) have found from hst observations a u - shaped dust lane circling around the nucleus . * ugc11178 :* this galaxy has no previous reported photometric study . the extinction map show a ring like structure in all the bands . the ellipticity abruptly changes between 5 and @xmath122 , which could be because of the ring like structure . the @xmath101 coefficient is significantly positive upto @xmath122 . the color maps are featureless . * ugc11781 :* this galaxy is classified as s0 in the ugc but as sab in the rc3 catalogue . the presence of a bar is clear in the direct as well as extinction images . there is a dip in ellipticity profile at @xmath119 , where there is a change in the position angle profile as well . the @xmath101 coefficient is positive in this region .
the conventional theory of homogeneous and heterogeneous nucleation in a supersaturated vapor is tested by monte carlo simulations of the lattice gas ( ising ) model with nearest - neighbor attractive interactions on the simple cubic lattice . the theory considers the nucleation process as a slow ( quasi - static ) cluster ( droplet ) growth over a free energy barrier , constructed in terms of a balance of surface and bulk term of a `` critical droplet '' of radius , implying that the rates of droplet growth and shrinking essentially balance each other for droplet radius . for heterogeneous nucleation at surfaces , the barrier is reduced by a factor depending on the contact angle . using the definition of `` physical '' clusters based on the fortuin - kasteleyn mapping , the time - dependence of the cluster size distribution is studied for `` quenching experiments '' in the kinetic ising model , and the cluster size where the cluster growth rate changes sign is estimated . these studies of nucleation kinetics are compared to studies where the relation between cluster size and supersaturation is estimated from equilibrium simulations of phase coexistence between droplet and vapor in the canonical ensemble . the chemical potential is estimated from a lattice version of the widom particle insertion method . for large droplets it is shown that the `` physical clusters '' have a volume consistent with the estimates from the lever rule . `` geometrical clusters '' ( defined such that each site belonging to the cluster is occupied and has at least one occupied neighbor site ) yield valid results only for temperatures less than 60% of the critical temperature , where the cluster shape is non - spherical . we show how the chemical potential can be used to numerically estimate also for non - spherical cluster shapes .
1305.0386
since the theory of nucleation phenomena was introduced a long time ago @xcite , the question under which conditions the `` conventional theory '' of nucleation is accurate has been debated ( see e.g. @xcite ) and this debate continues until today . for the simplest case of homogeneous nucleation ( by statistical fluctuations in the bulk ) of a one - component liquid droplet from the vapor , the basic statement of the theory is that under typical conditions nucleation processes are rare events , where a free energy barrier @xmath0 very much larger than the thermal energy @xmath4 is overcome , and hence the nucleation rate is given by an arrhenius law , @xmath5 here @xmath6 is the number of nuclei , i.e. droplets that have much larger radii @xmath7 than the critical radius @xmath1 associated with the free energy barrier @xmath0 of the saddle point in configuration space , that are formed per unit volume and unit time ; @xmath8 is a kinetic prefactor . now @xmath0 is estimated from the standard assumption that the formation free energy of a droplet of radius @xmath7 can be written as a sum of a volume term @xmath9 , and a surface term @xmath10 , i.e. @xmath11 since the liquid droplet can freely exchange particles with the surrounding vapor , it is natural to describe its thermodynamic potential choosing the chemical potential @xmath12 and temperature @xmath13 as variables , and expand the difference in thermodynamic potentials of liquid and vapor at the coexistence curve , @xmath14 , @xmath15 and @xmath16 denoting the densities of the coexisting vapor ( @xmath17 ) and liquid ( @xmath18 ) phases . according to the capillarity approximation , the curvature dependence of the interfacial tension @xmath19 is neglected , @xmath19 is taken for a macroscopic and flat vapor - liquid interface . then the critical radius @xmath1 follows from @xmath20 and the associated free energy barrier is @xmath21 however , since typically @xmath22 is less than 100 @xmath4 , the critical droplet is a nanoscale object , and thus the treatment eqs . - is questionable . experiments ( e.g. @xcite ) were not able to yield clear - cut results on the validity of eqs . - , and how to improve this simple approach : critical droplets are rare phenomena , typically one observes only the combined effect of nucleation and growth ; also the results are often `` contaminated '' by heterogeneous nucleation events due to ions , dust , etc . @xcite , and since @xmath6 varies rapidly with the supersaturation , only a small window of parameters is suitable for investigation . therefore this problem has been very attractive , in principle , for the study via computer simulation . however , despite numerous attempts ( e.g. @xcite ) , this approach is also hampered by two principal difficulties : * computer simulations can often only study a small number of decades in time , @xcite , which in typical cases correspond to small barriers @xmath23 rather than the larger ones which are of more interest in the context of experiments . * on the atomistic scale , it is a difficult and not generally solved problem to decide which particles belong to a droplet and which particles belong to its environment ; the vapor - liquid interface is diffuse and fluctuating @xcite . for these reasons , many of the available simulation studies have addressed nucleation in the simplistic ising ( lattice gas ) model , @xcite , first of all since it can be very efficiently simulated , and secondly because one can define more precisely what is meant by a `` cluster '' . associating ising spins @xmath24 at a lattice site @xmath25 with a particle , @xmath26 with a hole , originally `` clusters '' were defined as groups of up - spins such that each up - spin in a cluster has at least one up - spin as nearest neighbor belonging to the same cluster @xcite . however , now it is well understood that these `` geometrical clusters '' in general do not have much physical significance @xcite : e.g. , it is known that there exists a line of percolation transitions , where a geometrical cluster of infinite size appears , in the phase diagram @xcite . this percolation transition is irrelevant for statistical thermodynamics of the model @xcite . based on the work of fortuin and kasteleyn @xcite on a correlated bond - percolation model , it is now understood that physically relevant clusters in the ising model should not simply be defined in terms of spins having the same orientation and are connected by nearest neighbor bonds , as is the case in the `` geometrical clusters '' , but in addition one has to require the bonds to be `` active '' : bonds are `` active '' with probability @xmath27 @xmath28 @xmath29 being the ising model exchange constant . due to eq . , the `` physical clusters '' defined in this way are typically smaller than the geometrical clusters , and their percolation point can be shown to coincide with the critical point @xcite . a geometrical cluster hence can contain several physical clusters . note that to apply eq . , random numbers are used , and hence physical clusters are not deterministically defined from the spin configuration , but rather have some stochastic character . this presents a slight difficulty in using physical clusters in the study of cluster dynamics . while eq . has been used in the context of simulations of critical phenomena in the ising model , applying very efficient swendsen - wang @xcite and wolff @xcite simulation algorithms , this result has almost always been ignored in the context of simulations of nucleation phenomena @xcite . while it is allright to ignore the difference between geometrical and physical clusters in the limit @xmath30 ( obviously @xmath31 then , all bonds becoming active ) , this is completely inappropriate at higher temperatures . the present work hence reconsiders this problem , studying both dynamical aspects of nucleation in the framework of the kinetic ising model @xcite ( without conservation laws ) , and the static properties of large droplets , applying the definition of `` physical clusters '' based on eq . throughout . for comparison , we shall also occasionally use the `` geometrical '' cluster definition , to demonstrate that misleading conclusions would actually result in practice , for the temperatures that are commonly studied . the study will be generalized to ising systems with free surfaces , where a boundary field @xmath32 acts @xcite . first of all , in this way also a systematic investigation of heterogeneous nucleation at planar walls becomes feasible ; secondly , due to the reduction of the barrier @xmath33 in comparison to @xmath34 ; nucleation for reasonably large values of @xmath1 becomes accessible to study . in sec . [ sec2 ] , we consider the equilibrium of the lattice gas model for @xmath35 in systems in a @xmath36 geometry with periodic boundary conditions , to show that physical clusters do occupy precisely the volume predicted by the lever rule analysis , @xcite as they should when the thermodynamic limit is approached . we present evidence that physical clusters are correctly identified by both the lever rule method and the approach based on the `` atomistic '' identification of clusters based on eq . at all temperatures , from zero temperature up to the critical temperature @xmath37 . in contrast , eq . , which implies a spherical droplet shape , is found to work only at temperatures distinctly above the interface roughening transition temperature @xmath38 @xcite , even for very large radii @xmath7 . we attribute these discrepancies to the fact that due to the anisotropy of the interface tension for our lattice model pronounced deviations of the average droplet shape from a sphere occur @xcite , presenting data on the shape of large droplets . in sec . [ sec3 ] , we describe our results on the dynamics of the droplet size distribution and on the attempt to find @xmath1 from the size where growth and shrinking processes of clusters are balanced . this study is also carried out for systems with a free surface , for which the contact angles for various values of the surface field have been estimated previously @xcite , since in this case much lower barriers ( for large clusters ) result , which is crucial for making this study feasible with manageable effort . however , the radii @xmath1 predicted from this analysis of kinetics show slight deviations from the radii @xmath1 predicted from @xmath33 . possible reasons for this discrepancy will be discussed . finally , sec . [ sec4 ] summarizes our conclusions . as is obvious from eq . and the reasoning behind it , this approach is adequate when one deals with the description of macroscopically large domains in equilibrium with a surrounding bulk phase . however , one needs to find an extension of the concept that can be applied also to nanoscopically small droplets , `` clusters '' in the lattice gas model that contain perhaps only of the order of 100 fluid particles . in this section , we want to confirm the idea that one must use the concept of `` physical clusters '' based on eq . for this purpose , rather than the `` geometrical clusters '' that are so widely used when the lattice gas model is used to test nucleation theory concepts . while the geometrical clusters are appropriate if one works at extremely low temperatures where the clusters basically have the shape of small cubes @xcite , this region clearly is inappropriate when one has the application for vapor - to - liquid nucleation in mind , where droplets are spherical , and their interfaces are rough and fluctuating rather than smooth planar facets . in fact , many studies of nucleation in the lattice gas model have been made in @xmath39 dimensions at temperatures near @xmath40 or thereabout ; given the fact that the interfacial roughening transition of the ising model on the simple cubic lattice is known to occur at about @xcite @xmath41 , i.e. ( note @xmath42 @xcite ) @xmath43 , it is clear that temperatures much closer to @xmath37 must be studied to render the assumption of a spherical droplet shape accurate . in fact , this assumption of a spherical droplet shape is accurate when the difference between the interfacial stiffness @xcite and the interfacial free energy becomes negligibly small . numerical studies of hasenbusch and pinn @xcite indicate that this is only the case for @xmath44 . as a consequence , it is clear that most of the existing studies of nucleation phenomena in the ising model , that were based on geometrically defined clusters , and had to be done at much lower temperatures , are inconclusive : the deviation of the average droplet shape from a sphere enhances the surface term in eq . ; but the fluctuation corrections discovered for small droplets by the `` lever rule method '' @xcite show that @xmath45 for small @xmath7 and hence the surface term in eq . is decreased . thus , it hardly can be a surprise that some of the studies concluded that the nucleation barriers predicted by classical nucleation theory and the capillarity approximation ( that ignores the @xmath7-dependence of @xmath46 are too high , and others concluded they are too low , or even reported good agreement . we take the latter finding as indication that the two opposing effects have accidentally more or less canceled each other . this problem is the motivation for the present section , which attempts to show that `` physical clusters '' based on eq . are appropriate to identify clusters in the lattice gas model , irrespective of temperature and cluster size , and are equivalent to the droplets of the `` lever rule method '' , for large enough droplets . [ fig1 ] recalls this approach : one samples for a system of volume @xmath47 the effective thermodynamic potential @xmath48 per lattice site as a function of density @xmath49 , @xmath50/v\,,\ ] ] where @xmath51 is the number of occupied lattice sites ( @xmath52 with @xmath53 the spin variable at lattice site @xmath25 ) . since phase coexistence between bulk liquid ( at density @xmath16 ) and vapor ( at density @xmath15 ) occurs at a chemical potential @xmath54 that corresponds to the `` field '' ( in magnetic notation ) @xmath55 , @xmath16 and @xmath15 are simply related to the spontaneous magnetization of the ising ferromagnet as @xmath56 , @xmath57 , and @xmath58 translates into @xmath12 via @xmath59 . the accurate sampling of @xmath48 for large @xmath60 is a nontrivial task , it requires the use of advanced methods such as `` multicanonical monte carlo '' @xcite or `` wang landau sampling '' @xcite or `` successive umbrella sampling '' @xcite , see @xcite for background on such techniques . from eq . , one defines a chemical potential function as a derivative , @xmath61 one recognizes that the isotherms @xmath62 vs. @xmath49 exhibit a loop : the homogeneous vapor remains stable also for some region where @xmath63 , until a peak occurs , which indicates the `` droplet evaporation / condensation transition '' @xcite : in the first regime where @xmath64 decreases with @xmath49 , a ( more or less spherical or cubical ) droplet coexists with surrounding vapor . here , we are not interested in the further transitions that one can recognize from this curve , where the droplet changes shape from spherical to cylindrical , or to a slab configuration , etc . @xcite . instead , we emphasize the key idea of the `` lever rule method '' : one can identify a range of choices for the chemical potential @xmath12 where three states of the finite system can exist in equilibrium with the same chemical potential , namely a homogeneous vapor at density @xmath65 , a homogeneous liquid at density @xmath66 , and a state where two - phase coexistence between the droplet and surrounding vapor occurs . since the vapor in this case exists at the same chemical potential as the pure vapor , it must be of the same physical nature as the state with density @xmath67 , and similarly , the liquid in the droplet can be identified with the liquid at @xmath68 . making now use of the fact that for large enough systems a system can be suitably decomposed into independent subsystems , we write for the free energy , with @xmath69 the volume taken by the droplet , @xmath70 , @xmath71 where the free energy densities @xmath72 and @xmath73 are explicitly known , and also @xmath48 is known : thus , when the droplet volume @xmath69 is known , the surface free energy @xmath74 of the droplet , which is defined via eq . , is determined . a similar decomposition can readily be written down for the particle number , @xmath75 where we have allowed for an excess number @xmath76 of particles , to be associated with the interface . if we consider the definition of an `` equimolar dividing surface '' @xcite , @xmath77 , and then reading off @xmath67 and @xmath68 from the construction in fig . [ fig1 ] we see that eq . readily yields @xmath78 and @xmath69 for the considered density @xmath49 , and via eq . we can immediately extract @xmath74 from the data . note that these arguments do not invoke the assumption that the dividing surface needs to be a sphere . if one makes the assumption , @xmath79 , and then one can write also @xmath80 . thus , it is assumed that all interactions of particles inside the droplet ( volume region @xmath69 ) with particles inside the vapor ( volume region @xmath78 ) are restricted to the interfacial region , and hence can be accounted for by their contribution to the surface free energy @xmath74 . \(a ) ( b ) however , the method defined via eqs . , , and illustrated in fig . [ fig1 ] becomes difficult to apply for very large droplets , because the sampling of @xmath48 then becomes unreliable or would require an unaffordable effort . the method is also difficult to apply for small droplets , because then one must use relatively small simulation boxes to ensure the stability of the inhomogeneous state where a droplet coexists with surrounding vapor @xcite . thus , it is very desirable to complement the approach by a more `` microscopic '' identification of droplets , and this is possible via eq . . in particular , it has been shown that apart from finite size effects ( see e.g. @xcite ) that for @xmath81 the spontaneous magnetization of the ising ferromagnet @xmath82 coincides with the percolation probability @xmath83 , which is defined @xcite as the fraction of sites belonging to the largest `` physical cluster '' in the system . when we hence analyze a configuration at a density @xmath49 , where ( cf . fig . [ fig1 ] ) a large cluster is present in the system , using eq . to define clusters the largest cluster will include @xmath18 sites , which we hence can associate with its ( total ) magnetization @xmath84 where @xmath85 is then the magnetization per site ( note that the ising magnet / lattice gas isomorphism implies that @xmath86 ) . as a consequence , we can obtain the droplet volume @xmath69 from a `` measurement '' of the average size @xmath87 of the largest cluster in the system via @xmath88 note that this volume in general differs from the volume of a geometrical cluster : if @xmath89 counts all occupied sites belonging to the geometrical cluster , and noting that the density in a large geometrical cluster is just the bulk density , namely @xmath90 , the volume taken by the geometrical cluster is given by @xmath91 in the limit @xmath92 , where also @xmath87 and @xmath69 get macroscopically large , @xmath85 tends to @xmath82 , while for finite @xmath60 it is clear that @xmath85 slightly exceeds @xmath82 ( and @xmath68 exceeds @xmath16 , see fig . [ fig1 ] ) . however , recording the relation @xmath93 [ or the equivalent relation @xmath94 of the ising ferromagnet ] very precisely is an easy task , see fig . [ fig2 ] , since both states at densities @xmath95 ( fig . [ fig1 ] ) are homogeneous , not affected by heterophase fluctuations , and since the temperatures studied are still well below @xmath37 , statistical fluctuations are small , and finite size effects are negligible . fig . [ fig2 ] presents representative results for both @xmath96 and @xmath97 versus @xmath98 . note we also have used the lattice version of the widom particle insertion method @xcite to record the inverse function @xmath99 from simulations in the canonical ensemble , where @xmath49 was chosen as the independent control variable . the perfect agreement between both approaches not only serves as a test of the accuracy and correctness of our numerical procedures , but also shows that for the chosen temperatures and linear dimensions finite size effects on states in `` pure '' phases are completely negligible , since finite size effects are known to differ in the two ensembles @xcite , but are not detected here at all . \(a ) ( b ) ( c ) ( d ) to record @xmath69 as defined in eq . from the simulations , we performed simulations in the canonical ensemble , choosing various values of @xmath49 , and equilibrate a large droplet coexisting with surrounding vapor . the initial state is then chosen putting a droplet with the size predicted by the lever rule ( and density @xmath100 ) into the simulation box , which then is carefully equilibrated . the standard method to simulate the ising model with conserved magnetization ( which corresponds to the lattice gas model in the canonical @xmath101 ensemble ) is the `` spin exchange algorithm '' @xcite . however , the standard nearest neighbor exchange method implies that any local excess of magnetization ( or density , respectively ) can only relax diffusively , and the resulting `` hydrodynamic slowing down '' @xcite hampers the fast approach towards thermal equilibrium . we thus used instead a single spin - flip algorithm in which the total magnetization is restricted to two neighboring values @xmath102 : so if the system has magnetization @xmath103 , flipping a down - spin ( which would mean a transition @xmath104 ) is automatically rejected , and if the system is in the state @xmath105 , flipping of up - spins is forbidden . for large systems ( @xmath106 ) any corrections to a strictly canonical simulation at a magnetization per spin @xmath107 are of order @xmath108 and hence negligible . choosing very large systems ( up to @xmath109 , rather than @xmath110 as used in fig . [ fig1 ] ) the ratio of @xmath69 as found from eq . and @xmath111 from the lever rule ( with the assumption @xmath112 , i.e. @xmath113 is plotted vs. @xmath114 in fig . [ fig3 ] for several temperatures and various densities @xmath49 . these data show that @xmath115 as @xmath81 , irrespective of temperature and density ( in the density region where a droplet not affected by the periodic boundary conditions is present , as explained in fig . [ fig1 ] ) . the fact that @xmath116 extrapolates to unity for @xmath106 not precisely , but only within some error , is due to the fact that statistical errors affect both the estimation of @xmath87 and of @xmath111 ( via errors in the estimation of @xmath98 and hence @xmath67 ) . as expected , using the proper definition of physical clusters , one can work at arbitrary temperatures , both at @xmath117 ( a ) , @xmath13 slightly above @xmath38 ( b ) or @xmath13 rather close to @xmath37 ( d ) . while in cases ( a ) the simple geometrical cluster definition would also work , since essentially all bonds are `` active '' ( @xmath118 is almost unity \{eq . } ) , and the non - spherical shape of the clusters does not matter in this context . but the geometrical cluster definition would clearly break down in case ( d ) due to the proximity of the percolation transition that occurs for geometrical clusters at a density not much larger than those included in fig . [ fig3](d ) @xcite . on the other hand , we note from the fact that there always occurs asymptotically in the relation @xmath119 a correction of order @xmath114 , that for physical clusters the assumption @xmath77 , that is often ( but not always @xcite ) made in the lever rule method , does not hold : i.e. , when we assume that @xmath76 is proportional to the surface area of the droplet , we can write @xmath120 where @xmath121 is a geometrical factor @xmath122 for a spherical droplet , @xmath123 for a cube ) , and @xmath124 is an excess density of the particles due to the interface of the droplet . using @xmath111 from eq . as a first - order estimate in eq . , one readily finds that the term @xmath125 in eq . yields a @xmath114 correction , @xmath126 and hence eq . would yield instead of eq . @xmath127 which is qualitatively in accord with fig . [ fig3 ] . in order to test eq . and obtain estimates for the temperature dependence of @xmath124 , fig . [ fig4 ] plots our numerical results for the ratios @xmath116 versus @xmath128 . we see a very good data collapse at straight lines going through unity at the ordinate within numerical error at all studied temperatures . \(a ) ( b ) \(a ) ( b ) ( c ) ( d ) since the data in fig . [ fig3 ] suggest that for `` physical droplets '' in the lattice gas model an appreciable `` interfacial adsorption '' ( as expressed in @xmath129 in eq . ) occurs , it is of interest to not only study the average volume of the droplets @xmath87 but also their radial density profile ( fig . [ fig5 ] ) . it is seen that for radii @xmath7 which are in the range from 10 to 20 lattice constants ( corresponding to droplet volumes in the range from about 4500 to 36000 , so these are already clusters of a mesoscopic size , with a huge nucleation barrier far beyond observation in a simulation of nucleation events or in experiments ) the profiles are very broad , and their width increases slightly with increasing droplet size . for comparison , the prediction for the cluster radius resulting from the widely used standard geometrical definition of clusters is also included : it happens that this geometrical radius is still fairly close to the correct radius at @xmath130 . of course , closer to @xmath37 the geometrical cluster definition yields completely unreasonable results , due to the onset of percolation phenomena , and for @xmath131 and @xmath132 , the geometrical radii are indeed unreasonably large . it is interesting to note ( fig . [ fig5]d ) that the width of the interfacial profile increases with increasing @xmath7 . this phenomenon is well - known for planar interfaces and attributed to capillary waves . since for a large droplet the surface is locally planar , most of the capillary wave spectrum is not affected by the interface curvature . thus we may , as a first approximation , take over the result for the broadening of a planar interface of linear dimension @xmath60 , replacing @xmath60 by the droplet radius @xmath7 @xcite @xmath133 where @xmath134 is the `` intrinsic width '' of the interfacial profile , @xmath135 the `` interfacial stiffness '' ( note that a factor @xmath136 is absorbed in its definition ) and @xmath137 a short wavelength cutoff , whose precise value is not known . near @xmath37 the interfacial stiffness coincides with the interfacial free energy , while @xmath138 at @xmath139 , and then the capillary wave broadening disappears . if one accepts the above formula , and uses the data of hasenbusch and pinn @xcite , one predicts for the slope @xmath140^{-1 } \approx 2.5 $ ] ( @xmath131 ) or @xmath141 ( @xmath142 ) , respectively . the actually observed slopes of the @xmath143 term are actually somewhat smaller , namely about 2.13 at @xmath131 and @xmath144 at @xmath142 , but of the same order of magnitude . we deliberately do not discuss the radial droplet density profiles for @xmath145 and @xmath146 , however , since at these temperatures the droplet shape shows distinct deviations from the spherical shape . this can be checked directly by recording contours of constant density in slices of width @xmath147 ( taking into account 3 lattice planes through the droplet s center of mass , parallel to the @xmath148 plane , the @xmath149 plane and the @xmath150 plane , respectively ) . averaging these density contours over several hundred statistically independent observations the plots shown in fig . [ fig6 ] are obtained . they are all taken at the same fixed density @xmath151 . note that for this density , the stable state would be a cylindrical droplet ( stabilized by the periodic boundary conditions ) but for such large systems ( @xmath152 ) the compact droplet shapes shown here ( chosen by an appropriate initial condition ) are always perfectly metastable . at @xmath145 the cubic symmetry of these cross sections through the droplet is evident ( although the presence of facets parallel to the planes @xmath153 or @xmath154 is not evident , due to finite - size effects at the points where the facets join the round sections , which replace the sharp edges of the cubes at nonzero temperature ) . at @xmath146 , where we exceed the roughening temperature slightly , there clearly occur no longer any facets , but the density contours in fig . [ fig6 ] are still distinctly non - circular : the diameter in diagonal direction clearly is about 7% larger than in the lattice directions . even at @xmath130 , we find an enhancement of the diameter in diagonal direction of about 3% . at @xmath131 to 4.3 , however , no longer any statistically significant deviation from spherical droplet shapes ( and hence circular shape of the density contours in the cross sections ) can be detected . note that for @xmath142 ( case ( f ) ) the geometrical cluster definition would not be applicable due to the proximity of the percolation transition of geometrical clusters . \(a ) ( b ) ( c ) ( d ) ( e ) ( f ) \(a ) ( b ) ( c ) ( d ) as a final part of our analysis of static properties of physical droplets in the ising model , we exploit the fact that the chemical potential @xmath12 can be `` measured '' by the lattice version of the widom particle insertion method @xcite also in the state when the system is inhomogeneous , e.g. for the case of interest when a droplet coexists with surrounding vapor . actually , already in @xcite it was shown that @xmath12 actually stays spatially constant in such a situation . however , while the estimation of @xmath12 from eq . requires a very accurate estimation of the free energy density @xmath155 \{eq . , fig . [ fig1 ] } , and in practice this works only for not so large linear dimension @xmath60 ( such as @xmath110 in fig . [ fig1 ] ) , the estimation of @xmath12 from the particle insertion method still works for volumes that are orders of magnitude larger . as a consequence , we can use eq . to test whether or not the actual droplet volumes @xmath156 are compatible with conventional nucleation theory for large droplets ( assuming that the droplets are spherical ) so @xmath157 holds for a critical droplet . [ fig7 ] presents a plot of the cluster volume @xmath69 versus inverse linear dimension for several densities , at the temperatures @xmath146 , @xmath158 , @xmath159 and @xmath132 , using the classical nucleation theory prediction @xmath160 with @xmath1 given by eq . for comparison . for this purposes @xmath161 is taken from the work of hasenbusch and pinn @xcite , and so there occur no unknown parameters whatsoever . the geometrical cluster volume @xmath162 \{eq . } is close to the estimate based on the coniglio - klein - swendsen - wang `` physical cluster''-definition , eq . , at @xmath146 and @xmath130 , while at @xmath131 ( and higher ) the deviations become appreciable : @xmath163 then is systematically too high ( in comparison with all other estimates , including @xmath111 \{eq . } , which is used as a convenient normalization ) . it is interesting to observe that the classical nucleation theory estimates based on eq . are systematically too low for @xmath164 and @xmath165 , while for @xmath166 eqs . , are found to be in very good agreement . this discrepancy at the relatively low temperatures comes from the fact that using eq . for the estimation of @xmath69 we imply that the cluster volume is spherical and hence we underestimate the surface area : the geometrical factor @xmath167 introduced in eq . increases from about 4.836 for the spherical shape near @xmath37 up to 6.0 for the cube , as the temperature is lowered . since the non - spherical droplet shapes at @xmath146 and 3.0 have regions of rather large curvature near the parts of the droplet where at low temperature the edges of the cube will appear , the estimation of @xmath12 then yields a too low radius @xmath1 . the observation that the anisotropy of surface tension in the lattice gas model becomes noticeable for @xmath168 is consistent with the findings of hasenbusch and pinn @xcite . if all the methods to define clusters were correct , in the thermodynamic limit all data should extrapolate to @xmath169 for @xmath106 , since in this limit the lever rule \{eq . } is trivially true with @xmath170 and @xmath171 . the method based on the definition of physical clusters , eq . , indeed is nicely compatible with this expectation at all temperatures ; although it is somewhat unsatisfactory ( and unexpected ) that at finite @xmath60 there occurs a surface correction due to the surface excess @xmath76 noted in eq . ( and discussed above ) . however , it is clear that the two other methods do not give results that are correct for @xmath106 in general : while the method based on the geometric cluster definition still gives essentially correct results at @xmath145 ( not shown here ) , where the distinction between `` geometrical '' and `` physical '' clusters is irrelevant , for temperatures @xmath172 , the volume of geometrical clusters is systematically too large . at @xmath142 , the error is as large as 60 to 80% even asymptotically , and for the cluster sizes that were actually studied the overestimation actually is by a factor two to five ( fig . [ fig7]d ) ! in view of the fact that the geometric cluster definition must break down due to the percolation transition @xcite , this failure is not unexpected , but we are not aware that it ever has been quantified previously . in the regime from @xmath146 to @xmath159 , the error of the geometric cluster volume raises from a few percent to 15 to 40% . the results obtained from the classical nucleation theory via the `` measurement '' of the supersaturation @xmath98 , on the other hand , yield essentially the correct result at high temperatures ( @xmath173 ) , for all linear dimensions studied , since the data are essentially independent of @xmath60 , and hence @xmath1 , in the considered range . but it is remarkable that for @xmath130 we find @xmath174 ( fig . [ fig7]b ) and for @xmath146 we find @xmath175 ( fig . [ fig7]a ) . this discrepancy becomes worse at lower temperatures ( e.g. @xmath176 at @xmath145 [ not shown ] ) , and obviously this discrepancy must be attributed to the orientation dependence of the interfacial free energy and the resulting non - spherical droplet shapes ( fig . [ fig6 ] ) . at first sight , the result that @xmath177 is independent of @xmath1 seems to be at variance with the finding of a curvature - dependent surface tension @xmath178 due to winter et al . in fact , evidence was provided that @xmath179 where @xmath180 is a length proportional to the correlation length in the bulk @xcite . note that due to the spin reversal symmetry of the ising model one can show @xcite that a tolman correction ( @xmath181 ) must be absent for @xmath182 . however , from eqs . and , it is straightforward to show that @xmath183 \;.% has been checked \end{split}\ ] ] hence for @xmath184 it is clear that the result for @xmath1 is still given by eq . , and for @xmath185 we are safely in this regime , since the correlation length then does not yet exceed the lattice spacing . so the deviations of the ratio @xmath177 that are seen in fig . [ fig7 ] can be attributed fully to the deviation of the droplet shape from a perfect sphere , caused by the anisotropy of the interfacial free energy . in fig . [ fig : vcntvsvlr ] , we now present the ratio of the intercepts @xmath186 for @xmath187 as a function of temperature , since we know that \{eq . } @xmath188 ^ 3 $ ] , while eq . yielded @xmath189 for a spherical droplet . at @xmath190 , however , the droplet is a perfect cube , and for intermediate temperatures , its shape ( for @xmath191 ) is given by the wulff construction @xcite ( and hence not explicitly known ) . however , for large droplet volume @xmath192 we can write in general @xmath193 when we have assumed that the droplets of different linear dimension @xmath7 ( for @xmath194 ) have the same shape at fixed temperature , so we can write @xmath195 for the droplet volume ( @xmath196 is then formally the volume for @xmath197 ) and the surface area is @xmath198 . for instance , for a sphere we have @xmath199 and @xmath200 , and for the cube @xmath201 and @xmath202 . minimizing @xmath203 with respect to @xmath192 yields @xmath204 for a general shape , which is in between sphere and cube . the barrier then can be written as @xmath205 using now the fact that by choosing a particular large droplet volume in our simulation , @xmath58 is automatically fixed for any volume @xmath206 due to the thermal equilibrium situation constructed in our simulation . so it makes sense to estimate the ratio of barriers as @xmath207 for @xmath190 , we know that @xmath208 is again the interface tension of the planar surface , also used in the classical nucleation theory for the spherical surface . hence when we write the ratio of @xmath209 , using eq . , the term @xmath210 cancels , @xmath211 this asymptotic value of @xmath212 should be reached for @xmath213 , while @xmath214 as @xmath215 . our numerical results ( fig . [ fig : vcntvsvlr ] ) are compatible with this expectation . we now consider time - dependence of the system where we start the system at time @xmath216 in an equilibrated state in the vapor at @xmath217 , but switch on at time @xmath216 a chemical potential @xmath218 ( or equivalently , a positive magnetic field @xmath58 in the notation of ising ferromagnet ) at which the liquid is the stable phase . as mentioned in the introduction , we consider heterogeneous in addition to homogeneous nucleation , choosing a @xmath219 geometry with two walls at @xmath220 and @xmath221 , choosing surface fields @xmath222 such that @xmath223 favors the vapor but @xmath32 ( acting at the wall at @xmath220 ) favors the liquid . the reason for this choice is , that by proper choice of @xmath32 one can adjust the contact angle @xmath224 at which sessile wall - attached macroscopic droplets can occur . the barrier against heterogeneous nucleation @xmath225 is predicted to be very much reduced , in comparison to the barrier @xmath34 against homogeneous nucleation , if the contact angle is small , since @xcite @xmath226 as shown with the `` lever rule '' method @xcite indeed rather large wall attached droplets can be simulated in equilibrium with supersaturated vapor which have barriers of order @xmath227 or so only , for suitable choices of @xmath32 and @xmath58 , and so a comparison with kinetic studies then seems reachable , and varying @xmath32 over some range provides an additional variable to test the theory . \(a ) ( b ) as a first step , preliminary runs were performed with the single spin flip metropolis algorithm @xcite monitoring the average lifetime of the metastable vapor . this was done using a large sample @xmath228 ) of equilibrated initial states at @xmath55 , where the considered field @xmath58 ( or chemical potential @xmath229 , respectively ) was then switched on and the time recorded when the ( initially negative ) magnetization reaches the value @xmath230 for the first time . [ fig8 ] shows estimates for the resulting mean first passage times for a range of choices of @xmath32 as a function of the field . when this `` lifetime '' of the state with @xmath231 ( i.e. , vapor ) does not exceed @xmath232 , the system is rather unstable , nucleation occurs fast and is followed by fast domain growth as well ; such fast decays of unstable systems are not suitable for tests of nucleation theory . it is seen , that in a rather narrow interval of fields @xmath58 ( for each value of @xmath32 ) the lifetime increases from @xmath232 to @xmath233 . such parameter combinations @xmath234 will be studied in the following only ; if we would study cases where the lifetime is significantly larger than @xmath233 , no critical droplet would be formed during affordable simulation times . \(a ) ( b ) fig . [ fig9 ] then shows typical time evolutions of the size distribution of @xmath235 of physical clusters of size @xmath18 , normalizing them by the equilibrium cluster concentration @xmath236 for @xmath55 . here @xmath236 is defined as the average number of physical clusters per lattice site , and we have the sum rule @xmath237 since every site occupied by a particle must be part of some cluster . of course , here we are only interested in not too small clusters , and hence fig . [ fig9 ] focuses on clusters with @xmath238 . in the time evolution of @xmath239 we recognize three regimes : for times of order @xmath240 is rapidly rising : this period of time corresponds to the relaxation from the initial state ( where @xmath55 ) towards the metastable state . in the latter , @xmath235 is almost constant for at least one , or even several decades of time . then a decay of these plateau values sets in , which is due to the fact that too many much larger clusters have grown , the volume fraction of the system that is still in the metastable phase shrinks , and so less clusters of intermediate size ( as studied in fig . [ fig9 ] ) are observed . this behavior is qualitatively similar to previous studies ( e.g. @xcite ) which were based on the geometrical cluster definition , however . we also remark that in the case of fig . [ fig9]b the lifetime of the plateau extends up to @xmath241 , one decade only , as expected from fig . [ fig8]a , since this case corresponds to a `` lifetime '' of the metastable state of only @xmath242 , and it is clear that only times @xmath243 distinctly less than @xmath244 should be analyzed . while some of the previous work on the studies of the kinetics of cluster growth in metastable ising models ( e.g. @xcite ) tried to use directly @xmath235 to extract information on the validity of nucleation theory concepts , we here try to implement a different concept . namely , we follow the trajectories of individual ( large ) clusters with respect to their size in time , @xmath245 , where @xmath25 is an index to label individual clusters . to ensure that the @xmath25th cluster at time @xmath246 is actually a descendent of the @xmath25th cluster at the time @xmath243 , we have to choose @xmath247 small enough , and also record the location ( center of gravity @xmath248 and the components of the gyration radius @xmath249 where @xmath250 is the @xmath251th cartesian coordinate for the @xmath252th lattice site belonging to the cluster with label @xmath25 , consisting of @xmath253 lattice sites . in each time step in which an analysis of the clusters is performed , the set of coordinates @xmath254 is recorded . note that there occurs the difficulty that the number of large clusters is not constant during the simulation : clusters form and decay or split into parts , and since we know that @xmath255 exceeds @xmath18 for each cluster , and the assignment of the `` active bonds '' according to eq . to identify from the geometrical cluster the associate physical clusters is a random process , some random shift of @xmath256 would occur even if we carry out two successive cluster identifications from the same spin configuration @xmath257 ) . of course , such shifts should be small in comparison with @xmath258 , but as @xmath247 is chosen nonzero it is clear that useful results are only obtained if @xmath247 is small enough , and @xmath18 is large in comparison to clusters that correspond to typical thermal fluctuations @xcite . hence only clusters for which @xmath259 are considered ( for the temperature @xmath130 we chose arbitrarily @xmath260 ) . so if by such criteria ( for details see @xcite ) it is ensured that the @xmath25th cluster with size @xmath261 is a descendent of the @xmath25th cluster with size @xmath253 at time @xmath243 , we can define a reaction rate @xmath262 as @xmath263}{\delta t } \right\rangle _ { \ell_i(t)}\ ] ] here the index @xmath253 stands for an average over a sampling of all cluster trajectories recorded in the simulation , and a smoothing procedure of the ( otherwise too noisy ) data with a triangular smoothing function @xcite was applied . of course , in order to collect statistically significant data on @xmath264 , it is necessary to perform many runs for each parameter combination @xmath265 that is studied . we observe that the lifetime of the metastable stale in such runs is fluctuating dramatically , and so it is necessary to choose the run time of each run individually , rather than the same for all runs . it was decided to stop each run automatically when the largest cluster size @xmath266 was reached . of course , then only clusters with @xmath267 could be studied , to avoid artifacts caused by this cutoff . \(a ) ( b ) while eq . is based on using all clusters @xmath268 at each time @xmath243 , one can simplify matters by restricting the analysis only to the trajectory of the biggest cluster in the system @xcite . when one does this , one ignores possible problems from the fact that from time to time the identity of the largest cluster changes . [ fig10 ] shows now typical results for @xmath264 , using both this latter approximation and the method based on eq . . both methods yield similar trends , although they differ somewhat in detail ( particularly in the case of homogeneous nucleation in the bulk ) . \(a ) + ( b ) what one expects theoretically for @xmath264 is a monotonous increase of @xmath264 with @xmath18 , where @xmath264 is negative for clusters smaller than the critical cluster size @xmath3 while @xmath264 is positive for @xmath269 . the method yields a second positive part for @xmath18 slightly larger than @xmath270 . this is an artifact of ignoring clusters smaller than @xmath270 , which vanishes for @xmath271 @xcite . if we identify the critical cluster size with the zero crossing of @xmath264 at the right hand side , and convert to the cluster volume @xmath272 according to eq . , we obtain the data shown in fig . [ fig11 ] . it is seen that for a given value of the field @xmath58 ( or @xmath229 , respectively ) classical nucleation theory underestimates the volume of the critical cluster : in other words , a given volume @xmath272 leads to a larger value of @xmath58 . since ( according to classical nucleation theory and eq . ) the barriers scale as @xmath273 , this means when one studies nucleation barriers as function of cluster volume @xmath272 or cluster radius @xmath1 , one finds lower barriers in the simulation rather than predicted . qualitatively , the data from the present analysis of cluster kinetics confirm the findings from the lever rule method of winter et al . @xcite , as far as homogeneous nucleation is concerned ( fig . [ fig11]a ) , although some questions on systematic errors in both methods have not been fully settled . nevertheless , the qualitative agreement between these quite different approaches is satisfactory . for the case of heterogeneous nucleation , however , for a given supersaturation the critical cluster volume predicted from cluster kinetics ( fig . [ fig10 ] ) is distinctly larger than the corresponding results from the static methods . we have no explanation for this discrepancy . in the present work , we have studied aspects of nucleation theory by simulation of clusters and their dynamics , using the ising ( lattice gas ) model on the simple cubic lattice . both homogeneous nucleation and heterogeneous nucleation at planar walls ( where a `` surface field '' acts ) have been considered . although many aspects of this problem have been studied before in works of various groups extending over several decades , most of the previous work is inconclusive since it relied on the use of the `` geometric '' cluster definition . we have given evidence that this geometric cluster definition does not yield correct results for large clusters at the temperatures far above the roughening transition temperature where the clusters have spherical shape ; at temperatures below the roughening transition temperature the geometric clusters and the `` physical clusters '' are basically indistinguishable , but due to the pronounced anisotropy effects a simple analysis of nucleation phenomena is not possible . however , in the limit of large droplet volumes @xmath274 , where one can neglect any corrections to the decomposition of the droplet formation free energy into the bulk term plus a surface correction , one can compute the nucleation free energy barrier @xmath0 from measuring the excess chemical potential @xmath98 that is in equilibrium with a given @xmath192 . fig . [ fig : vcntvsvlr ] shows the enhancement of @xmath0 with respect to the standard result for spherical droplets ( using eqs . , ) . we thus show that in the ising ( lattice gas ) model this enhancement gradually rises from unity as the temperature is lowered from the critical temperature , reaches almost 10% at @xmath275 , and rises steeply below the roughening temperature towards the low temperature limit @xmath276 ( fig . [ fig : vcntvsvlr ] ) . this enhancement reflects the consequences of the anisotropy of the interfacial free energy , such as the gradual crossover of the droplet shape from a sphere to a cube ( fig . [ fig6 ] ) . on the other hand , we demonstrate that physical clusters do give consistent results , at least in the bulk when one is concerned with homogeneous nucleation . we show that in the limit where the droplets get macroscopically large , they converge against the simple lever rule predictions . however , we do find an ( unexpected ) surface excess in the particle number of such clusters also in this case . we also demonstrate the validity of the relation between chemical potential ( of the supersaturated vapor ) and the droplet radius that classical nucleation theory predicts for large droplets near the critical temperature . we also give evidence that the droplet - vapor interface is broadened due to capillary waves ; we remind the reader that mean - field type theories and density functional theories @xcite can not include such capillary wave effects ( which also should give rise to a correction term on the droplet formation free energy , not yet included in eq . ) . we would also like to stress that many of our considerations can be carried over to a study of clusters in @xmath277 dimensions , where a construction as in fig . [ fig1 ] also holds . however , we expect two distinctions : ( i ) the roughening transition temperature @xmath38 is zero , so the crossover of droplet shape from the circle to the square occurs without any singularity even for arbitrarily large droplets . ( ii ) percolation coincides with the critical point , but geometrical clusters still are too large , and to describe nucleation , physical clusters defined via eq . should also be used . of course , it would be very desirable to carry these considerations over to nucleation in off - lattice models of fluids . however , a precise analogue of eq . is still not known , and hence other concepts to define physical clusters @xcite need to be used , if one wishes to study nucleation near the critical point . in the second part we present a first study of the time evolution of the cluster population based on the `` physical cluster '' definition . however , due to the large computer resources needed for this study , only data at a single temperature ( @xmath130 ) are presented . in order to allow a comparison of this part of the study with our results on static properties of critical droplets , as studied in the first part of the paper , we use a criterion to estimate the critical droplet size from the balance between droplet growth and shrinking processes . in the case of homogeneous nucleation , the results obtained in this way are roughly compatible with the results obtained from the static lever rule method . studying droplet volumes in the range from 100 to 200 , clear deviations from the classical nucleation theory are seen , which can be attributed to a decrease of the nucleation barrier due to fluctuation effects . however , in the case of heterogeneous nucleation , a rather large discrepancy between the results of the statics and dynamics of droplets is found . this discrepancy is not understood yet , and must be left as a challenging problem for the future . * acknowledgements * : we thank d. winter for providing us with the data from ref . @xcite that were included in fig . [ fig11 ] for comparison . one of us ( f. s. ) thanks the deutsche forschungsgemeinschaft for partial support under grant no vi 237/4 - 3 ( spp 1296 ) . 999 m. volmer and a. weber , z. phys . chem . * 119 * , 277 ( 1926 ) r. becker and w. dring , ann . * 416 * , 719 ( 1935 ) yu . b. zeldovich , acta physicochim . urss * 18 * , 1 ( 1943 ) j. feder , k. c. russell , j. lothe , and g. m. pound , adv . phys . * 15 * , 111 ( 1966 ) h. reiss , j. l. katz , and e. r. cohen , j. stat . phys . * 2 * , 83 ( 1968 ) j. s. langer , ann . * 41 * , 108 ( 1967 ) ; ann . phys . * 54 * , 258 ( 1969 ) _ nucleation _ , edited a. c. zettlemoyer ( m. dekker , new york , 1969 ) f. f. abraham , _ homogeneous nucleation theory _ ( academic , new york , 1974 ) k. binder and d. stauffer , adv . phys . * 25 * , 343 ( 1976 ) k. binder , rep . * 50 * , 783 ( 1987 ) d. w. oxtoby and r. evans , j. chem . phys . * 89 * , 7521 ( 1988 ) a. dillmann and g. e. a. meier , chem . 89 * , 71 ( 1989 ) h. reis , a. tabazadeh , and j. talbot , j. chem . phys . * 92 * , 1266 ( 1990 ) d. w. oxtoby , j. phys . : condens . matter * 4 * , 7627 ( 1992 ) a. laaksonen , v. talanquer , and d. w. oxtoby , annu . phys . chem . * 46 * , 489 ( 1995 ) p. r. ten wolde and d. frenkel , j. chem . phys . * 109 * , 9901 ( 1998 ) d. kashchiev , _ nucleation , basic theory with applications _ ( butterworth - heinemann , oxford , 2000 ) a. c. pan and d. chandler , j. phys . b * 108 * , 19681 ( 2004 ) nucleation , _ comptes rendus physique _ , vol . 7 ( 2006 ) , special issue , edited by s. balibar and j. villain . k. binder , in _ kinetics of phase transitions _ , edited by s. puri and v. wadhavan ( crc press , boca raton , 2009 ) , chap . 2 m. schrader , p. virnau , and k. binder , phys . e * 79 * , 061104 ( 2009 ) s. ryu and w. cai , phys . e * 81 * , 030601(r ) ( 2010 ) b. j. block , s. k. das , m. oettel , p. virnau and k. binder , j. chem . phys . * 133 * , 154702 ( 2010 ) a. troester , m. oettel , b. block , p. virnau and k. binder , j. chem . phys . * 136 * , 064709 ( 2012 ) y. vilsanen , r. strey , and h. reiss , j. chem . phys . * 99 * , 4680 ( 1993 ) a. fladerer and r. strey , j. chem . phys . * 124 * , 164710 ( 2006 ) k. iland , j. wlk , r. strey and d. kashchiev , j. chem . phys . * 127 * , 154506 ( 2007 ) d. turnbull , j. appl . phys . * 21 * , 1022 ( 1950 ) d. turnbull , j. chem . * 18 * , 198 ( 1950 ) a. w. castleman , adv . colloid interface sci . * 10 * , 73 ( 1979 h. biloni , in _ physical metallurgy _ ( r. w. cahn and p. haaasen , eds ) amsterdam , north - holland , 1983 ) p. 477 j. curtius , comptes rendus phys . * 7 * , 1027 ( 2006 ) k. binder and h. mller - krumbhaar , phys . rev . b**9 * * , 2328 ( 1974 ) j. s. rowlinson and b. widom , _ molecular theory of capillarity _ ( oxford univ . press , oxford , 1982 ) k. binder and d. stauffer , j. stat . * 6 * , 49 ( 1972 ) k. binder and e. stoll , phys . lett . * 31 * , 47 ( 1973 ) k. binder and m. h. kalos , j. stat . phys . * 22 * , 363 ( 1980 ) d. stauffer , a. coniglio , and d. w. heermann , phys . * 49 * , 1299 ( 1982 ) d. stauffer , int . j. mod . phys . c**3 * * , 1071 ( 1992 ) h. furukawa and k. binder , phys . rev . a * 26 * , 556 ( 1982 ) j. marro and r. toral , physica * 122a * , 563 ( 1983 ) o. penrose , j. lebowitz , j. marro , m. kalos , and j. tobochnik , j. stat . phys . * 34 * , 399 ( 1984 ) d. w. heermann , a. coniglio , w. klein , and d. stauffer , j. stat . phys . * 36 * , 447 ( 1984 ) h. tomita and s. miyashita , phys . b * 46 * , 8886 ( 1992 ) p. a. rikvold , h. tomita , s. miyashita , and s. w. sides , phys . e * 49 * , 5080 ( 1994 ) p. a. rikvold , and b. m. gorman , in _ annual reviews of computational physics , vol . 1 _ ( d. stauffer , ed . ) ( world scientific , singapore , 1994 ) m. acharyya and d. stauffer , eur . j. b * 5 * , 571 ( 1998 ) h. vehkamaki and i. j. ford , phys . e * 59 * , 6483 ( 1999 ) v. a. shneidman , k. a. jackson , and k. m. beatty , phys . rev . b**59 * * , 3579 ( 1999 ) v. a. shneidman , k. a. jackson , and k. m. beatty , j. chem . * 11 * , 6932 ( 1999 ) r. a. ramos , p. a. rikvold , and m. a. novotny , phys . b * 59 * , 9053 ( 1999 ) s. wonczak , r. strey , and d. stauffer , j. chem . phys . * 113 * , 1976 ( 2000 ) a. t. bustillos , d. w. heermann and c. e. cordeiro , j. chem . phys . * 121 * , 4864 ( 2004 ) k. brendel , g. t. barkema , and h. van beijeren , phys . rev . e * 71 * , 031601 ( 2005 ) d. winter , p. virnau and k. binder , j. phys . : condens . matter * 21 * , 464118 ( 2009 ) d. winter , p. virnau and k. binder , phys . * 103 * , 225703 ( 2009 ) h. mller - krumbhaar , phys 50a * , 27 ( 1974 ) k. binder , ann . phys . * 98 * , 390 ( 1976 ) a. coniglio and w. klein , j. phys . a**13 * * , 2775 ( 1980 ) r. h. swendsen and j. s. wang , phys . rev . lett . * 58 * , 86 ( 1987 ) m. donorio de meo , d. w. heermann , and k. binder , j. stat . phys . * 60 * , 585 ( 1990 ) a. coniglio , j. phys . * a8 * , 1773 ( 1975 ) a. coniglio , f. peruggi , c. nappi and l. russo , j. phys . a**10 * * , 205 ( 1977 ) k. binder , solid state commun . * 34 * , 191 ( 1980 ) s. hayward , d. w. heermann , and k. binder , j. stat . phys . * 49 * , 1053 ( 1987 ) p. w. kasteleyn and c. m. fortuin , j. phys . japan * 26 * ( suppl ) 11 ( 1969 ) c. m. fortuin and p. w. kasteleyn , physica * 57 * , 536 ( 1972 ) u. wolff , phys . lett . * 62 * , 361 ( 1989 ) k. kawasaki , in _ phase transitions and critical phenomena , vol . 2 _ ( c. domb and m. s. green , eds . ) ( academic press , london , 1972 ) h. mller - krumbhaar and k. binder , j. stat . phys . * 8 * , 1 ( 1973 ) j. d. weeks , in _ ordering in strongly fluctuating condensed matter systems _ ( t. riste , ed . ) p. 293 ( plenum press , new york , 1980 ) h. van beijeren and i. nolden , in _ structure and dynamics of surfaces ii _ ( w. schommers and p. blanckenhagen , eds . ) p. 259 ( springer , berlin 1987 ) g. wulff , z. kristallogr . mineral . * 34 * , 449 ( 1901 ) c. herring , phys . 82 * , 87 ( 1951 ) c. rottman and m. wortis , phys . b * 24 * , 6274 ( 1981 ) m. holzer , phys . rev . b * 42 * , 10570 ( 1990 ) k. k. mon , s. wansleben , d. p. landau and k. binder , phys . rev . b * 39 * , 7089 ( 1989 ) m. hasenbusch and k. pinn , j. phys . a * 30 * , 63 ( 1997 ) a. m. ferrenberg and d. p. landau , phys . b * 44 * , 5081 ( 1991 ) v. privman , phys . rev . lett . * 61 * , 183 ( 1988 ) m. hasenbusch and k. pinn , physica a * 192 * , 342 ( 1993 ) m. hasenbusch and k. pinn , physica a * 203 * , 189 ( 1994 ) d. p. landau and k. binder , _ a guide to monte carlo simulation in statistical physics , 3rd ed _ ( cambridge univ . press , cambridge 2009 ) k. binder and d. w. heermann , _ monte carlo simulation in statistical physics : an introduction_. 5th ed . ( springer , berlin 2010 ) k. binder , physica a * 319 * , 99 ( 2003 ) d. stauffer and a. aharony , _ introduction to percolation theory _ ( taylor and francis , london , 1994 ) b. widom , j. chem . phys . * 39 * , 2808 ( 1963 ) b. widom , j. phys . chem . * 86 * , 869 ( 1982 ) f. schmitz , _ diploma thesis _ ( johannes gutenberg - universitt mainz , 2011 , unpublished . ) p. virnau , m. mller , j. chem . phys . * 120 * , 10925 ( 2004 ) p. virnau , m. mller , l. g. macdowell , k. binder , j. chem . phys . * 121 * , 2169 ( 2004 ) l. g. macdowell , p. virnau , m. mller , and k. binder , j. chem . phys . , 5293 ( 2004 ) b. a. berg , t. neuhaus , phys . lett . * 68 * , 9 ( 1992 ) f. wang and d. p. landau , phys . lett . * 86 * , 2050 ( 2001 ) f. wang and d. p. landau , phys . e * 64 * , 056101 ( 2001 ) s. k. das and k. binder , phys . lett . * 107 * , 235702 ( 2011 ) m. p. a. fisher and m. wortis , phys . b * 29 * , 6252 ( 1984 ) d. jasnov , rep . * 47 * , 1059 ( 1984 ) k. binder , m. mller , f. schmid , and a. werner , adv . colloid and interface science * 94 * , 237 ( 2001 ) k. binder and m. mller , int . c * 11 * , 1093 ( 2000 ) g. v. wulff , z. kristallogr . mineral . * 34 * , 449 ( 1901 )
we consider the supernova shock effects , the mikheyev - smirnov - wolfenstein ( msw ) effects , the collective effects , and the earth matter effects in the detection of type ii supernova neutrinos on the earth . it is found that the event number of supernova neutrinos depends on the neutrino mass hierarchy , the neutrino mixing angle , and neutrino masses . therefore , we propose possible methods to identify the mass hierarchy and acquire information about and neutrino masses by detecting supernova neutrinos . we apply these methods to some current neutrino experiments .
1003.1197
supernovas ( sns ) are extremely powerful explosions in the universe which terminate the life of some stars @xcite . they make the catastrophic end of stars more massive than 8 solar masses , leaving behind compact remnants such as neutron stars or black holes . the sn explosion , such as sn1987a @xcite , is one of the most spectacular comic events and the source of new physics ideas @xcite . detecting sn neutrinos on the earth has been a subject of active investigation in astroparticle physics @xcite@xcite@xcite-@xcite . we may obtain some information about the explosion mechanism of sn @xcite and the parameters of neutrinos @xcite@xcite@xcite@xcite by detecting sn neutrinos on the earth . several neutrino experiments , including the daya bay reactor neutrino underground laboratory @xcite , can be used to detect neutrino events from a sn explosion and serve as a sn early warning system @xcite . recently , several methods to obtain information on neutrino parameters by the study of sn neutrinos have been proposed . in ref . @xcite , dasgupta _ et al . _ proposed a method to identify the neutrino mass hierarchy for very small @xmath0 ( @xmath1 ) through the earth matter effects . in ref . @xcite , skadhauge _ et al . _ proposed methods to identify the neutrino mass hierarchy for larger @xmath0 ( @xmath2 ) and obtain information about @xmath0 by sn neutrinos . in ref . @xcite , using the event number of the delayed sn neutrinos , a method to measure neutrino masses was proposed by beacom _ et al._. in these works , the sn shock effects and the collective effects were not taken into account . furthermore , uncertainties of some of the parameters appearing in sn neutrino fluxes , such as the luminosities @xmath3 , temperatures @xmath4 , and pinching parameters @xmath5 , were not considered . in ref . @xcite , we studied the earth matter effects in the detection of sn neutrinos at the daya bay experiment and proposed a possible method to acquire information about the neutrino mixing angle @xmath0 smaller than @xmath6 which is difficult to access in reactor neutrino experiments . we defined the ratio @xmath7 as the event number of @xmath8 over that of @xmath9 . using the relation between the event number of sn neutrinos @xmath10 and @xmath0 , we can obtain information about small @xmath0 by measuring @xmath7 . in that paper , a simplified picture for the collective effects was used @xcite and the neutrino flavor transformation due to the sn shock effects was not considered . furthermore , it was assumed that different flavor sn neutrinos have the same time - independent luminosity . hence in that calculation there are several uncertainties which we will address in the present work . in this paper , we study the collective effects more accurately @xcite@xcite and consider the neutrino flavor transformation arising from the sn shock effects @xcite@xcite@xcite . in addition , according to current monte carlo simulations on sn neutrino spectra @xcite , we allow different flavor neutrinos to have different time - dependent luminosities . furthermore , in order to reduce the uncertainties of unknown parameters in sn neutrino fluxes such as luminosity , temperature , and pinching parameter , we will define the ratio of the event number of sn neutrinos for the charged - current reaction detected after one second over that detected before one second ( this is different from the ratio considered in ref . @xcite ) . with this ratio , we may identify the mass hierarchy , @xmath11 vs @xmath12 , and acquire information about @xmath0 . this method can be applied to neutrino detectors of various types . examples are the liquid scintillator detector of the daya bay experiment @xcite under construction , the decommissioned heavy water detector of the previous sudbury neutrino observatory ( sno ) @xcite , etc . the other main purpose of the present work is to propose a method to obtain some information about neutrino masses from sn neutrinos . since neutrinos have small masses , their transmit velocities are smaller than that of light . therefore , a massive neutrino will have an energy and mass dependent delay relative to a massless neutrino after traveling over a long distance @xcite . we will define another ratio , for the neutral - current reactions , as the event number of the delayed sn neutrinos over the total event number . with the relation between this ratio and neutrino masses , we will propose a possible method to acquire information about neutrino masses . this method will also be applied to daya bay and sno for examples . the paper is organized as follows . in sec . ii , we review the detection of sn neutrinos on the earth . in sec . iii , we propose a possible method to identify the mass hierarchy and acquire information about @xmath0 and apply the method to the daya bay and sno experiments . in sec . iv , a method to acquire information about neutrino masses is given and also applied to daya bay and sno . a summary is given in section v. in the explosion a typical type ii sn generates intensive neutrinos which carry about @xmath13 of the total energy and the explosion itself consumes about 1@xmath14 of the total energy . the vast amount of neutrinos are produced in two bursts . in the first burst which lasts for only a few milliseconds , electron neutrinos are generated by the electron capture by nuclei @xmath15 via the inverse beta decay @xmath16 which leads to a neutron rich star . in the second burst which lasts longer ( about 10 seconds ) , neutrinos of all flavors are produced through the electron - positron pair annihilation @xmath17 , the electron - nucleon bremsstrahlung @xmath18 , the nucleon - nucleon bremsstrahlung @xmath19 , the plasmon decay @xmath20 , and the photoannihilation @xmath21 @xcite . we assume a `` standard '' sn explosion at a distance @xmath22 from the earth , releasing a total energy @xmath23 ( similar to sn1987a ) and the luminosity flux of the sn neutrino @xmath24 , @xmath25 , distributed in time as @xcite@xcite@xcite @xmath26 where @xmath3 is the luminosity of the neutrino @xmath24 at @xmath27 and @xmath28 the luminosity decay time . the following range of @xmath28 was obtained by fitting the the experimental data of sn1987a with the exponentially decaying luminosity @xcite@xcite : @xmath29 the typical relations among @xmath3 obtained from numerical simulations are @xcite @xmath30 there are two parametrizations of sn neutrino fluxes used in general simulations , one used by the livermore group @xcite@xcite and the other by the garching group @xcite . in the present work , we use the livermore parameterization which is similar to the fermi - dirac distribution . for the sn neutrino of flavor @xmath24 , the time - integrated neutrino energy spectra is given by @xcite @xmath31 where @xmath32 is the neutrino energy , @xmath4 the temperature of the neutrino @xmath24 , @xmath33 the pinching parameter of the spectrum , @xmath25 the luminosity given by eq . ( [ l ] ) , and @xmath34 is defined by @xmath35 where @xmath36 is an integer . the spectra obtained from numerical simulations can be well fitted by @xcite @xmath37 while neutrinos propagate outward to the surface of the sn they could be subjected to the sn shock effects @xcite@xcite@xcite@xcite@xcite , the msw effects @xcite@xcite@xcite , and the collective effects @xcite@xcite@xcite@xcite@xcite@xcite@xcite@xcite@xcite . then , after travelling the cosmic distance to reach the earth , they go through a part of the earth and are subjected to the earth matter effects @xcite@xcite@xcite@xcite . in the following , we will consider the above four types of effects in detail . as was pointed out in ref . @xcite , the shock propagating inside the sn may reach the region of densities relevant for neutrino conversion during the period of neutrino emission . it modifies the density profile of the star , thus affecting neutrino flavor transitions after the core bounce @xcite@xcite . the time - dependent variations of the neutrino potential in matter along with the sn shock profile can leave distinctive imprints on the energy and time structure of the neutrino signal . the study of such effects may provide us with additional constraints for the determination of the neutrino mass hierarchy , the neutrino mixing angle @xmath0 , and neutrino masses . in ref . @xcite , the authors introduce a simplified , empirical parametrization of the shock density profile @xmath38 , which reproduces the main features of the graphical profile of ref . @xcite and is continuous in both @xmath39 and the supernova radial coordinate @xmath40 . in this way , numerical and analytical calculations of the crossing probability at the high resonance region @xmath41 can continuously cover the relevant parameter space . we follow the notation of ref . @xcite in the discussion below . for post - bounce time @xmath42 @xcite before the shock effects begin , the sn matter density is @xmath43 for @xmath44 the shock effects set in and the matter density is given by @xmath45 where @xmath46 is defined as @xmath47[\arcsin(1-r / r_s)]^{1.1}\ } , \label{f}\ ] ] and @xmath48 is a typical ratio of the potential across the shock front , @xmath49 which measures the sn matter potential @xmath50 drop from @xmath51 to @xmath52 the sn matter potential is related to the sn electron density @xmath53 by @xmath54 where @xmath55 is the avogadro s number and @xmath56 is the electron fraction . in the numerical calculations , we assume @xmath57 . in eqs . ( [ rho0])-([xi ] ) , a slightly accelerating shock - front position @xmath58 is assumed with the explicit time dependence , @xmath59 where @xmath58 is in units of @xmath60 and @xmath39 in units of @xmath61 . using eqs . ( [ rho0])-([xi ] ) , we plot in fig . 1 the density profile of the sn as a function of the radius at different times . these curves reproduce reasonably well the main features of the shock profiles as shown in ref . the straight line labelled by @xmath27 is the unmodified density profile given by eq . ( [ rho0 ] ) . once the shock effects begin , at a given time the shock radius @xmath58 is determined by eq . ( [ rs ] ) . the density profile is suppressed more and more in the region @xmath62 as the time and therefore @xmath58 increase . the density reaches a minimum and then rises relatively quickly to the value @xmath63 as @xmath64 . in the region @xmath65 the density profile is the unperturbed original form @xmath66 . the density suffers a sizable discontinuity at @xmath58 . and therefore the discontinuity is @xmath67 . ] for neutrinos emitted in the region @xmath65 , due to their much higher velocities , which is the order of speed of the light , than that of the shock wave propagation , which is the order of @xmath68 , they are not influenced by the shock effects . hence the majority of sn neutrinos which are emitted within a 10 second period will be able to semble the whole range of the profile of the sn density under the influence of the shock . an important consequence of the density modification by the shock wave is that some of the neutrinos can undergo multiple resonance flavor conversion when they traverse the sn . we consider them below . let us first consider the neutrino flavor conversion due to the density discontinuity at the shock front @xmath58 . this crossing probability of the neutrino flavor transformation which we denote as @xmath69 can be determined by the conservation of flavor across the matter potential discontinuity . by a simple calculation , we have @xcite @xmath70 where @xmath71 and @xmath72 are respectively the effective mixing angles in matter immediately before and after the shock front position @xmath58 . such angles are defined by @xmath73 the msw effects @xcite is the adiabatic or partially adiabatic neutrino flavor conversion in media with locality or time varying matter density . the main feature is that in a high density medium a flavor is also a mass state which propagates without flavor oscillation , until crossing a resonance where there is a finite probability for flavors change , i.e. , a jump from one mass eigenstate to another . for @xmath0 smaller than @xmath74 , using the landau - zener formula @xcite , one can obtain the crossing probability , @xmath75 , for neutrinos to jump from one mass eigenstate to another at the high resonance layer ( @xmath76 ) @xmath77 with @xmath78 where @xmath79 is the electron density @xcite and @xmath80 @xcite . similarly , one can calculate the expression of the crossing probability @xmath81 at the low resonance region inside the sn . however , due to the large angle solution of the solar neutrino , @xmath82 . in the presence of a propagating shock wave , the calculation of the global crossing probability , @xmath41 , is not straightforward . the shock modifies the density profile as discussed in the preceding subsection . within the shock wave and near the shock front the matter density is no longer monotonically decreasing , but rather goes through a minimum and a maximum . so the resonance condition can be satisfied possibly as many as three points : @xmath83 it is , of course , possible that there are only two or one resonance points , or even none at all . in the case of three resonance points , which we denote as @xmath84 , two of the points lie below the shock front and the third one beyond the shock front , i , e . , @xmath85 . the crossing probability @xmath75 at @xmath86 , @xmath87 , can be calculated from eq . ( [ ph0 ] ) . note that although the matter densities are the same at all resonance points as required by eq . ( [ resonance ] ) , their crossing probabilities , as given by eq . ( [ gamma ] ) , may be different because the slop of the matter density may be different at different resonance points . the global crossing probability can be expressed in terms of four crossing probabilities @xmath88 , @xmath89 , @xmath90 , and @xmath91 . to obtain the expressions of the global crossing probabilities , we first define four densities : @xmath92 is the resonance density which satisfies eq . ( [ resonance ] ) , @xmath93 the density at @xmath94 corresponding to the matter potential @xmath95 , @xmath96 the density at @xmath94 corresponding to the matter potential @xmath97 , and @xmath98 the density at the bottom of the camber which is the minimum of the density profile below the shock front . from fig . 1 , it can be seen that @xmath99 in the pre - bounce time period @xmath42 , the resonance condition occurs at only one point and there are no neutrino shock effects . then , @xmath100 in the post - shock period @xmath101 we have four different situations : \(1 ) for @xmath102 , the resonance condition does nt occur and then @xmath103 \(2 ) for @xmath104 , or @xmath105 and @xmath106 , or @xmath107 and @xmath108 , the resonance condition occurs at only one point , @xmath109 or @xmath110 , and then @xmath111 \(3 ) for @xmath112 and @xmath106 , or @xmath105 and @xmath108 , the resonance condition occur at two points , @xmath113 or @xmath114 , and then @xmath115 \(4 ) for @xmath116 , the resonance condition occur at three points , @xmath117 , and then @xmath118 in eqs . ( [ ph3]-[ph5 ] ) , @xmath119 is defined as @xmath120 ^ 2+(\delta{m}^2_{31 } \sin2\theta_{13})^2 } , \label{phiij}\end{aligned}\ ] ] with @xmath50 being the sn matter potential . a summary for neutrino flavor conversions due to the neutrino shock effects and the msw effects in various density regions is given in table i. due to the sn matter effects and the sn shock effects , the expressions of @xmath41 , eqs . ( [ ph1])-([ph5 ] ) , are different from eq . ( 25 ) in ref . @xcite and eq . ( 34 ) in ref . @xcite . because of the various effects they have been subjected to , the sn neutrinos reaching the earth have a very rich structures to be explored in their spectra in time , energy , and the mixing angle @xmath0 . in fig . 2 , we plot the crossing probability @xmath41 as a function of the neutrino energy @xmath32 at @xmath121 for two very small mixing angles @xmath122 and @xmath123 . it is seen that there are large variations in @xmath41 as the neutrino energy varies . these forms of behavior are different from those given in refs . @xcite and @xcite . in fig . 3 , we plot @xmath41 as a function of @xmath39 for @xmath124 at a fixed @xmath32 for three different values of @xmath32 . it can be seen that the shapes of the curves depend strongly on @xmath32 and there are many oscillations in @xmath41 in an interval around @xmath125 . the interval enlarges and moves to higher @xmath39 value as @xmath32 increases . this is different from that given in ref . @xcite . in fig . 4 , we plot @xmath41 as a function of @xmath0 for different values of @xmath32 and @xmath39 . it is found that at some values of @xmath0 , which depends on the energy of sn neutrinos , @xmath41 begins to change very rapidly with @xmath0 at @xmath121 , and the behavior of the @xmath41 dependence on @xmath0 is different from that given in ref . these differences are attributable to the different sn matter effects when the shock effects is included in the consideration . the relation between @xmath41 and @xmath0 will provide a way to obtain information on @xmath0 and the study of the shock wave effects . because of the neutrino - neutrino forward scattering or neutrino self - interaction , neutrinos can experience collective flavor transformation in the sn near the neutrinosphere where the neutrino number densities are very large @xcite . this phenomenon is different from the msw effects in that the flavor evolution histories of neutrinos in the collective oscillations are coupled together and must be solved simultaneously . recently , a significant amount of studies on collective effects have been made by a number of authors , e.g. , duan _ et al . _ @xcite@xcite@xcite@xcite , dasgupta _ et al . _ @xcite@xcite , hannestad _ et al . _ @xcite , raffelt _ et al . _ @xcite , and fogli _ et al . _ @xcite . defining @xmath126(@xmath127 ) as the probability that the ( anti-)neutrino @xmath128(@xmath129 ) remains as @xmath128(@xmath129 ) after the collective effects , dasgupta _ et al . _ @xcite introduced a much simplified picture to explain the characteristics of the collective effects which we followed in our earlier paper @xcite , i.e. @xmath130 in the case of normal mass hierarchy ; and @xmath131 , @xmath132 in the case of inverted mass hierarchy , where @xmath133 is a critical energy across which a stepwise flavor conversion of @xmath8 develops . however , from the numerical simulations of the collective effects obtained by duan _ @xcite , it is found that @xmath126(@xmath127 ) depends on the neutrino ( anti - neutrino ) energy @xmath32 , the sn radial coordinate @xmath40 , the sn neutrino emission angle , and it is very difficult to obtain an analytical expression for @xmath126(@xmath127 ) . it can be seen from movies 5 and 6 in ref . @xcite that when the sn radial coordinate @xmath40 increases , there are oscillations between @xmath8 ( @xmath9 ) and @xmath134 ( @xmath135 ) due to the collective effects . when @xmath136 , which is far away from the neutrinosphere , the collective effects for the neutrino amount to a swap between the @xmath8 and @xmath134 fluxes , irrespective of the mass hierarchy being normal or inverted . for the anti - neutrino , in the case of normal mass hierarchy , the fluxes of @xmath9 and @xmath135 are nearly unchanged and there is no swap between them . in the case of inverted mass hierarchy , the behaviors of the @xmath9 and @xmath135 fluxes are very complex and , for simplicity , they are generally approximated @xcite@xcite as that there is no swap between them . in this way , we apply the following new simplified picture to describe the characteristics of the collective effects : @xmath137 for neutrinos and @xmath138 for antineutrinos . it will be seen from our numerical calculations that the results obtained in secs . iii and iv hardly change for @xmath133 varying between @xmath139 and @xmath140 . therefore , similar to ref . @xcite , we take @xmath141 . neutrinos from a sn may travel through a significant portion of the earth before reaching the detector and are subjected to the matter effects @xcite . therefore , we need to consider the earth matter effects in studying sn neutrinos . suppose a neutrino reaches the detector at the incident angle @xmath142 ( see fig . then the distance that the neutrino travels through the earth to the detector @xmath143 and the distance of the neutrino to the center of the earth @xmath144 are given by @xmath145 where @xmath146 is the depth of the detector in the earth , @xmath147 the distance that the neutrino travels into the earth , and @xmath148 the radius of the earth . let @xmath149 be the probability that a neutrino mass eigenstate @xmath150 enters the surface of the earth and arrives at the detector as an electron neutrino @xmath8 , we have @xcite @xmath151 where @xmath152 is the potential @xmath8 experiences due to the matter density @xmath153 inside the earth @xcite@xcite@xcite @xmath154 and @xmath155 is defined as @xmath156 where @xmath157 with @xmath158 , @xmath159 , @xmath160 @xcite . when all the effects , including the shock , msw , collective , and the earth matter , are taken into account , the sn neutrino fluxes at the detector can be written as @xmath161 where the survival probabilities @xmath162 and @xmath163 are given by @xmath164 , \nonumber\\ \bar{p}&=&(1-\bar{p}_{2e})\bar{p}_{\nu\nu } , \label{pn}\end{aligned}\ ] ] for the normal mass hierarchy and @xmath165 , \label{pi}\end{aligned}\ ] ] for the inverted mass hierarchy . therefore , the event numbers @xmath166 of sn neutrinos in the various reaction channels `` @xmath167 '' can be calculated by integrating over the neutrino energy , the product of the target number @xmath168 , the cross section of the given channel @xmath169 , and the neutrino flux function at the detector @xmath170 , @xmath171 where @xmath24 stands for the neutrino or antineutrino of a given flavor , and @xmath172 is the distance between the sn and the earth . in the following section , we propose a method to identify the mass hierarchy and acquire information about @xmath0 by detecting sn neutrinos . in the above section , we have discussed the sn shock effects , the msw effects , the collective effects , and the earth matter effects in detail for the detection of sn neutrinos . from eqs . ( [ ps])-([gamma ] ) and ( [ ph1])-([ntotal ] ) , it can be seen that the event number of sn neutrinos depends on the mixing angle @xmath0 for different mass hierarchies . therefore , we may identify the mass hierarchy and acquire information about @xmath0 by detecting sn neutrinos . in ref . @xcite , we defined a ratio of the event number of @xmath8 over that of @xmath9 and studied the method to acquire information about @xmath0 . however , in that paper , we ignored the sn shock effects and only a very simple model for the collective effects was considered . in addition , with the naive luminosity of sn neutrinos employed there , the uncertainties in @xmath0 will be very large . in the present work , we take into account the sn shock effects and consider the collective effects more realistically . in order to further reduce the influence of the uncertainties in the parameters of the sn neutrino fluxes , we define a more appropriate ratio as the event number of sn neutrinos for the charged - current reaction detected after @xmath173 over that detected before @xmath173 . using this new ratio , we can better acquire information about @xmath0 and identify the neutrino mass hierarchy . from ref . @xcite and detailed numerical studies related to the present work , we found that the ratios of event numbers of sn neutrinos are nearly independent of the sn neutrino incident angle . therefore , in the following we will only consider the case for the incident angle of @xmath174 . there will be eight detectors located at the near and far sites of the daya bay experiment @xcite . the total detector mass is about 300 tons and the depth of the detector @xmath175 m. the daya bay collaboration has decided to use linear alkyl benzene ( lab ) as the main part of the liquid scintillator . lab has a chemical composition including @xmath176 and @xmath177 and the ratio of the number of @xmath176 and @xmath177 is about 0.6 . then , the total numbers of target protons , electrons , and @xmath178 are @xmath179 in the daya bay experiment , there are three reactions which can be used to detect sn neutrinos : the inverse beta decay , neutrino - electron reactions , and neutrino - carbon reactions . however , since the event number observed in the channel of neutrino - electron reactions at daya bay is very small @xcite , these reactions will not be considered in the present work . therefore , we only consider the inverse beta decay and the neutrino - carbon reactions , their effective cross sections are given as follows @xcite@xcite : \(1 ) the inverse beta decay is @xmath180 where the reaction threshold is @xmath181 . \(2 ) the neutrino - carbon reactions are @xmath182 for the charged - current capture , and @xmath183 for the neutral - current capture . the effective cross sections in eq . ( [ sigma cb ] ) are valid for sn neutrinos without oscillations . when neutrino oscillations are taken into account , the oscillations of higher energy @xmath184 into @xmath8 result in an increased event rate since the expected @xmath8 energies in the absence of oscillations are just at or below the charged - current reaction threshold . this leads to an increase by a factor of 35 for the cross section @xmath185 if we average it over a @xmath8 distribution with @xmath186 rather than @xmath187 . similarly , the cross section @xmath188 is increased by a factor of 5 . in order to reduce the uncertainties of the unknown parameters appearing in the sn neutrino fluxes , including those in the luminosities @xmath3 , temperatures @xmath4 , and pinching parameters @xmath5 , we define the ratio @xmath189 as the event number of @xmath8(@xmath9 ) for the charged - current reaction detected after @xmath173 over that detected before @xmath173 , @xmath190 with eqs . ( [ l ] ) , ( [ foa ] ) , ( [ ps])-([gamma ] ) , ( [ ph1])-([ntotal ] ) , and ( [ ratio1 ] ) , we can see that @xmath191 depends on the mixing angle @xmath0 and the luminosity decay time @xmath28 . form a detailed numerical calculation , we found that @xmath191 which changes with @xmath0 has also a significant dependence on @xmath28 . this can lead to large uncertainties in @xmath191 when @xmath28 varies in the range given in eq . ( [ tau ] ) . therefore , it will be difficult to identify the mass hierarchy and acquire information on @xmath0 from @xmath191 when @xmath28 has a significant uncertainty . however , if the value of @xmath191 is determined through sn neutrinos , we can obtain a rather tight relation between @xmath28 and @xmath0 . we demonstrate this relation in fig . 6 , for the neutrino - carbon reactions . in fig . 6 we plot @xmath28 as a function of @xmath0 for a fixed value of @xmath192 . we take @xmath193 and @xmath194 as an example and @xmath195 take other values , the qualitative results do not change . ] . from this figure , we see that the value of @xmath28 lies in a small range , i.e. , @xmath196 when @xmath0 varies from 0 to @xmath197 for the normal mass hierarchy . this is much smaller than the presently allowed range given in eq . ( [ tau ] ) . it should be remarked that several authors @xcite@xcite@xcite have typically used the best fit value , i.e. , @xmath198 @xcite in their study of sn neutrinos . first consider the channel of the inverse beta decay in which only @xmath199 is relevant . we plot @xmath199 in fig . 7 as a function of @xmath0 for the two mass hierarchies , and for @xmath195 , @xmath4 , and @xmath5 taking their allowed extreme values in the ranges given in eqs . ( [ l0 ] ) and ( [ t ] ) . for the normal mass hierarchy @xmath199 , having the value of 2.4 , is independent of @xmath0 . so it is impossible to obtain any information on @xmath0 for the normal mass hierarchy in the inverse beta decay process . for the inverted mass hierarchy , for a given value of @xmath200 , @xmath199 lies in a band bounded by the curves marked as `` inverted & max '' and `` inverted & min '' , which are obtained by setting the values of both @xmath201 and @xmath33 at their upper and lower bounds , respectively . from this plot , it can be seen that if the value of @xmath199 is away from 2.4 , that is statically significant , the mass hierarchy must be inverted and a band of values of @xmath0 and @xmath200 can be determined . but it is difficult to determine @xmath0 from @xmath199 alone . it is interesting to observe that , since the value of @xmath199 falls bound in the limited range of 2.1 - 2.8 , it can serve as a test of the validity of the overall parametrization of the sn explosion mechanism . for the neutrino - carbon reactions , both @xmath192 and @xmath199 can be used to identify the mass hierarchy and obtain information on @xmath0 . in fig . 8a , we plot @xmath192 vs @xmath0 similar to fig . 7 . here for the inverted mass hierarchy , @xmath192 has the constant value 2.4 independent of @xmath0 . if @xmath192 is larger than 2.4 , the mass hierarchy must be normal . it can also be seen from fig . 8a that @xmath192 depends on @xmath0 sensitively for the normal mass hierarchy and the uncertainties of @xmath192 due to @xmath3 , @xmath4 , and @xmath5 are small . therefore , in the case of normal mass hierarchy , we can restrict @xmath0 to a small range of values by measuring @xmath192 . similar plots are made for @xmath199 in fig . now for the normal mass hierarchy , @xmath199 is independent of @xmath0 . for the inverted mass hierarchy @xmath199 varies with @xmath0 and is smaller than 2.4 . hence we may also be able to identify the mass hierarchy by @xmath199 . however , as shown in the plot , in the case of inverted mass hierarchy the large uncertainties of @xmath199 due to @xmath3 , @xmath4 , and @xmath5 make it difficult to constrain @xmath0 by @xmath199 . the daya bay experiment has the sensitivity limit of 0.01 for @xmath202 , i.e. , @xmath197 for @xmath0 . however , if a sn explosion takes place during the operation of the experiment , it is possible to reach a much smaller value of @xmath0 with the benefit of determining the mass hierarchy by measuring @xmath192 for the neutrino - carbon reactions . now let us consider the heavy water detector , for an example , used in the previous sno experiment @xcite . the detector mass was made of 1000 tons of heavy water located in the depth of @xmath203 m.w.e . we consider only the neutrino - deuterium reactions . the total number of deuterium is @xmath204 . scaling the experimentally measured energy values from the decay of the muon at rest to the energy scale of sn neutrinos , one can obtain the effective cross sections of neutrino - deuterium reactions as follows @xcite@xcite : @xmath205 for the charged - current capture , and @xmath206 for the neutral - current capture , where @xmath207 . in figs . 9a and 9b we plot @xmath192 and @xmath199 of the neutrino - deuterium reactions in analogy with figs . 8a and 8b . similar to the neutrino - carbon case given in fig . 9a for @xmath192 can be used to determine the neutrino mass hierarchy . but it is difficult to pin down the value of @xmath0 . figure 9b for @xmath199 is similar to fig . 7 and the discussion can be applied to the present situation , with the allowed value of @xmath199 between 1.9 to 2.6 in the case of inverted mass hierarchy . in the above discussions , we have applied our method to three reactions : inverse beta decay , neutrino - carbon , and neutrino - deuterium . a summary is given in table ii . from this table , it can be seen that the mass hierarchy may be identified by measuring @xmath192 in the neutrino - carbon reactions and the neutrino - deuterium reactions , and by measuring @xmath199 in the inverse beta decay , the neutrino - carbon reactions , and the neutrino - deuterium reactions . in addition , in the case of normal mass hierarchy , @xmath192 measured in the channel of neutrino - carbon reactions can provide the most accurate information about @xmath0 . similarly , the present method can also be applied to other neutrino experiments , such as kamland @xcite , minboone @xcite , borexino @xcite , and double - chooz @xcite in the channel of neutrino - carbon reactions . in the future , if a sn explosion takes place within the cosmic distance considered here , it is possible to identify the mass hierarchy and reach a much smaller value of @xmath0 than current neutrino experiments through the measurement of @xmath192 in the channel of neutrino - carbon reactions . all of our understanding of neutrino properties has been derived in the terrestrial environment , with input from solar neutrinos , and our study of sn neutrinos has been made under the reasonable assumption that the terrestrial properties can be applied unaltered to the extreme conditions of the cosmos , such as those in the sn , e.g. , high temperature , high density , etc .. if an observation of sn neutrinos deviates from the prediction , it will be argued that our picture of sn requires modification . however , it would be prudent to check if some aspects of the properties of neutrinos may have been modified under the unusual cosmic conditions . in this section we propose a measurement , the observation of which , can only be attributed to some change of the neutrino property , albeit a very specific one . since neutrinos , at least some of the species , have finite masses , their transmitting velocities are smaller than that of light . then , neutrinos from a sn explosion , while traveling through a large distance to the earth , can experience a delay in the arrival time , which depends on their energies and masses and may be observable @xcite@xcite@xcite . suppose a neutrino from a sn , of mass @xmath208 ( in ev ) and energy @xmath32 ( in mev ) , travels over a distance @xmath172 ( in 10kpc ) to the earth , it will experience an energy - dependent time delay of @xmath209 ( in second ) relative to a massless neutrino @xcite : @xmath210 where only the lowest order term in the neutrino mass expansion has been kept . in ref . @xcite , a method of measuring the neutrino mass is proposed , using the detection of the event number of the delayed sn neutrinos . in the papers , a simple model of the luminosities of sn neutrinos was assumed , in the absence of various effects considered in recent sn neutrino studies . the neutrino luminosities were assumed to have a common step - function cutoff at a fixed time . this form of luminosities was derived from the argument that the neutrino gravitational redshifts become severe enough only near the sharp cutoff time @xmath211 , where @xmath211 can be determined from the observed time profile of sn neutrinos . therefore , the luminosities , which are constant before @xmath211 and vanishes abruptly afterward , take the form @xmath212 . in the present work , we take into account the full complexity of the neutrino flavor transformation due to the effects of the shock , msw , collective , and the earth matter . also , we consider the time - dependence of the sn neutrino luminosities given in eq . ( [ l ] ) and the uncertainties in the sn neutrino flux parameters given in eqs . ( [ l0 ] ) and ( [ t ] ) . since the time delay @xmath209 depends on the neutrino mass @xmath208 , we can express the event number of the delayed sn neutrinos as a function of @xmath208 . therefore , information about @xmath208 can be obtained by detecting the event number of the delayed sn neutrinos . from ref . @xcite there are the following neutrino mass bounds @xmath213 for the sn electron antineutrino the averaged energy is @xmath214 , which together with @xmath22 , gives @xmath215 from eq . ( [ dt ] ) . this time delay of the @xmath9 is much smaller than the time scale of the sn explosion of about @xmath216 . and it is also too short to be be observable . similarly , for the electron neutrino @xmath217 , we find @xmath218 , which is again very short and much smaller than the time scale of sn explosion . therefore , unless their effective mass is somehow significantly enhanced , @xmath8 and @xmath9 can be considered as massless for the time delay observation . for the muon and tau neutrinos , according to the oscillation experiments their effective masses in the terrestrial setting will also be bound by @xmath219 because of the much smaller values of @xmath220 and @xmath221 , , @xmath222 , @xmath223 , and @xmath28 , to those of the regular neutrino masses , @xmath224 , @xmath225 , and 3 , by @xmath226 where @xmath227 is the neutrino mixing matrix . ] their time delay will be very small too . in the following , we will tentatively ignore these terrestrial bounds to form the basis of our proposal for a test of neutrino properties under the cosmic setting of extreme conditions . after the sn explosion the emission of different flavors of neutrinos lasts for a period of about @xmath216 . the arrival to the earth of those flavors of neutrinos and antineutrinos whose effective masses are not affected will last about the same period of time . for other flavors which are affected there may be a delay in the arrival time , depending on the values of their effective masses . we look for measurements which are sensitive to both charge and neutral current events , such as the detector types of the daya bay @xcite , lvd @xcite , and sno @xcite experiments , which involve the neutrino - carbon and neutrino - deuterium reactions . the experimental information that these detectors can potentially provide include the event numbers of the electron antineutrino from charged - current reactions and the total event number of all the three flavors of neutrinos from the neutral - current reactions . we shall assume that the electron neutrino mass is not affected so as to avoid any possibility of influencing the existing result of nucleosynthesis . but the effective masses of the muon and tau neutrinos might be affected . then , there might be a time delay @xmath228 between the charged - current events and some of the neutral - current events if the muon and tau flavors of sn neutrinos appear to have sufficiently sizable effective masses . denoting the effective mass of @xmath184 ( @xmath229 , @xmath230 , @xmath134 , or @xmath135 ) by @xmath231 , we have @xmath232 where we have dropped the units for @xmath233 , @xmath234 , @xmath32 , and @xmath172 as indicated in eq . ( [ dt ] ) . @xmath235 is the average time of sn neutrinos of the neutral - current reactions to arrive at the detector and @xmath236 that of the charged - current reactions . a nonvanishing @xmath233 signalizes a significant larger effective mass for muon or tau neutrino , independent of parameters of the sn and neutrinos . to gain a sense of how @xmath233 may look like and how sensitive it is to the effective mass of the muon and tau neutrinos , we have to do some model calculation for quantities which can be measured in a sn neutrino experiment . we found that such a quantity is available as described below . we use the charged - current events which are from the electron antineutrino to monitor the events that are not delayed , and the neutral - current events which involve all flavors to measure the possible time delay . we divide the total neutral - current events into two groups : @xmath237 and @xmath238 , where @xmath237 is the total neutral - current events in accompanying with the charged - current events and are therefore not delayed , and @xmath238 are all the total neutral - current events including both delayed and not delayed . we define the event number of the delayed @xmath184 plus @xmath239 from the neutral - current events as @xmath240 the quantity to be measured is the new ratio , @xmath241 this ratio is very insensitive to the values of @xmath220 and @xmath221 . is insensitive to variations of @xmath220 and @xmath221 in a range of 10 and 8 orders of magnitude respectively . ] it should also be remarked that because of the ratio , the effect of the various uncertainties involved , such as the neutrino and sn parameters , will be reduced in @xmath242 . using the charged current events to monitor the undelayed neutral current events will bypass the question of the uncertainty in the total emission time of sn neutrinos . since we can not tag the flavor of neutral - current events the information on the masses of the muon and tau neutrinos are correlated . for the purpose of illustration we assume @xmath229 and @xmath134 to have the same effective mass . we keep all other features of neutrinos and the sn unchanged . for the neutrino - carbon reactions at daya bay , we plot in fig . 10 two sets of curves which describe @xmath242 as a function of @xmath231 , each for a fixed value of @xmath0 . in fig . 10(a ) for @xmath243 , it can be seen that the curves are insensitive to @xmath195 and the mass hierarchy . for @xmath244 , @xmath242 is sensitive to @xmath231 . a measurement of @xmath242 can restrict @xmath234 in a distinctive range of values determined by the uncertainties in the sn neutrino temperature and chemical potential . in fig . 10(b ) for @xmath245 , @xmath242 is again shown to be sensitive to @xmath234 for @xmath244 . the two mass hierarchies overlap for small @xmath234 but fall into well separated regions for @xmath246 . in figs . 11 we plot @xmath242 for the neutrino - deuterium reactions in analogy to figs . 11 of the neutrino - carbon reactions . figure 11(a ) is very similar to fig . figure 11(b ) is also similar to fig . 10(b ) for @xmath231 smaller than @xmath247 , but there are some differences at larger @xmath234 . in this latter case , unlike fig . 10(b ) , fig . 11(b ) shows that the two mass hierarchies can no long be separated even for @xmath231 greater than @xmath248 . the method can also be applied to kamlamd , minboone , borexino , and double - chooz in the channel of neutrino - carbon reactions , and the results are similar to those given for the daya bay experiment . in this paper , we considered the sn shock effects , the msw effects , the collective effects , and the earth matter effects in the detection of type ii sn neutrinos . also , we considered the uncertainties in the neutrino luminosities @xmath3 , their temperatures @xmath4 , and their pinching parameters @xmath5 in the calculation of different flavors of sn neutrinos . we found that quantities suitably defined in terms of event numbers of different flavors of sn neutrinos are sensitive to the neutrino mass hierarchy , the mixing angle @xmath0 , and neutrino masses . therefore , it is possible to identify the mass hierarchy , acquire information about @xmath0 , and look into certain neutrino mass patterns of sn neutrinos . firstly , we defined the ratio @xmath189 as the event number of @xmath8(@xmath9 ) for the charged - current reaction detected after one second over that detected in the first one second , and discussed the dependence of @xmath189 on @xmath0 for different mass hierarchies . with this , we may identify the mass hierarchy and obtain information about @xmath0 . this method can be applied to several detector types relying on specific physical processes , including the inverse beta decay ( fig . 7 ) , neutrino - carbon reactions ( figs . 8(a ) and 8(b ) ) , and neutrino - deuterium reactions ( figs . 9(a ) and 9(b ) ) . from our numerical calculations , the mass hierarchy may be identified by measuring @xmath192 in the neutrino - carbon reactions and the neutrino - deuterium reactions , @xmath199 in the inverse beta decay , the neutrino - carbon reaction , and the neutrino - deuterium reactions . in addition , in the case of normal mass hierarchy , @xmath192 measured in the neutrino - carbon reactions provides the most restrictive information about @xmath0 . therefore , we can both identify the mass hierarchy and measure @xmath0 from @xmath192 in the channel of neutrino - carbon reactions at several experiments such as daya bay , kamland , minboone , borexino , and double - chooz . secondly , we defined another ratio , @xmath242 , for the neutral - current reactions , as the event number of the delayed neutrinos over the total event number , with the charge current events as a monitor of undelayed events . the significance of @xmath242 is that it offers a possibility to detect possible difference between the electron neutrino and its other flavors counterparts , in particular , any significant difference in their effective masses under the cosmic setting of extreme conditions . with the plot of @xmath242 vs @xmath231 , we can study @xmath231 vs @xmath249 . this method is applied to both neutrino - carbon and neutrino - deuterium reactions and the result is given respectively in figs . 10 and 11 . what we have presented for @xmath242 is a case study with rather restricted input . in case there is an indication of non - vanishing @xmath242 a detailed refined calculation should be made . in conclusion , if a sn explosion takes place within the cosmic distance considered here , it is possible to identify the mass hierarchy and obtain information on @xmath0 . with suitable set up of the detector , a study of some unusual properties of neutrinos under the cosmic setting is possible . a summary of the detector types and their capabilities in association with the studies presented in this article is given in table ii . we would like to thank f .- g . cao and c .- g . yang for helpful discussions . this work was supported in part by national natural science foundation of china ( project numbers 10535050 , 10675022 , 10975018 , and 10890092 ) and the special grants from beijing normal university . k. hirata _ et al . _ , lett . * 58 * , 1490 ( 1987 ) ; r.m . et al . _ , lett . * 58 * , 1494 ( 1987 ) ; v. trimble , rev . phys . * 60 * , 859 ( 1988 ) ; w.d . arnett , j.n . bahcall , r.p . kirshner , and s.e . woosley , ann . . astrophys . * 27 * , 629 ( 1989 ) ; w. hillebrandt and p. h@xmath250flich , rep . . phys . * 52 * , 1421 ( 1989 ) ; d.n . schramm , phys . rep . * 189 * , 89 ( 1990 ) ; m. koshiba , phys . * 220 * , 229 ( 1992 ) ; r. mccray , ann . . astrophys . * 31 * , 175 ( 1993 ) . [ daya bay collaboration ] x .- h . guo _ _ , _ a precision measurement of the neutrino mixing angle @xmath0 using reactor antineutrinos at daya bay _ , arxiv : hep - ex/0701029 . p. antonioli _ et al . _ , new j. phys . * 6 * , 114 ( 2004 ) [ arxiv : astro - ph/0406214 ] . guo , m .- y . huang , and b .- young , phys . d * 79 * , 113007 ( 2009 ) [ arxiv:0806.2720 ] . b. dasgupta and a. dighe , phys . d * 77 * , 113002 ( 2008 ) [ arxiv:0712.3798 ] ; s. chakraborty , s. choubey , b. dasgupta , and k. kar , j. cosmol . . phys . * 09 * , 013 ( 2008 ) [ arxiv:0805.3131 ] . kneller , g.c . mclaughlin , and j. brockman , phys . d * 77 * , 045023 ( 2008 ) [ arxiv:0705.3835 ] . duan , g.m . fuller , and j. carlson , comp . @xmath251 disc . * 1 * , 015007 ( 2008 ) [ arxiv:0803.3650 ] . wilson , _ numerical astrophysics _ , edited by j.m . centrella , j.m . leblanc and r.l . bowers , ( boston : jones and bartlett 1985 ) pp 422 - 434 ; j.r . wilson , r.w . mayle , s.e . woosely , and a.t . weaver , n. y. acad . 470 * , 267 ( 1986 ) . schirato and g.m . fuller , _ connection between supernova shock , flavor transformation , and the neutrino signal _ , arxiv : astro - ph/0205390 . fogli , e. lisi , a. mirizzi , and d. montanino , phys . d * 68 * , 033005 ( 2003 ) [ arxiv : hep - ph/0304056 ] . keil , g.g . raffelt , and h .- janka , astrophys . j. * 590 * , 971 ( 2003 ) [ arxiv : astro - ph/0208035 ] ; g.g . raffelt , m.t . keil , r. buras , h .- t . janka , and m. rampp , _ supernova neutrinos : flavor - dependent fluxes and spectra _ , arxiv : astro - ph/0303226 . j. boger _ et al_. , nucl . methods phys . res . , sect . a * 449 * , 172 ( 2000 ) . c. giunti and c.w . kim , _ fundamentals of neutrino physics and astrophysics _ ( oxford , new york 2007 ) ; s.l . shapiro and s.a . teukolsky , _ black holes , white dwarfs , and neutron stars _ ( wiley , new york 1983 ) ; r.n . mohapatra and p.b . pal , _ massive neutrino in physics and astrophysics _ ( world scientific , 2004 ) , 3nd ed .. d.n . spergel , t. piran , a. loeb , j. goodman , and j.n . bahcall , scince * 237 * , 1471 ( 1987 ) . loredo and d.q . lamb , ann . sci . * 571 * , 601 ( 1989 ) , phys . d * 65 * 063002 ( 2002 ) . y. totsuka , rept . phys . * 55 * , 377 ( 1992 ) . wilson and r.w . mayle , phys . rep . * 163 * , 63 ( 1988 ) , phys . rep . * 227 * , 97 ( 1993 ) . t. totani , k. sato , h.e . dalhed , and j.r . wilson , astrophys . j. * 496 * , 216 ( 1998 ) [ arxiv : astro - ph/9710203 ] . janka and w. hillebrandt , astron . astrophys . * 224 * , 49 ( 1989 ) ; astron . ser . * 78 * , 375 ( 1989 ) ; h .- t . janka , astron . astrophys . * 244 * , 378 ( 1991 ) . fogli , e. lisi , a. mirizzi , and d. montanino , j. cosmol . . phys . * 04 * , 002 ( 2005 ) [ arxiv : hep - ph/0412046 ] ; j. cosmol . phys . * 06 * , 012 ( 2006 ) [ arxiv : hep - ph/0603033 ] . c. lunardini and a.y . smirnov , j. cosmol . phys . * 06 * , 009 ( 2003 ) [ arxiv : hep - ph/0302033 ] . k. takahashi , k. stao , h.e . dalhed , and j.r . wilson , astropart . 20 * , 189 ( 2003 ) [ arxiv : astro - ph/0212195 ] ; r. tom@xmath252s , m. kachelrie@xmath253 , g.g . raffelt , a. dighe , h .- janka , and l. scheck , j. cosmol . . phys . * 09 * , 015 ( 2004 ) [ arxiv : astro - ph/0407132 ] . l. wolfenstein , phys . d * 17 * , 2369 ( 1978 ) , phys . d * 20 * , 2634 ( 1979 ) ; s. p. mikheyev and a. y. smirnov , sov . * 42 * , 913 ( 1985 ) , nuovo cimento c * 9 * , 17 ( 1986 ) , zh . . fiz . * 91 * , 7 ( 1986 ) [ arxiv : hep - ph/0706 . 0454 ] , sov . phys . * 30 * , 759 ( 1987 ) , prog . * 23 * , 41 ( 1989 ) . t.k . kuo and j. pantaleone , phys . * 61 * , 937 ( 1989 ) , phys . d * 39 * , 1930 ( 1989 ) . a.yu . smirnov , phys . t * 121 * , 57 ( 2005 ) [ arxiv : hep - ph/0412391 ] , _ recent developments in neutrino phenomenology _ , arxiv : hep - ph/0702061 . duan and j.p . kneller , j. phys . g : nucl . part . phys . * 36 * , 113201 ( 2009 ) [ arxiv:0904.0974 ] . duan , g.m . fuller , and y .- z . qian , phys . d * 74 * , 123004 ( 2006 ) [ arxiv : astro - ph/0511275 ] , phys . d * 76 * , 085013 ( 2007 ) [ arxiv:0706.4293 ] ; h .- y . duan , g.m . fuller , j. carlson , and y .- z . qian , phys . d * 75 * , 125005 ( 2007 ) [ arxiv : astro - ph/0703776 ] , phys . lett . * 99 * , 241802 ( 2007 ) [ arxiv:0707.0290 ] . duan , g.m . fuller , and y .- z . qian , j. phys . g : nucl . part . phys . * 36 * , 105003 ( 2009 ) [ arxiv:0808.2046 ] . duan , g.m . fuller , and y .- z . qian , _ collective neutrino oscillations _ , arxiv:1001.2799 . s. hannestad , g.g . raffelt , g. sigl , and y.y.y . wong , phys . d * 74 * , 105010 ( 2006 ) [ arxiv : astro - ph/0608695 ] , phys . d * 76 * , 029901 ( 2007 ) . raffelt and a.y . smirnov , phys . d * 76 * , 081301(r ) ( 2007 ) [ arxiv:0705.1830 ] , phys . d * 76 * , 125008 ( 2007 ) [ arxiv:0709.4641 ] . fogli , e. lisi , a. mirizzi , and d. montanino , j. cosmol . * 12 * , 010 ( 2007 ) [ arxiv:0707.1998 ] ; g.l . fogli , e. lisi , a. mirizzi , and i. tamborra , j. cosmol . phys . * 04 * , 030 ( 2009 ) [ arxiv:0812.3031 ] , j. cosmol . * 10 * , 002 ( 2009 ) [ arxiv:0907.5115 ] . a.s . dighe and a.y . smirnov , phys . d * 62 * , 033007 ( 2000 ) [ arxiv : hep - ph/9907423 ] . guo and b .- young , phys . d * 73 * , 093003 ( 2006 ) [ arxiv : hep - ph/0605122 ] . a.n . ioannisian and a.yu . smirnov , phys . lett . * 93 * , 241801 ( 2004 ) [ arxiv : hep - ph/0404060 ] ; a.n . ioannisian , n.a . kazarian , a.yu . smirnov , and d. wyler , phys . d * 71 * , 033006 ( 2005 ) [ arxiv : hep - ph/0407138 ] . c. lunardini and a.yu . smirnov , nucl . b * 616 * , 307 ( 2001 ) [ arxiv : hep - ph/0106149 ] , phys . d * 63 * , 073009 ( 2001 ) [ arxiv : hep - ph/0009356 ] ; e. k. akhmedov , c. lunardini , and a.yu . smirnov , nucl . b * 643 * , 339 ( 2002 ) [ arxiv : hep - ph/0204091 ] . a.m. dziewonski and d.l . anderson , phys . earth planet . inter . * 25 * , 297 ( 1981 ) ; f.d . stacey , _ physics of the earth _ ( wiley , new york , 1977 ) , 2nd ed .. l. cadonati , f.p . calaprice , and m.c . chen , astropart . * 16 * , 361 ( 2002 ) [ arxiv : hep - ph/0012082 ] . a. burrows , s. reddy , and t.a . thompson , nucl . a * 777 * , 356 ( 2006 ) [ arxiv : astro - ph/0404432 ] . beacom and p. vogel , phys . d * 58 * , 093012 ( 1998 ) [ arxiv : hep - ph/9806311 ] ; j.f . beacom and l.e . strigari , phys . c * 73 * , 035807 ( 2006 ) [ arxiv : hep - ph/0508202 ] . n. tatara , y. kohyama , and k. kubodera , phys . c * 42 * , 1694 ( 1990 ) ; s. ying , w.c . haxton , and e.m . henley , phys . d * 40 * , 3211 ( 1989 ) , phys . c * 45 * , 1982 ( 1992 ) . _ et al_. , ( kamland collaboration ) , proposal for us participation in kamland , march 1999 ; j. busenitz , int . j. mod . a * 16 * , 742 ( 2001 ) ; k. eguchi _ et al_. , phys . * 90 * , 021802 ( 2003 ) [ arxiv : hep - ex/0212021 ] ; t. araki _ et al_. , nature * 436 * , 499 ( 2005 ) . et al_. , phys . rev . lett . * 89 * , 011301 ( 2002 ) [ arxiv : nucl - ex/0204008 ] ; s.n . ahmed _ et al_. , phys . lett . * 92 * , 181301 ( 2004 ) [ arxiv : nucl - ex/0309004 ] ; b. aharmim _ et al_. , phys . c * 75 * , 045502 ( 2007 ) [ arxiv : nucl - ex/0610020 ] . 4 the crossing probability at the high resonance region @xmath41 as a function of the neutrino mixing angle @xmath0 : ( a ) for three different times at @xmath255 . the solid , dashed , and dotted curves are for @xmath256 , respectively ; ( b ) for three different neutrino energies at @xmath121 . the solid , dashed , and dotted curves are for @xmath257 , respectively . fig . 5 illustration of the path of the sn neutrino reaching the detector in the earth . @xmath172 is the location of the detector , @xmath142 is the incident angle of the neutrino , @xmath258 is the center of the earth , @xmath143 is the distance the neutrino travels through the earth matter , and @xmath144 is the instantaneous distance of the neutrino to the center of the earth . 6 the luminosity decay time @xmath28 as a function of the mixing angle @xmath0 in the channel of neutrino - carbon reactions at daya bay when @xmath193 and @xmath194 . the `` max '' ( `` min '' ) corresponds to the maximum ( minimum ) values of @xmath4 and @xmath5 . the ratio @xmath199 of the event number of @xmath9 for the charged - current reaction detected after 1s over that detected before 1s , as a function of the mixing angle @xmath0 in the channel of inverse beta decay at daya bay . the solid curves correspond to @xmath259 , the dashed curves @xmath194 , and the dot - dashed curves @xmath260 . the `` max '' ( `` min '' ) corresponds to the maximum ( minimum ) values of @xmath4 and @xmath5 . ( a ) similar to fig . 7 but the electron neutrinos @xmath8 are observed in the channel of neutrino - carbon reactions at daya bay ; ( b ) similar to fig . 7 but the anti - electron neutrinos @xmath9 are observed in the channel of neutrino - carbon reactions at daya bay . fig . 9 ( a ) similar to fig . 7 but the electron neutrinos @xmath8 are observed in the channel of neutrino - deuterium reactions at sno ; ( b ) similar to fig . 7 but the anti - electron neutrinos @xmath9 are observed in the channel of neutrino - deuterium reactions at sno . fig . 10 for the neutral - current reactions @xmath242 in the channel of neutrino - carbon reactions at daya bay , the ratio of the event number of delayed neutrinos over the total event number , as a function of @xmath231 at two values of @xmath0 , each has the two mass hierarchies with @xmath195 , @xmath4 , and @xmath5 taking their limiting values . the solid curves correspond to @xmath259 , the dashed curves @xmath194 , and the dot - dashed curves @xmath260 . the `` max '' ( `` min '' ) corresponds to the maximum ( minimum ) values of @xmath4 and @xmath5 . figure ( a ) shows that @xmath242 is nearly independent of @xmath195 and the mass hierarchy .
assuming a tricritical point of the two flavor qcd in the space of temperature , baryon number chemical potential and quark mass , we study the change of the associated soft mode along the critical line within the ginzburg landau approach and the nambu jona - lasinio model . the ordering density along the chiral critical line is the scalar density whereas a linear combination of the scalar , baryon number and energy densities becomes the proper ordering density along the critical line with finite quark masses . it is shown that the critical eigenmode shifts from the sigma like fluctuation of the scalar density to a hydrodynamic mode at the tricritical point , where we have two ordering densities , the scalar density and a linear combination of the baryon number and energy densities . we argue that appearance of the critical eigenmode with hydrodynamic character is a logical consequence of divergent susceptibilities of the conserved densities .
hep-ph0402263
at high temperature and/or baryon density , the system governed by qcd will show a transition from an ordinary hadronic phase to a chirally symmetric , deconfined plasma phase@xcite . the main objective of the heavy ion programs at rhic and at future lhc is to create this long - sought plasma state and to study collective properties of this many body assembly@xcite . these two phases would have to be separated by a boundary with singularity if chiral symmetry and/or confinement of qcd were exact symmetry . in reality , dynamical quarks with finite masses @xmath0 make both symmetries only approximate , and their order parameters , the quark condensate and the polyakov loop , have non vanishing values everywhere in the phase diagram . thus the plasma state may be smoothly connected with the ordinary hadronic state , even though they would possess qualitatively different properties from each other . recently a strong possibility of a critical point in the real qcd phase diagram was suggested@xcite , based on model calculations@xcite as well as lattice qcd results@xcite . it is the endpoint of the first order line , inferred from the crossover behavior along the temperature ( @xmath1 ) axis and the first order transition along the axis of the baryon number chemical potential ( @xmath2 ) , and is a genuine singular point with the same criticality as the z@xmath3 ising model . its location , which is sensitive to the strange quark mass @xmath4 , is expected to be within the reach of current experimental facilities . observable implications of this z@xmath3 critical point ( z@xmath3cp)cp instead of cep , to indicate the symmetry of the point . ] in heavy ion experiments have been discussed in the literature@xcite such as large fluctuations of the low momentum particle distributions , and the limitations on them due to the finite space time geometry of collision events . this z@xmath3cp will become a critical cornerstone in the qcd phase diagram once its location is confirmed in experiments . based on the approximate chiral symmetry , the scalar density is usually taken as the order parameter of the ginzburg landau ( gl ) effective potential to describe the critical behavior at the z@xmath3cp . in this description all the singularities associated with the z@xmath3cp seemingly originate from softening of the scalar density fluctuations as the effective potential becomes flat there . especially , it might be concluded that the sigma meson becomes massless as an immediate consequence of this critical point . as a basic fact , however , we should strictly distinguish between the chiral critical point with @xmath5 and the z@xmath3cp with @xmath6 even within the chiral effective models . when the chiral symmetry is exact , the @xmath1@xmath2 plane is divided into two domains of the symmetric and broken phases with a boundary _ line_. but the symmetry argument is unable to fix the order of the singularity of this line , especially the possible existence of the tri critical point ( tcp ) on this line . since the z@xmath3cp at finite quark mass is the remnant of this tcp , the relation of the z@xmath3cp to the chiral symmetry is very obscure . in fact , z@xmath3 is not the symmetry of the underlying interactions , but of the thermodynamic potential at this particular isolated _ point _ in the @xmath1@xmath2 phase diagram@xcite . from this point of view the z@xmath3cp is different from the chiral critical point . the flat effective potential indicates the large fluctuation of the scalar density . at the chiral critical point , this is related to softening of the sigma meson mode , so as to form a symmetric excitation spectrum together with the pions . on the other hand , there is no such symmetry constraint at the z@xmath3cp . actually the scalar fluctuation _ linearly _ mixes with fluctuations of baryon number density and energy density , and therefore not only the sigma mode but also the hydrodynamic mode are to be taken into account there to study the associated soft mode . consequently the fate of the sigma meson mode at the z@xmath3cp is non trivial . recent calculations of the dynamic mode in the scalar channel using the chiral models@xcite indeed showed that the sigma meson is massive at the z@xmath3cp . furthermore , another scalar mode with space like momentum dispersion is identified as the soft mode associated with the z@xmath3cp in the nambu jona - lasinio ( njl ) model@xcite . in this paper we shall confirm the result of ref . @xcite on the more general ground using the time dependent ginzburg landau ( tdgl ) approach , and extend the study to discuss the changeover of the soft modes along the critical line in the @xmath1@xmath2@xmath7 space within the tdgl approach as well as the njl model . our investigation is based on two fundamental observations about the z@xmath3cp . the first point is that the proper ordering density at the z@xmath3cp is a linear combination of the scalar , baryon number and internal energy densities@xcite , as mentioned above . because of this mixing all the susceptibilities of these densities diverge with the same critical exponent at the z@xmath3cp . in contrast , in the chiral critical transition , the susceptibility of the scalar density diverges with exponent @xmath8 of the o(4 ) model in the two flavor case , while the other susceptibilities of the baryon number and the energy have the smaller exponent @xmath9 . the second is a consequence on the dynamics following from the conservation of the baryon number and the energy . the fluctuations of these conserved densities are intrinsically soft and constitute the hydrodynamic modes , whose excitation energies vanish as the wavevector @xmath10 goes to zero . susceptibilities of these conserved densities in turn have the spectral contributions solely from these hydrodynamic modes when expressed as a sum of mode spectra@xcite . hence the divergence of the susceptibility of a conserved density must be accompanied by critical slowing of a hydrodynamic mode . the spectral contribution from this hydrodynamic mode may well be involved in the scalar susceptibility through the mixing at the z@xmath3cp . at an o(4 ) critical point ( o(4)cp ) the importance of the hydrodynamic mode depends on which phase we start from . the hydrodynamic mode plays no critical role in the symmetric phase whereas the scalar condensate makes the mixing possible in the broken phase . the situation becomes more subtle at the tcp , where the o(4 ) critical line shifts to the first order line . only the scalar susceptibility diverges due to the softening of the sigma meson at the tcp if it is approached from the symmetric phase . otherwise , the hydrodynamic soft mode causes the divergence in the susceptibilities of the baryon number and energy as well as the scalar one . this paper is organized as follows . in the next section we briefly review generic properties of the phase diagram of qcd with two flavors near the tcp using the gl effective potential . it is stressed that at the tcp there are two relevant order parameters , the scalar condensate and a conserved density which is a linear combination of the baryon number and entropy densities . then we include the dynamics using the tdgl model . writing the susceptibilities as a spectral sum , we discuss the relative weight of the spectral contributions from the sigma and hydrodynamic modes . it is pointed out that the hydrodynamic contribution generates the discontinuity of the baryon number and entropy susceptibilities at the o(4)cp , and that this hydrodynamic mode gives the divergence at the tcp approached from the broken phase and also at the z@xmath3cp . in iii we perform the same analysis using the njl model as an illustration . the gl effective potential with two ordering densities are numerically constructed there . the flat directions at the critical points are shown and discussed in relation to the divergences of the susceptibilities . the spectral origins of these divergences are studied with the relative weight of the mode spectra , and in detail based on the poles and residues of the scalar response function . sections iv and v are devoted to discussions and summary . in appendix a we prove the relation between the susceptibility and the response function , and in appendix b we present the explicit formulas of the response functions in the njl model . the results with the chiral quark model is briefly reported in appendix c. let us briefly review the phase structure near the tcp@xcite . it is known that the critical properties near the tcp are described , up to logarithmic corrections , with the ginzburg landau effective potential @xmath11 where @xmath12 ( @xmath13 ) denotes the ( non-)singular part of the effective potential , and @xmath14 . the pseudo scalar density is set to zero and neglected here in the mean field approximation . the critical exponents can be easily found from ( [ eq : gleffpot ] ) at the mean field level . along the line of the first order transition within the symmetry plane ( @xmath15 ) , we have @xmath16 where three minima with @xmath17 coexist ( dashed line in fig . 1 ) . the baryon number and entropy densities are functions of @xmath18 due to symmetry , and discontinuous across the boundary between the symmetric phase ( @xmath19 ) and the broken phase ( @xmath20 ) . at the tcp , where @xmath21 , these _ three _ phases coalesce and the first order line ( @xmath22 ) smoothly joins with the o(4 ) critical line ( @xmath23 ) . once a small explicit breaking field @xmath24 is exerted , the o(4 ) critical line disappears and the tcp is lifted to the z@xmath3cp . the line of z@xmath3cp as a function of @xmath25 is determined by the condition , @xmath26 ( @xmath27 ) , which is solved for a negative @xmath28 with @xmath29 , @xmath30 and @xmath31 . two lines of z@xmath3cp with @xmath32 form the edge of the wing - like surface of the first order transition in the @xmath33@xmath28@xmath25 space , and these lines connect smoothly to the o(4 ) critical line at the tcp . this wing structure is mapped into the physical phase space of @xmath1 , @xmath2 and @xmath34 ( see fig . 1 ) . the slope of the first order boundary can be related to the discontinuities of the densities across the boundary via the clapeyron clausius relation@xcite , @xmath35 with baryon number density @xmath36 and entropy density @xmath37 . the chiral broken phases with @xmath38 coexist within the symmetry plane @xmath15 , and accordingly there is no gap in @xmath36 and @xmath37 across this symmetry plane . only the scalar density @xmath39 bifurcates as the ordering density at the o(4)cp approached from the symmetric phase . its correlations with the `` energy like '' densities vanish @xmath40 because of the symmetry in @xmath41 in the symmetric phase . = 0.55 = 0.45 from the relation ( [ eq : ccrel ] ) we know that all of @xmath39 , @xmath36 and @xmath37 generally have discontinuities across the wing because there is no reason for any of these slopes to vanish once @xmath42 . let us discuss the energy like and ordering densities around the z@xmath3cp@xcite . first we introduce the `` temperature like '' field as a vector tangential to the coexistence boundary . then the energy - like density is defined as the thermodynamic variable conjugate to this temperature like field . this density has no discontinuity in the vicinity of the critical point . since the boundary is two dimensional , there are two independent temperature like fields and correspondingly two energy like densities . next the ordering density is defined as the density whose correlations with the energy like densities vanish at the critical point approached from the `` symmetric '' phase along the temperature like direction . the conjugate field of this ordering density is no longer tangential to the coexistence boundary . there is a single ordering density at the z@xmath3cp , which is in general a linear combination of @xmath39 , @xmath36 and @xmath37 . since all the susceptibilities of these densities include the same singular fluctuation , they diverge with the same critical exponent at the z@xmath3cp . the coexistence wing is squeezed to be one dimensional at the tcp , where two lines of the z@xmath3cp and the line of the first order transition with @xmath15 merge and smoothly connect to the single o(4 ) critical line . thus at the tcp we have only one energy like density , which will be a linear combination of @xmath36 and @xmath37 . accordingly there are _ two _ ordering densities from dimensionality . the obvious one is the scalar density @xmath39 related to the chiral symmetry and the other @xmath43 is another linear combination of @xmath36 and @xmath37 representing the z@xmath3 symmetry of the potential at this particular point . it is sometimes useful to construct the effective potential with two ordering densities , @xmath39 and @xmath43 , which become soft at the tcp . the same observation can be made by looking at the susceptibilities directly . there are three fields @xmath25 , @xmath33 and @xmath28 in the effective potential ( [ eq : gleffpot ] ) . the singular parts of the corresponding susceptibilities form a 3-by-3 matrix ( @xmath44 ) , @xmath45 with @xmath46 the scalar susceptibility , @xmath47 where @xmath39 takes the value at the extremum of the potential . when the o(4)cp is approached from the symmetric phase , @xmath48 . in the broken phase the situation is different . the singular part of ( 1 ) gives a finite contribution @xmath49 as the o(4)cp is approached with @xmath50 , although the divergent susceptibility is still the scalar one alone . the singular contribution to @xmath51 eventually blows up at the tcp approached along the o(4 ) critical line . in fact , the tcp may be understood as a usual critical point with the ordering density @xmath43 conjugate to @xmath33 , sitting on the chiral phase boundary . when the tcp is approached from the broken phase with @xmath52 , the scalar ordering density vanishes slowly @xmath53 , and @xmath46 and @xmath51 diverge like @xmath54 and @xmath55 , respectively , while @xmath56 is still non - singular . note that the divergence of @xmath51 at the tcp indicates the infinities in the baryon number and energy susceptibilities , or equivalently the isothermal compressibility and specific heat , respectively . all these susceptibilities in ( [ eq : chimatrix ] ) diverge at the z@xmath3cp , where @xmath57 . with the finite condensate @xmath39 we can diagonalize this matrix of the susceptibilities , leaving only one singular susceptibility . the resulting eigenvalues are ( 0,0,@xmath58 ) , with eigenvectors @xmath59 , @xmath60 , @xmath61 , respectively . for small @xmath39 or @xmath25 , we see that the ordering density is approximately a linear combination of the densities @xmath39 and @xmath43 . the fluctuations of these two ordering densities @xmath39 and @xmath43 become large near the tcp as explained above , and should be included as the soft degrees of freedom , especially when we discuss the dynamic aspects . we generalize the free energy so as to have two ordering densities , @xmath62 coupling between @xmath39 and @xmath43 must respect the underlying chiral symmetry and the simplest coupling is @xmath63 . a flat direction of this potential appears at a critical point in the @xmath39@xmath43 plane . in the case of the o(4)cp / tcp it is in the @xmath39 direction reflecting the symmetry while the direction will become a linear combination of the two densities at the z@xmath3cp . eliminating the density @xmath43 by @xmath64 , we recover the original form ( [ eq : gleffpot ] ) of the free energy with @xmath65 and @xmath66 , up to an analytic term . we may introduce the dynamics to the system described by the free energy ( [ eq : gl2var ] ) phenomenologically@xcite . we have seen in the previous subsection that there are two ordering densities conjugate to the fields @xmath25 and @xmath33 at the tcp , and that a _ linear _ combination of these two densities will become the relevant ordering density at the z@xmath3cp for small @xmath25 or @xmath39 . we should include at least these two densities in order to describe the soft dynamics . furthermore , it is known that ( non linear ) mode mode coupling between the fluctuations of the ordering densities and other ( non - critical ) hydrodynamic modes are important in general to describe the dynamics in the critical region@xcite , which is beyond the scope of this work@xcite . we will see , however , that the coupled system of the two ordering densities in the mean field approximation yields already a non - trivial result@xcite . deviation of the densities from the absolute equilibrium gives rise to time evolution of the system . here we assume simple phenomenological equations of motion for densities @xmath39 and @xmath43 as @xmath67 where @xmath68 and @xmath69 are the differential operators . appropriate forms of @xmath70 and @xmath71 are unknown in this description . but as a strong constraint we know that the operator @xmath71 must be consistent with the conservation of the density @xmath43 and describe the hydrodynamic motion . as a typical hydrodynamic evolution , we consider here the diffusion motion @xmath72 with wavevector @xmath10 . note that the diffusion is time irreversible . we assume propagating motion @xmath73 for the scalar density , identifying this mode as the sigma meson which degenerates with the propagating pions at the o(4)cp / tcp . other possible forms are considered below in this section . the transport coefficients @xmath74 are treated as constants here . for small deviations @xmath75 and @xmath76 from the equilibrium values , we linearize these equations of motion with respect to @xmath77 and @xmath78 to obtain @xmath79 where @xmath80 , etc . the soft eigenmodes of the system are determined by the condition @xmath81 where @xmath82 is the scalar susceptibility given in ( [ eq : chih ] ) , and we introduced a term @xmath83 in @xmath84 . the eigenmodes for small @xmath10 are found as @xmath85 with @xmath86 where @xmath87 is the susceptibility of the density @xmath43 . the eigenmode @xmath88 is oscillating while @xmath89 has the diffusion - like hydrodynamic character . the @xmath88 vanishes at the o(4)cp , being the critical eigenmode . although the hydrodynamic mode @xmath89 is an intrinsic soft mode of the system , it does not show the critical slowing there . when the tcp is approached from the symmetric phase , the situation is the same . on the other hand , at the tcp approached from the broken phase , both frequencies slow down , reflecting the divergence of the susceptibilities , which seems reflecting the existence of two independent ordering densities there . at the z@xmath3cp the susceptibilities @xmath82 and @xmath87 diverge with the same exponent due to the linear mixing . the propagating @xmath88 is a fast mode there due to the non - zero condensate @xmath90 , whereas the hydrodynamic slow mode @xmath91 becomes the critical mode ( @xmath92 ) associated with the z@xmath3cp . this result is similar to the level - crossing phenomenon where the mode coupling makes the lower energy mode lowered further . the explicit @xmath93 factor of @xmath89 stemming from the hydrodynamic character results in the larger dynamic critical exponent @xmath94 in the mean field level , which makes @xmath89 apparently slower than the @xmath88 mode . in contrast , at the o(4)cp , the linear mixing is banned by the underlying chiral symmetry . inverse of the differential operator ( [ eq : normal ] ) with the retarded boundary condition is the response function @xmath95 which characterizes the time dependent response of these densities to the external fields , @xmath25 and @xmath96 , within the linear approximation . the susceptibility is obtained in the limit of @xmath97 . the response function is analytic in the upper complex@xmath98 plane , which fact allows us to express generally the susceptibility as a sum of the mode spectra@xcite : @xmath99 where a ultra violet regularization is understood if necessary . this expression shows that the divergence at a critical point should come from an infrared enhancement of the spectral function because the spectral function itself is usually integrable . using this expression we can examine the relative weight of each mode contribution to the susceptibility . in our case , the oscillating and diffusion modes give spectral contributions as @xmath100 and @xmath101 here the first term in the bracket originates from the poles @xmath102 and the second from @xmath89 . first , we note that only the diffusion like @xmath89 pole contributes to the susceptibility of @xmath43 . this is a robust result following from the conservation of the density @xmath43 . existence of the current @xmath103 such that @xmath104 dictates that the frequencies of the modes contributing to the @xmath43 susceptibility must vanish as @xmath10 goes to zero . we can formally show that the spectrum of the @xmath43 response function behaves as @xmath105 ( see appendix a ) . conversely , we can state that softening of the hydrodynamic mode must accompany the divergence of @xmath87 . second , the ratio of the @xmath89 spectral contribution to the total in the scalar susceptibility , @xmath106 goes to unity at the tcp approached from the broken phase and at the z@xmath3cp , which means that the leading divergence of the scalar susceptibility is also generated by the @xmath89 spectrum at these critical points . even at the o(4)cp approached from the broken phase the @xmath89 spectrum gives a finite portion of the divergence @xmath107 since @xmath108 . this result can be understood by rewriting the scalar response function as @xmath109 where @xmath110 and @xmath111 . the denominator expresses the linear mixing between the `` bare '' @xmath77 and @xmath78 modes through the coupling @xmath112 . even though the coupling becomes smaller as the o(4)cp is approached from the broken phase , softening of the mediating `` bare '' @xmath77 propagator provides @xmath113 factor , which results in the finite mixing of the @xmath114 mode in the scalar channel . this is a simple example indicating the importance of the mode coupling near the critical point . in summary , the @xmath77 and @xmath78 fluctuations mix and form two kinds of eigenmodes , @xmath88 and @xmath89 . we find that along the o(4 ) critical line approached from the broken phase , the critical eigenmode shifts from the sigma meson like @xmath88 to the diffusion like @xmath89 mode at the tcp . in contrast , when we approach the tcp from the symmetric phase , the scalar susceptibility @xmath82 is given completely by the critical @xmath88 spectrum without any mixing of the @xmath78 fluctuation . at the z@xmath3cp the @xmath88 mode becomes a fast mode whereas the whole divergence comes from the critical softening of the @xmath89 spectrum with the hydrodynamic character . in more microscopic njl model calculation in the later section , the mode with the hydrodynamic character is provided as the mode of landau - damping type , contrary to the macroscopic analysis in the previous subsection , and we may change the time evolution operator accordingly as @xmath115 . more generally , the sigma mode may be oscillating or relaxing ( @xmath116 or @xmath117 ) while the possible hydrodynamic fluctuation can be the diffusion , landau - damping type , or sound motion ( @xmath118 , or @xmath119 ) . furthermore , couplings with other hydrodynamic modes , if exist , are to be included to describe the correct dynamic behavior of the system@xcite . we stress here that our result on the critical eigenmodes at the critical points are independent on these ambiguities . the only important fact is that the operator @xmath71 has the hydrodynamic character , and therefore intrinsically soft mode of the system . in any of these choices , we find the eigenmodes for small @xmath93 as @xmath120 note that the typical mode frequencies of the diffusion , landau - damping type and sound - like dispersions vanish @xmath121 as @xmath122 , and satisfy the spectral property following from the conservation law @xmath123 finally we note that these hydrodynamic modes drops out in the @xmath98 limit of the response function : @xmath124 as an definite illustration , we shall study the spectral contributions of the collective modes at critical points in the njl model , and confirm that the result is consistent with the tdgl approach . we remark here that , unlike in the tdgl approach , there are no bare bosonic modes . the bosonic modes are dynamically generated through the interaction between the quarks and their softening causes the divergences at the critical points . we analyze the simplest version of the njl model @xcite @xmath125 $ ] in the mean field approximation ( @xmath126=const , @xmath127=0 ) . the thermodynamics is described by the effective potential@xcite , @xmath128 + { 1\over 4 g}(2g\sigma)^2 , \label{eq : njlfree}\end{aligned}\ ] ] where @xmath129 , @xmath130 , @xmath131 , and @xmath132 with @xmath133 and @xmath134 the numbers of flavor and color , respectively . here @xmath135 is the quark chemical potential . the true thermodynamic state is determined by the extremum condition , @xmath136 , and the corresponding grand potential is @xmath137 . we define the model with the three momentum cutoff @xmath138 and with the coupling constant @xmath139 which allows the tcp . in the following , all the dimensionful quantities are expressed in the units of @xmath138 . expansion of the effective potential around @xmath19 with @xmath5 gives rise to @xmath140 + \frac{1}{2}(\frac{1}{2g}-j^0)(2g\sigma)^2 + \frac{1}{2\cdot 4 } i^0(2g\sigma)^4 + \cdots \nonumber \\ & = & \omega(t,\mu,0;0)/v + a(t,\mu)(2g\sigma)^2 + b(t,\mu)(2g\sigma)^4 + \cdots , \label{eq : njl0effpot}\end{aligned}\ ] ] where the superscript 0 indicates the quantity evaluated in the massless limit . the first term is the non - singular part of the free energy in the gl description . the integrals @xmath141 and @xmath142 are given in appendix b. the tcp determined by @xmath21 appears at @xmath143 and @xmath144 . = 0.7 as explained in ii , it is useful to introduce the effective potential with another relevant ordering density besides the @xmath39 in studying the behavior of the quark number susceptibility and specific heat near the tcp and the z@xmath3cp . from the physical grand potential @xmath137 , we can construct the landau effective potential with two ordering densities @xmath145 and @xmath39 in the following way : first we introduce the free energy @xmath146 via @xmath147 where @xmath148 and @xmath149 are defined by inverting the functions @xmath150 then introducing new parameters @xmath135 and @xmath7 , we _ define _ the landau - type effective potential as @xmath151 the extremum condition for the densities @xmath145 and @xmath39 yields @xmath152 and @xmath153 , recovering the physical grand potential @xmath137 . use of the entropy density @xmath37 instead of the quark number @xmath145 is straightforward . it is known that the effective potential constructed in this way must be convex and can not be defined in the mixed phase . fortunately in the njl model we can bypass this difficulty by supplementing the unphysical grand potential @xmath137 defined with the unstable solutions of the gap equation , which corresponds to the non convex part of the potential @xmath154 . one can easily show that the susceptibilities @xmath155 ( @xmath156 ) are equal to the inverse of the curvature matrix of the gl effective potential at the extremum point . from this fact the divergence of the susceptibilities at a critical point is related to the appearance of a particular flat direction in the gl effective potential . in fig . [ fig : glpotential ] we show the effective potential with two ordering densities at three critical points , ( a ) o(4)cp with @xmath157 , ( b ) tcp , and ( c ) z@xmath3cp with @xmath158 in the units of @xmath138 . the critical instability at these points is usually discussed using the effective potential ( [ eq : njlfree ] ) with a single order parameter @xmath39 based on chiral symmetry , which is shown in the upper panels . the flat curvature of this potential means the divergence of the scalar susceptibility . at the o(4)cp it is clear from the potential with two ordering densities ( @xmath159 ) , or ( @xmath160 ) , that the @xmath39 axis is indeed the symmetry direction of the system . the densities @xmath145 and @xmath37 depend on @xmath18 when calculated from ( [ eq : njlfree ] ) with @xmath5 . this fact is seen here as a quadratic bending of the potential valley . thus the fluctuation of these densities are weaker than that of @xmath39 , and the susceptibilities of the quark number and the entropy have the smaller exponent @xmath9 . = = at the z@xmath3cp , on the other hand , the flat direction of the gl potential is not parallel to the @xmath39 axis in the @xmath39@xmath145 and @xmath39@xmath37 planes . the proper flat direction is a _ linear _ combination of the three densities of @xmath39 , @xmath145 and @xmath37 , and all the susceptibilities of them diverge with the same exponent at the z@xmath3cp . it will be very instructive to introduce the gl function with single ordering density by eliminating @xmath39 by @xmath161 in favor of @xmath145 , as shown in fig . [ fig : rhopot ] . the curvature at the extremum coincides with the inverse of the quark number susceptibility . in case of the o(4)cp the curvature does not vanish , implying the finite susceptibility @xmath162 . it takes different values depending on from which side we approach the equilibrium value of @xmath145 . since the @xmath18 and @xmath163 terms of the potential ( [ eq : njl0effpot ] ) disappear at the tcp and @xmath145 changes with @xmath18 along the potential valley ( see fig . [ fig : glpotential ] ( b ) ) , the @xmath145 potential becomes flat on the side corresponding to the broken phase as seen in fig . [ fig : rhopot ] ( b ) . this indicates the critical point for @xmath145 . on the higher density side , in contrast , the curvature is non vanishing . at the z@xmath3cp , the potential fig . [ fig : rhopot ] ( c ) is essentially the same as the potential ( [ eq : njlfree ] ) , and we may equally well choose @xmath145 or @xmath37 as the ordering density instead of @xmath39 to describe this criticality . we show in fig . [ fig : chimual ] the quark number susceptibility @xmath162 as a function of @xmath135 along the o(4 ) critical line , across which @xmath162 is discontinuous . the value of @xmath162 on the o(4 ) critical line approached from the broken phase grows up toward the tcp and eventually diverges there as is described with the gl potential ( @xmath164)@xcite . the @xmath165 also behaves in the same way . the only qualitative difference is that at the point @xmath166 the @xmath162 is continuous across the phase boundary because no linear coupling with @xmath39 is allowed due to the symmetry under @xmath167 . the spectral origin of the critical divergence can be investigated by studying the spectral function which is obtained as the imaginary part of the response function . we discuss here the structure of the collective eigenmode , which couples with the relevant susceptibilities and shows softening at the critical point . the response functions in the njl model are calculated as@xcite @xmath168 here the polarizations are defined with the imaginary time quark propagator @xmath169 as @xmath170 where @xmath171 @xmath172 , @xmath173 , @xmath174 is an appropriate dirac matrix , and the trace is taken over the flavor , color and dirac indices . @xmath175 for the scalar , @xmath176 for the baryon number , and @xmath177 for @xmath178 with @xmath179 we calculate here the response function with @xmath177 the energy operator in the mean field approximation instead of the entropy , because the entropy has no microscopic expression as it is defined only in equilibrium . the real time response function is obtained from the imaginary time propagator through the usual replacement @xmath180 in the final expression . the static response function in the long wavelength limit reduces to the corresponding susceptibility , @xmath181 . especially , @xmath182 . the one loop polarizations @xmath183 are the response functions of the free quark gas with mass @xmath184 , and contain no contributions from the collective mode . in our simple njl model , the collective mode is generated by the bubble sum encoded in the denominator of eq . ( 24 ) and there are two kinds of collective motion@xcite : the sigma meson mode and the particle hole ( p h ) mode . in ref . @xcite it is argued that the soft mode associated with the z@xmath3cp is not the sigma meson but the p h motion . = 0.45 = 0.45 = 0.45 = 0.45 = 0.47 = 0.47 the spectral function @xmath185 of the scalar response function@xcite yields @xmath186 ^ 2 + [ 2g{\rm i m } \pi_{mm}(\omega , { \bf q})]^2 } . \label{eq : spectrum}\end{aligned}\ ] ] we notice that the spectrum @xmath187 of the free quark gas is enhanced by the bubble type correlation in the denominator . the scalar spectral functions are shown at @xmath188=0.350 and 0.339 near the o(4)cp with @xmath189=(0.3419 , 0.3 ) in fig . [ fig : spfuncchiral ] . one should keep in mind that the @xmath190 spectrum of the sigma meson at @xmath191 in the broken phase is hard to be seen in this figure . the sigma meson spectrum is softening just above and below the o(4)cp ( see also [ subsec : d ] ) . besides the sigma spectrum we clearly find the p h mode spectrum in the space like momentum region , whose strength looks stronger in the broken phase . as we approach the tcp , as is shown in fig . [ fig : spfunctcp ] , this p h mode spectrum grows in the small @xmath10 region in the broken phase , while in the symmetric phase it does not show such a enhancement . we show in fig . [ fig : spfunc ] the spectral functions of the scalar channel as well as the vector channel ( quark number response ) at the z@xmath3cp with @xmath192 . in the scalar channel clearly seen are the two spectral peaks of the sigma meson and the p h motion in medium , respectively@xcite . this spectral structure is to be compared with that of the free quark case given in appendix b. the most significant feature in @xmath193 is the critical enhancement in the @xmath194 region provided by the p h mode , which gives rise to the divergence of the scalar susceptibility . although the sigma meson shows the clear spectral peak in this model , the mode is massive due to the explicit symmetry breaking by the current quark mass . the spectral function of the quark number response , @xmath195 , also contains these two spectral contributions , but the sigma spectrum strongly diminishes as @xmath196 . it is worthwhile to note the fact that @xmath197 and @xmath198 as @xmath196 , which reflects the conservation of the quark number . thus the response function @xmath199 obtained in ( [ eq : response ] ) in the random phase approximation ( rpa ) , too , shares the same property , and the sigma meson can not couple to the quark number susceptibility @xmath162 at @xmath191 . therefore the divergence of @xmath162 at the z@xmath3cp must comes solely from the softening of the p h spectrum . the same is true for @xmath165 . = 0.45 in the previous subsection we have identified the soft mode associated with z@xmath3cp as the p h motion @xcite generated in the scalar channel whereas at the o(4)cp the sigma meson mode becomes soft . let us examine how the changeover of the soft mode from the sigma meson to the p h one occurs along the critical line . once we noticed that the difference between the two limits , @xmath200 and @xmath201 , is caused by the hydrodynamic mode spectrum @xmath202 as @xmath196 , it is easy to calculate the ratio of the spectral strength of the two types of spectral contributions . in these limits the explicit form of the rpa scalar response functions in the broken phase ( with @xmath5 ) yields , respectively , @xmath203 where function @xmath204 is given in appendix b. then we can define the ratio @xmath205 of the hydrodynamic spectrum to the total strength of the scalar susceptibility in the njl model as @xmath206 on the other hand , these limits in the symmetric phase result in the same value @xmath207 which means no hydrodynamic contribution to the scalar susceptibility there ( @xmath208 ) . h mode must be decoupled from @xmath209 in the symmetric phase . meanwhile we know that the corresponding ratios for the susceptibilities of the conserved quantities are always unity ( @xmath210 ) , which can be explicitly seen with the expressions given in appendix b. the ratio @xmath205 ( [ eq : r - njl ] ) is shown in fig . [ fig : ratio ] as a function of @xmath135 along the critical line . we find that even in the o(4 ) chiral transition at zero baryon number density ( @xmath166 ) the hydrodynamic spectrum contributes to the divergence by a finite fraction . this contribution of the hydrodynamic spectrum increases toward the tcp , and eventually gives the leading divergence at the tcp , where @xmath211 but @xmath212 . this behavior is completely in parallel with the tdgl approach . the fact that the p h mode gives a finite fraction of the divergence at the o(4)cp might be again unexpected from the viewpoint of the sigma meson as the associated soft mode there . indeed , the sigma meson spectrum generates the total divergence when the critical point is approached from the symmetric phase . we should note here that the mixing of the scalar fluctuation in the broken phase is the origin of the discontinuity of the baryon number and energy susceptibilities across the boundary and that only the scalar p h mode with the hydrodynamic character can couple with the fluctuations of these conserved quantities . since the transitions between the scalar and other channels is proportional to @xmath184 , the p h spectral strength in the scalar is necessarily of order @xmath213 so as to bring a finite contribution to @xmath162 and @xmath165 . we note that the _ scalar _ p h motion of the njl model is possible only in medium , but always possible in medium even in the symmetric phase , where the p h contribution should be decoupled from the scalar susceptibility . one may ask the reason for this decoupling of the p h spectrum . the absorption amplitude of the collective p h mode with momentum @xmath10 by a left handed quark @xmath214 is proportional to a spinor product @xmath215 . in the symmetric phase this coupling must contain a helicity flip because chirality and helicity are identical for a massless quark , and vanishes in the @xmath122 limit . it is easy to show that @xmath216 . for a massive quark in the broken phase , each of the left and right handed states has both helicity components , and thus @xmath217 in the @xmath122 limit . let us discuss a little more details of the spectral contributions to the susceptibility , studying the the poles and the residues of the scalar response function @xmath218 near the njl critical points . it is useful to represent the spectral contributions as @xmath219 with the poles corresponding to the sigma meson ( @xmath220 ) near @xmath221 and the p h ( ph ) mode on the negative imaginary axis@xcite . we would obtain the susceptibility as @xmath196 after setting @xmath222 . as for the p h contribution , however , we take into account the kinematic condition @xmath223 for the spectrum via @xmath224 the scalar susceptibility in this approximation is expressed as a sum of the sigma and p h pole contributions : @xmath225 in fig . 9 we show numerical results of the poles and residues of the scalar response function ( 32 ) as functions of @xmath226 with fixed @xmath227 . the behavior of them can be understood as follows : _ across the o(4)cp _ in the broken phase ( @xmath228 ) , the sigma pole with @xmath229 locates at @xmath230 on the real axis , whose residue is @xmath231 . these quantities scale as @xmath232 . in the symmetric phase ( @xmath233 ) , the complex sigma meson pole appears at @xmath235 with the residue @xmath236 . in both cases , the sigma mode gives the appropriate strength of the divergence @xmath237 . h mode arises from the @xmath98 dependence of the function @xmath204 . since we are interested in the behavior in the small @xmath98 and @xmath10 region with @xmath238 , we may approximate the function @xmath239 as @xmath240 where @xmath241 . in the broken phase , the condition , @xmath242 gives the pole as @xmath243 , whose residue @xmath244 . similarly in the symmetric phase the condition @xmath245 fixes the pole position as @xmath246 with the residue @xmath247 . then according to eq . ( 34 ) the spectral contribution is estimated to be @xmath248 for the broken and symmetric cases , respectively , as @xmath249 . we note that the p h mode gives finite portion of the divergence at the o(4)cp approached from the broken phase due to the enhancement of the residue by @xmath213 , despite that the frequency @xmath250 shows no critical slowing . the decoupling of the p h mode in the symmetric phase is correctly described by the behavior of the pole . = 0.3 = 0.3 = 0.3 _ across the tcp_at the tcp , @xmath251 . when we approach from the broken phase @xmath252 while @xmath253 . then the sigma mode slows down , but can not generate the leading divergence because @xmath254 whereas @xmath255 . the fact @xmath256 changes the scaling of the p h mode into @xmath257 with the residue @xmath258 , which gives rise to the correct order of the divergence @xmath259 the p h mode must correspond to the critical eigenmode . on the other hand , if the tcp is approached from the symmetric phase , the p h mode is decoupled from the scalar susceptibility and the critical sigma mode generates the total divergence . it is very interesting that the soft mode associated with the tcp is different between the symmetric and broken phases . _ at the z@xmath3cp_the sigma meson mode has a finite energy gap of order @xmath260 in our model . the pole position of the p h mode can be evaluated as @xmath261 with its residue @xmath262 , which gives rise to @xmath263 from this estimate , we see that the p h mode properly accounts the divergence of the scalar susceptibility at the z@xmath3cp , and therefore the softening of the p h mode is the origin of the critical divergence at the z@xmath3cp . in the microscopic calculation with the njl model , we have seen that the collective p h mode , besides the sigma meson mode , is generated in the scalar channel and brings the spectral contributions to the channels of the conserved quantities through the mixing when @xmath264 . h spectrum makes the susceptibilities of @xmath162 and @xmath165 discontinuous across the o(4 ) critical line , and eventually gives rise to the critical divergences at the tcp and the z@xmath3cp . this role of the p h mode is consistent with the behavior of the hydrodynamic mode in the tdgl analysis . we remark here that this p h mode in the njl model is the time reversible landau damping type . in the phenomenological tdgl approach , on the other hand , we assumed the time irreversible diffusion motion for the conserved density , which seems more appropriate to the non equilibrium soft dynamics . it would be very interesting to study how a time irreversible equation of motion emerges out of the time reversible microscopic theory ( see , e.g. , refs . @xcite ) . the flat curvature of the effective potential is usually referred to as the vanishing screening mass , which naively hints the reduction of a kind of particle mass . as we approach the o(4)cp , the sigma meson mass actually gets reduced to cause the critical divergence . however , approaching the z@xmath3cp , we see that the flat curvature leads to the vanishing diffusion constant of the hydrodynamic mode@xcite . in general , the potential curvature expresses the stiffness of the system with respect to the variation of the ordering density . the dynamic quantity related to this stiffness can be the particle mass , sound velocity , relaxation constant or diffusion constant , depending on the equation of motion of the critical eigenmode . in the z@xmath3cp case , the linear mixing of the conserved densities in the proper ordering density dictates that the critical eigenmode should have hydrodynamic character . we have seen that the sigma meson like mode is massive and is decoupled from the slow dynamics . it is explicitly argued in ref.@xcite that the remaining set of slow modes is equivalent to that of the liquid gas critical point@xcite . in the course of heavy ion events passing by the z@xmath3cp , it is important to study the observable implications of critical slowing of hydrodynamic fluctuations in baryon number and entropy densities@xcite . the growth of the diffusive fluctuations within the finite space time would get renewed interest@xcite . the critical soft mode of the tcp is different between the symmetric and broken phases . the sigma mode becomes the critical eigenmode and the hydrodynamic one behaves normally if the critical point is approached from the exactly symmetric phase . otherwise the critical eigenmode bears the hydrodynamic character and the sigma mode slows down only moderately , which implies the importance of the hydrodynamic fluctuations near the tcp : the endpoint of the first order line sitting on the chiral phase boundary . theoretically and also experimentally it is worthwhile to elaborate and classify the dynamic critical behavior at the tcp as well as other critical points of qcd@xcite . in the qcd thermodynamics the first order transition is believed to occur at finite temperature in the massless three flavor case@xcite and in the pure gluonic case@xcite , where the chiral symmetry and the @xmath265 center symmetry are exact , respectively . as varying masses of the quarks from these two limits , we will have critical points ( line ) at the edge of the first order region . for example , in ref . @xcite , a z@xmath3cp is studied in the @xmath1@xmath7 plane with three quark flavors of equal mass in lattice qcd . there a particular linear combination of the quark condensate and the energy density is identified as a proper order parameter , which should be mapped to the magnetization in the z@xmath3 ising model . we expect that a hydrodynamic mode related to the energy fluctuation shows critical slowing at this z@xmath3cp , whose spectrum may be detected in the lattice qcd . the dynamic critical behavior of this point is also of importance . it could be different from that of the z@xmath3cp at finite @xmath2 because of no linear mixing with the baryon number fluctuation due to the symmetry @xmath266 . furthermore , the lines and surfaces of the qcd critical points in the @xmath1@xmath2@xmath267@xmath4 space are speculated@xcite . one can also extend the space to the isospin channel@xcite . generally at such a critical point the proper ordering density becomes a linear combination involving conserved densities . since the critical eigenmode must have the hydrodynamic character in this case , the dynamic critical nature would be quite different from the case with ( e.g. ) the exact chiral symmetry . our identification of the soft modes along the critical line is done within the mean field approximation . fluctuations around the mean fields are known to become crucial for describing the singular behavior at the critical points correctly . to this end we should utilize mode - coupling theory or dynamic renormalization group method . the mean field analysis provides a good starting point to identify an appropriate set of the slow modes . the fundamental points about the z@xmath3cp are following : ( 1 ) in the absence of the chiral symmetry , the ordering density becomes a _ linear _ combination of the scalar density , the baryon number density and the energy density , in general , and their susceptibilities have the same critical exponent . in describing the static property of this critical point one may equally well take any of these densities as the ordering density . ( 2 ) then the critical eigenmode must be the hydrodynamic one which can cause the critical divergence of the susceptibility of the conserved density . on the other hand , in the chiral transition approached from the symmetric phase , the exact chiral symmetry prohibits the linear mixing of the hydrodynamic mode in the fluctuation of the ordering density . we have showed these points using the tdgl approach as well as the microscopic njl model . we have studied the change of the critical eigenmode along the o(4 ) critical line . when the critical point is approached from the symmetric phase , the soft mode is indeed the sigma meson mode . on the other hand , approaching from the broken phase we see the scalar condensate allows the linear mixing between the sigma and hydrodynamic modes , and eventually at the tcp the hydrodynamic mode turn out to be the critical eigenmode which generates the leading critical divergence . thus the shift of the critical mode from the sigma meson to the hydrodynamic mode occurs at the tcp . and the soft mode at the tcp crucially depends on from which phase one is approaching the point . the criticality of the z@xmath3cp is given by the softening of a hydrodynamic mode . the sigma mode remains as a fast mode due to the explicit breaking @xmath7 and is decoupled from the soft dynamics . based on this understanding , we should study fluctuations with the hydrodynamic character , such as baryon number and entropy fluctuations@xcite in locating the z@xmath3cp experimentally . in the qcd thermodynamics with three quark favors , several kinds of end points are speculated . one should keep in mind that at these points the hydrodynamic mode will become the critical eigenmode once a conserved density is linearly mixed in the ordering density and also that the _ approximate _ symmetry of the underlying interactions does not provide any reason for singularity in the phase diagram . the authors thank k. ohnishi for discussions in the early stage of this work . they are very grateful to t. matsui , m. rho and m.a . stephanov for fruitful discussions and comments . an important comment by t. kunihiro on the susceptibility of the conserved density is sincerely acknowledged . thanks the warm hospitalities extended to him at yitp and at kias , where parts of this work were performed . this work is supported in part by the grants - in - aid for scientific research of monka - sho ( grant no . 13440067 ) . the susceptibility of a ( bosonic ) density @xmath268 in a system described with the grand potential @xmath269 with @xmath270 is defined as @xmath271 this susceptibility is obtained as @xmath272-limit of the response function@xcite because @xmath273 where we used a formula , @xmath274 with a matrix valued function @xmath275 , and the time dependence of the operators are defined by @xmath276 . the imaginary time correlation and the response function are introduced as @xmath277 and @xmath278 \rangle_c$ ] , respectively . once we establish the relation between the susceptibility and the response function , it is easy to express the susceptibility as an integral over the spectral density : @xmath279 specifically , when @xmath268 is conserved , @xmath280=0 $ ] , we can freely change the position of the operator @xmath281 in the trace and the susceptibility is directly related to the equal time correlation function @xmath282 via@xcite @xmath283 in the last equality we used the fluctuation dissipation theorem , which relates the fluctuation @xmath282 to the dissipative part @xmath284 of the response function . noting the spectral condition , @xmath285 , we conclude that these two expressions for the susceptibility coincide if and only if @xmath286 . physically this is a consequence of the existence of the current @xmath287 such that @xmath288 . we present the explicit formulas and procedures to evaluate for the one loop polarization functions . first working in the imaginary time formalism@xcite , we derive the expressions for the polarization functions with make use of the frequency sum formulas : @xmath289 \nonumber \\ & & -{1 \over 4 e_1}\left [ { 1-n_{+1}-n_{-2 } \over i\omega_l- e_1 -e_2 } - { n_{-1}-n_{-2 } \over i\omega_l+ e_1 -e_2 } - { n_{+1}-n_{+2 } \over i\omega_l- e_1 + e_2 } + { 1-n_{-1}-n_{+2 } \over i\omega_l + e_1 + e_2 } \right ] \end{aligned}\ ] ] where @xmath290 is the bosonic matsubara frequency , @xmath291 the quark momentum with @xmath292 , @xmath293 , @xmath294 , @xmath295 , and @xmath296 . the @xmath10 and @xmath297dependences are implicit through these quantities . then the analytic continuation to the real frequency is done by the replacement @xmath298 , which uniquely gives the retarded functions with the asymptotic behavior @xmath299 as @xmath300 . since the retarded function is analytic in the upper half plane , we can reconstruct it from the imaginary part using the dispersion integral , @xmath301 this relation is quite useful when we evaluate the polarization functions with finite @xmath10 because the imaginary part is easier to calculate . the imaginary part comes from two physical processes in our model : the @xmath272@xmath302 creation / annihilation and the mode - absorption / emission by a quark @xmath272 or an anti - quark @xmath302 as shown in fig . [ fig : cut ] . kinematically the former occurs for the time like momentum with @xmath303 . the latter is possible for @xmath304 , resulting in the landau damping . then the complex @xmath98 plane has two kinds of cuts in the case of finite @xmath10 . analytic continuation of the retarded function to the lower half plane ( @xmath305 ) across one of these cuts is defined by @xmath306 where @xmath307 is the imaginary part in the time - like or the space - like region , depending on across which cut the function is continued . substituting these expressions to eq . ( [ eq : response ] ) , we can obtain the response functions with finite @xmath10 and search the poles on the unphysical riemann sheet ( see fig . [ fig : cut ] ) . = 0.35 = 0.35 the one loop polarization function in the scalar channel yields @xmath308 where we have shifted the momentum as @xmath309 and @xmath310 , and then introduce the cutoff at @xmath311 for this new @xmath297 . in the second line we defined the functions @xmath312 and @xmath239 , whose massless limits appear in the expansion series of the njl effective potential in terms of @xmath39 , eq.([eq : njl0effpot ] ) . the integral @xmath141 is seemingly divergent logarithmically in the infrared region , but is actually finite because @xmath313 as @xmath314 . after performing the angular integration using the delta function imposed by the on shell condition , the imaginary part of the scalar polarization for @xmath315 yields @xmath316 @xmath317 where @xmath318 and @xmath319 . due to the on - shell condition of the imaginary part , the quark energies in @xmath320 are set to be @xmath321 for @xmath322 and @xmath323 for @xmath324 . the last @xmath325integration can be done analytically . the imaginary part for @xmath326 is obtained by @xmath327 . similarly , other polarization functions and their imaginary parts for @xmath315 are found and given below : @xmath328 @xmath329 @xmath330 , \nonumber \\ \\ { \rm im}\pi_{\beta\beta}(q_0,{\bf q } ) & = & { \nu \over 16 \pi } \int { dk k\over |{{\bf q}}| } \left [ { ( q_0 ^ 2 - 4e^2)(e^2-q^2/4+\mu^2 ) \over \sqrt{e^2-q^2/4 } } { \cal d}(q_0,q^2 ) + 2\mu ( q_0 ^ 2 - 4e^2 ) { \cal d}_\omega(q_0,q^2 ) \right ] , \end{aligned}\ ] ] @xmath331 , \\ { \rm im}\pi_{m\beta}(q_0,{\bf q ) } & = & { m\nu \over 16 \pi } \int { dk k\over |{{\bf q}}| } [ 4\sqrt{e^2-q^2/4}\ , { \cal d}(q_0,q^2 ) -4\mu { \cal d}_\omega(q_0,q^2 ) ] , \end{aligned}\ ] ] @xmath332 , \\ { \rm im}\pi_{\mu\beta}(q_0,{\bf q } ) & = & { \nu \over 16 \pi } \int { dk k\over |{{\bf q}}| } \left [ \mu { q_0 ^ 2 - 4 e^2+{{\bf q}}^2 \over \sqrt{e^2-q^2/4}}\ , { \cal d}(q_0,q^2 ) + ( q_0 ^ 2 - 4e^2 ) { \cal d}_\omega(q_0,q^2 ) \right ] .\end{aligned}\ ] ] in these expressions we introduced a function @xmath333 with the sign ` @xmath334 ' for @xmath335 and ` @xmath336 ' for @xmath324 . it is known that the response functions have a non analytic property at the origin of the @xmath98@xmath272 plane when the p h mode spectrum exists . for demonstration we show in fig . [ fig : singular ] the real and imaginary parts of @xmath337 with @xmath338 . an abrupt change of the real part is seen in the region @xmath339 , which is clearly caused by the p h spectrum in accord with the dispersion relation . in the @xmath196 limit , the imaginary part becomes proportional to @xmath202 , which leads to the discontinuity of the real part of @xmath340 as mentioned in eq . ( 19 ) and shown in fig . [ fig : singular ] . in the case of the massless quarks , on the other hand , the scalar channel does not couple to the p h motion in the @xmath196 limit due to the chiral symmetry . hence the real part is non - singular at the origin as shown in fig . [ fig : scalarpol ] . = 0.25 = 0.37 = 0.37 = 0.37 = 0.37 in appendix a it is shown that the spectral function for a conserved density fluctuation must be proportional to @xmath341 as @xmath196 in general . we confirm that this requirement is fulfilled by these one - loop polarizations , @xmath342 , @xmath343 , @xmath344 , and @xmath345 . the susceptibilities in free quark gas with mass @xmath184 are found as @xmath346 , \nonumber \\ \chi_{m\mu}^{(0)}&=&\nu \int\frac{d^3k}{(2\pi)^3}\frac{m}{e}(n_+'-n_- ' ) , \nonumber \\ \chi_{mt}^{(0)}&=&\nu \int\frac{d^3k}{(2\pi)^3}\frac{m}{e } \left [ \frac{e-\mu}{t } n_+ ' + \frac{e+\mu}{t } n_- ' \right ] , \nonumber \\ \chi_{\mu\mu}^{(0)}&=&-\nu \int\frac{d^3k}{(2\pi)^3}(n_+'+n_- ' ) , \nonumber \\ \int\frac{d^3k}{(2\pi)^3}\left [ \left ( \frac{e-\mu}{t}\right ) ^2 n_+ ' + \left ( \frac{e+\mu}{t}\right ) ^2 n_- ' \right ] , \nonumber \\ \chi_{t\mu}^{(0)}&=&-\nu \int\frac{d^3k}{(2\pi)^3 } \left [ \frac{e-\mu}{t } n_+ ' -\frac{e+\mu}{t } n_- ' \right ] . \nonumber \\\end{aligned}\ ] ] these expressions coincide with the static one - loop polarizations in the @xmath196 limit . through this limiting procedure we find that the terms containing @xmath347 are related with the p h spectrum parts in functions @xmath348 and @xmath349 , and that all susceptibilities , except for the scalar , must accompany @xmath347 because of their hydrodynamic nature . we also notice the fact of no mixing of the scalar fluctuation with others in the massless quark gas @xmath233 . in the case of @xmath166 , the vector fluctuation does not mix with others . both of these originate from the symmetries . if we write down the gl effective potential , it must be invariant under @xmath350 and/or @xmath351 , respectively , and therefore any linear mixing with other densities is impossible . the chiral quark model can be used to perform the same analysis as in the njl model : @xmath352q , \end{aligned}\ ] ] where @xmath353 , @xmath354 , and @xmath355 . the meson mode is introduced here as an elementary field with the kinetic term . integrating out the quark field , we obtain the effective potential within the mean field approximation for @xmath39 and @xmath356 as @xmath357 \label{eq : chiqfree}\end{aligned}\ ] ] with @xmath130 and @xmath358 . this potential is almost the same as that of the njl model ( [ eq : njlfree ] ) , and this model is expected to have the same phase structure . the subtle point is that the divergent vacuum quark fluctuation in the integrand of eq . ( [ eq : chiqfree ] ) requires a regularization and renormalization . instead of the three momentum cutoff used in our njl model calculation , here we adopt a simple prescription following ( e.g. ) ref . @xcite ; we assume that the renormalization is already done in the vacuum and discard the vacuum polarization term . then the parameters are chosen so as to reproduce the pion decay constant , the pion and sigma masses , and the constituent quark mass in the vacuum . we found the z@xmath3cp at @xmath359=(117.7 , 176.2 ) mev . within the same level of the approximation , the scalar response function is calculated as @xmath360 where the polarization @xmath340 is defined in eq . ( [ eq : polfuncs ] ) with @xmath175 , but whose vacuum part is removed . the four point interaction of the njl model is replaced by the non local one here . other response functions @xmath361 have the same structure as the njl result ( [ eq : response ] ) because we assume the same scalar type interaction between quarks . in the numerical calculation with this chiral quark model at the z@xmath3cp , we confirmed the spectral enhancement in the space like momentum region , similar to fig . [ fig : spfunc ] , and found a pole responsible for this enhancement on the negative imaginary axis in the complex@xmath98 plane , just as in the njl model . the ratio @xmath205 defined in eq . ( [ eq : r - njl ] ) also goes to unity as the z@xmath3cp approached . therefore our conclusion on the importance of the hydrodynamic mode at the z@xmath3cp is unaltered here . we should note , however , that the semi positivity condition on the spectral function is violated in our numerical result in the time like momentum region . this is because we replaced the term @xmath362 in the expression of @xmath340 with @xmath363 to remove the divergence ( see appendix b ) . this simple regularization breaks the detailed balance relation which is essential to assure thermal equilibrium . hence the result of the spectrum in this chiral quark model should be interpreted with caution . the vacuum subtraction also results in the unexpected infra red divergence of the quartic term , @xmath364 , in the expansion of ( [ eq : chiqfree ] ) around @xmath19 . because of this difficulty , we could not find the tcp in this model with the regularization adopted here . in order to properly discuss the spectral structure in the chiral quark model we need the regularization scheme which satisfies the condition of thermal equilibrium@xcite . k. paech , h. stocker and a. dumitru , phys . c * 68 * , 044907 ( 2003 ) . c. e. aguiar , e. s. fraga and t. kodama , arxiv : nucl - th/0306041 . y. hatta and m.a . stephanov , phys . lett . , 102003 ( 2003 ) ; erratum - ibid , 129901 ( 2003 ) .
we consider approximate distance oracles for edge - weighted-vertex undirected planar graphs . given fixed , we present a-approximate distance oracle with space and query time . this improves the previous best product of query time and space of the oracles of thorup ( focs , j. acm ) and klein ( soda ) from to .
1601.00839
given an @xmath0-vertex edge - weighted undirected planar graph @xmath10 , a distance oracle for @xmath10 is a data structure that can efficiently answer distance queries @xmath11 between pairs of vertices @xmath12 in @xmath10 . one way of achieving this is to simply store an @xmath13-distance matrix where @xmath0 is the number of vertices . each query can be answered in constant time but the space requirement is large . if one is willing to settle for approximate distances , much more compact oracles exist . it has been shown that for any @xmath1 , there is a @xmath2-approximate distance oracle for @xmath10 of size @xmath14 which for any query pair @xmath12 outputs in time @xmath15 an estimate @xmath16 such that @xmath17 ( thorup @xcite and klein @xcite ) . the oracles of thorup and of klein both rely on a recursive decomposition of @xmath10 using shortest path separators from a shortest path tree @xmath18 : first @xmath10 is decomposed into two subgraphs with such a separator and then the two subgraphs are recursively decomposed . an important observation is that for any vertex @xmath19 and any shortest path separator @xmath20 , there is a size @xmath15 set @xmath21 of so called portals on @xmath20 which are vertices such that for any @xmath22 , there exists a @xmath23 such that @xmath24 . thus , to get an approximate distance from @xmath19 to any @xmath22 only @xmath15 distances @xmath25 need to be stored in addition to distances in @xmath18 . the oracle stores distances from @xmath19 to portals on each of the @xmath26 separators above @xmath19 in the recursive decomposition tree , giving a total space of @xmath27 . to answer a @xmath28-query , the oracle identifies the nearest common ancestor separator @xmath29 of @xmath19 and @xmath30 in the recursive decomposition . as @xmath29 separates @xmath19 and @xmath30 , distances from @xmath19 and from @xmath30 to their respective portal sets on @xmath29 can be combined to obtain a @xmath2-approximate distance estimate in @xmath15 time . additional oracles for planar graphs have since been presented . kawarabayashi , klein , and sommer @xcite showed how to improve space to @xmath31 at the cost of an increase in query time to @xmath32 and gave generalizations to bounded - genus and minor - free graphs . kawarabayashi , sommer , and thorup @xcite focused on improving the space - query time tradeoff and gave an oracle with @xmath33 space and @xmath34 query time , where @xmath35 hides @xmath36 and @xmath37 factors , thereby essentially improving the query time - space product from @xmath38 to @xmath14 . they also showed that if the average edge weight is poly - logarithmic , @xmath39 space and @xmath34 query time can be obtained . except for planar graphs with poly - logarithmic average edge weights , every oracle presented so far has a query time - space product of order @xmath40 ( ignoring the dependency on @xmath41 ) . we finally break this barrier by giving an oracle with @xmath3 space and @xmath4 query time . the exact bounds are given in the following theorem . [ thm : main ] let @xmath10 be an @xmath0-vertex undirected edge - weighted planar graph . for any @xmath42 , there is a @xmath2-approximate distance oracle of @xmath10 with query time @xmath43 and space @xmath44 . our dependency on @xmath41 in the query time - space product is worse than in @xcite but still only a low - degree polynomial in @xmath45 ; it is roughly @xmath46 when @xmath47 and roughly @xmath48 otherwise , compared to @xmath49 in @xcite . focus in this paper is on improving the dependency on @xmath0 and not @xmath41 which we regard as fixed . our data structure uses randomization due to hashing and fast integer sorting . space and query time for hashing can be made worst - case with expected construction time . for sorting , we use the algorithm of han and thorup @xcite to get the bound in theorem [ thm : main ] . to make our data structure deterministic , we can use a standard optimal comparison sort or the slightly faster deterministic integer sorting algorithm of fredman and willard @xcite . with the latter , we get a deterministic query time of our data structure of @xmath50 . a main difference between our oracle and those of thorup and of klein is that we do not store distances from each vertex @xmath19 to portals on all @xmath26 separators above @xmath19 . instead we save space by introducing a shortcutting system to the recursive decomposition tree so that we can get from @xmath19 to any separator above it using @xmath51 shortcuts . each shortcut corresponds to a pair of separators on a root - to - leaf path in the tree and we essentially store approximate distances from vertices on one separator to portals on the other separator closer to the root of the recursive decomposition tree . this complicates the query algorithm and its analysis since approximate distances are found in @xmath51 steps instead of just one . thorup @xcite also gave an oracle for planar _ digraphs _ which for polynomially bounded edge weights achieves @xmath52 space and close to @xmath15 query time . exact oracles for planar digraphs with tradeoff between space and query time have been studied but require near - quadratic space for constant or near - constant query time @xcite . for a general undirected @xmath0-vertex graph @xmath10 , thorup and zwick showed that for any parameter @xmath53 , there is a @xmath54-approximate distance oracle for @xmath10 with @xmath55 space and @xmath56 query time which is believed to be essentially optimal due to a girth conjecture of erds @xcite . variations and slight improvements have since been presented ; see , e.g. , @xcite . in section [ sec : prelim ] , we give some basic definitions , notation , and a variant of a standard recursive decomposition of planar graphs with shortest path separators . we then present our oracle in sections [ sec : phasei ] and [ sec : phaseii ] . section [ sec : phasei ] presents the first phase of the query algorithm . this phase computes approximate distances from query vertices @xmath19 and @xmath30 to certain portals on the nearest common ancestor separator @xmath29 of @xmath19 and @xmath30 in the recursive decomposition tree but where approximate shortest paths are restricted to the child regions of @xmath29 in the recursive decomposition . the second phase in section [ sec : phaseii ] uses the output of the first phase to then find approximate distances from @xmath19 and @xmath30 to portals on @xmath29 in the entire graph @xmath10 . a main challenge is that the vertices of @xmath29 are not represented explicitly on all recursion levels but on various levels on the path from @xmath29 to the root of the recursive decomposition ; the second phase traverses this path to find the desired approximate distances and portals . from the output of phase ii , obtaining an approximate @xmath28-distance can then be done efficiently , as we show in section [ sec : approxdist ] . finally , we make some concluding remarks in section [ sec : conclrem ] . for a graph @xmath10 , denote by @xmath57 and @xmath58 its vertex set and edge set , respectively . when convenient , an edge @xmath12 with weight @xmath59 is denoted @xmath60 . for a rooted tree @xmath18 and two nodes @xmath61 , denote by @xmath62 the nearest common ancestor of @xmath19 and @xmath30 in @xmath18 . for a path @xmath21 and for two vertices @xmath63 , @xmath64 $ ] denotes the subpath of @xmath21 between @xmath19 and @xmath30 . as in previous papers on approximate distance oracles for planar graphs , we assume the word - ram model with standard instructions . in the following , @xmath65 denotes an @xmath0-vertex , undirected , edge - weighted planar embedded graph and @xmath18 is a shortest path tree in @xmath10 rooted at a source vertex @xmath66 . by performing vertex - splitting , we may assume that @xmath10 has degree three . the oracle of thorup keeps a recursive decomposition of @xmath10 consisting of shortest path separators . our oracle obtains a similar decomposition but we need it to have some additional properties which we focus on in the following . denote by @xmath67 an arbitrary triangulation of @xmath10 where edges of @xmath68 are called _ pseudo - edges _ and are given infinite weight . a shortest path separator of @xmath67 w.r.t . an assignment of weights to triangles of @xmath67 consists of a ( possibly non - simple ) cycle @xmath69 defined by two shortest paths @xmath70 and @xmath71 in @xmath18 and a non - tree edge @xmath12 ; the total weight of triangles on each side of @xmath69 is at most @xmath72 of the total weight of all triangles of @xmath67 . see @xcite for details . first , we decompose @xmath67 into two subgraphs enclosed by @xmath69 ; both subgraphs inherit the edges and vertices of @xmath69 , for a suitable weight function on triangles . degree two vertices @xmath19 are removed from each subgraph by replacing their incident edges @xmath73 and @xmath74 with a single edge @xmath75 whose weight is the sum of weights of @xmath73 and @xmath74 . then the two subgraphs are recursively decomposed until constant - size subgraphs are obtained . for each subgraph @xmath76 obtained in the above recursive procedure , we form a _ region _ @xmath77 as follows . subgraph @xmath76 contains @xmath26 separators of the form @xmath78 , namely those formed from the root of the recursion down to @xmath76 . these are separators formed from the root of the recursion down to @xmath76 . region @xmath77 is obtained from @xmath76 by removing pseudo - edges , except those contained in the separators that formed @xmath76 , and then removing degree two vertices as above . let @xmath79 be the faces of @xmath76 containing vertices / edges of @xmath10 not belonging to @xmath76 . each @xmath80 is a separator @xmath78 ( possibly with some degree two vertices removed ) and we call @xmath80 a _ hole _ ( of @xmath77 ) . ancestor / descendant relations between regions are defined according to their nesting in the recursive decomposition tree which we denote by @xmath81 . in section [ subsec : constructrecdecomp ] , we show how to pick shortest path separators such that 1 . there are @xmath31 regions in total each having @xmath82 holes , 2 . for each child @xmath77 of each region @xmath76 , every hole of @xmath76 ( regarded as the closed set of the plane inside the hole ) is fully - contained in a hole of @xmath77 , 3 . the height of @xmath81 is @xmath26 . when we refer to a recursive decomposition in the following , we assume it has these properties . observe that for all regions @xmath76 and @xmath77 where @xmath76 is a descendant of @xmath77 , @xmath83 . this follows since @xmath76 is obtained from @xmath77 by eliminating subgraphs of @xmath77 and degree two vertices . in the following , refer to the triangulated subgraphs obtained when recursively decomposing @xmath67 as _ @xmath84-regions _ ; holes of @xmath84-regions are defined to be the holes of the corresponding regions with pseudo - edges added to form the triangulation . we now show how to pick the separators so that the following _ region conditions _ are satisfied : 1 . each @xmath84-region of even resp . odd depth in @xmath81 has at most three resp . four holes , 2 . for each @xmath84-region of even depth in @xmath81 , if it contains exactly @xmath85 faces of @xmath86 , each of its grandchildren contain at most @xmath87 faces of @xmath86 , 3 . for each child @xmath77 of each @xmath84-region @xmath76 , every hole of @xmath76 ( regarded as the closed set of the plane inside the hole ) is fully - contained in a hole of @xmath77 . furthermore , we show that the number of regions is @xmath31 and that the height of @xmath81 is @xmath26 . this gives the desired properties for a recursive decomposition , as stated above . all separators are formed from shortest path tree @xmath18 . the recursion stops once @xmath84-regions with at most two faces of @xmath86 are obtained . for a @xmath84-region @xmath76 of even depth in @xmath81 , assume it has at most three holes ( this trivially holds for @xmath67 at the root of @xmath81 ) . we assign a unit of weight to each face of @xmath76 that is also a face of @xmath67 . all other faces of @xmath76 are given weight @xmath88 . using the subtree of @xmath18 in @xmath76 , we find a balanced shortest path separator w.r.t . this weight function . if @xmath85 is the number of faces of @xmath86 in @xmath76 then each of the subgraphs formed contain at most @xmath87 of these faces . furthermore , each of these subgraphs have at most four holes since at most one new hole is formed when removing one side of the separator . now consider a @xmath84-region @xmath76 of odd depth in @xmath81 and assume it has at most four holes . if @xmath76 has at most two holes , we decompose it as described above for even - depth @xmath84-regions ; the sub-@xmath84-regions formed will have at most three holes . otherwise , we define a different weight function than that above : for each hole @xmath89 of @xmath76 , exactly one of the triangles of @xmath76 contained in @xmath89 is given unit weight . all other triangles of @xmath76 are given weight @xmath88 . the shortest path separator w.r.t . this weight function ensures that each of the two sub-@xmath84-regions of @xmath76 formed will have at most three holes , namely at most @xmath90 holes inherited from @xmath76 and one hole formed by removing one side of the separator . it is now clear that the first region condition holds . the second region condition holds as well from the above and from the fact that each child of an odd - depth @xmath84-region @xmath76 can not contain more faces of @xmath86 than @xmath76 . assume for the sake of contradiction that the third region condition does not hold for some @xmath84-region @xmath76 and one of its children . then the separator that was used to decompose @xmath76 must have used one of the pseudo - edges in the triangulated hole @xmath89 of @xmath76 . since the boundary of @xmath89 consists of two shortest paths from @xmath18 and a single pseudo - edge , one of the children of @xmath76 must be fully contained in @xmath89 , which for both choices of weight function above gives an unbalanced separator , a contradiction . we conclude that the third region condition holds . the number of @xmath84-regions is asymptotically bounded by the number of leaves of the recursive decomposition tree . each leaf @xmath84-region has a parent with at least three faces of @xmath86 . for any two distinct @xmath84-regions @xmath76 and @xmath77 that are parents of leaf @xmath84-regions , no face of @xmath86 belongs to both @xmath76 and @xmath77 . it follows that there are only @xmath31 parents of leaf @xmath84-regions . since the decomposition tree @xmath81 is binary , the total number of leaf @xmath84-regions is @xmath31 . this implies that the total number of @xmath84-regions , and hence regions , is @xmath31 . it follows from the second region condition and the termination condition for the recursion that @xmath81 has height @xmath26 . let @xmath91 be a path in @xmath18 from @xmath66 to some vertex . for any subpath @xmath92 $ ] of @xmath91 , let @xmath93 be those interior vertices of @xmath91 having an incident edge of @xmath94 emanating to the left of @xmath92 $ ] when looking in the direction from @xmath19 to @xmath30 . we order the vertices such that @xmath92 = u\leadsto v_1\leadsto\cdots\leadsto v_k\leadsto v$ ] . the _ left side _ of @xmath92 $ ] is the @xmath28-path with edges @xmath95 where each edge has weight equal to the weight of the corresponding subpath of @xmath91 . we define the _ right side _ of @xmath92 $ ] similarly . for a region @xmath76 , denote by @xmath96 the _ boundary _ of @xmath76 which is the subgraph of @xmath76 contained in the @xmath82 holes of @xmath76 . for each hole @xmath89 , it will be convenient to regard the two shortest paths in @xmath18 bounding @xmath89 as disjoint in @xmath96 by replacing one path with its left side and the other with its right side ; vertices shared by the original two paths are regarded as distinct in the two new paths , see figure [ fig : eulertour ] . note that @xmath96 represented in this way is now a single face of @xmath76 . cutting open each hole like this ensures that paths in @xmath76 do not cross its boundary . a vertex ( edge ) of @xmath76 that is not contained in @xmath96 is called _ interior_. denote by @xmath97 the @xmath82 shortest paths from the shortest path tree @xmath18 that bound @xmath96 . a vertex which is an endpoint of a path in @xmath97 is called a _ corner _ of @xmath76 ; see figure [ fig : eulertour ] . we let @xmath98 denote the set of @xmath82 corners of @xmath76 . in some places , we will instead consider the face @xmath99 obtained from @xmath96 by replacing each edge with its corresponding path in @xmath10 . the following lemma gives the structure of @xmath96 . [ lem : deltar ] for any region @xmath76 , @xmath96 consists of @xmath82 subpaths each of which is either a single ( possibly pseudo-)edge between two corners , an edge corresponding to a path in @xmath18 from @xmath66 to some corner of @xmath76 , or the left or right side of a path in @xmath18 . consider partitioning @xmath96 into subpaths each starting and ending in a corner and with no interior corner vertices . there are only @xmath82 such subpaths and each of them is either an edge or pseudo - edge ending in corners or the subpath corresponds to a path in @xmath18 incident to one or two holes . if it is incident to one hole , it is the left or right side of a path in @xmath18 and if it is incident to two holes , it is a single edge corresponding to a path ending in @xmath66 . when convenient , we identify regions with their corresponding nodes in @xmath81 . for two regions @xmath76 and @xmath77 , denote by @xmath100 the simple path from @xmath76 to @xmath77 in @xmath81 . let @xmath65 be an undirected edge - weighted ( not necessarily planar ) graph , let @xmath101 , and let @xmath91 be a shortest path in @xmath10 . thorup @xcite showed that for any given @xmath1 , @xmath102 contains a set @xmath21 of @xmath15 _ portals _ such that for any @xmath103 , there is a @xmath23 such that @xmath104 . let @xmath76 be a region and let @xmath105 . given a value @xmath106 , a vertex @xmath101 , and an undirected ( possibly non - planar ) graph @xmath89 with @xmath107 where every edge @xmath108 in @xmath89 corresponds to a path @xmath109 in @xmath10 of the same weight . then @xmath21 is a _ @xmath110-portal set of @xmath96 _ if for any @xmath111 , there exists a vertex @xmath23 such that @xmath112 . we call @xmath113 a _ portal _ ( of @xmath21 ) . applying the portal construction of thorup @xcite ( see also the proof of lemma [ lem : portal ] ) to each path in @xmath97 gives a @xmath110-portal set of @xmath96 of size @xmath114 . define @xmath110-portal set of @xmath99 similarly ; its size is @xmath115 as well . we need a slightly more general result regarding portals than that of thorup which we state in the following somewhat technical lemma . it roughly says that if we have a graph @xmath89 representing some subgraph of @xmath10 such that distances in @xmath10 from a vertex @xmath19 to a shortest path @xmath91 in @xmath10 are approximated well in @xmath89 , then we can pick a small number of portals from @xmath89 along @xmath91 such that these distances are also approximated well with shortest paths in @xmath89 from @xmath19 to @xmath91 through these portals . [ lem : portal ] let @xmath91 be a shortest path in an edge - weighted undirected graph @xmath65 and let @xmath101 and @xmath116 be given . let @xmath89 be an undirected graph with @xmath117 such that for any @xmath118 , @xmath119 . assume that for any @xmath103 , there is a @xmath120 such that @xmath121 . then there is a subset @xmath122 of @xmath123 of size @xmath124 such that for any @xmath103 there is a @xmath125 such that @xmath126 . the construction is similar to that of thorup . the first portal @xmath127 added to @xmath122 is the vertex @xmath128 minimizing @xmath129 . let @xmath130 be an endpoint of @xmath91 . we show how to construct @xmath131 $ ] ; the same construction is done for the other subpath of @xmath91 . let @xmath132 be the latest portal added to @xmath122 and traverse @xmath133 $ ] towards @xmath130 until encountering a vertex @xmath134 such that @xmath135 . portal @xmath136 is then the next portal added to @xmath122 . the process stops when reaching the vertex of @xmath123 closest to @xmath130 ; this vertex is added as the final portal @xmath137 to @xmath122 . let @xmath103 be given . by assumption , there is a @xmath120 such that @xmath121 . the above construction ensures that there is a @xmath125 such that @xmath138 it remains to prove that @xmath139 . by symmetry , it suffices to show @xmath140| = o(1/\epsilon'')$ ] . for any @xmath141)$ ] , define potential @xmath142 . for @xmath143 , @xmath144 so the potential is reduced by @xmath145 since @xmath146 and @xmath147 ( because @xmath148 ) , we have @xmath149 , implying that @xmath150 . our data structure answers a query for vertex pair @xmath12 in two phases , _ phase i _ and _ phase ii_. in this section , we describe the preprocessing for phase i and then the query part . the output and performance of phase i applied to @xmath19 is stated in the following lemma ( @xmath30 is symmetric ) . phase i starts with @xmath151 which is an arbitrary region @xmath76 ( among at most two choices ) such that @xmath152 and @xmath153 where @xmath77 is the parent of @xmath76 in @xmath81 . region @xmath154 is defined similarly . [ lem : phasei ] phase i for vertex @xmath19 can be implemented to run in @xmath155 time using @xmath156 space . for the output @xmath157 , we have that for all @xmath158 , there is a @xmath159 such that @xmath160 where @xmath161 is the child of @xmath162 on the path in @xmath81 from @xmath163 to @xmath164 . furthermore , @xmath165 . note that any @xmath28-path of @xmath10 must intersect @xmath166 . phase i computes approximate distances to this separator but with the restriction that paths must be contained in @xmath161 . phase ii is considered in section [ sec : phaseii ] and it extends the output of phase i to approximate distances to @xmath167 in the entire graph @xmath10 . we start by constructing a recursive decomposition of @xmath10 and the associated decomposition tree @xmath81 . in order to traverse leaf - to - root paths of @xmath81 efficiently , we set up a shortcutting system for @xmath81 . for any region @xmath168 , let @xmath169 be the largest integer such that the depth of @xmath76 in @xmath81 is divisible by @xmath170 . for any integer @xmath171 between @xmath88 and @xmath169 , we add a pointer from @xmath76 to the ancestor @xmath77 @xmath172 levels above @xmath76 . we refer to this pointer as a _ shortcut _ and denote it by @xmath173 . we can get from any region @xmath174 to any proper ancestor @xmath175 of @xmath174 by traversing only @xmath51 shortcuts : first traverse the shortcut @xmath176 where @xmath77 is the closest region to @xmath175 which is either @xmath175 or one of its descendants . then recurse on pair @xmath177 until reaching @xmath175 . before describing the preprocessing for phase i , we need the following lemma . for any vertex @xmath101 , define @xmath178 as the set of regions @xmath76 where @xmath179 . [ lem : regionpath ] for all @xmath101 , regions of @xmath178 form a subpath of a leaf - to - root path in @xmath81 . let @xmath76 be a region in @xmath178 . non - corner vertices of shortest paths in @xmath97 have degree three in @xmath10 ( otherwise , their incident edges would have been merged into one in the construction of @xmath76 ) so @xmath19 must be incident to an interior edge @xmath180 of @xmath76 . this edge can not be a pseudo - edge since @xmath181 but must be an edge of @xmath10 . the same can not be true for both child regions of @xmath76 in @xmath81 since @xmath10 has degree three so at most one of these regions belongs to @xmath178 . we sometimes regard @xmath178 as the subpath from the lemma . for any shortcut @xmath182 , define @xmath183 as the set of @xmath82 vertices @xmath184 such that @xmath19 is the last vertex from @xmath66 on a path of @xmath185 satisfying @xmath186 ; see figure [ fig : phaseidata ] . for each shortcut @xmath182 and each @xmath187 , we construct and store a size @xmath188 @xmath189-portal set @xmath190 of @xmath191 together with distances @xmath192 for each @xmath193 ( figure [ fig : phaseidata ] ) ; @xmath194 will be specified precisely in section [ subsec : timestretch ] below . this completes the description of the preprocessing . [ lem : typeiportalset ] the total space required for phase i is @xmath195 . it suffices to give an @xmath195 bound on the total size of portal sets defined above . let @xmath196 be given . by lemma [ lem : regionpath ] , there can only be @xmath51 shortcuts @xmath182 where @xmath197 and @xmath198 . hence there are only @xmath51 shortcuts @xmath182 where @xmath199 . the total number of sets @xmath183 and the total number of corners of @xmath174 over all shortcuts @xmath182 is @xmath200 and @xmath201 . as each set @xmath190 has size @xmath188 , total size of portal sets is @xmath195 . in this subsection , we present phase i for query vertices @xmath19 and @xmath30 . figure [ fig : phaseidata ] is useful to consult in the following . we assume that starting regions @xmath151 and @xmath154 are not on the same leaf - to - root path in @xmath81 . as we will see later , the other case is easily handled . pseudocode for vertex @xmath19 can be found in figure [ fig : phasei ] ; the same call is made with @xmath19 replaced by @xmath30 . ' '' '' + ddd============**phase i * * for @xmath19 : + + 1 . let @xmath202 be the shortcuts from @xmath203 to @xmath204 + 2 . let @xmath205 + 3 . for each @xmath206 , let @xmath207 + 4 . for @xmath208 to @xmath209 + 5 . let @xmath210 + 6 . let @xmath211 + 7 . let @xmath212 be face @xmath213 restricted to vertices that are either in @xmath214 or + are incident to edges in @xmath215 + 8 . construct the graph @xmath80 consisting of the edges @xmath216 + 9 . for each @xmath217 , let @xmath218 + 10.if @xmath219 , construct @xmath220-portal set @xmath221 of @xmath213 of size @xmath188 + 11.construct @xmath222-portal set @xmath223 of @xmath224 of size @xmath225 + 12.output @xmath226 , where @xmath227 + ' '' '' let @xmath161 be defined as in lemma [ lem : phasei ] ( @xmath228 is defined similarly for @xmath30 ) . let @xmath202 denote the sequence of shortcuts from @xmath229 to @xmath204 . to simplify the code , we assume @xmath230 ; the other case is straightforward . note that @xmath231 . in lines @xmath232 and @xmath233 , we obtain the precomputed portal set @xmath205 as well as distances @xmath234 for each portal @xmath206 . note that @xmath235 is well - defined by definition of @xmath151 . in the @xmath169th iteration of the for - loop , we are given @xmath236 constituting portals for @xmath237 and we form a graph @xmath80 containing @xmath19 and a subset of @xmath238 such that all distances from @xmath19 to @xmath213 in @xmath239 can be approximated by going through @xmath80 and then along @xmath213 . an illustration of @xmath80 can be seen in figure [ fig : hiphaseidata ] . edges of @xmath240 in line @xmath241 are added to @xmath80 and these represent the approximate distances to @xmath236 in @xmath242 found in the previous iteration . to have @xmath80 approximate distances from @xmath19 in @xmath239 , we add another edge set @xmath243 , defined in line @xmath244 ; for every vertex in @xmath245 , we add to @xmath80 the edge @xmath246 for every @xmath247 , representing a shortest @xmath248-path in @xmath239 . to allow @xmath80 to traverse @xmath213 , we add @xmath212 in line @xmath249 . this is a compact representation of face @xmath213 restricted to a subset of its vertices ; each subpath of @xmath213 between two consecutive vertices in this subset is a single edge of the same weight in @xmath212 . we show below that distances in @xmath80 from @xmath19 to @xmath250 approximate distances from @xmath19 to @xmath213 in @xmath239 . in order to avoid an explosion in the size of future portal sets , we form a @xmath220-portal set @xmath221 of @xmath213 of size only @xmath188 in line @xmath251 which is then used in the next iteration . line @xmath252 is identical to line @xmath251 except that we use a value @xmath253 instead of @xmath254 . we shall pick @xmath255 which gives a much smaller portal set @xmath256 output in line @xmath257 ; this will help speed up phase ii . we do not use @xmath258 inside the for - loop in line @xmath251 since the approximation error builds up over each iteration so we need the smaller value @xmath254 there . lemma [ lem : phasei ] follows from the following invariant for the for - loop in lines @xmath259@xmath251 : invariant : : : at the start of the @xmath169th iteration of the for - loop in lines @xmath259@xmath251 of figure [ fig : phasei ] , for all @xmath260 , there is a @xmath261 such that @xmath262 , and @xmath263 . note that the invariant holds initially when @xmath208 since @xmath235 is a @xmath264-portal set of @xmath191 and its size is @xmath188 . [ [ maintenance - of - invariant ] ] maintenance of invariant : + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath265 be given and assume the invariant holds at the beginning of the @xmath169th iteration of the for - loop . we show that it also holds at the beginning of the @xmath266th iteration . let @xmath91 be a shortest path in @xmath239 from @xmath19 to a vertex @xmath267 . we show that there is a @xmath268 such that @xmath269 approximates the weight of @xmath91 up to a factor of @xmath270 . the second inequality of the invariant will then follow from lemma [ lem : portal ] . the first inequality follows since @xmath271 is the weight of an actual path in @xmath239 from @xmath19 to @xmath113 for each @xmath272 . let @xmath273 be the last vertex on @xmath91 such that the subpath of @xmath91 from @xmath19 to @xmath273 is contained in @xmath242 . in particular , @xmath274 . by the invariant at the beginning of the @xmath169th iteration , there is a portal @xmath275 such that @xmath276 . assume first that @xmath277 . then @xmath91 is contained in @xmath242 and @xmath80 approximates the distance from @xmath19 to @xmath278 up to a factor of @xmath279 with the path starting with @xmath280 and followed by edges of @xmath212 . now assume that @xmath281 . consider first the case where @xmath282 ( figure [ fig : phaseidata ] with @xmath59 playing the role of @xmath283 ) . we have the precomputed portal set @xmath284 and it contains a portal @xmath285 such that @xmath286 . hence , @xmath287 now consider the other case where @xmath281 and @xmath288 . then a shortest path from @xmath283 to @xmath273 in @xmath237 contains a vertex @xmath289 and the subpath from @xmath283 to @xmath290 is contained in @xmath291 . we have a precomputed portal set @xmath292 containing a portal @xmath285 such that @xmath293 . this gives @xmath294 as desired . [ [ termination ] ] termination : + + + + + + + + + + + + the invariant shows that at the beginning of the @xmath209th iteration , for all @xmath295 , there is a @xmath296 such that @xmath297 . line @xmath252 is identical to line @xmath251 for @xmath298 except that @xmath254 is replaced by @xmath258 so line @xmath252 increases the approximation by a factor of @xmath299 . below we choose @xmath254 and @xmath258 such that @xmath300 . this will imply lemma [ lem : phasei ] . obtaining shortcuts in line @xmath301 can be done in @xmath302 time . lines @xmath232 and @xmath233 take @xmath303 time as @xmath304 and distances @xmath234 for @xmath206 are precomputed . we first show how a single iteration @xmath169 of the for - loop in lines @xmath259@xmath251 can be implemented to run in @xmath305 time . then we show how to improve it to @xmath306 . finding @xmath240 takes @xmath307 time . as @xmath308 , @xmath243 can be found in time @xmath309 . face @xmath212 is obtained in @xmath310 time by sorting the vertices according to their cyclic ordering in an euler tour of face @xmath213 . graph @xmath80 and distances @xmath311 for @xmath217 can then be obtained in @xmath312 time . to obtain @xmath313 , apply the portal construction algorithm in the proof of lemma [ lem : portal ] to each shortest path in @xmath314 restricted to @xmath80 . this takes @xmath315 time . line @xmath252 takes @xmath225 time . we improve the time bound to @xmath306 by avoiding the sorting step above . instead , we omit adding edges of @xmath212 to @xmath80 and apply a variant of the portal construction algorithm in the proof of lemma [ lem : portal ] to each shortest path @xmath316 restricted to @xmath80 . we describe this variant in the following . first observe that @xmath317 can be partitioned into @xmath188 subsets @xmath318 where each subset @xmath319 is either a singleton set consisting of a vertex in @xmath214 , a singleton set consisting of an endpoint of an edge in @xmath240 , or a set @xmath320 for each @xmath321 . assume that each set @xmath319 is sorted along @xmath322 in non - decreasing distance from the source @xmath66 in shortest path tree @xmath18 ; this sorting can be done in the preprocessing step . the first portal @xmath127 to be added to @xmath313 is the vertex @xmath323 minimizing @xmath324 . we can identify @xmath127 in @xmath325 time . let @xmath130 be the vertex of @xmath322 farthest from @xmath66 . as in the proof of lemma [ lem : portal ] , we only describe the algorithm for adding portals to @xmath326 $ ] ; adding portals along @xmath327 $ ] is symmetric . for each set @xmath319 , we keep a pointer to the first vertex in its sorted order . we then make a single pass over the sets @xmath319 and for each of them move its pointer forward to the first vertex in @xmath326 $ ] ( if any ) for which the distance to it can not be approximated by going through the previously added portal @xmath127 , i.e. , the first vertex @xmath328 $ ] such that @xmath329 . among the vertices with pointers to them , the one closest to @xmath127 in @xmath322 is then added to @xmath313 as the next portal , and @xmath127 is updated to this vertex . additional passes are made until the pointers have moved past all vertices of their respective @xmath319 sets . correctness follows since the set of portals formed is the same as that obtained by the portal - construction algorithm of thorup . running time is @xmath306 . to see this , note that each pass ( except possibly the last ) adds at least one portal to @xmath313 so the number of passes is @xmath188 . furthermore , each pass takes @xmath330 time where @xmath331 is the total number of vertices visited in that pass over all sets @xmath319 . since the total number of vertices visited over all passes is @xmath332 , the time bound follows . we can obtain a stretch of @xmath333 for the approximate distances obtained in the final iteration of phase i as follows . since @xmath334 when @xmath335 , we pick @xmath336 to obtain @xmath337 which is at most @xmath338 when @xmath339 . picking @xmath340 gives @xmath341 for @xmath41 smaller than some positive constant . this shows lemma [ lem : phasei ] . phase ii takes as input the sets @xmath256 and @xmath342 with associated approximate distances @xmath343 and @xmath344 that were output by phase i. the output of phase ii has the properties stated in the following lemma . denote by @xmath29 the shortest path separator in @xmath10 that separates @xmath163 into @xmath161 and @xmath228 . in section [ sec : approxdist ] , we efficiently obtain from this output an approximate @xmath28-distance . [ lem : phaseii ] phase ii for @xmath19 can be implemented to run in @xmath345 time using @xmath346 space , given the output from phase i. for the output @xmath347 from phase ii , we have @xmath348 and for any @xmath349 , there is a vertex @xmath350 such that @xmath351 . the preprocessing for phase ii consists of the following four steps : [ [ step-1 ] ] step @xmath301 : + + + + + + + + + + + + + + + + + + + + + for each @xmath196 and each of the at most two regions @xmath352 with @xmath353 and @xmath354 where @xmath355 is the parent of @xmath352 in @xmath81 , we form a @xmath356-portal set @xmath357 of @xmath358 . from this we form and store a subset @xmath359 of @xmath360 . this subset contains @xmath361 . in addition , for every @xmath362 and every left or right side @xmath363 ( see section [ subsec : recdecomp ] ) such that @xmath91 and @xmath113 are contained in the same path of @xmath18 , @xmath359 contains the successor and predecessor ( if any ) of @xmath113 on @xmath91 ; figure [ fig : pw ] gives an illustration . note that @xmath364 so @xmath365 . for any @xmath366 there is a @xmath367 with @xmath368 . since @xmath369 the shortest path in @xmath358 from @xmath370 to @xmath30 intersects @xmath360 . hence there is a @xmath371 which is either @xmath370 , the successor or predecessor of @xmath370 on @xmath91 , or a corner in @xmath372 such that @xmath373 . we also have @xmath374 . [ [ step-2 ] ] step @xmath232 : + + + + + + + + + + + + + + + + + + + + + for each shortcut @xmath182 and each @xmath375 , store a @xmath376-portal set @xmath377 of @xmath378 of size @xmath225 ; call this a _ type 1 _ portal set . [ [ step-3 ] ] step @xmath233 : + + + + + + + + + + + + + + + + + + + + + for any shortcut @xmath182 , _ dual portal set _ @xmath379 is the set of vertices @xmath380 for which a vertex @xmath59 exists with @xmath381 , @xmath382 , such that @xmath383 ; see figure [ fig : phaseiidata ] . define @xmath384 . note that @xmath385 . for each @xmath386 , store a @xmath387-portal set @xmath388 of @xmath378 of size @xmath225 together with distances @xmath389 for each @xmath390 . refer to it as a _ type 2 _ portal set . the definition of type 2 portal sets is rather technical so let us give the high - level idea for introducing them ; see figure [ fig : phaseiidata ] in the following . as in phase i , we jump along shortcuts @xmath391 in phase ii ; a detailed description is given in the next subsection . in phase ii , we need approximate distances in @xmath10 from certain portals @xmath113 in @xmath237 to @xmath392 . however , @xmath113 might also be present in @xmath213 . in this case , we can not afford to associate portal sets with @xmath113 and shortcut @xmath391 since @xmath113 may occur in several regions of @xmath81 ( see lemma [ lem : regionpath ] ) . however , vertices @xmath59 for which @xmath352 is sandwiched in between @xmath242 and @xmath239 can pay for dual portal set @xmath393 and the associated type 2 portal sets . as we show below , we can obtain an approximate distance from @xmath113 to any @xmath394 by first going along @xmath18 from @xmath113 to a nearby @xmath395 , then along a shortest path in @xmath10 from @xmath396 to a portal @xmath397 in the type 2 portal set of @xmath396 , and finally from @xmath370 to @xmath398 along @xmath18 . using hashing , we can access each type 1 and type 2 portal set in @xmath82 time from the vertex and the shortcut defining it . [ [ step-4 ] ] step @xmath259 : + + + + + + + + + + + + + + + + + + + + + for each shortcut @xmath182 and for any shortest path @xmath399 , we keep a veb - tree , allowing us to find the successor / predecessor of any vertex of @xmath91 in the subset @xmath400 in @xmath51 time . with hashing , space required for the veb - tree is @xmath401 @xcite . as mentioned in @xcite , both space and query bounds can be made deterministic . [ lem : typeiiandiiiportalset ] the total space required for phase ii is @xmath402 . the total size of portal sets @xmath403 over all @xmath101 is @xmath404 . a proof similar to that of lemma [ lem : typeiportalset ] shows that the total size of all type @xmath301 portal sets is @xmath405 . to bound the size of type 2 portal sets , consider two shortcuts @xmath182 and @xmath406 . by lemma [ lem : regionpath ] , @xmath379 and @xmath407 are disjoint if @xmath408 and @xmath409 are not contained in the same leaf - to - root path @xmath21 in @xmath81 . for any @xmath196 , there are only @xmath51 shortcuts @xmath410 where @xmath411 . hence @xmath412 summed over all shortcuts @xmath182 is @xmath413 . hence , the total size of all dual portal sets @xmath379 , and hence also the total size of all veb trees , is @xmath405 . recall that the number of regions is @xmath31 . each region has @xmath82 corners and each set @xmath183 has constant size so the total size of all sets @xmath414 is @xmath405 . each element of these sets has a type 2 portal set of size @xmath225 . let @xmath415 be the path from @xmath161 to the root @xmath10 of @xmath81 . in the following , for any vertex @xmath59 , denote by @xmath352 the region ( if any ) such that @xmath416 , @xmath353 , and @xmath354 where @xmath355 is the parent of @xmath352 in @xmath81 . observe that @xmath417 . phase ii for @xmath19 takes the output from phase i and produces output satisfying lemma [ lem : phaseii ] . ' '' '' + ddd============**phase ii * * for @xmath19 : + + 1 . let @xmath202 be the shortcuts from @xmath418 to @xmath419 + 2 . let @xmath420 ( portal set output from phase i ) + 3 . let @xmath89 be the graph initially consisting of edges @xmath421 for all @xmath422 + 4 . for @xmath423 to @xmath209 + 5 . add to @xmath89 edges @xmath424 for all @xmath425 and @xmath426 + 6 . for each @xmath261 and each @xmath427 + 7 . add to @xmath89 edge @xmath428 for all @xmath429 + 8 . let @xmath430 + 9 . + ' '' '' we give a high - level description of phase ii before going into details . pseudocode can be seen in figure [ fig : phaseii ] . the algorithm traverses shortcuts @xmath202 from @xmath418 to the root @xmath419 of @xmath81 and incrementally constructs a graph @xmath89 which at termination will satisfy lemma [ lem : phaseii ] . in line @xmath233 , edges of @xmath89 represent approximate paths found in phase i. these paths correspond to subpaths of the final full paths in @xmath10 ( corresponding to the final @xmath89 ) and the subpaths are prefixes of these full paths that are contained in @xmath161 . consider the @xmath169th iteration of the for - loop . in line @xmath241 , we check if any subpath endpoint @xmath261 disappears as a boundary vertex when jumping from @xmath242 to @xmath239 . if so , we can extend the subpath to full paths @xmath432 for each @xmath426 . the other interesting case is when @xmath433 . then we do not have a type @xmath301 portal set associated with @xmath113 and @xmath391 but it might be that some separator vertices @xmath59 of @xmath29 that are present in @xmath392 are no longer present in @xmath213 and we need to ensure that there is a good path in @xmath434 from @xmath113 to @xmath59 . this case is handled in lines @xmath244 and @xmath249 where we ensure such a good path from @xmath113 to @xmath59 by using the type @xmath232 portal sets associated with vertices of @xmath435 that are close to @xmath113 . to show correctness , i.e. , that the set output in line @xmath436 satisfies lemma [ lem : phaseii ] , let @xmath59 be any vertex on @xmath29 and let @xmath21 be a shortest path in @xmath10 from @xmath19 to @xmath59 . let @xmath398 be the last vertex on @xmath21 such that @xmath437 $ ] is contained in @xmath161 . note that @xmath438 . by lemma [ lem : phasei ] , there is a @xmath439 such that @xmath440 . since @xmath417 , we have @xmath416 . since @xmath441 , we also have @xmath442 . let @xmath443 and @xmath444 be the shortcuts such that @xmath445 , @xmath446 , and @xmath447 , @xmath448 . we consider two cases in the following : @xmath449 and @xmath450 . [ [ case-1-i_p_1leq - i_w - figurefigcase1 ] ] case @xmath301 , @xmath449 ( figure [ fig : case1 ] ) : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider iteration @xmath451 of the for - loop . since @xmath452 , we have @xmath453 . since @xmath448 , it follows that @xmath454 . hence , the final @xmath89 contains an edge @xmath455 for each portal @xmath456 ( line @xmath241 ) . since @xmath457 and since @xmath458 , we must have @xmath459 so there is a portal @xmath460 satisfying @xmath461 . note that @xmath462 . the path in the final graph @xmath89 consisting of edges @xmath463 and @xmath464 followed by the path in @xmath18 from @xmath465 to @xmath59 has weight at most @xmath466 [ [ case-2-i_p_1-i_w - figurefigcase2 ] ] case @xmath232 , @xmath450 ( figure [ fig : case2 ] ) : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + consider iteration @xmath467 of the for - loop . since @xmath468 , there is a portal @xmath469 such that @xmath470 ( see step @xmath301 of the preprocessing ) . if @xmath471 then @xmath472 otherwise , there is a vertex @xmath473 such that the path in @xmath18 from @xmath474 to @xmath475 contains @xmath370 . hence , @xmath476 it follows from the above that there is a vertex @xmath477 such that @xmath478 . thus , for one of the two choices of @xmath396 in line @xmath244 , we have @xmath479 for that choice of @xmath396 , let @xmath480 be the portal in @xmath481 such that @xmath482 . the path in @xmath89 consisting of edges @xmath463 ( added in line @xmath233 ) and @xmath483 ( added in line @xmath249 ) followed by the path in @xmath18 from @xmath480 to @xmath59 has weight at most @xmath484 to show lemma [ lem : phaseii ] , note that @xmath485 is at most @xmath486 for any @xmath41 less than some positive constant . as for query time , lines @xmath301 and @xmath232 can be executed in @xmath302 time . by lemma [ lem : phasei ] , adding edges to @xmath89 in line @xmath233 can be done in @xmath487 time . the total time to find successors and predecessors over all iterations of the for - loop is @xmath488 . the additional time spent in the for - loop is bounded by the number of edges added to @xmath89 . the total number of edges added in line @xmath249 is @xmath489 . note that every @xmath113 considered in line @xmath241 is not included in @xmath313 in line @xmath490 . hence , we add a total of @xmath491 edges in line @xmath241 . in this section , we show how to obtain an approximate @xmath28-distance within the time and space stated in theorem [ thm : main ] , given the output of phase ii . we execute phase ii for both @xmath19 and @xmath30 , getting outputs @xmath492 and @xmath493 , respectively . to find an approximate @xmath28-distance , assume first that @xmath494 . let @xmath91 be one of the two shortest paths from @xmath66 in @xmath18 bounding @xmath29 and sort the vertices of @xmath495 along @xmath91 . to do the sorting efficiently , we make a dfs traversal of @xmath18 during preprocessing and label each vertex with an integer time stamp denoting when it was first visited in the traversal . now , sorting the vertices of @xmath495 along @xmath91 corresponds to integer sorting their precomputed labels . with the algorithm of han and thorup @xcite , this takes @xmath496 time . we then remove all @xmath497 for which there is another @xmath498 such that @xmath499 . this can be done in @xmath500 time with two linear scans over @xmath495 , one in sorted order and the other in reverse sorted order ; the first resp . second scan removes @xmath59 if there is a @xmath398 before resp . after @xmath59 in the order considered such that the inequality holds . let @xmath501 be the resulting subset and form a similar subset @xmath502 of @xmath503 . let @xmath504 be the minimum of @xmath505 over all pairs @xmath506 where @xmath507 $ ] has no interior vertices belonging to @xmath508 . this takes @xmath500 time . compute a similar value @xmath509 for the other shortest path @xmath510 of @xmath18 bounding @xmath29 . the approximate distance output is @xmath511 . we need to show @xmath512 . pick @xmath349 such that @xmath513 and assume that @xmath514 ; the case where @xmath515 is symmetric . by lemma [ lem : phaseii ] , there are vertices @xmath516 and @xmath517 such that @xmath518 and @xmath519 . pick a pair @xmath520 belonging to @xmath507 $ ] such that they occur on this path in the order @xmath521 and such that @xmath522 $ ] has no interior vertices belonging to @xmath508 . then @xmath523 showing the desired . above , we assumed that @xmath524 . if this is not the case , we modify @xmath525 as follows . partition @xmath526 into maximal - size groups where in each group @xmath527 , all vertices @xmath59 have the same nearest neighbor @xmath398 in @xmath18 belonging to @xmath29 . replace vertices of @xmath527 by @xmath398 in @xmath525 and instead of approximate distances @xmath528 for @xmath529 , use instead @xmath530 for the approximate distance for @xmath398 . a similar update is done to @xmath503 , ensuring that @xmath524 . it is easy to see that the above analysis still carries through . we have shown theorem [ thm : main ] in the case where @xmath164 and @xmath531 are not on the same leaf - to - root path in @xmath81 . if instead , say , @xmath532 then @xmath533 and phase i and ii for @xmath19 gives a portal set of @xmath29 . our algorithm above is modified to find the portal @xmath113 nearest to @xmath30 on @xmath29 and outputs @xmath534 , giving the desired stretch . this shows theorem [ thm : main ] in the remaining case where @xmath164 and @xmath531 are on the same leaf - to - root path . we gave a @xmath2-approximate distance oracle for undirected @xmath0-vertex planar graphs and fixed @xmath1 with @xmath3 space and @xmath4 query time which improves the previous best query time - space product from @xmath8 to @xmath9 . we have not focused on preprocessing time . with a simple implementation , we should get near - quadratic preprocessing time and it is possible that the exact space - efficient oracle in @xcite can speed this up further to @xmath535 as the number of precomputed distances required by our oracle is only @xmath536 . with techniques from , e.g. , @xcite , we can likely get down to @xmath536 . the dependency on @xmath41 in the query time - space product is slightly worse ; it is roughly @xmath48 ( @xmath46 when @xmath47 ) compared to @xmath49 in @xcite and roughly @xmath45 in @xcite ( where the latter has a slightly worse dependency on @xmath0 than @xcite ) . using mainly monge properties , we believe it should be possible to replace at least one @xmath45 factor by @xmath37 . getting @xmath537 query time and @xmath538 space for some constant @xmath539 seems problematic with our techniques due to the @xmath540 bottleneck from the use of veb trees when answering queries . 99 s. chechik . approximate distance oracle with constant query time . stoc@xmath541 , pp . p. erds . extremal problems in graph theory . in theory of graphs and its applications ( proc . smolenice , @xmath544 ) , czechoslovak acad . prague , @xmath545 , pp . m. l. fredman and d. e. willard . surpassing the information theoretic bound with fusion trees . journal of computer and system sciences , @xmath548(@xmath233):@xmath549@xmath550 , @xmath551 . y. han and m. thorup . integer sorting in @xmath552 expected time and linear space . @xmath553rd annual symposium on foundations of computer science ( focs ) , pp . @xmath554@xmath555 , @xmath7 . k. kawarabayashi , p. n. klein , and c. sommer . linear - space approximate distance oracles for planar , bounded - genus , and minor - free graphs . in @xmath556th international colloquium on automata , languages and programming ( icalp ) , pp . @xmath554@xmath557 , @xmath558 . k. kawarabayashi , c. sommer , and m. thorup . more compact oracles for approximate distances in undirected planar graphs . in @xmath559th acm - siam symposium on discrete algorithms ( soda ) , @xmath560@xmath561 , @xmath562 . p. n. klein . preprocessing an undirected planar network to enable fast approximate distance queries . in @xmath563th acm - siam symposium on discrete algorithms ( soda ) , @xmath564@xmath565 , @xmath7 . r. j. lipton and r. e. tarjan . a separator theorem for planar graphs siam journal on applied mathematics , @xmath547(@xmath232):@xmath566@xmath567 , @xmath568 . m. mendel and a. naor . ramsey partitions and proximity data structures . journal of the european mathematical society , @xmath436(@xmath232):@xmath569@xmath570 , @xmath571 . announced at focs@xmath572 . s. mozes and c. sommer . exact distance oracles for planar graphs . in @xmath573rd acm - siam symposium on discrete algorithms ( soda ) , @xmath574@xmath575 , @xmath576 . m. ptracu and l. roditty . distance oracles beyond the thorup - zwick bound . in @xmath577st ieee symposium on foundations of computer science ( focs ) , @xmath578 . m. ptracu and m. thorup . time - space trade - offs for predecessor search . @xmath556th annual acm symposium on theory of computing ( stoc ) , pp . @xmath579@xmath580 , @xmath581 . m. thorup . compact oracles for reachability and approximate distances in planar digraphs . journal of the acm , @xmath577(@xmath244):@xmath582@xmath583 , @xmath6 . announced at focs @xmath5 . m. thorup and u. zwick . approximate distance oracles . journal of the acm , @xmath584(@xmath301):@xmath301@xmath559 , @xmath585 . announced at stoc@xmath586 . d. e. willard . log - logarithmic worst - case range queries are possible in space @xmath587 . information processing letters , @xmath588(@xmath232):@xmath589@xmath590 , @xmath591 . c. wulff - nilsen . algorithms for planar graphs and graphs in metric spaces . phd thesis , university of copenhagen , @xmath578 . c. wulff - nilsen . approximate distance oracles with improved preprocessing time . in @xmath573rd acm - siam symposium on discrete algorithms ( soda ) , pp . @xmath592@xmath593 , @xmath576 . c. wulff - nilsen . approximate distance oracles with improved query time . in @xmath559th acm - siam symposium on discrete algorithms ( soda ) , @xmath594@xmath595 , @xmath562 .
we have obtained spectra through small apertures centered on the nuclear region and major axis of m32 , with the faint object spectrograph ( fos ) on the hubble space telescope ( hst ) . a detailed analysis and reduction of the data is presented , including : ( i ) new calibrations and modeling of the fos aperture sizes , point - spread - function and line - spread - functions ; ( ii ) determination of the aperture positioning for each observation from the observed count rate ; and ( iii ) accurate wavelength calibration , template matching , and kinematical analysis of the spectra . this yields measurements of the stellar rotation velocities and velocity dispersions near the center of m32 , with five times higher spatial resolution than the best available ground - based data . the inferred velocities provide the highest angular resolution stellar kinematical data obtained to date for any stellar system . the hst observations show a steeper rotation curve and higher central velocity dispersion than the ground - based data . the rotation velocity is observed to be at from the nucleus . this is roughly twice the value measured from the ground at this distance . the nuclear dispersion measured through the smallest fos aperture ( square ) is . the average of four independent dispersion measurements at various positions inside the central is , with a rms scatter of . the nuclear dispersion measured from the ground is only 8595 , whereas the dispersion outside the central arcsec is only . these results significantly strengthen previous arguments for the presence of a massive nuclear black hole in m32 . detailed dynamical models are presented in a series of companion papers . # 1#2
astro-ph9702147
astronomers have sought for two decades for dynamical evidence for the presence of massive black holes in galaxies by studying the dynamics of gas and stars in their nuclei . rapid motions provide the main signature of a black hole . if these are indeed observed , the main difficulty is to rule out alternative interpretations . gas motions can be due to hydrodynamical processes ( inflow , outflow , turbulence , etc . ) in addition to gravity . large stellar velocities in galactic nuclei can be the result of an overabundance of stars on radial orbits . the primary observational tool to discriminate between different models is to obtain data of the highest possible spatial resolution . in recent years much progress has been made in all areas of this field . for active galaxies , the existence of a central dark mass in at least some galaxies is now well established . the rotation velocities of nuclear gas disks detected with the hst can be used to study the central mass distribution . this has yielded evidence for a central dark mass of @xmath9 in m87 ( ford 1994 ; harms 1994 ) , and of @xmath10 in ngc 4261 ( ferrarese , ford & jaffe 1996 ) . even higher spatial resolution vlba radio observations of water masers in the nucleus of the active galaxy ngc 4258 ( miyoshi 1995 ) have revealed a torus in keplerian rotation around a dark mass of @xmath11 . for the case of ngc 4258 , there are strong theoretical arguments that this mass is indeed a black hole ( maoz 1995 ) . the density of quasars at high redshifts suggests that many currently normal galaxies had an active phase in the past ( chokshi & turner 1992 ; haehnelt & rees 1993 ) . hence , black holes are believed to be common in quiescent galaxies as well . in these galaxies only stellar kinematics are generally available to study the nuclear mass distribution . the evidence for a black hole in our own galaxy is now very strong , due to proper motion measurements for individual stars near sgr a@xmath12 ( eckart & genzel 1996 , 1997 ) . for a handful of other , nearby galaxies , evidence for a central dark mass was obtained from ground - based measurements of line - of - sight velocities ( see kormendy & richstone 1995 for a review ) , but it remained difficult to rule out all alternative models . it was always foreseen to be a main task for the hst to improve the evidence , by providing spectra of superior spatial resolution . stellar kinematical studies with hst became possible after the refurbishment mission in 1993 . the first results were presented by kormendy ( 1996a , b ) , for ngc 3115 and ngc 4594 . spectra near the nucleus with @xmath13 resolution confirmed previous arguments for black holes of @xmath14 and @xmath15 , respectively . the quiescent e3 galaxy m32 has long been one of the best - studied black hole candidate galaxies . the presence of a central dark mass of @xmath16@xmath17 has been argued on the basis of ground - based data with continuously increasing spatial resolution . the most recent work indicates a central dark mass of @xmath18@xmath19 ( van der marel et al . 1994b ; qian et al . 1995 ; dehnen 1995 ; bender , kormendy & dehnen 1996 ) . however , the best ground - based kinematical data still have a spatial resolution of ` only ' @xmath20 . goodman & lee ( 1989 ) showed that this is insufficient to rule out a cluster of dark objects , as opposed to a central black hole , on the basis of theoretical arguments . in addition , none of the previous dynamical studies has considered axisymmetric stellar dynamical models with fully general phase - space distribution functions . hence , it has not been shown convincingly that no plausible model can be constructed that fits the data without requiring a central dark mass . higher spatial resolution data are therefore highly desirable . this paper is part of a series in which we present the first hst spectra and new dynamical models for m32 . we obtained spectra of the nuclear region with the hst / fos , and we give here a detailed description of the acquisition and analysis of these data . dynamical models are presented in van der marel ( 1997b ) and cretton ( 1997 ) . the main results of the project are summarized and discussed in van der marel ( 1997a ) . the instrumental resolution of the fos has a gaussian dispersion of @xmath21 , while the wavelength scale can vary at the @xmath22 level from orbit to orbit . a study of a low mass galaxy such as m32 ( the main body of which has a velocity dispersion of @xmath23 and a rotation velocity amplitude of @xmath24 ) is therefore significantly more complicated observationally , than that of more massive galaxies ( which have higher dispersions ) . an additional complication is that the ` sphere of influence ' of the suspected black hole in m32 is smaller than that of most other candidate galaxies . it is thus necessary to use the smallest fos apertures , for which it is more difficult to do an accurate target acquisition . in addition , for a proper interpretation of the results it is necessary to determine the aperture position for each observation _ post facto _ , with an accuracy of @xmath25 . to deal with these complications it proved necessary to study the instrument and analyze the data in more than the usual detail . parts of this paper are therefore of a somewhat technical nature . readers interested mostly in the stellar kinematical results may wish to skip directly to section [ ss : results ] . the paper is organized as follows . section [ s : obsetup ] summarizes the observational setup and strategy . section [ s : pos ] discusses the aperture positions for the observations . section [ s : datareduc ] describes some aspects of the data reduction . section [ s : lsf ] presents calculations of the line - spread function for each observation . section [ s : temp ] discusses the template spectrum used in modeling the m32 spectra . section [ s : kinematics ] discusses the kinematical analysis . section [ s : discussion ] summarizes the main observational results . in the appendices , new calibrations and modeling are presented of the fos aperture sizes , point - spread - function ( psf ) and line - spread - functions , for which no sufficiently accurate determinations were previously available . we observed m32 with the red side detector of the hst / fos for nine spacecraft orbits on august 22 , 1995 ( project go-5847 ) . the costar optics corrected the spherical aberration of the hst primary mirror . telescope tracking was done in ` fine lock ' . the rms telescope jitter was typically @xmath26 milli - arcsec ( mas ) . each orbit consisted of @xmath27 minutes of target visibility time , followed by @xmath28 minutes of earth occultation . the first two orbits were used to accurately acquire the nucleus of m32 . subsequently , spectra were taken at various positions near the nucleus . the g570h grating was used in ` quarter - stepping ' mode , yielding spectra with 2064 pixels covering the wavelength range from 4572 to 6821 . periods of earth occultation were used to obtain wavelength calibration spectra of the arc lamp . in the last orbit of the observations the fos was used in a special mode to obtain an image of the central arcsec of m32 , to verify the telescope pointing . two different square apertures were used to obtain the spectra : the ` 0.1pair ' and the ` 0.25pair ' . these are the smallest apertures available on the fos . although they are paired ( two square openings separated by several arcsec ) , they were used exclusively in single - aperture mode . all data were collected through the upper aperture . the aperture names are based on their size in arcsec _ before _ the installation of costar . their nominal post - costar sizes are smaller : @xmath29 square and @xmath30 square , respectively . evans ( 1995a ) presented calibration observations in which the light throughput was measured for a point source as function of position in the apertures . models for these observations presented in appendix [ s : appa ] indicate somewhat smaller aperture sizes : @xmath2 square and @xmath31 square , respectively . these sizes are adopted in the remainder of the paper . the natural coordinate system of the fos is the cartesian @xmath32 system defined in the fos instrument handbook ( keyes 1995 ) . the apertures have their sides parallel to the axes of this system . the grating disperses the light in the @xmath33-direction . the @xmath32 system is ` left - handed ' when projected onto the sky . in the following we adopt a more convenient ` right - handed ' system @xmath34 , defined through @xmath35 and @xmath36 . during the observations , the direction of the @xmath37-axis was fixed at a position angle of @xmath38 on the sky . this coincides with the major axis position angle of m32 ( cf . lauer 1992 ) . the position of the m32 center in the hst guide star coordinate system is ra @xmath39 0h 42 m 41.82s , @xmath40 . the positional uncertainty is dominated by that in the hst guide star catalog itself , which is @xmath20 rms . hence , some form of acquisition is required to properly position the galaxy in the small apertures . for an extended target with a sharp and well defined brightness maximum , such as m32 , the method of choice is a so - called ` peak - up ' acquisition , which consists of a number of ` stages ' . each stage adopts a rectangular grid of @xmath41 points on the sky , with inter - point spacings of @xmath42 and @xmath43 , respectively . an fos aperture is positioned at each of the grid points , and the total number of counts is measured in some exposure time . the grid point with the most counts is adopted as new estimate of the target position . subsequent stages in a sequence use smaller and smaller apertures , with each stage increasing the accuracy of the target positioning . various standard choices exist for the number of stages , and for the aperture , the grid and the exposure time for each stage in the acquisition ( keyes 1995 ) . these standard sequences work well for point - source targets . however , they have not been particularly well tested for extended sources , especially not for acquisitions into the small 0.1-pair aperture . therefore , as part of the preparation of the observations , a software package was developed to simulate fos peak - up acquisitions of arbitrary targets in a monte - carlo manner ( van der marel 1995 ) . guided by simulations done with this package , a non - standard 5-stage peak - up sequence was constructed for m32 . the parameters of this sequence are listed in table [ t : peakup ] . the sequence was executed with the g570h grating in place , because m32 is too bright to be acquired with the fos mirror . the accuracy of the sequence is such that in an idealized situation ( no poisson noise , no telescope jitter , etc . ) @xmath44 and @xmath45 , where @xmath46 is the difference between the adopted pointing at the end of the acquisition and the true position of the galaxy center . the observed intensities in the acquisition stages can be analyzed _ post facto _ , to determine the extent to which the acquisition was successful . most important are the results of the fifth and final stage , in which the 0.1-pair aperture was sequentially positioned at the points of a @xmath47 grid on the sky , with spacing of @xmath48 between the grid points . the left panel of figure [ f : peakup5 ] shows a grey - scale representation of the observations . the solid dot marks the position with the highest intensity , which was adopted by the telescope software as its best estimate for the position of the galaxy center . to model the observed intensities , the ` cusp model ' for the unconvolved m32 surface brightness presented by lauer ( 1992 ) was used , which is based on pre - costar hst / wfpc images . the point - spread - function ( psf ) of the hst+fos was approximated by a sum of gaussians , as determined in appendix [ s : appa ] . the total magnitude at each grid point was calculated using the equations in the appendix . the offset @xmath46 was varied to optimize the fit . the best fit is displayed in the right panel of figure [ f : peakup5 ] . it has @xmath49 , with a formal error in either coordinate ( based on the chi - squared surface of the fit ) of approximately 1 mas . the target acquisition was thus successful . an fos image of the central region of m32 was obtained in the final orbit of the observations . the telescope was commanded to position the upper ` 1.0-pair ' aperture , which has a nominal post - costar size of @xmath50 square , on the galaxy center , after which the intensity on the photocathode was scanned with the diode array of the detector ( with no grating in the light path ) . the scan pattern was chosen to provide @xmath51 pixels . however , the spatial resolution of the image is poor , @xmath52 , corresponding to the size of one fos diode ( koratkar 1994 ; evans 1995b ) . the resolution can be improved by deconvolution . figure [ f : grey_image ] shows the result of lucy deconvolving the raw image with a boxcar psf with the size of one diode . the cross marks the galaxy center , while the dot marks the aperture center , i.e. , the position where the telescope thought the galaxy center would be . the latter is offset from the actual galaxy center by @xmath53 ( with a formal error of 3 mas in each coordinate ) . this offset at the end of the observations is much larger than that achieved by the target acquisition . hence , positional errors must have accumulated during the observations . the reason for this is not well understood . however , the most likely cause is a drift of the telescope pointing due to thermal effects related to the heating and cooling of the spacecraft and fine guidance sensors ( fgss ) during the 14.4 hours of the observations . such thermal effects exist ( lupie , priv . comm . ) , but were not previously reported to complicate observations with the fos ( keyes , priv . comm . ) . nine spectra were taken . those obtained with the 0.1-pair aperture are referred to as s1s4 , those obtained with the 0.25-pair aperture as l1l5 . table [ t : obsetup ] summarizes the observational setup for each spectrum . it also lists the position with respect to the galaxy center at which the telescope was instructed to place the aperture . these _ intended _ aperture placements are illustrated in the left panel of figure [ f : aper_plot ] . in reality , the apertures were not placed exactly at their intended positions . the actual position is a sum of two vectors : ( i ) an intentional offset from the ( telescope s estimate of the ) galaxy center ( table [ t : obsetup ] ) ; and ( ii ) an unintentional offset @xmath46 of the telescope s estimate of the galaxy center from the actual galaxy center . the intentional offsets are applied by slewing the telescope , which should be very precise ( @xmath54 mas positional error ) . to determine the actual aperture positions one thus needs to know the offset @xmath46 for each observation . these can be determined by modeling the observed intensities of the spectra , which are listed in table [ t : applaces ] . the spectra s1 and s2 , taken in the subsequent orbits # 3 and # 4 , were scheduled to have identical aperture positions ( cf . table [ t : obsetup ] ) . the same holds for the observations l4 and l5 , taken in the subsequent orbits # 8 and # 9 . in both cases , however , the observed intensities in the two spectra are significantly different . this indicates that the aperture positions must have changed ( by @xmath55 mas , cf . figure [ f : telpoint ] below ) between the orbits # 3 and # 4 , and between the orbits # 8 and # 9 . there was no motion of either the grating wheel or the aperture wheel of the instrument between the different observations . the guide star reacquisition at the beginning of each orbit is normally accurate to a few mas . hence , the inferred pointing drifts are indeed most likely due to thermal effects on the fgs . to proceed we make a number of simplifying assumptions about the offset @xmath46 : ( i ) the offset for the first spectrum , s1 , taken in orbit # 3 , was identical to the offset determined from the final peak - up stage ( figure [ f : peakup5 ] ) , executed in orbit # 2 ; ( ii ) the offset for the last spectrum , l5 , taken in orbit # 9 , was identical to the offset determined from the fos image ( figure [ f : grey_image ] ) , taken in that same orbit ; ( iii ) the observations s2s4 , taken in orbits # 4 , # 5 and # 6 , have a common offset ; and ( iv ) the observations l1l4 , taken in orbits # 7 and # 8 , have a common offset . these assumptions do not allow for positional errors between orbits # 4 and # 5 , between orbits # 5 and # 6 , and between orbits # 7 and # 8 . even though errors could have occurred , they are not required to fit the observed intensities . the assumptions do allow for an error between orbits # 6 and # 7 , when the 0.1-pair aperture was replaced by the 0.25-pair aperture . this is because these apertures might not be exactly concentric on the aperture wheel ( evans 1995 ) , and because the motion of the aperture wheel might have a positional non - repeatability ( dahlem & koratkar 1994 ) . with these assumptions , the offset @xmath46 can be determined for each observation by fitting to the observed intensities , using the lauer ( 1992 ) cusp model for the m32 surface brightness , and the aperture sizes and psf determined in appendix [ s : appa ] . the results are displayed in figure [ f : telpoint ] . the corresponding aperture placements for the individual spectra are listed in table [ t : applaces ] and are displayed in the right panel of figure [ f : aper_plot ] . the fits to the intensities constrain only the absolute value of @xmath56 , and not its sign . this introduces an ambiguity for l1l4 , which was resolved by adopting the same sign for @xmath56 as measured in figure [ f : grey_image ] ( so as to minimize the size of the positional error between orbits # 8 and # 9 ) . at the positions of the apertures ( along the major axis ) , the surface brightness changes more rapidly along the @xmath37 direction than along the @xmath57-direction hence , the observed intensities constrain the @xmath58 more tightly than the @xmath56 . other model assumptions to fit the observed intensities were also studied . in all cases the results for @xmath58 agreed to within a few mas , while the @xmath56 could differ by as much as @xmath59 . most of the necessary data reduction steps are performed by the hst calibration ` pipeline ' . for the m32 data , three issues required additional attention : ( i ) wavelength calibration ; ( ii ) absolute sensitivity calibration ; and ( iii ) flat - fielding . the wavelength scale provided by the calibration pipeline is not accurate enough for our project . arc lamp spectra were therefore obtained in each orbit during occultation . the emission line centers were determined , yielding for each arc spectrum @xmath60 and for each line @xmath61 with vacuum wavelength @xmath62 , the observed wavelength @xmath63 on the scale provided by the calibration pipeline . subsequently , the observed offsets @xmath64 were fit as @xmath65 the constant offset @xmath66 is the result of non - repeatability in the fos grating and aperture wheels , and was measured to be @xmath67 . the third order polynomial @xmath68 with zero mean accounts for a slight ( @xmath69 ) non - linearity of the wavelength scale . the offsets @xmath70 account for constant shifts of the wavelength scale from orbit to orbit , mostly due to the ` geo - magnetically induced image motion problem ' ( keyes 1995 ) . these shifts ranged between @xmath71 , and could be determined with @xmath72 ( @xmath73 ) accuracy . the wavelength scale of the m32 spectra was corrected using the above fits to the arc spectra . an additional shift of @xmath74 was applied to correct for the fact that the light path for external targets differs from that for the internal arc lamp ( keyes 1995 ) . for each m32 spectrum the @xmath70 were used that was / were determined from the arc spectrum / spectra obtained immediately preceding or following the m32 spectrum . the fact that the arc spectra are never obtained completely simultaneously with galaxy spectra introduces a slight additional uncertainty . the cumulative uncertainty in the mean streaming velocities of the stars as a result of uncertainties in the wavelength scale is estimated to be @xmath75 . this is always smaller than the formal errors in these velocities as a result of photon noise ( cf . table [ t : kinresults ] below ) . after wavelength calibration ( and absolute sensitivity calibration and flat - fielding as described below ) the galaxy spectra were rebinned logarithmically in wavelength , as required for the stellar kinematical analysis . a scale of @xmath76pixel was used . the fos is a photon counting detector , so the formal errors in the spectra follow directly from poisson statistics . on average , the ( logarithmic ) rebinning decreases the formal errors , since it reduces the scatter between neighboring pixels . the decrease can be calculated for each pixel , and was taken into account in the subsequent analysis . the correlation between the errors in neighboring pixels induced by the rebinning was neglected . the calibration pipeline multiplies the observed count rate spectra by a so - called ` inverse sensitivity file ' ( ivs ) , to obtain fluxes in @xmath77@xmath78 . the pipeline ivs files contain a wavelength dependent aperture throughput correction , based on calibration observations of point sources . these , however , are not applicable to extended sources . therefore , instead of the standard pipeline ivs file a more recent ivs file with no aperture throughput correction was used , provided to us by fos instrument scientist tony keyes . each wavelength in the m32 spectra samples a slightly different region of the galaxy , because of the wavelength dependence of the psf . this influences only the continuum slope of the spectra , because the psf varies very slowly with wavelength . in the stellar kinematical analysis one is only interested in the absorption lines , which are not influenced . the underlying continuum is subtracted . hence , no attempt was made to construct aperture throughput corrections . ( in fact , this would require accurate knowledge of the wavelength dependence of the psf , which is not readily available for the fos ; see appendix [ s : appa ] ) . a g570h flat - field based on multiple 0.25-pair ( upper ) aperture observations of a star was used , as provided to us by tony keyes . no flat - fields obtained explicitly with the 0.1-pair ( upper ) aperture were available . a star illuminates the photocathode of the detector differently than an extended source . to test the appropriateness of the flat - field for the m32 data it was cross - correlated with the continuum subtracted normalized galaxy spectra . this yielded clear peaks , indicating that the flat - field and the m32 spectra share the same features ( as they should ) . the peak was only @xmath7910% lower for the 0.1-pair spectra than for the 0.25-pair spectra . this indicates that the flat - field can be properly used for the 0.1-pair ( upper ) aperture as well , although a flat - field obtained specifically with that aperture would have been preferable . shifts between the galaxy spectra and flat - field were identified from the positions of the cross - correlation peaks , and were corrected for ( see also kormendy 1996a ) . the flat - fielding removes most of the pixel - to - pixel sensitivity variations , but some small residual variations might remain . the observed spectrum of a source is the convolution of its actual spectrum with the line - spread - function @xmath80 . the lsf can be written as the convolution of an ` illumination function ' @xmath81 and an ` instrumental broadening function ' @xmath82 ( cf . appendix [ s : appb ] ) . the function @xmath81 is the normalized intensity distribution of the light that falls onto the grating . it depends on the target brightness and choice of aperture . the direction of dispersion is parallel to the @xmath57 direction . hence , @xmath81 is the integral over the @xmath37 direction of the two - dimensional brightness distribution that falls onto the grating . for the psf and aperture sizes derived in appendix [ s : appa ] , @xmath81 is given by equations ( [ blamdef ] ) and ( [ bkernel ] ) in appendix [ s : appb ] . it can be calculated for each observation upon substitution of the lauer ( 1992 ) cusp model for the m32 surface brightness , and the aperture positions in table [ t : applaces ] . the normalized function @xmath82 accounts for the instrumental broadening due to the grating and finite size of a detector diode ( the resolution element ) . it was determined empirically from fits to the emission line shapes in the arc spectra , as discussed in appendix [ s : appb ] . the lsf has zero mean if the galaxy light is distributed symmetrically within the aperture . this is not generally the case , because the aperture centers are not exactly at @xmath83 ( figure [ f : aper_plot ] ) . calculations show that the intensity weighted mean position is within @xmath84 from the aperture center for all the observations . nonetheless , there are noticeable wavelength shifts , because @xmath84 projects onto @xmath85 in the wavelength direction ( @xmath86 at 5170 ) . table [ t : applaces ] lists the lsf mean as calculated for each of the m32 spectra , both in and in @xmath6 at 5170 . errors in these values due to errors in the aperture positions do not exceed 0.04 ( @xmath87 at 5170 ) . the width and shape of the lsf are determined mainly by the aperture size . the calculated lsfs for all the m32 spectra are shown in figure [ f : lsf ] . the lsf shapes for the 0.1-pair observations are very similar to each other , as are the lsf shapes for the 0.25-pair observations . the shapes are also similar to the emission line shapes observed directly in arc spectra ( figure [ f : arc ] ) . the 0.25-pair lsf can be well approximated by convolving the 0.1-pair lsf by a gaussian with a dispersion of @xmath88 ( @xmath89 at 5170 ) , as illustrated by the dotted curve in figure [ f : lsf ] . the best - fitting gaussian to the 0.1-pair lsf has a dispersion of @xmath90 ( @xmath91 at 5170 ) . for the 0.25-pair lsf it has a dispersion of @xmath92 ( @xmath93 at 5170 ) . however , these numbers are of limited use in characterizing the lsfs , since these have noticeably broader wings than a gaussian . for stellar kinematical analysis a template spectrum is needed to compare the m32 spectra to . to avoid systematic errors it is important to minimize template mismatching . this is best done by constructing a composite template which contains an appropriate mix of spectral types . observing template stars with the hst is inefficient and time consuming ( mainly because of the lengthy target acquisitions that are required ) . to date less than a handful different template stars have been observed with the hst , all of similar spectral type . these hst spectra are insufficient to construct an optimal template . a ground - based template library of spectra of 27 stars of different spectral types was therefore used , obtained in february 1990 by m. franx at the 4 m telescope of the kpno with the rc spectrograph ( as discussed previously by van der marel & franx 1993 ) . the lsf of these spectra is approximately gaussian with a dispersion of @xmath94 ( @xmath95 at 5170 ) . the spectra cover the spectral range from 4836 to 5547 , centered on the mg b triplet at @xmath96 . this is the most useful wavelength range for stellar kinematical analysis , so it is no drawback that this range is smaller than the full range covered by the fos g570h grating . the stellar spectra were similarly rebinned logarithmically as the galaxy spectra , and shifted to a common velocity . no attempt was made to construct a different composite template for each m32 spectrum ( which is reasonable , given that there are no strong color gradients in the central arcsec ; lugger 1992 ) . instead , the best fit was sought to a grand - total m32 spectrum , constructed by summing the hst spectra . the resulting composite template is a weighted mix of the individual templates , with the weights determined using the method outlined in appendix a.3 of van der marel ( 1994a ) . the mix contains giants , sub - giants and dwarfs of spectral types g and k. the composite template is used in the remainder of the paper . however , other templates were studied as well . for example , a stellar kinematical analysis was performed with a k - star template spectrum obtained from the hst data archive , taken with the fos circular 0.3 aperture ( @xmath97 diameter ) by h. ford and collaborators before the installation of costar . this template provides a poor fit to the spectrum of m32 . nonetheless , the stellar kinematical properties inferred with this template were found to be consistent with those derived using the composite template . the stellar kinematical analysis was performed with the method presented in van der marel ( 1994a ) . it fits the convolution of a parametrized velocity profile and a template spectrum to a galaxy spectrum in pixel space , using chi - squared minimization . the formal errors in the fit parameters follow from the shape of the chi - squared surface near its minimum . the method has been well tested , and its results agree with those from other methods for extracting stellar kinematics from galaxy spectra . deviations of the line - of - sight velocity profiles from gaussians contain useful information on the dynamical structure of galaxies . unfortunately , the velocity resolution of the fos is too poor to extract any reliable velocity profile shape information from the m32 data . the analysis was therefore restricted to gaussian velocity profile fits . data obtained with the fos is time resolved . the red side detector reads and stores the data every @xmath98 minutes . between 6 and 20 spectra were therefore available per aperture position . the aperture positions for the observations l4 and l5 differ by @xmath99 , but both are at a distance of @xmath20 from the galaxy center , where kinematical gradients are small . in the kinematical analysis l4 and l5 were therefore treated as a single observation , referred to as l45 . especially for the 0.1-pair observations , the signal - to - noise ratio ( s / n ) of the individual read - outs is low . individual read - outs for a given aperture position were therefore added to achieve an average s / n of @xmath100 per ( logarithmically rebinned ) pixel . this resulted in 4 independent spectra for each of the observations s1s4 with the 0.1-pair aperture , and between 9 and 11 independent spectra for the observations l1 , l2 , l3 and l45 with the 0.25-pair aperture . a stellar kinematical analysis was performed on each of these spectra . the kinematical results for each aperture position were then averaged together , weighted with the errors . it was found that this yields slightly more accurate results than a kinematical analysis of the sum of the individual spectra for a given aperture position . no systematic trends with time in the orbit were found in the kinematical results . the stellar kinematical analysis was performed over the wavelength range 48595520 . regions influenced by so - called ` noisy ' diodes were masked . the fit to the galaxy spectrum includes a low order polynomial to account for continuum differences between the galaxy and template spectrum . this polynomial is fit simultaneously with the velocity profile . the parameters of the best - fitting velocity profile are virtually independent of the choice of polynomial order . a polynomial of order 5 was used . the mean velocity and velocity dispersion of the best - fitting gaussian velocity profiles must be corrected for instrumental effects . the mean velocity is biased because the lsf mean is non - zero . this bias is removed by subtracting the corrections @xmath101 listed in table [ t : applaces ] . these were calculated for the wavelength 5170 of the mg b triplet , but the wavelength dependence of the velocity corrections over the fit range is negligible ( @xmath102 in absolute value ) . further , to obtain velocities in the m32 frame one must subtract a constant offset , determined by the difference between the systemic velocity of m32 and the template velocity . these were not known accurately enough to determine the offset directly , and it was therefore estimated from the data itself . somewhat arbitrarily , it was fixed to the intercept of the best least - squares fit line in a plot ( see figure [ f : allres]c below ) of rotation versus position @xmath37 along the m32 major axis for the observations s4 , s1 and s2 ( for which @xmath103 , @xmath104 and @xmath105 , respectively ) . the velocity dispersions @xmath106 of the best - fitting gaussians must be corrected for differences in the lsfs of the template and galaxy spectra . for this it is useful to consider not the observed dispersions themselves , but rather differences in the observed dispersions between two observations . let @xmath107 and @xmath108 be the best - fitting dispersions for observations at positions a and b. assume that the lsf of observation b can be obtained from the lsf of observation a through convolution with a gaussian of dispersion @xmath109 . the difference between the actual dispersions at positions a and b can then be estimated as : @xmath110 ( which uses the fact that the convolution of two gaussians is again a gaussian , the dispersion of which is the rms sum of the input dispersions ) . this approach has two important advantages . first , the lsf of the template influences the kinematical analysis at points a and b in the same way , so it does not enter into @xmath111 . second , the differences in the lsfs for the hst observations can in fact to good approximation be accounted for by convolution with a gaussian ( cf . figure [ f : lsf ] ) . one does _ not _ have to assume that the lsfs themselves are gaussian , which would be a poor approximation . in the following , the observation l45 at @xmath20 from the galaxy center is used as ` reference point ' . equation ( [ sigdef ] ) then yields @xmath112 where @xmath113 for an observation with the 0.25-pair aperture , and @xmath114 for an observation with the 0.1-pair ( based on the results of section [ s : lsf ] ) . the latter value was calculated for the wavelength 5170 of the mg b triplet , which is roughly at the center of the fit range . the actual value of @xmath115 varies from @xmath116 to @xmath117 over the fit range , which can be taken into account by assigning a ` formal ' error of @xmath118 to @xmath109 . equation ( [ sigcor ] ) can be used to estimate the stellar velocity dispersion @xmath119 for any of the observations s1s4 and l1l3 , if one assumes that the dispersion @xmath120 at the position of observation l45 is known a priori . at @xmath121 along the major axis one is beyond the region most influenced by seeing in ground - based data . based on the data of van der marel ( 1994a ) and bender , kormendy & dehnen ( 1996 ) it was assumed that @xmath122 ( see figure [ f : allres]d below ) . it is demonstrated in appendix [ s : appc ] that the velocity dispersions thus obtained have no systematic biases , and that even velocity dispersions as small as @xmath23 can be measured reliably , in spite of the large fos instrumental dispersion of @xmath123 . the formal errors in dispersions @xmath119 obtained from equation ( [ sigcor ] ) follow from the errors in @xmath106 , @xmath124 , @xmath125 and @xmath126 using a standard first order error analysis . systematic errors due to template mismatching are not likely to exceed a few @xmath6 , and are smaller than the formal errors in the measurements . table [ t : kinresults ] lists the line strengths , rotation velocities and velocity dispersions , after correction for all instrumental effects . these results are discussed below . figure [ f : allres ] displays the results as function of the position @xmath37 of the aperture center along the m32 major axis . the best spatial resolution ground - based m32 data are also shown . these are the data from van der marel ( 1994a ) obtained at the william herschel telescope ( wht ) , and those of bender , kormendy & dehnen ( 1996 ) obtained at the canada france hawaii telescope ( cfht ) . a comparison of these ground - based data to older ( lower spatial resolution ) ground - based data can be found in kormendy & richstone ( 1995 ) . the spatial resolution for each of the observations in figure [ f : allres ] is characterized by a convolution kernel , which results from the psf and finite aperture size ( for ground - based long - slit data the effective aperture is the spatial ccd pixel size by the slit width ) . the top left - panel illustrates the difference in spatial resolution between the data sets . it displays the projection of the convolution kernel along the m32 minor axis , i.e. , the probability @xmath127 that a photon observed in a given aperture was emitted at distance @xmath37 from the center of that aperture , measured along the m32 major axis . these probability distributions have fwhm values of @xmath128 , @xmath129 , @xmath130 and @xmath131 , for the hst / fos/0.1-pair , hst / fos/0.25-pair , cfht and wht data , respectively . the hst data therefore provide a factor 5 increase in spatial resolution over even the best available ground - based data . the line strengths shown in figure [ f : allres]b were normalized to an average of unity ( line strengths are measured with respect to the template spectrum , and are hence on an arbitrary scale ) . there is no obvious trend with radius . the observations with the 0.1-pair aperture closest to the nucleus have relatively low line strength , but this is unlikely to be real . metallicities and line strengths generally increase towards the centers of galaxies . the observed low values could be an artifact due to the presence of a black hole . the stars that move rapidly near the black hole produce broad velocity profile wings . these can be mistaken for an enhanced continuum , which causes the line strength to be underestimated ( van der marel 1994b ) . the broad - band colors of m32 are constant in the central arcsec ( lugger 1992 ) . the true line strengths in the central @xmath132 covered by the hst observations are therefore probably also close to constant . the observed line strengths have a sizable rms scatter of @xmath133 . this exceeds the formal errors , which are on average @xmath134 . the measurements are therefore not statistically consistent with a constant line strength ( as is confirmed by a chi - squared test ) . the scatter in the observed line strengths could be the result of minor inaccuracies in the data , such as unidentified bad pixels or flat - field errors , or it could be real , and due to shot noise in the contributions from individual stars . the total v - band luminosity inside the aperture varies from @xmath135 ( for observation s3 ) to @xmath136 ( for observation l3 ) . most of the observed light comes from giants , some of which can be as luminous as @xmath137 ( freedman 1989 ) . the luminosity of a ` typical ' giant as measured from surface brightness fluctuations ( tonry , ajhar & luppino 1990 ) is @xmath138 . this implies that there are @xmath139 typical stars in each hst aperture . therefore , @xmath140 fluctuations can be of order a few percent of the intrinsic line strength differences between stars . figure [ f : allres]c shows the rotation velocities . the hst rotation curve is smooth . it is significantly steeper in the central @xmath141 than seen in the ground - based data , consistent with the presence of a central black hole . the rotation velocity is observed to be @xmath142 at @xmath1 from the nucleus . this is roughly twice the value measured from the ground at this distance . the scatter in the line strength measurements ( figure [ f : allres]b ) induces scatter in the velocity dispersion measurements , because the velocity dispersion and line strength of a gaussian fit are statistically correlated . tests indicate that a line strength error @xmath143 induces an error @xmath144 in the velocity dispersion . rotation velocity measurements are not influenced . to avoid unwanted errors in the velocity dispersion measurements , they were determined by fitting gaussian velocity profiles to the data while keeping the line strength constant to the mean of the measurements in figure [ f : allres]b . this yields velocity dispersions that differ only at the @xmath145 level from those obtained by gaussian fits with variable line strength , so the main results do not depend on this correction . figure [ f : allres]d shows the resulting velocity dispersions . the dispersion measured with the 0.1-pair aperture closest to the nucleus is @xmath3 . the profile of the velocity dispersion with radius is smooth , with the exception of the measurement @xmath146 from observation s2 , which is significantly lower than that for the other observations in the central @xmath1 . we doubt the reality of this , but have not been able to attribute it to any known instrumental effect . the average of the four dispersion measurements inside @xmath1 is @xmath4 , with a rms scatter of @xmath5 . hence , the central velocity dispersion is definitely higher than the 8595@xmath6 determined from ground - based measurements . it thus appears that a keplerian increase in the velocity dispersion close to the center of m32 has been resolved . the difference in the dispersions measured for observations s1 and s2 , which were obtained with the same aperture at similar positions , indicates that uncertainties in the velocity dispersions are dominated by systematic errors , rather than poisson noise . systematic errors in the rotation velocity measurements are believed to be much smaller , primarily because these are less sensitive to the poor instrumental resolution of the fos . either way , the data clearly imply that the dispersion in the central @xmath1 is higher than measured from ground - based data . whether the nuclear dispersion seen through a small aperture is indeed as high as @xmath147 should await verification by observations with the future hst spectrograph stis . hst / fos spectra were obtained through small apertures centered on the nuclear region and major axis of m32 . kinematical analysis of these data is more complicated than it is for ground - based data : the fos is a low - resolution spectrograph , and is not particularly well suited for dynamical studies of low - dispersion objects such as m32 . one of the main results of this paper is therefore that stellar kinematical quantities _ can _ in fact be determined from the data , provided that a detailed study is made of various technical issues , including : calibration and modeling of the fos aperture sizes , psf and lsfs , determination of the aperture positions from the data , accurate wavelength calibration , template matching , and corrections for instrumental resolution . the resulting kinematical quantities are displayed in figure [ f : allres ] . the main result is that in the central @xmath1 the m32 rotation curve is significantly steeper , and the velocity dispersion significantly higher , than measured from even the highest resolution ground - based data . these observational results are robust . they do not depend critically on the corrections made in the analysis ; even a rough analysis without any template matching and corrections for instrumental effects shows these main features . the results are exactly what would be expected if m32 does indeed have a massive dark object in its nucleus , as suggested previously on the basis of ground - based data . in the companion papers in this series ( van der marel 1997a , b ; cretton 1997 ) we construct dynamical models to address the presence of such a dark object quantitatively . we also address the question whether the dark object has to be a black hole , or whether plausible alternatives still exist . the authors are grateful to the fos instrument scientists tony keyes , anuradha koratkar and michael dahlem for helpful discussions , and to bill workman and jean surdej for successful scheduling and implementation of the observations . tony keyes provided preliminary instrument calibration data and marijn franx provided the template library . simon white helped with the early stages of the project . support for this work was provided by nasa through grant number # go-05847.01 - 94a , and through a hubble fellowship # hf-1065.01 - 94a awarded to rpvdm , both from the space telescope science institute which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . some calibrations exist of the hst+fos psfs and lsfs ( koratkar 1996 ) and of the fos aperture sizes ( evans 1995 ) , but these are too crude for the purposes of this paper . to obtain more accurate calibrations we present detailed models for the calibration observations obtained by evans ( 1995a ) after the installation of costar . he centered a star in fos apertures , and stepped the star across each aperture in the fos @xmath57 and @xmath37 directions , respectively . at each step the total intensity was measured . here the results are considered for the fos red detector with three apertures : the square upper aperture of the 0.1-pair , the square upper aperture of the 0.25-pair , and the circular 1.0 . the names of these apertures are based on their size in arcsec , _ before _ the installation of costar . their nominal post - costar sizes are smaller by a factor @xmath148 . figure [ f : transmis ] shows for each aperture the observed intensity as function of distance from the aperture center , normalized by the observed intensity with the star at the center of the circular 1.0 aperture . the aperture length or diameter is easily determined from these data for apertures much larger than the psf core : it is then twice the radius at which the intensity has fallen to 50% of its central value . this was used by evans ( 1995 ) to estimate the size of all fos apertures . although this is reasonable for the larger apertures , it is not for the smaller apertures . for those a more careful analysis is required , in which the aperture sizes and psf are fitted to the data simultaneously . for a target with surface brightness @xmath149 , the intensity @xmath150 observed through an aperture centered on @xmath34 is @xmath151 where @xmath152 is a convolution kernel that depends on the psf and the aperture size and geometry . the psf is assumed to be a circularly symmetric sum of gaussians , @xmath153 , \label{psfgauss}\ ] ] where the @xmath154 must satisfy @xmath155 . for a perfectly rectangular aperture with sides @xmath156 and @xmath157 the convolution kernel is then @xmath158 -\mathop{\rm erf}\nolimits \left [ \fr{x-(a_x/2)}{\sqrt{2}\ > \sigma_i } \right ] \right \rbrace \ > \left \lbrace \mathop{\rm erf}\nolimits \left [ \fr{y+(a_y/2)}{\sqrt{2}\ > \sigma_i } \right ] -\mathop{\rm erf}\nolimits \left [ \fr{y-(a_y/2)}{\sqrt{2}\ > \sigma_i } \right ] \right \rbrace , \ ] ] while for a perfectly circular aperture with diameter @xmath66 it is @xmath159 \int_0^{d/2 } \exp \left [ - \fr{1}{2 } ( \fr{r}{\sigma_i } ) ^2 \right ] \ > i_0(\fr{r \rho}{\sigma_i^2 } ) \ > r \ > dr , \ ] ] ( van der marel 1995 ) . the function @xmath160 is the error function and @xmath161 is a bessel function . this model was used to fit the data in figure [ f : transmis ] . for a point source at @xmath162 , the intensity observed through an aperture centered at the origin is simply @xmath163 . the solid curves in the figure show the best chi - squared fit to the data for a psf that is a sum of three gaussians . the aperture sizes were treated as free parameters . the square apertures were assumed to have equal sizes in the @xmath57 and @xmath37 directions . the model was forced to fit 2 additional constraints , obtained from theoretical modeling ( as quoted in keyes 1995 ) : ( i ) the throughput ( for a centered point source ) of the circular 1.0 aperture is @xmath164 times that of a rectangular aperture of size @xmath165 ; and ( ii ) the absolute throughput of the @xmath165 aperture is @xmath166 . the best - fitting model parameters are listed in table [ t : psfapfit ] . the fit is robust , in spite of the fact that the model parameters are highly correlated . in particular : ( i ) no good fit can be obtained if the aperture sizes are held fixed to their nominal post - costar values ( which are @xmath29 square , @xmath30 square , and @xmath50 circular , respectively ) ; and ( ii ) the fit does not improve significantly if the number of gaussians in the psf is increased . the formal errors in the best fit parameters are @xmath167 for the @xmath154 , and @xmath18% for the @xmath168 and aperture sizes . these numbers are not too meaningful , however , because systematic errors probably exceed the formal errors . figure [ f : psffit ] shows the psf for the best fit , and the encircled flux curve @xmath169 . the latter is given by @xmath170 . \label{encircled}\ ] ] the psf has 73% of the light within a circle of radius @xmath1 . this is smaller than the 84% measured from star images at visual wavelengths obtained with the foc , but larger than the 65% measured with the wfpc2 ( which suffers from additional pixel scattering ) . it should be noted that the psf derived here pertains only to the upper position of the square paired fos apertures . the aperture transmissions , and thus the psf , are somewhat different at the lower positions of the square paired apertures ( evans 1995a ) . the above analysis ignores the effects of diffraction at the aperture edges . the derived aperture sizes are therefore not the geometrical apertures sizes . this might explain why the derived sizes are smaller than the nominal post - costar sizes . non - circularly symmetric features in the psf are also ignored . such features might in fact be present , given that the results of scanning a star across the aperture in the @xmath57 and @xmath37 directions differ slightly ( figure [ f : transmis ] ) . either way , both the aperture sizes and the psf enter into the data analysis only through the convolution kernels @xmath152 . these kernels are adequately fit by the model , independent of whether or not the aperture sizes and psf of the model are unbiased estimates of their true values . the observations in figure [ f : transmis ] were derived from so - called ` white light ' images , i.e. , with no grating in the light path . the psf with the parameters of table [ t : psfapfit ] is therefore an average over all wavelengths to which the detector is sensitive . the m32 observations were obtained with the g570h grating . observations with this grating ( bohlin & colina 1995 ) show fractional throughputs for centered point sources that are larger by a few percent than those in figure [ f : transmis ] . this indicates that the core of the psf for the g570h grating might be @xmath171% smaller than that of the psf derived here . on the other hand , in fitting the m32 peak - up acquisition observations ( section [ s : acq_results ] ) by convolving the lauer ( 1992 ) cusp model for the m32 surface brightness , it was found that the chi - squared of the fit could be improved by _ increasing _ the width of the psf core by @xmath172% . this could mean that the psf does in fact have a broader core than derived here , or alternatively , that the lauer model , based on wfpc1 measurements , is not a perfectly adequate representation of the m32 surface brightness in the central @xmath1 . in the absence of theoretical constraints on the wavelength dependence of the psf from accurate wavefront modeling , it was decided not to make any corrections to the psf derived from the white light images ( table [ t : psfapfit ] ) . the psf remains somewhat uncertain , but it was verified that changes in the psf core width between @xmath173% and 20% do not change any of the major conclusions of our paper(s ) . the observed spectrum @xmath174 of a source is the convolution of its actual spectrum @xmath175 with the line - spread - function @xmath80 , @xmath176 the lsf can be written as the convolution @xmath177 where the ` illumination function ' @xmath81 is the normalized intensity distribution of the light that falls onto the grating , and the normalized ` instrumental broadening function ' @xmath82 accounts for the broadening due to the grating and the detector . the @xmath34 coordinate system is defined such that the @xmath57 direction is parallel to the direction of dispersion . the function @xmath81 is therefore proportional to the integral over the @xmath37 direction of the two - dimensional brightness distribution that falls onto the grating . all the light in the aperture is detected , since the apertures used for the observations are smaller than the @xmath37-size of the diode array of the detector ( @xmath178 ) . for the case of observations of a target with surface brightness @xmath149 through an aperture centered on position @xmath162 , @xmath179 where @xmath180 is defined in equation ( [ itot ] ) , and @xmath181 is a convolution kernel that depends on the psf and the aperture size and geometry . for the g570h grating the wavelength @xmath182 is related to the position @xmath57 according to @xmath183(x / arcsec ) . if the aperture is rectangular of size @xmath184 , and the psf is as given in equation ( [ psfgauss ] ) , then @xmath185 \ > \left \lbrace \mathop{\rm erf}\nolimits \left [ { { y'+(a_y/2 ) } \over { \sqrt{2}\ > \sigma_i } } \right ] -\mathop{\rm erf}\nolimits \left [ { { y'-(a_y/2 ) } \over { \sqrt{2}\ > \sigma_i } } \right ] \right \rbrace , & \quad \vert x \vert \leq { { a_x}\over2 } ; \\ 0 , & \quad \vert x \vert > { { a_x}\over2 } . \end{array } \right . \label{bkernel}\ ] ] the grating disperses the image of the aperture onto the photocathode of the detector , which is scanned using an array of diodes . a simple model for the instrumental broadening function @xmath82 is therefore : @xmath186 where @xmath187 is the normalized response function of a detector diode and @xmath188 is the normalized broadening function of the grating . let @xmath187 be a top - hat function with full width @xmath189 , and let the grating broadening function be @xmath190 , where @xmath191 is a normalized function , and @xmath192 is a free parameter that measures the width of @xmath193 . the notation @xmath194 is used for the primitives of @xmath191 . the function @xmath195 is then @xmath196 .\ ] ] it can be determined empirically by fitting to the emission lines in the arc spectra . if the arc lamp illuminates the aperture homogeneously , then @xmath81 is a normalized top - hat function of full width @xmath156 . for an emission line that is intrinsically a delta - function at @xmath197 , the observed line shape will be @xmath198 . \label{oarc}\ ] ] several functional forms were considered for @xmath191 . a gaussian did not yield good fits to the observed line shapes . satisfactory results were obtained with a lorentzian , for which @xmath199 .\ ] ] with this choice , the function @xmath195 has two parameters , @xmath192 and @xmath189 . the best fit is generally attained for values of @xmath189 that differ from the actual size of one diode , which is @xmath200 for the g570h grating . this probably indicates that the true diode response function is not a perfect top - hat . however , this does not invalidate the model . one is interested only in a fit to the function @xmath82 , not in a determination of either @xmath187 or @xmath188 . therefore , both @xmath192 and @xmath189 are treated as free fitting parameters . the best fits were determined to the emission line shapes in arc spectra obtained with the 0.1-pair and the 0.25-pair apertures , while keeping the values of @xmath156 fixed to their values in table [ t : psfapfit ] . figure [ f : arc ] shows the fits to the arc lines and the best - fitting function @xmath195 . no evidence was found for a wavelength dependence of the lsf over the wavelength range covered by the grating . the best - fit parameters are : @xmath201 and @xmath202 for the 0.1-pair aperture , and @xmath203 and @xmath204 for the 0.25-pair aperture . hence , the function @xmath82 is broader for the 0.25-pair aperture than for the 0.1-pair . the difference is too large to be attributed to small errors in the assumed aperture sizes . hence , the instrumental broadening is more complicated in reality than in our simplified model , in which one would have expected @xmath82 to be independent of the aperture size . however , the only important point for the interpretation of the m32 spectra is that an accurate empirical description of the lsf is available . the fits in figure [ f : arc ] suggest that the model is satisfactory for this purpose . tests with artificial galaxy spectra were performed to determine whether systematic biases could be present in the velocity dispersion measurements . artificial galaxy spectra were created from each of the 27 stellar spectra in the template library , by broadening them with gaussian velocity profiles with dispersions @xmath205 . twenty copies were made of each broadened template , and noise was added to yield a s / n of 10 per pixel . a stellar kinematical analysis was then performed on each copy , using the composite spectrum discussed in section [ s : temp ] as template . the twenty velocity dispersion measurements where then averaged to yield an output velocity dispersion @xmath206 . figure [ f : kintest ] plots @xmath206 as function of @xmath205 for each of the input spectra . the results lie closely along the line @xmath207 . the minor deviations can be attributed entirely to the template mismatch in the simulations ( the input stellar spectra are giants , sub - giants and dwarfs of spectral types g , k and m ) . at dispersions below @xmath208 there is some tendency for the dispersion to be underestimated . however , for the m32 observations one is always in the situation where the difference in dispersion between the galaxy and template spectrum is @xmath209 ( because the lsf of the galaxy spectra is broader than that of the template , in addition to the kinematical doppler broadening ) . at these dispersions there are no significant biases in the results of the kinematical analysis , but only small ( @xmath210 ) errors due to template mismatching . these errors are nearly independent of the velocity dispersion . since the approach used in this paper for determining velocity dispersions relies on differences between velocity dispersion measurements at different positions , it is expected that even these small errors largely cancel out . a crude consistency check on our approach for determining velocity dispersions can be obtained by making the ( poor ) assumption that the lsfs of both the galaxy and template spectra are gaussian , with dispersions @xmath211 and @xmath212 , respectively . the stellar velocity dispersion @xmath119 then follows from the dispersion @xmath213 obtained by fitting to the galaxy spectrum , according to : @xmath214 for the observation l45 the dispersion measured directly from the spectra is @xmath215 . upon substitution in equation ( [ sigcorgau ] ) of @xmath216 ( cf . section [ s : temp ] ) , and @xmath217 ( cf . section [ s : lsf ] ) one obtains the estimate @xmath218 for the stellar velocity dispersion at @xmath20 along the m32 major axis . this agrees reasonably well with the value of @xmath219 adopted in section [ s : correc ] on the basis of ground - based measurements . this confirms the result of appendix [ s : ctests ] that there are no systematic biases in the velocity dispersion measurements from the spectra , and that the differences in the lsfs of the template and galaxy spectra can be fully corrected for . evans , i. n. , koratkar , a. p. , keyes , c. d. , & taylor , c. j. 1995 , fos instrument science report cal / fos-138 , ` smov report vii : fos aperture alignments ii . small apertures and adopted alignments ' ( baltimore : space telescope science institute ) koratkar , a. , wheeler , t. , evans , i. , lupie , o. , taylor , c. , keyes , c. , & kinney , a. 1994 , fos instrument science report cal / fos-123 , ` smov report v : fos plate scale and orientation ' ( baltimore : space telescope science institute ) ccccccccr 1 & 4.3 & 3.66 & 1.29 & 2 & 7 & 3.36 & 0.80 & 4.7 + 2 & 1.0pair & 0.86 & 0.86 & 6 & 2 & 0.61 & 0.65 & 7.4 + 3 & 0.5pair & 0.43 & 0.43 & 3 & 3 & 0.29 & 0.29 & 9.1 + 4 & 0.25pair & 0.215 & 0.215 & 3 & 3 & 0.143 & 0.143 & 8.5 + 5 & 0.1pair & 0.086 & 0.086 & 5 & 5 & 0.043 & 0.043 & 38.7 + ccclcccc s1 & y2uf0107 t & 3 & 0.1-pair & @xmath220 & @xmath221 & @xmath222 & 2400 + s2 & y2uf0109 t & 4 & 0.1-pair & @xmath220 & @xmath221 & @xmath222 & 2400 + s3 & y2uf010bt & 5 & 0.1-pair & @xmath220 & @xmath221 & @xmath223 & 2400 + s4 & y2uf010dt & 6 & 0.1-pair & @xmath220 & @xmath221 & @xmath224 & 2400 + l1 & y2uf010gt & 7 & 0.25-pair & @xmath225 & @xmath221 & @xmath226 & 1105 + l2 & y2uf010ht & 7 & 0.25-pair & @xmath225 & @xmath221 & @xmath227 & 1105 + l3 & y2uf010jt & 8 & 0.25-pair & @xmath225 & @xmath221 & @xmath222 & 1395 + l4 & y2uf010kt & 8 & 0.25-pair & @xmath225 & @xmath221 & @xmath228 & 800 + l5 & y2uf010mt & 9 & 0.25-pair & @xmath225 & @xmath221 & @xmath228 & 1790 + llcccccc s1 & 0.1-pair & 0.068 & 218.1 & @xmath229 & @xmath230 & @xmath231 & @xmath232 + s2 & 0.1-pair & 0.068 & 203.6 & @xmath233 & @xmath234 & @xmath233 & @xmath235 + s3 & 0.1-pair & 0.068 & 139.7 & @xmath233 & @xmath236 & @xmath233 & @xmath235 + s4 & 0.1-pair & 0.068 & 199.2 & @xmath233 & @xmath237 & @xmath233 & @xmath235 + l1 & 0.25-pair & 0.191 & 686.0 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + l2 & 0.25-pair & 0.191 & 1133.7 & @xmath238 & @xmath242 & @xmath243 & @xmath244 + l3 & 0.25-pair & 0.191 & 1269.3 & @xmath238 & @xmath245 & @xmath246 & @xmath247 + l4 & 0.25-pair & 0.191 & 435.5 & @xmath238 & @xmath248 & @xmath249 & @xmath250 + l5 & 0.25-pair & 0.191 & 427.7 & @xmath251 & @xmath252 & @xmath253 & @xmath254 + lrrrrrr s1 & 0.941 & 0.038 & @xmath255 & 10.0 & 156.4 & 10.2 + s2 & 0.957 & 0.033 & @xmath256 & 6.4 & 95.6 & 9.8 + s3 & 1.014 & 0.046 & @xmath257 & 9.7 & 111.9 & 11.7 + s4 & 1.049 & 0.039 & @xmath258 & 8.6 & 127.0 & 9.6 + l1 & 0.927 & 0.027 & @xmath259 & 5.3 & 78.3 & 11.0 + l2 & 0.989 & 0.022 & @xmath260 & 4.3 & 83.4 & 9.3 + l3 & 1.104 & 0.022 & @xmath261 & 4.8 & 124.2 & 7.0 + l45 & 0.973 & 0.022 & @xmath262 & 4.3 & 70.0@xmath12 & 5.0 +
we report on the results of an extensive monitoring campaign of the radio afterglow of grb 970508 , lasting 450 days after the burst . the spectral and temporal radio behavior indicate that the fireball has undergone a transition to sub - relativistic expansion at days . this allows us to perform `` calorimetry '' of the explosion . the derived total energy , erg is well below the erg inferred under the assumption of spherical symmetry from-ray and early afterglow observations . a natural consequence of this result , which can also account for deviations at days from the spherical relativistic fireball model predictions , is that the fireball was initially a wide - angle jet of opening angle . our analysis also allows to determine the energy fractions carried by electrons and magnetic field , and the density of ambient medium surrounding the fireball . we find that during the sub - relativistic expansion electrons and magnetic field are close to equipartition , and that the density of the ambient medium is . the inferred density rules out the possibility that the fireball expands into a strongly non - uniform medium , as would be expected , e.g. , in the case of a massive star progenitor . #1#2#3#1 , a&a , # 2 , # 3 . # 1#2#3#1 , a&as , # 2 , # 3 . # 1#2#3#1 , aj , # 2 , # 3 . # 1#2#3#1 , _ ap . j. _ , * # 2 * , # 3 . # 1#2#3#1 , _ ap . j. ( letters ) _ , * # 2 * , # 3 . # 1#2#3#1 , apjs , # 2 , # 3 . # 1#2#3#1 , ara&a , # 2 , # 3 . # 1#2#3#1 , baas , # 2 , # 3 . # 1#2#3#1 , icarus , # 2 , # 3 . # 1#2#1 , iau circ . no . # 2 # 1#2#1 , gcn circ . no . # 2 # 1#2#3#1 , _ m.n.r.a.s . _ , * # 2 * , # 3 . # 1#2#3#1 , _ nature _ , * # 2 * , # 3 . # 1#2#3#1 , pasj , # 2 , # 3 . # 1#2#3#1 , pasp , # 2 , # 3 . # 1#2#3#1 , qjras , # 2 , # 3 . # 1#2#3#1 , science , # 2 , # 3 . # 1#2#3#1 , soviet astr . , # 2 , # 3 . # 1#2#3#1 , soviet astr . lett . , # 2 , # 3 . # 1#2#3#4#1 , # 2 , # 3 , # 4 . # 1#1 .
astro-ph9910319
the gamma - ray burst ( grb ) of may 8 , 1997 was a watershed event in the study of these intriguing objects . there were two major advances resulting from afterglow observations of : ( 1 ) the first unambiguous evidence that some grbs are a cosmological population was obtained with the detection of red - shifted metal lines in absorption against the optical afterglow ( metzger et al . the resulting lower limit on the distance , combined with the measured gamma - ray fluence , established an approximate energy scale for these events . ( 2 ) the discovery of the first radio afterglow and the demonstration of relativistic expansion of the fireball ( frail et al . the latter was made possible by the observations of distinctive variations in the radio flux and attributed to interstellar scintillation ( goodman 1997 ) . it is worth noting that grbs now join quasars and galactic micro - quasars as sources for which superluminal motions have been inferred . the afterglow from grb 970508 was particularly bright and long - lived . indeed , to this date , this grb remains unusual in both these respects . consequently , astronomers carried out intensive observations of the afterglow across the electromagnetic spectrum . the principal motivation of the observations were to infer the fundamental parameters of the explosion : the total energy of the explosion ( @xmath7 ) , the distribution of the circumburst medium ( density , @xmath8 and possible radial dependence as would be the case if the explosion took place in a circumburst medium shaped by mass loss from the progenitor ) and the geometry of the afterglow ( sphere versus collimated flow or `` jets '' ) . of all the parameters listed above , @xmath7 is perhaps the most eagerly sought parameter . after all , in a manner similar to supernovae , it is @xmath7 which sets the scale for the entire grb phenomenon and thus it is a truly fundamental parameter of the explosion . with few exceptions ( katz & piran 1997 , brainerd 1998 ) most of the early estimates for @xmath7 ( waxman 1997 , wijers & galama 1999 , granot et al . 1999 ) lie within a factor of three of @xmath9 erg . although their precise methodologies differ , all of these determinations use information gleaned from the early afterglow observations . unfortunately , the lorentz factor is large during the early phase and due to relativistic beaming the observer sees only solid angle @xmath10 of the emitting surface . thus the inferred energy is necessarily the `` isotropic equivalent '' value . however , if fireballs are jet sources ( as appears to be the case ) , such `` isotropic equivalent '' estimates are upper limits and in some cases can be gross upper limits ( e.g. grb 990510 , harrison et al . 1999 ) radio observations are essentially immune from the geometry of the fireball and thus offer us the best method to infer @xmath7 since much of the radio afterglow is emitted at later epoch when @xmath11 is falling . indeed , over the duration of the radio emission , the emitting material typically becomes sub - relativistic . once the flow becomes sub - relativistic it will in due course also become spherical . thus observations of radio afterglows allow us the opportunity of inferring @xmath7 without the usual concerns of the geometry of the fireball . radio observations offer two additional advantages : ( 1 ) at x - ray and optical wavelengths , the afterglow emission rapidly enters the monotonic decaying regime within minutes to hours of the burst . unfortunately , at the current time , logistical difficulties prevent us from responding so rapidly to grb events . x - ray and optical afterglow observations are typically initiated hours after the burst . however , at radio wavelengths , the entire afterglow phenomenon is stretched in time . thus monitoring the radio afterglow observations allows us to see all the important transitions ( save the cooling transition ) which in turn provides key diagnostics to infer the fundamental parameters of the explosion . ( 2 ) radio observations offer us , via interstellar scintillations , the only way to measure the tiny angular size ( few microarcseconds ) of grb afterglows . the measured size in turn allows us to verify the dynamics of the explosion . in this paper , we present the complete radio light curves from the radio afterglow of beginning 3.5 hrs after the burst and ending 450 d later when the source was no longer detectable . this is the third and final paper of an ambitious monitoring program that we initiated following the discovery of radio afterglow from . in the first paper ( frail et al . 1997 ) we reported the discovery of the radio afterglow and the first 90 d of observations , and we interpreted the strong variations as being due to diffractive scintillation . in the second paper ( waxman , kulkarni & frail 1998 ; hereafter , wkf98 ) we compared these observations to theoretical models and showed that the size estimated from diffractive scintillation was in excellent accord with that expected from hydrodynamical models . in the same paper we noted that the observed flux at late times ( @xmath12 d ) was well below that predicted by a spherical model and suggested that this deviation was most likely explained by a jet - like geometry for the fireball rather than a sphere . we suggested that @xmath7 would be significantly lower than the @xmath9 erg isotropic equivalent estimate from the early afterglow observations and the isotropic equivalent @xmath3-ray energy release and further suggested that the fireball would undergo a transition from relativistic to sub - relativistic expansion at @xmath0 d. the organization of the paper is as follows . the observations are summarized in [ sec : observations ] . in [ sec : results ] we present the basic results in the form of light curves in three bands , 8.46 ghz , 4.86 ghz and 1.43 ghz . in [ sec : riss ] we note a transition of the variations in the light curve from the diffractive to the refractive regimes . in [ sec : spherical ] we compare the data to the expectations from the simple adiabatic fireball model . in [ sec : deviations ] we point out important deviations from model predictions and argue that the afterglow was not spherical but was a jet with an opening angle of about 30 degrees . finally , in [ sec : sub - relativistic ] we propose a detailed sub - relativistic model which provides a satisfactory explanation to the very late time observations . the transition to non - relativistic regime allows us to carry out proper calorimetry of the explosion ( [ sec : calorimetry ] ) . except where noted , all observations reported here were obtained at the very large array ( vla ) of the national radio astronomy observatory ( nrao ) . the very first observation was made in the 1.4-ghz band , only 3.5 hrs after the initial gamma - ray detection of ( costa et al . table 1 lists each observing session giving the starting date , the time elapsed since the burst , the 1.43 ghz flux density of ( f@xmath13 ) and the vla configuration . the first detection of the source was made on 1997 may 13.96 ut at 8.46 ghz . except for the first week , it was customary to observe at 8.46 ghz and 4.86 ghz simultaneously , so we include them together in table 2 . all flux density measurements prior to 1997 august were previously compiled in frail et al . ( 1997 ) but are listed here for completeness . calibration of the array phase was accomplished using extragalactic radio sources with well - known positions . j0410 + 769 was used primarily at 1.43 ghz while j0726 + 791 was the calibrator of choice for 4.86 ghz and 8.46 ghz . since these compact sources usually vary on time scales of a month or more , the absolute flux scale ( accurate to better than @xmath142% ) at each epoch for all frequencies was fixed with observations of 3c , 3c , and/or 3c . the off - line analysis of the vla data proceeded following standard practice using the aips data reduction package . when the radio afterglow from was clearly detected , its flux density was measured by a gaussian fitting procedure which solves for the peak and integrated intensity , the source position and its angular size . when the source is weak ( @xmath15 ) and no reliable fit was possible , the peak flux density was measured from the image at the reported position of ( taylor et al . 1997 ) . the error estimates in tables 1 and 2 are based on the thermal noise fluctuations in the final image and are estimated by calculating the rms fluctuations in a region free of any obvious radio sources . in the d - array and to some extent in the cs array , where the resolution at 1.43 ghz is poor , the limiting sensitivity is not due to the receiver noise but to the contribution of unresolved , faint sources within the beam . this quantity varies from point to point in the sky but is of order 80 @xmath16jy for a 45 beam at 1.4 ghz ( condon et al . improperly cleaned sidelobes from the myriad sources in the d - array 1.43 ghz images constitute a further source of confusion . together these effects can amount to 100 @xmath16jy of uncertainty added in quadrature to the error estimates of the d - array measurements given in table 1 . while the source confusion discussed above is negligible at 4.86 and 8.46 ghz , there is a second source of error in the largest array configurations ( a and b ) which can also be difficult to quantify . atmospheric phase instabilities occurring on a time scale shorter than or comparable to the cycle time between calibrator and source will `` scatter '' flux from and may introduce slight position offsets . large and erratic excursions of the calibrator phase over the course of an observing run are a useful diagnostic to indicate when this is occurring . suspect data is revealed in off - line processing by sudden jumps in the flux density of the phase calibrator from its long - term average and a significant difference between the peak and integrated flux densities in the gaussian fits of the radio emission from . following these criteria it was necessary to discard data from only four epochs in 1997 ( sept . 19 , sept . 21 , dec . 23 and dec . * late - time light curves . * in figure [ fig : ltcurve ] we show the entire light curve of all three frequencies at which was monitored by the vla . the behavior of the radio light curves in the first few months has been discussed elsewhere ( frail et al . 1997 , wkf98 ) ; galama et al . ( 1998a , b ) and shepherd et al . ( 1998 ) report a few additional points at low and high frequencies as well . for the sake of completeness , we now summarize the early time behavior here . as shown in figure [ fig : cxlight ] , the erratic flux variations reported in frail et al . ( 1997 ) do not die away entirely after the first month but their character changes . between 30 d and 85 d the 4.86 ghz and 8.46 ghz light curves have a mean flux density of 555@xmath14108 @xmath16jy and 630@xmath14107 @xmath16jy , respectively . the modulation indices ( defined as the ratio of the rms flux variations to the mean ) are modest ( 1520% ) and the variations are well correlated between the two frequencies . this is very different from the narrow bandwidth , short time scale fluctuations seen in the first month and attributed to diffractive interstellar scattering ( frail et al . we return to this point in [ sec : riss ] . the light curve at 1.43 ghz ( table 1 and figure [ fig : ltcurve ] ) is markedly different than the other two . as noted elsewhere ( frail et al . 1997 , shepherd et al . 1998 , galama et al . 1998a ) the 1.43 ghz flux density is initially weak ( 125 @xmath16jy ) . this can best be understood if the source is optically thick at 1.43 ghz ( katz & piran 1997 ) . near day 50 the source brightens considerably . between 50 and 300 days its average flux density is 300 @xmath16jy but with considerable scatter . most of this scatter likely originates from source confusion , particularly in the more compact array configurations ( see [ sec : observations ] ) . near day 90 the light curves at 8.46 ghz and 4.86 ghz begin a power - law decay ( figure [ fig : decay ] ) , similar to that which occurred much earlier at optical and x - ray wavelengths ( pian et al . 1998 , piro et al . 1998 ) . power - law fits to the light curves in figure [ fig : decay ] for @xmath17 d give values of @xmath18 and @xmath19 ( defined as @xmath20 ) . the decline at 1.43 ghz is not as well constrained but we estimate from figure [ fig : ltcurve ] that the decay begin near day 300 . * spectral index . * many of the features in the high frequency light curves have analogs in the spectral domain . for example , the narrow bandwidth ( @xmath21 ghz ) diffractive scintillation seen by frail et al . ( 1997 ) is best demonstrated in figure [ fig : cxespec ] by the large swings in the spectral slope @xmath22 ( defined as @xmath23 ) between 8.46 ghz and 4.86 ghz . the spectral index appears to settle down by day 35 and we interpret this to mean that the diffractive scintillation has died down . indeed , this is to be expected since only sources smaller than a critical size can undergo diffractive scintillation . thus at sufficiently late times , expansion will ensure that the source enters the refractive regime ( [ sec : riss ] ) . since refractive scintillation is broadband ( see [ sec : riss ] ) we see the true radio spectrum of the afterglow past day 35 . excluding two outlier points , the mean @xmath22 for 11 measurements between day 40 and day 85 is + [email protected] ( figure [ fig : cxespec ] ) . galama et al . ( 1998b ) determined @xmath24 near day 12 . since their estimate is based on extrapolations of the radio flux density when the source was scintillating strongly , the value of @xmath22 given here should be a more accurate one . in addition , the transition of the light curve to a power - law decay has a corresponding change in the spectral index plot of figure [ fig : cxspec ] , where @xmath22 undergoes an abrupt drop from positive to negative values . the weighted mean @xmath22 from day 95 to day 310 is @xmath250.3 . a joint fit to the spectrum and high frequency light curves of the form @xmath26 for @xmath27 days yields more accurate values of @xmath28 and @xmath29 * evolution of the radio spectrum . * in figure [ fig : allspec ] we follow the evolution of the radio afterglow by plotting all radio frequencies at different time intervals . these spectral `` snapshots '' show that the spectrum initially rises with frequency , flattens and then finally exhibits a power - law slope which declines with frequency . in frail et al . ( 1997 ) , we attributed the observed strong variability in the 8.46 ghz and 4.86 ghz bands over the course of the first month , to diffractive scintillation . the primary evidence for this inference is the observed strong , rapid ( decorrelation time , @xmath30 hr ) variability which is uncorrelated at 8.46 ghz and 4.86 ghz . these two characteristics , strong and chromatic variability , are the hallmarks of diffractive scintillation ( rickett 1977 ) . in contrast , beyond day 30 the variations are not only milder but also correlated between the two frequencies ( see [ sec : results ] ) . we attribute these variations to `` refractive '' scintillation . diffractive scintillation is caused by interference between rays diffracted by small - scale irregularities in the ionized interstellar medium . the resulting interference is narrow band and highly variable ( with the modulation index approaching unity ) . diffractive scintillation occurs only when the source size is smaller than a characteristic size , the so - called `` diffractive '' angle . thus as the fireball expands we expect the diffractive scintillation to be quenched . however , as explained by goodman ( 1997 ) , focusing and defocusing of the wavefront by large scale inhomogeneities in the ionized interstellar medium will also result in scintillation , albeit at a reduced level on a longer time scale compared to diffractive variations . this `` refractive '' scintillation , unlike the earlier diffractive scintillation , is broad - band , and , most importantly , is much less sensitive to the size of the source . thus , not surprisingly , refractive variability has been seen in many sources , whereas only the most compact sources pulsars and grb afterglows show diffractive scintillation . the light curves in figure [ fig : ltcurve ] and [ fig : cxlight ] show that the expected transition from the diffractive to the refractive regime did take place . the flux variations decrease , but do not vanish , after the first month . during this period , the modulation index ( defined as the ratio of the rms flux variations to the mean ) is more modest ( 1520% ) compared to the first month when the modulation index was about 80% . the change in the nature of the flux variations is also seen in the spectral index curve ( figure [ fig : cxespec ] ) . over the first month , the flux variations were uncorrelated between 4.86 ghz and 8.46 ghz ( decorrelation bandwidth , @xmath31 ghz ) , as best demonstrated by the large swings in the spectral slope @xmath22 whose values varied randomly between @xmath32 to + 1.6 from one day to the next ( figure [ fig : cxespec ] ) . starting around day 40 , @xmath22 converges to a constant value . we identify these slow , broad - band low - amplitude flux variations as being due to refractive scintillation , with the expansion of taking it through a transition from the diffractive to refractive regimes . in principle , an independent estimate of the angular size could be determined from the refractive scintillation , but our sampling of the light curve is not sufficient to accurately measure the refractive decorrelation time scale or the modulation index . nonetheless , as a check for consistency , we now estimate the expected modulation index and refractive time scales and compare the same to the observations . we use the angular size of the fireball inferred from the diffractive scintillations , @xmath33arcsec at @xmath34 weeks ( wkf98 ) , and extrapolate this size forward in time as @xmath35 to obtain a size of @xmath36arcsec at @xmath37 weeks . from this expected angular size , and using the formulation of goodman ( 1997 ) , we infer the refractive time scales to be between 10 and 20 hrs and the modulation indices of order 40% to 25% at these frequencies . these estimates are consistent with our admittedly meager data . as remarked in [ sec : introduction ] the principal motivation for afterglow observations is to deduce the fundamental parameters of the explosion . to this end , we now compare the radio observations with the expectation of the simplest model , the spherical adiabatic model . other authors have carried out similar analyses but based only on early - time ( @xmath38 d ) multi - wavelength observations ( waxman 1997 , wijers & galama 1999 , granot et al . 1999 ) . they found good agreement with the basic expectation of spherical adiabatic models . the first predictive afterglow models ( mszros & rees 1993 , paczyski & rhoads 1993 , katz 1994 , mszros & rees 1997 ) actually preceded the discovery of the afterglow emission and have survived with relatively little modification ( sari 1997 , vietri 1997 , waxman 1997 , wijers , rees & mszros 1997 ) . in its most basic form the cosmological fireball model is a spherical blast wave , expanding adiabatically into a homogeneous medium of density @xmath8 . a fixed fraction @xmath39 of the blast wave energy @xmath7 goes into accelerating a power - law distribution of electrons with energy slope @xmath40 ( where the number of electrons with lorentz factor @xmath3 is @xmath41 for @xmath42 ) . in the presence of a magnetic field , which itself is a fixed fraction @xmath43 of the energy density of the blast wave , the electrons emit synchrotron radiation . the instantaneous synchrotron spectrum emitted by the electrons consists of four distinct spectral regimes and is given by sari , piran & narayan ( 1998 ) for the `` slow cooling '' case when the radiative cooling time for the majority of electrons with lorentz factor @xmath44 is large compared to the dynamical time : @xmath45 there are three transition frequencies : @xmath46 , @xmath47 and @xmath48 . they correspond to ( 1 ) the synchrotron self - absorption frequency @xmath46 below which the source becomes optically thick , ( 2 ) the peak frequency @xmath47 where the spectrum peaks at a flux density of f@xmath49 and ( 3 ) the cooling frequency @xmath48 above which radiative losses steepen the observed spectrum . in time , as the blast wave interacts with the surrounding medium , the shape of the spectrum is preserved but the peak of the spectrum shifts to lower frequencies . the time - dependence of the three transition frequencies for an adiabatically expanding blast wave is given by @xmath50 , @xmath51 , and @xmath52 with @xmath53 for most of the the time over which the afterglow is detectable . in addition , f@xmath49 is expected to remain constant with frequency . describing the evolution of the radio flux density as f@xmath54 for @xmath55 ( i.e. the optically thin regime ) we expect to see a slow rise to f@xmath49 as @xmath56 for @xmath22=+1/3 , followed by a power - law decline going as @xmath57 , where @xmath58 and @xmath59 when @xmath60 ; and @xmath61 steepening a further @xmath62 when @xmath63 . the observations presented in [ sec : results ] show substantially all of the gross spectral and temporal behavior expected from this basic model . in detail we find the following agreements with the model . * low energy spectral index . * at early times synchrotron self - absorption is clearly detected ( figure [ fig : allspec ] and shepherd et al . 1998 ) with a synchrotron self - absorption frequency @xmath46 near 3 ghz ( granot et al . 1999 ) . after day 35 when the diffractive interstellar scintillation between 4.86 ghz and 8.46 ghz has quenched , the low energy regime of the synchrotron spectrum ( i.e. @xmath64 ) is revealed . the measured slope of + [email protected] ( figure [ fig : cxspec ] ) is close to the predicted asymptotic value of 1/3 . this slope is a generic prediction of relativistic shocks which accelerate electrons with a power - law distribution of energies and have sharp cutoffs at low energies ( katz 1994 ) . * temporal decay : indices and time scales . * the decay in the 8.46 ghz and 4.86 ghz light curves ( figure [ fig : decay ] ) begins near day 90 and can be most easily seen in the spectral index plot of figure [ fig : cxspec ] , where the spectral slope undergoes an abrupt change from positive to negative values corresponding to the regime @xmath65 . the corresponding light curves have a power - law decay with a temporal slope near @xmath32 , comparable to that inferred from optical and x - ray observations ( see table 4 ) . as expected , the decay of the light curve for the lower frequency of 1.43 ghz begins much later ( figure [ fig : ltcurve ] ) . we can compare the time scale for the onset of the decay with the predicted @xmath66 dependence for @xmath47 . in table 3 we give estimates of @xmath67 between 1 and 100 ghz . the @xmath67 from the 86 ghz data of bremer et al . ( 1998 ) would predict that the onset of the power - law decay would start between 5070 d at 8.46 ghz and between 7095 d at 4.86 ghz . similarly , from the 8.46 ghz and 4.86 ghz data we would predict a decay at 1.43 ghz between 200300 d after the burst . given the uncertainties in the estimates of @xmath67 this agreement is surprisingly good . * lack of cooling break . * there is no evidence for the existence of an additional cooling break in the light curves or spectra of figures [ fig : decay ] and [ fig : cxspec ] . in contrast , galama et al . ( 1998b , c ) have identified passage of the cooling break at higher frequencies , specifically in the optical band around day 1.4 . accepting this result , the expected epoch of cooling break in the 8.46 ghz band is @xmath68 , which is considerably larger than one year ; here @xmath69 is the frequency in the optical band , @xmath70 hz , @xmath71 is the typical radio frequency and @xmath72 d. thus , the absence of any clear cooling break in our radio light curves is not surprising . in the previous section we found that the spherical adiabatic fireball model provided a satisfactory explanation for the observations at a gross level . however , on closer inspection there are several important differences between the radio afterglow of and the model predictions . indeed , on theoretical grounds we expect deviations from the spherical relativistic model due to a transition from relativistic to sub - relativistic regimes and possibly jet geometry ( wkf98 ) . we summarize the observed deviations below and offer explanations for them in [ sec : sub - relativistic ] . * deviation in the light curves . * as already noted in wkf98 , the spherical model which provided an excellent fit to the early time ( less than 3 weeks ) multi - wavelength data ( x - ray , optical and radio ) , predicted radio emission that was well above the observed value for @xmath12 d ; see figure 1 of wkf98 . * non - constancy of peak flux . * another important prediction of the spherical , adiabatic model is the constancy of the peak flux density f@xmath49 with time . while the peak frequency @xmath47 evolves as @xmath66 , the value of f@xmath49 is expected to remain constant as @xmath47 sweeps through the x - ray , optical and radio bands . this assumption underlies all attempts to fit `` snap - shot '' broad - band spectral distribution of grb afterglows ( e.g. galama et al . 1998b , wijers & galama 1999 ) and the subsequent derivation of the shock parameters . in table 3 there is a very clear trend for successive peaks in f@xmath49 to reach lower peak flux densities as @xmath47 moves towards lower frequencies . a least squares fit between 1 and 90 ghz gives f@xmath73 . * non - constancy of absorption frequency . * in the simple adiabatic fireball model , provided that the fundamental shock parameters ( e@xmath74 , @xmath8 , @xmath39 and @xmath43 ) do not change , @xmath46 is expected to remain constant until such time as @xmath75 ( waxman 1997 , granot et al . near day 50 the 1.43 ghz flux density rises from @xmath76125 @xmath16jy to @xmath76300 @xmath16jy ( see figure [ fig : ltcurve ] ) . such behavior is expected for an expanding optically thick source ( see frail et al . 1997 ) simply because of the increase in the radiative surface area . however , the spectra in figure [ fig : allspec ] suggest a different interpretation . the initial attenuation at @xmath77 d is well described by synchrotron self - absorption with unity optical depth at @xmath46=3 ghz ( shepherd et al . 1998 , granot et al . 1999 ) . in time , the spectra in figure [ fig : allspec ] clearly show an evolution from an optically thick to an optically thin source , with @xmath46 shifting to lower frequencies . * relation between @xmath61 and @xmath22 . * the final and perhaps the most significant deviation from the model concerns the measured values for @xmath61 and @xmath22 ( where @xmath78 ) of @xmath29 and @xmath28 from the decay of the radio light curve . in table 4 these values are compared to measurements made at optical wavelengths at different times in the evolution of the light curve . the temporal slope of the optical light steepens from its initial value of @xmath79 ( djorgovski et al . 1997 ) to @xmath32 ( e.g. zharikov et al . 1998 , garcia et al . 1998 , castro - tirado et al . 1998 , bloom et al . 1998 ) . similarly , the spectral slope undergoes an abrupt transition at optical wavelengths between day 1 and day 2 from @xmath80 ( djorgovski et al . 1997 ) to @xmath81 ( galama et al . 1998c , sokolov et al . galama et al ( 1998b ) have made a strong case that this spectral and temporal break corresponded to the passage of the cooling break @xmath48 through the optical passbands at roughly 1.4 d after the burst . prior to the passage of @xmath48 the relativistic model predicts @xmath58 , followed by a steepening ( by 1/2 ) of @xmath22 giving @xmath58 + 1/2 . the measured @xmath61 and @xmath22 on either side of @xmath48 values are consistent with these relationships . bolstering this claim , galama et al . ( 1998a ) derived two independent measures of the electron energy power - law index ( @xmath82 ) . in contrast , in the framework of the relativistic adiabatic model , no single value of @xmath40 can explain the values of @xmath61 and @xmath22 that we obtained from our radio observations . the radio @xmath61 is identical to the late - time optical value but the radio @xmath22 is more similar to the early optical estimates before @xmath48 moved through the optical band . the relativistic spherical model is the simplest afterglow model and as such has received considerable attention from observers . apart from the assumptions of constant @xmath39 , @xmath43 and acceleration to a power law spectrum , the main assumptions of the model are spherical geometry and relativistic outflow . hints of deviations from the expectations of the spherical model were already apparent in the wkf98 paper ( see [ sec : introduction ] ) . in this section , we first qualitatively expand on these two ideas and find that the four problems identified in the previous section can be satisfactorily explained by a model incorporating jets . an inevitable consequence of a jet geometry is early ( compared to spherical geometry ) transition to the non - relativistic regime . we develop and present in appendix [ sec : nrmodel ] a formulation describing non - relativistic fireballs , and then apply this formulation to the light curves ( figure [ fig : ltcurve ] ) and extract the basic parameters of the explosion in [ sec : calorimetry ] . * deficit of flux and lack of constancy of @xmath83 : jets . * a jet with an opening angle of @xmath84 behaves as if it were a conical section of a spherical fireball , and can not be distinguished by a distant observer from a spherical fireball , as long as the expansion lorentz factor is larger than the inverse of the opening angle , @xmath85 . once the jet decelerates to @xmath86 , it expands sideways , leading to a decrease in blast wave energy per unit solid angle , which results in a deviation from spherical fireball behavior . the decrease in energy per unit solid angle leads to a decrease with time of @xmath83 , and suppresses the @xmath56 flux increase expected in the spherical model at frequencies @xmath87 . thus a jet geometry provides a reasonable explanation for the flattening of the radio light curves and the lack of constancy of @xmath83 ( [ sec : deviations ] ) . we set @xmath88 , the epoch when the jet becomes apparent to the observer , as 25 d ( see figure 1 of wkf98 ; also discussion in [ sec : deviations ] ) . the lorentz factor corresponding to this epoch is @xmath89 , or equivalent to a jet opening angle @xmath90 . it was shown in wkf98 that on a time scale of @xmath91 , where @xmath88 is the time at which the jet decelerates to @xmath92 , the fireball approaches spherical , sub - relativistic expansion . a simple argument to understand this is as follows . the time at which we see deviation from spherical symmetry is @xmath93 , where @xmath94 is the radius at which @xmath95 . the transition to spherical symmetry occurs over time scale @xmath96 , i.e. @xmath97 . in addition , since the mass enclosed within a sphere of radius @xmath94 , @xmath98 , is comparable to the total jet energy @xmath99 , where @xmath100 is the energy inferred under the assumption of spherical symmetry , the fireball becomes sub - relativistic as it approaches spherical symmetry . using our @xmath10125 d and @xmath1022 we expect the fireball to approach spherical sub - relativistic expansion on a time scale of @xmath103100d . * relation between @xmath61 and @xmath22 : sub - relativistic fireball*. as discussed in appendix [ sec : nrmodel ] , in the sub - relativistic regime , the afterglow emission scales as @xmath104 with @xmath105 and @xmath106 ( compared to @xmath58 for relativistic expansion ) . the behavior of the radio flux observed at @xmath107 d is consistent with this expectation since past day 100 ( approximately the transition to sub - relativistic regime ) we obtain @xmath108 and @xmath109 ( [ sec : results ] ) . if the afterglow was in the sub - relativistic regime , then we expect the following closure relation , @xmath110 , whereas in the relativistic regime , @xmath111 . we find @xmath112 whereas @xmath113 . thus we can reject the relativistic model at the @xmath763-@xmath114 level . more importantly , a single value of @xmath40=2.2 , explains , in a self - consistent way , the x - ray ( piro et al . 1998 ) , optical ( galama et al . 1998b ) and radio light ( this paper ) afterglow observations over several decades in time through radiative transitions ( adiabatic to cooling ) and dynamical transitions ( relativistic to sub - relativistic ) . * non - constancy of @xmath46 . * the non - constancy of @xmath46 can be explained as follows . there are two consequences in the sub - relativistic model when @xmath47 falls below @xmath46 occurs ( see wkf98 and appendix [ sec : nrmodel ] ) . first , @xmath46 is no longer constant but evolves as @xmath115 . second , the peak in the spectrum is no longer at @xmath47 but is at @xmath46 . the peak flux then evolves as @xmath116 . by itself , the observed decrease in @xmath46 is consistent with the expectations of the sub - relativistic model . however , since one expects a similar temporal evolution of @xmath46 for @xmath75 in the relativistic model , it should not be taken as strong evidence in favor of the transition to sub - relativistic expansion . the above conclusions , the manifestation of a jet at @xmath117 d and the sub - relativistic transition at about epoch 100 d have been arrived at using extensive radio data presented in this paper . independent support for this can be obtained from the optical data . as noted by rhoads ( 1999 ) the optical data can not be fitted by a single power law decay functional form . rhoads derives a minimum beaming angle of @xmath118 , a value in complete agreement with our own . the principal strength of the radio result is the photometric uniformity of the data the data were obtained with one instrument , the vla , in a standard mode . in contrast , the optical data come from diverse sources and vexing cross - calibration issues still remain . in our model , we would predict that the optical flux would steepen beyond @xmath88 . a sharp break , such as the one seen in the light curve of grb ( e.g. harrison et al . 1999 ) , is not expected in the case of , given the wide opening angle . numerical and analytical studies which track the dynamical evolution of the beamed ejecta ( moderski , sikora , & bulik 1999 , wei & li 1999 ) find that unless the opening angle is very small or that lateral expansion is unimportant , a smooth and gradual transition is expected . in addition , past day 100 , our model would predict a further , slight steepening [ cf . ( [ eq : f_nuc ] ) ] of the optical flux as the shock makes the transition from the relativistic to sub - relativistic regimes . the combination of these two factors is the likely origin for the apparent disappearance of the afterglow at @xmath119 d in a recent hst image ( fruchter et al . in appendix [ sec : nrmodel ] , we present a rigorous formulation of a sub - relativistic fireball model . in wkf98 we considered sub - relativistic fireball models in the asymptotic limit of @xmath46 being very different from @xmath47 . however , for the data presented here , @xmath47 and @xmath46 are comparable , and the model presented in appendix [ sec : nrmodel ] does not rely on asymptotic limits but is an exact model . the radio flux and spectrum produced by the sub - relativistic fireball are determined by three parameters , @xmath120 , @xmath121 and @xmath122 ( see appendix [ sec : nrmodel ] ) . @xmath122 is @xmath123 at epoch @xmath124 . here we set @xmath125 d and note that the redshift of the host galaxy of is @xmath126 ( bloom et al . the @xmath127100 d radio data allow an accurate ( @xmath128 ) determination of the parameters @xmath120 and @xmath121 , which determine the high frequency flux normalization and the self - absorption suppression at 1.43 ghz : @xmath129 , @xmath130 . the spectral change at @xmath131 d ( figure [ fig : cxspec ] ) implies that the peak frequency @xmath47 decreases through the frequency range of 8.46 ghz to 4.86 ghz at @xmath132 d , i.e. that @xmath133 ghz . however , it should be noted that since in our model the fireball at time @xmath134 d is in transition to sub - relativistic spherical expansion ( see below ) , significant deviation from the sub - relativistic model prediction is possible at @xmath134 d. thus , the determination of @xmath122 is only approximate ( see figure [ fig : ewmodel ] ) . our conclusions remain unchanged , however , even with a factor of a few change in @xmath122 ( see below ) . the parameters @xmath120 , @xmath121 , and @xmath122 are determined ( see appendix [ sec : nrmodel ] ) by the four parameters , which completely specify a sub - relativistic fireball model : the strength @xmath135 of the magnetic field at @xmath136 , the minimum lorentz factor @xmath137 of the radiating electrons at @xmath136 , the ( time independent ) number density @xmath138 of the shocked plasma , and the shock radius @xmath139 at @xmath136 . using the three constraints provided by the determination of @xmath120 , @xmath121 and @xmath122 we first derive @xmath135 , @xmath137 , and @xmath138 as functions of @xmath139 . we then constrain @xmath139 by requiring that the derived minimum energy of the radiating electrons and magnetic field does not exceed the thermal energy of the sub - relativistic fireball , given by the sedov - von neumann - taylor formulation . using equation ( [ eq : a ] ) we find @xmath140 we use the definitions @xmath141 , @xmath142 , @xmath143 , @xmath144 ghz , and @xmath145 cm . as demonstrated above , observations require @xmath146 , and @xmath147 . @xmath148 for a flat universe with zero cosmological constant and @xmath149 . the minimum lorentz factor is obtained from equation ( [ eq : nu0 ] ) , @xmath150 finally , substitution of the above results into equation ( [ eq : b ] ) yields , @xmath151 where @xmath152 ; here @xmath153 is the thickness of the shell of the fireball . the ratio of electron to magnetic field energy densities is @xmath154 the minimum combined electron and magnetic field energy , @xmath155 is obtained for a radius @xmath156 , where the equipartition radius @xmath157 is the radius for which @xmath158 . using the above equations we obtain @xmath159 and @xmath160 during sub - relativistic expansion , the fireball hydrodynamics is described by the self - similar sedov - von neumann - taylor solution . we now compare the equipartition energy with the energy of a sedov - von neumann - taylor shock which reaches radius @xmath139 at time @xmath124 . this energy is given through the relation @xmath161 here @xmath8 is the ambient medium density , @xmath162 is the adiabatic index of the gas and @xmath163 is close to unity for @xmath162 in the range of @xmath164 ( relativistic fluid ) and @xmath165 ( non relativistic fluid ) . in what follows we use @xmath166 , appropriate for pressure equilibrium between relativistic electrons and non relativistic protons , for which @xmath167 . using the above equations we have @xmath168 at @xmath169 we find that the sedov - taylor energy is about twice the minimum electron and magnetic field energy required to account for the radio emission , @xmath170 . the thermal energy in a sedov - taylor solution constitutes approximately half the total energy . the energy in the shocked electrons and in magnetic field clearly can not exceed the thermal energy . thus , we obtain an additional constraint , @xmath171 . as can be seen from the above equations this constraint can only be satisfied with @xmath172 . for larger or smaller @xmath139 , @xmath173 would exceed the shock thermal energy , as @xmath174 , while @xmath175 and @xmath176 . thus , although radio observations provide only three constraints on the four model parameters , a self - consistent solution is obtained only for @xmath177 this implies that electrons and magnetic fields are close to equipartition , that the fireball energy is @xmath178 and that the ambient density is @xmath179 interstellar scintillations provide a direct and independent confirmation of the derived radius . as noted in wkf98 , the observations of scintillation yield a fireball radius , @xmath180 cm at @xmath181 d which compares favorably with that stated in equation [ eq : r_0 ] . furthermore , the shock radius derived in equation ( [ eq : r_0 ] ) implies a shock velocity @xmath182 thus , our model predicts relativistic to sub - relativistic transition at @xmath0 d as was indeed inferred from the @xmath4 d data . we also note that , assuming jet expansion at @xmath183 d , the inferred fireball energy is @xmath184 erg , rather than the isotropic @xmath185 erg , and in excellent agreement with the value obtained using the @xmath127100 d data . based on these same radio data , chevalier & li ( 1999a ) have proposed a model in which the afterglow is produced by the expansion of a shock into a circumburst medium , shaped by the mass loss history of the massive progenitor star . in this `` wind '' model , one expects the ambient density to be @xmath186 whereas the density is expected to be constant for an explosion in the typical interstellar medium . there is increasing support for such models , both in direct evidence linking supernovae with gamma - ray bursts ( galama et al . 1998d , kulkarni et al . 1998 ) , and indirect evidence in the optical and radio light curves of the afterglows ( e.g. bloom et al . 1999 , reichart 1999 , galama et al . 1999 , chevalier & li 1999b , frail et al . we find , however , that a wind model is inconsistent with data . let us first consider the temporal dependence of the optical and radio flux . of particular importance , is the location and subsequent evolution of the cooling frequency at early times . for an electron power - law index @xmath187 , implied by early afterglow observations ( galama _ et al . _ 1998a , b ) and used by chevalier & li , the flux at frequencies above the peak frequency makes a transition in the wind model from a @xmath188 decline to @xmath189 , as the cooling frequency @xmath48 passes through the band ( with @xmath190 ) . in contrast , for the same value of @xmath40 , in the constant density model , the cooling frequency decreases in time ( as @xmath191 ) and the flux at frequencies above the peak frequency makes a transition in the wind model from a @xmath192 decline to @xmath188 . early afterglow observations ( galama et al . 1998b ) yield two independent estimates of @xmath48 , the first at @xmath193 hz at @xmath194 d and another at @xmath195 hz at @xmath196 d. chevalier & li ( 1999a ) have questioned the identification of @xmath48 at @xmath194 d , since the optical light curve prior to this time deviates significantly from light curves of afterglow models which account for @xmath197 d data . however , the existence of @xmath48 dropping through the optical band on time scale of days following the burst can also be inferred from infra - red observations ( pian et al . 1998 , chary et al . 1998 ) at @xmath198 d. the galama et al . observations imply that @xmath48 _ decreased _ in frequency from 1.4 days to 12.1 days , in clear contradiction with expectations for the wind model . furthermore , since these observations imply that @xmath48 had passed through the optical bands in the first few weeks , in the wind model both the optical flux ( at @xmath199 d ) and the radio flux ( at @xmath107 d ) should decline as @xmath189 . such behavior is in contradiction with observations , which show a @xmath188 flux decline at both optical and radio frequencies ( note that the optical flux decline in the wind model can be modified to follow a @xmath188 power law by modifying model parameters to give @xmath200 hz , so as to have @xmath48 below the optical band . however , such choice would be inconsistent with the value of @xmath48 inferred from observations ) . moreover , for the energy and density required to account for @xmath201 d observations , @xmath202 erg and @xmath203 , the wind lorentz factor at @xmath134 d is @xmath204 ( the lorentz factor is only weakly dependent on @xmath205 and @xmath138 , and for the parameter choice of chevalier & li @xmath206 ) . thus , the fireball becomes sub - relativistic at @xmath107 d and the flux scaling with time , @xmath207 , which applies for @xmath208 , changes to the sub - relativistic scaling @xmath209 , making the discrepancy with the observed scaling even larger . second , let us consider the self - absorption frequency . our data clearly shows that the flux at 1.43 ghz is self absorbed up to at least @xmath134 d. the time dependence of the self absorption frequency in a wind model , @xmath210 , therefore implies that for this model to fit the late time radio data @xmath211 ghz is required ( indeed the self - absorption frequency for the parameters chosen in chevalier & li is @xmath212 ghz ) . this is in marked contradiction with the value inferred from the data , @xmath213 ghz ( note that this discrepancy in @xmath46 implies a factor of @xmath214 discrepancy between model and observations in the 1.43 ghz to 8.46 ghz flux ratio ) . in this paper we present the 450-day radio afterglow curve of . this is by far the longest light curve obtained for any grb . we also present a comprehensive model to explain the entire light curve . in our model , the fireball is a wide - angle jet which gradually expands sideways and becomes a spherical sub - relativistic shell by about day 100 . the spherical symmetry of the fireball and the sub - relativistic expansion allow us to carry out calorimetry of the explosion , unhindered by unknown geometry . to our knowledge , this represents the first time that one has been able to carry out absolute calorimetry of a grb explosion . below we summarize the salient features of the model and the principal results presented in this paper . the comprehensive model we present has three regimes : ( 1 ) early afterglow , @xmath215 d , ( 2 ) jet afterglow , @xmath216 d and ( 3 ) spherical sub - relativistic afterglow . several authors , in addition to us , have modeled the x - ray , optical and radio - afterglow observations of the early afterglow ( e.g. waxman 1997 , wijers & galama 1999 , granot et al . 1999 , wkf98 ) . the modeling yield electron and magnetic field equipartition factors of @xmath217 . estimates of the total explosion energy depend on the assumed fireball geometry ( spherical versus jet ) . the high lorentz factor hides the true geometry of the fireball . the inferred isotropic afterglow energy is @xmath185 erg which is comparable to the isotropic @xmath3-ray energy release of @xmath218 erg . the spherical model which explains the early afterglow quite nicely over - predicts the observed radio flux by day 25 . we had earlier proposed that this discrepancy could be explained by invoking a jet geometry for the fireball ( wkf98 ) . to our knowledge this was the first explicit suggestion for a jet in a grb . we argue that the jet starts significant sideways expansion at @xmath219 d , and that by day 100 , the fireball becomes essentially a spherically symmetric sub - relativistically expanding shell . the primary observational evidence supporting the transition to the sub - relativistic regime is relation between power law temporal and spectral indices , which is inconsistent with relativistic expansion , and consistent with sub - relativistic expansion . the radio observations presented here of the sub - relativistic phase of fireball expansion are unique . in this phase , the geometry of the fireball is no longer an issue , and the fireball dynamics is described by the well known sedov - von neumann - taylor self - similar solution . thanks to the extensive radio data ( multi - band and long - term monitoring ) we are able to deduce three out of the four fundamental parameters of the synchrotron emitting fireball shell : @xmath135 , the magnetic field strengths ; @xmath8 , the ambient density ; @xmath139 , the radius of the shell ; and @xmath137 , the lorentz factor of the emitting electrons . leaving the radius as the free parameter we estimate the total energy of the emitting particles ( including the energy in magnetic fields ) . we then use the constraint that this energy must not exceed the thermal energy of the a sedov - von neumann - taylor expanding shell . this constraint then yields the size , and hence the total energy , of the expanding shell . we find the following : @xmath1 erg and @xmath220 @xmath221 . furthermore , we find that the electrons and magnetic field are close to equipartition with @xmath222 . confidence in this `` absolute '' calorimetry is obtained when we note that the inferred size of the shell is comparable to that measured from interstellar scintillations ( during the early phase ) . furthermore , our model for @xmath107 d predicts that the transition to sub - relativistic regime takes place at @xmath0 , as is indeed observed ( see above ) . without any doubt , the most important result of this paper is the estimate of the total energy released in the explosion . this energy , @xmath1 erg is well below @xmath9 erg , determined from early afterglow observations and the isotropic @xmath3-ray energy release of @xmath218 erg . thus the energy release in may not be much larger than that in ordinary sne . srk and daf are indebted to barry clark for his generous scheduling of the vla for this long project . daf thanks j. katz and b. paczyski for useful discussions . srk s research is supported by nsf and nasa . ew s research is supported by aec 38/99 and by bsf 9800343 . since the fireball expands at a sub - relativistic velocity in our model , we assume to leading order that the radio flux observed at any given time is emitted by a static shell of radiating electrons . we denote the shell radius by @xmath223 . during sub - relativistic expansion , the dynamics of the shell are described by the self - similar sedov von neumann taylor solution . the shell radius and its velocity @xmath224 scale with time as @xmath225 here @xmath124 is some ( arbitrary ) reference time , and the @xmath226 subscripts denote parameter values at @xmath136 . in what follows we use the normalization @xmath227 d. the compressed , shock heated , material behind the shock occupies a thin shell of width @xmath153 , with @xmath228 . we assume that plasma parameters are uniform within the shell . electrons are assumed to be shock accelerated to a power - law energy distribution , @xmath41 for @xmath42 . we choose @xmath187 , as required by the high frequency observations . the magnetic field energy density , @xmath229 , and electron energy density , @xmath230 are assumed to constitute fixed ( time - independent ) fractions of the shock thermal energy density . under these assumptions , and recalling that the thermal energy density is proportional to @xmath231 , we have @xmath232 the synchrotron power per unit frequency @xmath233 emitted by a single electron of lorentz factor @xmath3 is given by ( rybicki & lightman 1979 ) , @xmath234 , \label{eq : single_e}\ ] ] where @xmath235 and @xmath236 are the electron charge and mass , and @xmath237 the function @xmath238 describes the synchrotron power spectrum ( rybicki & lightman 1979 ) , averaged over an isotropic distribution of pitch angles . to obtain the observed flux at a distance @xmath239 one needs to integrate over the the angular distribution of the intensity @xmath240 at the surface of the shell , @xmath241 . for a thin , @xmath242 , uniform shell , this integral is well approximated by @xmath243 provided the effective optical depth is chosen as @xmath244 . here @xmath245 is the synchrotron emissivity , @xmath246 , and @xmath247 is the synchrotron self - absorption coefficient . for a power - law electron distribution we obtain @xmath248^{5/2 } f_3\left({\nu\over \nu_m}\right ) f_2^{-1}\left({\nu\over \nu_m}\right ) , \label{eq : totalflux}\ ] ] @xmath249^{-(p+4)/2 } f_2\left({\nu\over \nu_m}\right ) . \label{eq : tau}\ ] ] here @xmath250 the functions @xmath251 and the constants @xmath252 @xmath253 @xmath254 the model spectrum is determined by three parameters , @xmath120 , @xmath121 and @xmath122 . these parameters , in turn , are determined by four model parameters , @xmath135 , @xmath139 , @xmath137 and @xmath138 . we define the self - absorption frequency @xmath46 by @xmath255 . for @xmath256 , i.e. when the self - absorption frequency is much smaller than the characteristic synchrotron radiation frequency of the lowest energy electrons , @xmath257 peaks at @xmath258 . the peak frequency and flux scale as @xmath259 and @xmath46 scales as @xmath260 the temporal and frequency dependence of @xmath257 is for @xmath262 , @xmath257 peaks at frequency @xmath263 . the optical depth at the peak frequency , @xmath264 , satisfies @xmath265 for @xmath266 this implies @xmath267 , @xmath268 , and for @xmath269 we have @xmath270 , @xmath271 . in this case , @xmath272 and the temporal and frequency dependence of @xmath257 is @xmath273 in our analysis we have so far neglected the effects of synchrotron cooling of electrons . synchrotron cooling is negligible for electrons radiating at radio frequencies , since @xmath274 hz at @xmath134 d ( see 5 ) , but may be important for electrons radiating at higher frequencies . the electron synchrotron cooling time is proportional to @xmath275 , where @xmath3 is the electron lorentz factor , while the shell expansion time is proportional to @xmath276 . thus , the lorentz factor @xmath277 of electrons for which the cooling time is comparable to the expansion time scales as @xmath278 , and the frequency @xmath48 , beyond which electron cooling affects the observed radiation , scales as @xmath279 at frequencies higher than @xmath48 , electron cooling steepens the spectrum and @xmath280 note that for @xmath266 , @xmath281 ( at @xmath282 ) for both relativistic and sub - relativistic expansion . lrcrc 1997 may 9.05 & 0.15 & & 45 & b 1997 may 9.84 & 0.93 & & 89 & b 1997 may 15.09 & 6.19 & 100 & 32 & b 1997 may 18.83 & 9.93 & & 63 & b 1997 jun . 1.00 & 23.10 & & 52 & cnb 1997 jun . 2.32 & 24.41 & 135 & 45 & cnb 1997 jun . 12.83 & 34.93 & 130 & 20 & cnb 1997 jul . 18.62 & 70.72 & 305 & 48 & c 1997 jul . 28.94 & 81.03 & 485 & 45 & cs 1997 aug . 19.02 & 102.11 & 250 & 70 & cs 1997 aug . 22.52 & 105.62 & 230 & 59 & cs 1997 sep . 3.47 & 117.56 & 206 & 63 & cs 1997 sep . 8.42 & 122.51 & 216 & 67 & cs 1997 sep . 12.68 & 126.78 & 333 & 65 & cs 1997 sep . 26.35 & 140.45 & & 83 & dnc 1997 oct . 4.63 & 148.73 & 217 & 105 & dnc 1997 oct . 13.22 & 157.32 & 381 & 127 & dnc 1997 oct . 21.39 & 165.48 & 250 & 66 & dnc 1997 nov . 8.09 & 183.19 & 338 & 63 & d 1997 nov . 11.33 & 186.42 & 497 & 51 & d 1997 nov . 13.97 & 189.06 & 376 & 93 & d 1997 nov . 17.65 & 192.75 & & 93 & d 1997 nov . 19.46 & 194.55 & 303 & 53 & d 1997 nov 22.46 & 197.56 & 199 & 62 & d 1997 nov . 23.05 & 198.14 & 271 & 74 & d 1997 nov . 24.96 & 200.05 & 208 & 77 & d 1997 nov . 26.17 & 201.26 & 155 & 43 & d 1997 nov . 27.12 & 202.21 & 170 & 50 & d 1997 nov . 28.40 & 203.50 & & 62 & d 1997 nov . 30.32 & 205.42 & 184 & 42 & d 1997 dec . 1.40 & 206.49 & & 71 & d 1997 dec . 6.65 & 211.75 & & 68 & d 1997 dec . 11.16 & 216.25 & & 64 & d 1998 feb . 23.44 & 290.53 & 274 & 27 & a 1998 mar . 18.06 & 313.16 & 434 & 24 & a 1998 apr . 27.66 & 353.76 & 249 & 60 & a 1998 may 2.52 & 358.62 & 204 & 30 & a 1998 jul . 19.78 & 436.89 & & 32 & b 1998 jul . 28.73 & 445.83 & & 33 & b lrcrcrc 1997 may 13.96 & 5.06 & & & 430 & 25 & b 1997 may 15.13 & 6.23 & 330 & 33 & 610 & 33 & b 1997 may 16.49 & 7.59 & & & 520 & 18 & b 1997 may 16.79 & 7.89 & & & 602 & 24 & v 1997 may 17.30 & 8.40 & & & 604 & 56 & v 1997 may 18.85 & 9.95 & 480 & 64 & 610 & 51 & b 1997 may 21.00 & 12.10 & & & 1270 & 120 & v 1997 may 21.13 & 12.23 & & & 930 & 120 & v 1997 may 22.10 & 13.20 & 410 & 47 & 570 & 42 & b 1997 may 22.49 & 13.59 & 360 & 31 & 880 & 33 & b 1997 may 23.00 & 14.10 & & & 1260 & 100 & v 1997 may 23.13 & 14.23 & & & 960 & 100 & v 1997 may 24.48 & 15.58 & & & 550 & 44 & b 1997 may 24.96 & 16.06 & & & 470 & 43 & b 1997 may 25.85 & 16.95 & & & 480 & 45 & b 1997 may 27.67 & 18.77 & 770 & 54 & 500 & 49 & b 1997 may 28.10 & 19.20 & & & 538 & 100 & v 1997 may 28.23 & 19.33 & & & 835 & 100 & v 1997 may 29.10 & 20.20 & 1350 & 55 & 1200 & 47 & cnb 1997 may 30.95 & 22.05 & 740 & 46 & 810 & 48 & cnb 1997 may 31.99 & 23.09 & 550 & 43 & 290 & 46 & cnb 1997 jun . 2.29 & 24.39 & 650 & 44 & 720 & 62 & cnb 1997 jun . 2.41 & 24.51 & & & 940 & 27 & cnb 1997 jun . 2.50 & 24.60 & & & 960 & 40 & v 1997 jun . 2.90 & 25.00 & 540 & 46 & 760 & 45 & cnb 1997 jun . 3.94 & 26.04 & 530 & 51 & 670 & 43 & cnb 1997 jun . 5.97 & 28.07 & 850 & 36 & 690 & 28 & cnb 1997 jun . 9.74 & 31.84 & 470 & 38 & 530 & 38 & cnb 1997 jun . 12.99 & 35.09 & 640 & 43 & 600 & 39 & cnb 1997 jun . 14.73 & 36.83 & 590 & 69 & 460 & 66 & cnb 1997 jun . 17.92 & 40.02 & 400 & 49 & 560 & 40 & cnb 1997 jun . 18.96 & 41.06 & 640 & 55 & 630 & 63 & cnb 1997 jun . 20.86 & 42.96 & 550 & 36 & 800 & 32 & cnb 1997 jun . 22.93 & 45.03 & 425 & 43 & 645 & 38 & cnb 1997 jun . 26.63 & 48.73 & 610 & 70 & 680 & 46 & cnb 1997 jun . 28.68 & 50.78 & 500 & 34 & 565 & 36 & c 1997 jul . 2.36 & 54.46 & 805 & 38 & 805 & 43 & c 1997 jul . 4.59 & 56.69 & 605 & 50 & 710 & 49 & c 1997 jul 11.10 & 63.20 & 630 & 32 & 725 & 40 & c 1997 jul . 14.61 & 66.71 & 585 & 48 & 680 & 46 & c 1997 jul . 15.33 & 67.43 & 550 & 31 & 630 & 37 & c 1997 jul . 18.44 & 70.54 & 505 & 62 & 575 & 61 & c 1997 jul . 18.64 & 70.74 & 535 & 43 & 515 & 41 & c 1997 jul . 22.55 & 74.65 & 385 & 51 & 440 & 51 & cs 1997 jul . 30.46 & 82.56 & 360 & 70 & 455 & 85 & cs 1997 aug . 4.01 & 87.11 & 370 & 38 & 380 & 42 & cs 1997 aug . 5.35 & 88.45 & 550 & 61 & 445 & 50 & cs 1997 aug . 7.23 & 90.33 & 515 & 37 & 430 & 39 & cs 1997 aug . 11.26 & 94.36 & 615 & 37 & 640 & 40 & cs 1997 aug . 15.13 & 98.23 & 770 & 57 & 565 & 49 & cs 1997 aug . 16.62 & 99.72 & 610 & 55 & 400 & 60 & cs 1997 aug . 18.95 & 102.05 & 500 & 42 & 300 & 35 & cs 1997 aug . 22.50 & 105.60 & 710 & 52 & 470 & 44 & cs 1997 aug . 31.82 & 114.92 & & & 315 & 70 & dnc 1997 sep . 3.45 & 117.55 & 425 & 57 & 355 & 47 & dnc 1997 sep . 4.71 & 118.81 & & & 370 & 70 & dnc 1997 sep . 8.40 & 122.50 & & & 350 & 47 & dnc 1997 sep . 12.64 & 126.74 & 400 & 41 & 370 & 39 & dnc 1997 sep . 26.36 & 140.46 & 380 & 40 & 330 & 45 & dnc 1997 oct . 4.62 & 148.72 & 420 & 44 & 310 & 56 & dnc 1997 oct . 13.23 & 157.33 & 350 & 40 & 280 & 37 & dnc 1997 oct . 21.40 & 165.50 & 300 & 34 & 240 & 34 & dnc 1997 oct . 26.56 & 170.66 & 280 & 31 & 300 & 37 & dnc 1997 nov . 6.19 & 181.29 & 385 & 32 & 240 & 24 & d 1997 nov . 19.15 & 194.25 & 325 & 37 & 240 & 30 & d 1997 dec . 9.14 & 214.24 & 180 & 35 & 170 & 30 & d 1998 jan . 2.08 & 238.18 & 220 & 35 & 160 & 30 & d 1998 jan . 18.90 & 255.00 & 275 & 53 & & & d 1998 jan . 20.98 & 257.08 & 235 & 41 & 150 & 36 & d 1998 jan . 27.77 & 263.87 & 220 & 50 & & 50 & d 1998 jan . 31.29 & 267.39 & & 80 & & 50 & d 1998 feb . 8.04 & 275.14 & 200 & 24 & 95 & 24 & a 1998 mar . 14.09 & 309.19 & 165 & 17 & 105 & 15 & a 1998 may 13.52 & 369.62 & 125 & 28 & & 25 & a 1998 jun . 21.83 & 408.93 & 104 & 44 & & 37 & bna ccccl 86.2 & 10 - 14 & 1620 & 250 & bremer et al . ( 1998 ) 15.0 & 8 - 13 & 660 & 110 & pooley & green ( 1997 ) 8.46 & @xmath7690 & 630 & 107 & this paper 4.86 & @xmath7690 & 555 & 108 & this paper 1.43 & 310 - 490 & 300 & 150 & this paper cccl 2 - 4 & @xmath283 & @xmath284 & djorgovski et al . ( 1997 ) 2 - 25 & @xmath285 & @xmath286 & pian et al . ( 1998 ) 2 - 120 & @xmath287 & @xmath288 & galama et al . ( 1998c ) 2 - 90 & @xmath289 & @xmath290 & sokolov et al . ( 1998 ) 2 - 100 & @xmath291 & @xmath292 & zharikov et al . ( 1998 ) 90 - 300 & @xmath293 & @xmath294 & this paper
in this second paper of a series of papers based on the first and sdss surveys we investigate the evolution of galaxy morphology and nuclear activity in the look - back time of the sdss ( 2 gyrs ) for a sample of 150000 galaxies in the local universe . we demonstrate an evolution in the strength of the radio power and the spectroscopic emission - lines typical of agn , as well as in the morphology of their hosts . such evolution appears more substantial for less luminous systems , and is possibly the low - redshift tail of the downsizing in star - formation , agn activity and supermassive black hole build - up observed in higher redshift surveys.this suggests that the differences in intrinsic properties of galaxies along the hubble sequence may arise from the difference in the depth of their potential wells which leads to different evolutionary paths because of different timescales for gas infall . this primordial infall and the following secular evolution mediated by bar and density wave instabilities may bring galaxies of different mass to have the different activity levels and morphological features in the local universe shown in this study . in agreement with such a hypothesis , we find that star - formation as traced by radio emission is progressively more centrally concentrated in more evolved star - forming galaxies and we show that the environment in which a galaxy resides plays a lesser role in shaping the features and activity for the majority of systems .
0903.0634
understanding the activity of galaxies , their morphologies , and the interplay between the two are key problems in galaxy evolution and are tightly correlated with the problem of understanding whether such properties depend on the initial physical conditions of a galaxy ( e.g. , its mass ) or , rather , result from evolutionary processes driven by the environment in which they live , such as mergers , gravitational and gas - gas interactions . models of galaxy formation and evolution have tried to explain the differences in the features of galaxies for decades , mostly unsuccessfully . hierarchical cold dark matter models , which have been shown to describe successfully large - scale structure formation , currently fail to describe properly many salient properties of galaxies , such as the color - magnitude relation , the bimodal distribution in colors , the luminosity function , and downsizing " ( i.e. , the earlier quenching of star formation in more massive objects ) . furthermore such models inadequately describe the formation of small bulge - less spirals , the most common galaxies in the universe . it is unclear if the problems that these models encounter are given by a simplistic treatment of the complex baryonic physics governing the observable parameters or if the models are not appropriate for the description of galaxy formation and evolution . explaining the mechanisms which initiate and regulate star - formation and agn activity in galaxies along the hubble diagram has proven as challenging . galaxies show very different rates and histories of star - formation activity and host agn with very different characteristics ( narrow and/or broad emission lines , point - like radio emission and large radio jets , to name a few ) , making the modeling of such activity quite complex . the interplay between star - formation and agn activity is also poorly understood . using the sdss , @xcite showed that active galactic nuclei selected on the basis of their emission lines preferentially reside in massive , bulge - dominated systems with traces of recent star - formation , possibly a consequence of the presence of a reservoir of cold gas which triggers both star - formation and agn activity . using the goods survey , however , @xcite showed that x - ray agn activity persists in many @xmath1 red - sequence galaxies , well after the star - formation has quenched , suggesting that the presence of a cold reservoir of gas may not be a necessary ingredient in the triggering of agn activity . the role of the environment surrounding a galaxy in initiating activity is also quite controversial . interactions and mergers among galaxies are known to trigger bursts of star formation in galaxies ( e.g. , * ? ? ? * ) , yet the trigger for the onset of star - formation on galactic scales in isolated spiral galaxies calls for a different explanation . in clusters , @xcite have shown that the environment produces a significant effect in the later evolution of galaxies , lowering their star - formation rates substantially , while @xcite using the sdss sample conclude the opposite , i.e. that the cluster environment may be irrelevant to the bulk of the transformation of galaxies . mergers and close interaction between galaxies have also been suggested as possible triggers of agn activity @xcite ; however , @xcite using the goods survey and two chandra deep fields have shown that there is no close connection between galaxy mergers and moderate - luminosity agn activity out to appreciable look - back times approaching the epoch of peak agn activity in the universe ( @xmath2 ) . this result is supported by the study of @xcite who showed that x - ray - selected agn tend to reside in undisturbed pairs of early type galaxies with no direct evidence of gravitational perturbation or interactions , suggesting that the activation mechanism of agn is not primarily galaxy interaction . the role of secular evolution in triggering star - formation and agn activity , as well as the morphological transformation of galaxies by means of a re - arrangement of energy and mass through collective phenomena such as bars @xcite and/or spiral waves @xcite is unclear , but there is increasing observational evidence of its importance in shaping galaxy features @xcite . interestingly , many properties observed in the local universe are observed also in the high- redshift population of galaxies investigated by recent deep galaxy surveys . using the goods survey , @xcite have shown that the well - known correlation between size of a galactic bulge and the mass of the supermassive black hole residing at its center @xcite is in place at early epochs ( [email protected] ) . @xcite , using the cosmos survey , have studied the morphological content and environmental dependence of the galaxy color - magnitude relation at @xmath4 and found that the red and blue sequences of galaxies found in the nearby universe are found at @xmath4 as well . by comparing the high- and low - redshift populations , they conclude that the local red sequence is adequately reproduced by passive evolution , suggesting that a fraction of the galaxies in the high - redshift blue sequence eventually evolve into red - sequence galaxies . they also confirm the result of @xcite , who showed that the slope and the normalization of such color - magnitude relations are substantially invariant with respect to local density when galaxies are separated into early and late types , arguing that secular star - formation is driven more by galaxy mass than by environment . the color - density correlation found by other authors ( e.g. , * ? ? ? * ) at high redshift , with bluer galaxies found in regions of greater over - density , is interpreted as an excess of more late - type galaxies in denser environments which evolve into red - sequence galaxies at the present time . these studies are in agreement with older results @xcite which showed a variation in the fractional abundance of different morphological types in galaxy cluster populations as a function of redshift . the study of @xcite is not limited to clusters , suggesting that a transformation of galaxies might also occur in the absence of mechanisms such as the gas - gas and gravitational interactions typical of dense cluster environments . these results are at odds with the work of @xcite who have used the sdss sample to show that close interaction among galaxies is the most likely mechanism leading to the density - morphology correlation observed . in a series of three papers based on the sloan digital sky survey @xcite and the first survey @xcite we examine the population of broad and narrow emission - line active galactic nuclei ( agn ) in the local universe in order to explore the physical and environmental properties of galaxies . in paper i @xcite we described the multiwavelength database built for this analysis and the systematics affecting the spectroscopic selection of agn . we presented evidence of the existence of a large population of radio - emitting radio agn which lack optical spectroscopic signatures and show that , at all luminosities , galaxies with redder colors are more likely to show weaker emission lines . in this second paper we further explore the interplay between galaxy properties and agn features , focusing on the interplay between morphology , star - formation , and nuclear activity , including its evolution over cosmic times . in a following paper we will present a comparative analysis of the radio and spectroscopic properties of narrow- and broad - lined agn and their evolution . the structure of the paper is the following : in section [ sample_ch3 ] we present the sample used for this analysis , while in [ evolution ] we show our evidence of co - evolution of galaxy activity and morphology , and in [ aging_radio ] we map the aging of the radio power associated with agn . in [ aging_spectro ] we illustrate the aging of the spectroscopic features of agn , in [ environment ] we study the correlation between environment , activity and morphology . we discuss our results in [ discussion_ch3 ] and present our conclusions in [ conclusion_ch3 ] . in this paper we further analyze a sample of 151815 galaxies with petrosian r magnitude in the range 14.5@[email protected] drawn from the sdss second data release , for which we have constructed a multiwavelength database , gathering information about their radio emission from the first and nvss surveys , the far - infrared emission from iras , and x - ray emission from the rosat surveys . details on the database construction are given in paper i. the density of the environment surrounding each galaxy has been calculated using the tridimensional density estimator discussed in paper i ; further description of this method can be found in @xcite . the spectroscopic classification of galaxies is adopted from @xcite and @xcite , along with the extinction - corrected fluxes for the main spectroscopic lines , their equivalent widths , and the extinction - corrected estimates of the stellar masses of the galaxies in the sample . as discussed in paper i , we have modified the classifications to correct for effects of dilution and host light pollution which tend to bias the classifications with increasing redshifts . about 6% of the sample shows detectable radio emission in either first or nvss . for these radio - emitting galaxies , the spectroscopic classification has been complemented with an independent classification based on the radio and optical morphology of the sources , as discussed in paper i and in @xcite . the look - back time of the sdss is about 2 gyrs in standard cosmological models ( @xmath5 and @xmath6 ) . the typical timescales for the processes associated with agn activity and star - formation are much shorter , typically a few tens of millions of years . therefore we wish to investigate whether evolution of activity and morphology can be detected over the look - back time of the sdss . in order to compare the properties of galaxies at different redshifts it is crucial to minimize selection effects given by the survey systematics . in paper i we showed how dilution significantly biases the selection of active galaxies with increasing redshift , and devised a method to correct for this bias . this correction is essential , since dilution is a redshift - dependent systematic and if not properly corrected , could mimic evolution of the sample with redshift ( cf . paper i for a more detailed discussion ) . in order to explore the properties of agn and host galaxies over the look - back time of the sdss , we have focused on volume - limited samples of galaxies which ensure completeness of the optical sample within the redshift and absolute magnitude cuts applied . we have constructed a chain of eight volume - limited samples covering the absolute magnitude range @xmath7 at intervals of 0.5 in magnitude . this ensures that we are looking at optically complete samples of agn hosts with similar properties . in figure [ abs_r_agn ] we show the scatter plot in redshift and absolute magnitude of galaxies belonging to each volume - limited sample . for the agn - classified galaxies , we show in figure [ lum_ha_z_agn ] the scatter plot of h@xmath8 luminosity versus the luminosity distance , and we note how the h@xmath8 luminosity decreases over time . the effects of dilution have been corrected for ( no lower cut appears in the plots ) but even if they had not , this would not account for the lack of strong emission - line agn at lower redshifts . differences in the volumes sampled at various redshifts , however , might explain the lack of high - luminosity systems at lower redshifts . a better way to explore this trend is to consider for each volume - limited sample the ratio of galaxies with h@xmath8 luminosity above the median of the sample to those with a luminosity below the median . this method allows us to eliminate the possibility that the trends observed in the scatter plots simply result from the fact that at higher redshifts larger volumes are sampled and therefore a larger number of high luminosity systems is to be expected ; with our procedure the volume effect is removed . in figure [ lum_ha_z_agn_frac ] we show that the strength of the h@xmath8 line has indeed decreased over time . in the next section we will demonstrate that this decrease is accompanied by a decrease in the radio luminosity of these active nuclei . as noted earlier , this is part of a more general evolution of agn strength in radio and x - rays observed from the high- to the low - redshift universe ( @xcite ) . the trends observed are unlikely the result of the inclusion in the fiber , with increasing redshift , of more h@xmath8 emission from star - forming regions residing in the outer part of the host : in order to explain the increase in h@xmath8 luminosity observed in each volume - limited sample , these star - forming regions would need to account for as much as three times the h@xmath8 emission observed in the nuclear region at lower redshifts . this would severely affect the spectral classification of these objects , making them fall in composite or sfg class . because of the requirement that each sample is volume - limited , the look - back time is shorter for less luminous systems . in order to compare the various samples , we have extrapolated the evolution in the look - back time considered to the same time interval @xmath9gyr . in figure [ rate_halpha_drop ] we show the trend for the parameter @xmath10 , where @xmath11 and @xmath12 refer to the fourth ( highest redshift ) and first ( lowest redshift ) bin in each sub - sample and @xmath13 is the associated cosmic time in a standard @xmath5 , @xmath6 cosmological model . this quantifies the order of magnitude by which the luminosity would increase over a 1 gyr time interval , assuming a first - order interpolation of the trends shown in fig.[lum_ha_z_agn].note that the bins for the lowest luminosity galaxies are both sparsely populated _ and _ extend only over a modest range of redshift , so the large changes seen may well be exaggerated ; over most of the luminosity range , the results suggest a factor of 2 - 3 decrease in h@xmath8 luminosity over the past gigayear . the evolution of the agn h@xmath8 power in less luminous galaxies is stronger than the evolution of this quantity in higher luminosity systems . this finding is in agreement with the evidence of downsizing in agn activity , with agn activity peaking at earlier epochs in more massive systems ( @xcite ) . according to our results , the most massive galaxies , have decreased their h@xmath8 by circa two orders of magnitude in the past 10 gyr , which means that at early times they had luminosities typical of strong agn and quasars , a fact that we will further discuss in paper iii . we note , however , that if black holes in less massive systems have been growing significantly only in recent epochs as suggested by downsizing , low - mass systems have not been steadily decreasing their luminosity over the past @xmath14 gyr like the most massive ones ( which would make these low - mass systems surprisingly bright at very high redshifts ) but only in a much more recent past . interestingly , this change in strength is accompanied by an increase in the concentration of the light as shown in figure [ conc_agn_frac ] and an increase in dn@xmath15 parameter describing the average time since the last burst of star formation ( figure [ d4000_agn_frac ] ) . this suggests a co - evolution of star - formation , agn and the morphological properties of galaxies over cosmic time . the transformation of the light concentration is stronger in less luminous systems : this is shown in figure [ rate_conc_drop ] where we plot the trend for the parameter @xmath16 which quantifies the factor by which the concentration parameter c increases over 1 gyr , assuming a first - order linear interpolation . we note that the c parameter spans a range in values between 0.6 for disks and 0.2 for ellipticals a change of @xmath0 0.04 in such parameter is therefore a @xmath0 10% effect . this suggests that bulges in less massive systems are forming at recent epochs , in agreement with the findings of @xcite . we further inspected the variation in the light - concentration for the other spectroscopic types ; since star - forming and composite systems are associated with lower luminosity systems , we have selected different intervals in magnitude : -22.0@xmath17r@xmath17 - 18.0 . while for composite systems we do observe a change in the number of systems with concentration above the median of each sample in their respective look - back times ( fig.[conc_com_frac ] ) , the same is not observed in sfg galaxies ( fig . [ conc_sfg_frac ] ) . this is in agreement with the fact that agn activity is strongly correlated with an enhancement in the concentration of galaxy light , typical of the presence of a bulge , and suggests that the disk / sfg phase is a relatively stable phase for galaxies , while the agn / bulge phase is not : i.e. , the transition from a disk to a bulge - dominated system must be rapid . from this point of view , the existence of the red and blue sequences and a clean cut in morphology for galaxies at u - r = 2.22 @xcite can be interpreted as a consequence of the rapid evolution of systems of intermediate size , from star - forming disks to early - type systems , in agreement with the suggestions @xcite . in paper i we have shown that @xmath18 of emission - line agn at z @xmath17 0.2 show radio emission detectable by first and/or nvss surveys . the undetected emission line population has a mean radio flux density detectable by stacking first images that is only an order of magnitude below the detection threshold . furthermore we have shown that there exists a spectroscopically unremarkable population of agn detected in first which accounts for 4% of the early - type population . since radio emission and optical emission lines are independent features of an active nucleus , with one not necessarily requiring the other @xcite , we have explored the evolution of agn activity in the sdss look - back time for agn radio emission , in order to see if the tail of the well - known decrease in radio luminosity over cosmic time ( e.g. , * ? ? ? * ) can be detected in the look - back time of the sdss . using the technique described in the previous sections , we have selected a chain of volume - limited samples and we have plotted radio luminosity in each redshift range selected . since nvss and first are both flux - limited surveys , selecting a volume - limited sample in the optical does not remove selection effects in the radio band : brighter radio sources are expected to be selected at higher redshifts in all volume - limited samples . in order to avoid this effect produced by the incompleteness of the radio sample at higher redshifts , we have used the aforementioned stacking procedure @xcite to evaluate the median luminosity of radio agn in bins of redshift for each volume - limited sample . the procedure effectively removes the selection given by the flux - limited nature of the radio surveys since all fields , including those with source flux densities below the survey flux - limit , are used for obtaining the median luminosity . for each volume - limited sample we have thus considered four bins in redshift and plot the radio luminosity obtained by stacking all fields containing emission - line agn . the median luminosity was obtained by stacking in luminosity space and then fitting a two - dimensional gaussian profile to the median image obtained ( see paper i for details ) . the results are shown in figure [ stacks_evolution_agn ] . we find that the average radio luminosity of emission - line systems has decreased over time in nearly all volume - limited samples . the average k - correction is of order 10% , too small to account for the decrease in radio luminosity found in the data . the stacked radio sources were unresolved ( or marginally resolved ) : the radio luminosity is therefore equivalent to the peak radio emission coming from the central radio agn . the fact that the first survey resolves out part of the flux of extended sources therefore does not affect our analysis . we note a relationship between the strength of the radio emission and the optical luminosity of its host : galaxies with lower optical luminosities harbor less intense radio agn , a fact we will investigate further in paper iii . the decrease in radio power is more pronounced in galaxies of lower luminosity , as we found for the h@xmath8 emission : as shown in figure [ stacks_evolution_agn ] , no evolution is seen in either the -23.5@xmath17r@xmath17 - 23.0 bin or the -20.0@xmath17r@xmath17 - 19.5 : a lack of evolution in the concentration parameter was also noted for the highest luminosity bin ( fig . [ conc_agn_frac ] ) . in the -20.0@xmath17r@xmath17 - 19.5 bin the look - back - time may be too short for observing an actual change . if we consider spectroscopically passive galaxies which show no significant signs of agn activity in their spectra , we find a less marked decrease in the median power of their agn from higher to lower redshift . this is shown in figure [ stacks_evolution_abg ] . for absolute magnitudes fainter than -21.5 , no flux is detected in our stacks ( @xmath17 10 @xmath19jy ) , showing that low - luminosity hosts lacking emission lines harbor fainter , or non - existent , radio agn . in current radio agn production models , the radio power is connected with accretion rate @xcite . if these systems have used up most of their gaseous reservoir at earlier times as suggested by the lack of emission lines , it is likely that their accretion rates are lower and their radio emission weaker . we will investigate this fully in paper iii . in summary , it appears that more evolved , more luminous galaxies show less sign of evolution in radio emission over the past 1 gyr , consistent with the trends found for the emission lines and the morphology of the host . in paper i we showed that systems with unremarkable lines are often weak agn with weak emission lines that are misclassified as passive galaxies because of nuclear light dilution by the host galaxy stellar light . we have shown that this population is on average redder than the population with detected lines , and suggested that there is a physical difference between the two populations . in figure [ color_trend_agn_new ] we show that this is part of a more general trend of line diminution in redder systems . if we select all systems at [email protected] with the four standard btp lines detected ( requiring that their ew is @xmath201.0 ) and divide these systems between seyfert ii galaxies and liners using the usual cut [ oiii]/h@xmath213 , we find that the seyfert ii population is on average bluer than the liner population . if we consider the class of agn with h@xmath8 and [ nii ] detected ( ew @xmath201 ) , but [ oiii ] and h@xmath22 marginally detected or undetected ( ew@xmath171 ) , which we will refer to as les ( low excitation systems ) , we find that this class is redder than the liners . if we then consider the class with all four btp lines marginally detected or not detected at all ( ew of all four lines @xmath171 ) which we reclassified as agn according to the criteria discussed in paper i , this class is redder than the les class . finally , if we consider the sub - sample of truly spectroscopically passive galaxies with all lines not detected , but showing x - ray or radio agn , we find that these galaxies are redder than all other classes . this strongly suggests that the line strength in active galaxies is strongly linked with the galaxy color . the colors adopted have been corrected for dust reddening by means of the estimate of the extinction coefficient a@xmath23 obtained by @xcite . estimate of the extinction in the u and r bands from the a@xmath23 coefficient have been obtained by means of the extinction law calculated by @xcite , where the a@xmath23/a@xmath24 value needed for the conversion has been obtained by interpolating the values for the i and j band , since the average wavelength of the z band is 8931 angstrom , in between the values for the standard j and the i bands . the trends found are therefore not an effect of dust extinction , i.e. , they are not given by a correlation between the nuclear extinction and the global extinction in the galaxy that would imply that more dusty galaxies also have more dusty nuclei . in figure [ color_trend_agn_split_new ] we show that the same trends are found when hosts with similar luminosities ( and therefore with similar supermassive black hole mass ) are chosen by restricting the sample to volume - limited sub - samples spanning intervals of only 0.5 in absolute @xmath25 magnitude . this suggests that the mass of the supermassive black hole is unlikely to be the parameter that drives the difference in the prominence of the lines . as we will fully explore in paper iii , the accretion rate onto the supermassive black hole could be the parameter that regulates the prominence of the lines . it is reasonable to expect that , as systems become more and more gas depleted and therefore redder as a consequence of their decrease in star - formation rate , their accretion rates drop . lower accretion rates would power less luminous nuclear emission , leading to less prominent recombination lines . if galaxies progressively use up their gaseous reservoir , then more gas - depleted galaxies should have quenched their star formation at earlier times . in this case , the prominence of their lines in spectra should be correlated not only with the color of the galaxy , but also with the dn@xmath15 @xciteand the h@xmath26 @xcite parameters , which are indicators of the age of their stellar population . quantifies the break in galaxy spectra occurring at 4000 , a feature which arises because of the accumulation of a large number of absorption lines in a narrow wavelength range . these absorption lines arise in stellar atmospheres . young stars , having highly ionized atmospheres , are ineffective at absorbing photons , so galaxies with young stellar population have weaker dn@xmath15 breaks . h@xmath27 , on the contrary , quantifies the strength of the h@xmath26 absorption line , which is typical of a stars . h@xmath26 therefore peaks in galaxies in which the light is dominated by such a population and then decreases as the a - star population fades and the galaxy becomes redder . ] in figure [ d4000_trend_agn ] and [ hdelta_trend_agn ] we show the median values of the dn@xmath15 and h@xmath26 parameters for the five different classes of objects discussed above . it is clear that weaker lines are associated with galaxies with older stellar populations , since the trend for dn@xmath15 is decreasing and the trend for h@xmath28 is increasing . if we split our sample into four classes of luminosity as before , we find that such trends persist for all types of galaxies confirming that , at all galaxy luminosities , lines become weaker as the star - formation is quenched and the galaxy stellar population becomes redder ( fig . [ d4000_trend_agn_split ] and [ hdelta_trend_agn_split ] ) . if we consider the evolution of the dn@xmath15 parameter after a burst of star - formation as shown in kauffmann et al . 2003 ( we show their plot in fig . [ kauff_d4000_time ] ) we find that a variation from 1.6 to 1.9 in the d4000 parameter corresponds approximately to an interval of 4@xmath29 years . at each luminosity , the abg galaxies had their last burst of star formation @xmath30 years earlier than the seyfert population . if we consider seyferts and liners we have a variation of dn@xmath15 of @xmath31 , which corresponds to @xmath32 yrs . this implies that on timescales of @xmath33 yrs a significant variation of the nuclear spectrum can be observed . the look back time for the sdss survey is @xmath34 . therefore , if this line of reasoning is correct , a decline in the strength of lines for a fixed host should be observable , as shown in section [ evolution ] . differences in star - formation and agn activity in galaxies may depend on their intrinsic properties , such as the depth of their potential well or their richness in gas , or by the effects of the environment in which they reside . denser environment favor close encounters among galaxies which are known to disturb the distribution of their interstellar medium , triggering star - formation activity and possibly agn activity . we have therefore explored further the dependence of activity and morphology on environment and galaxy mass / luminosity to clarify this issue . we have associated all galaxies with a density calculated by sampling the volume defined by the tenth nearest neighbor to a galaxy and statistically correcting for unseen galaxies , as described in @xcite and in paper i. this provides us with an estimate of the environmental density on a few megaparsec scale , effectively dividing the sample into the different regions of large - scale structure : voids , filaments , and clusters . the estimator chosen is not sensitive to substructure on smaller scales and therefore does not `` see '' cluster cores or tight pairs , the study of which is beyond the scope of this paper . in figure [ density_morphology ] we show that the well - known density - morphology relation , originally observed in clusters by @xcite , is observed also in the lower density regime of filaments and voids , with an increase in late type systems with decreasing large - scale environmental density . this trend is observed both both if we divide early- and late - types according to color following @xcite or on the basis of the likelihood of an exponential or de - vaucouleurs fit to their light profiles . this change in the typical morphology of galaxies with increasing density is accompanied by a change in activity as well . in figure [ fraction_rho_final ] we show the fractional abundance as a function of density for the four main spectroscopic types , after correction for dilution as discussed in paper i : star - forming galaxies ( sfg ) , agn , composite galaxies ( star - forming plus agn ) and passive galaxies showing only absorption lines in their spectra ( abg ) . it is evident how galaxies with different types of activity inhabit different environments : star - forming galaxies preferentially reside in low - density environments , galaxies showing both star - formation and agn activity signatures in environments of intermediate density , galaxies with spectroscopic agn activity in dense environments and finally , galaxies lacking emission lines in very dense environments . as discussed in paper i , 4% of these passive galaxies do harbor radio agn at the sensitivity threshold of first and nvss , and the majority of such galaxies host radio agn typically one order of magnitude fainter . the abg class , therefore , should be considered as passive only spectroscopically . we note that the smooth transformation of activity with environment from star - formation - dominated to agn - dominated and eventually into galaxies lacking signs of nuclear recombination lines is not only accompanied by a change in the relative abundance of early- to late - type galaxies in denser environments ( as described by the usual density - morphology relation ) , but also by a smooth transition in the morphology of galaxies , with an increasing prominence of the bulge relative to the disk . in order to quantify this variation we have considered two different indicators . the first takes advantage of the fact that the light profile of disk galaxies is well - described by an exponential law , while the light profile of bulges is well - described by a de - vaucouleurs law . as a consequence , for each galaxy , the parameter @xmath35 defined as the arcotangent of the ratio of the exponential fit likelihood @xmath36 to the de - vaucouleurs fit likelihood @xmath37 , @xmath38 , is a measure of the probability that the system is disk or bulge dominated . values of @xmath35 close to @xmath39 describe galaxies with a very high probability of being bulge - dominated , while values of @xmath35 close to zero describe galaxies with a very high probability of being disk - dominated . values in between represent systems with intermediate features or irregular galaxies . the smooth variation from disk - dominated systems to bulge - dominated systems with environment is shown in figure [ bulgy_density ] . a transformation from disk- to bulge - dominated systems along the hubble sequence can also be traced using a different estimator the concentration parameter c , given by the ratio of the radius containing 50% of the galaxy light to the radius containing 90% of the light . moving from low to high values of c signifies a change in the concentration of the light which describes a change in morphology from bulge - dominated to disk - dominated systems , since the parameter is inversely proportional to the concentration of the light . the smooth variation of the value of c with environment is shown in figure [ conc_density ] . this shows that the correlation of density with morphology is not simply given by a different fraction of ellipticals to spirals in different environments , but that the bulginess of systems varies smoothly across environments . this change accompanies the variation in the type of activity with environment noted above . since the average mass and luminosity of galaxies increases with increasing environmental density and the prominence of bulges increases with increasing luminosity / mass as shown in figure [ conc_abs_r ] , the trends observed may result from either environmental effects @xcite or the different evolution of galaxies of different mass @xcite . in order to explore further this degeneracy , we compared the actual effect of environment on the morphology of galaxies , selecting three environments ( low density @xmath40 , medium density @xmath41 and high density @xmath42 ) and evaluating the average bulginess of systems with different luminosities in each environment . if the environment plays a crucial role in shaping the properties of galaxies , we would expect a significant difference in the trends across different environments : all galaxies in denser environments must have undergone more interactions than galaxies in low density environments and at all luminosities they should be significantly more evolved towards earlier types . the results are shown in figure [ morpho_lum ] . we find that a similar correlation between luminosity and bulge prominence is in place in high- , medium- , and low- density environments : higher luminosity systems show increased bulginess and the trends are quite smooth . this is in agreement with the claim by @xcite that the color - magnitude relation for galaxies does not depend significantly on the environment and suggests that whatever mechanism transforms galaxies into bulgier systems must be effective in all types of environments from low to high density . from this point of view , the environment seems to play a lesser role in shaping the morphology of galaxies ; the magnitude of its effect can be evaluated by comparing the offsets in the trends of figure [ morpho_lum ] as shown in the lower right corner of figure [ morpho_lum ] . this offset accounts for just a few percent variation : the bulk of the correlation must be produced by a mechanism that works equally well in all environments . we have conducted a similar analysis on star - forming galaxies and agn hosts in order to evaluate whether or not the dependence of activity on density is primarily an effect of the density - morphology correlation . in figure [ sfg_late_agn_early ] we show that the fraction of agn among early - type galaxies in different environments is remarkably flat , suggesting that there is no significant variation in the counts of passive and active ellipticals across different environments . the fraction of star - forming galaxies among late - type galaxies is also constant across densities ranging over three orders of magnitude . the fact that this fraction is @xmath0 1.0 reflects the fact that disk galaxies are mostly ( 96% ) forming stars and are not in a composite ( agn plus star - formation ) or agn stage . ( note that bins where the ratio is @xmath43 are the consequence of the existence of star - forming early - type galaxies : since we are normalizing by the number of late - type galaxies , the ratio can be greater than 1.0 ) . the results do not change significantly if we divide the sample into early and late types by requiring the early - type galaxies to have the likelihood of a de - vaucouleurs profile higher than the likelihood of a disk profile . this suggests that on the scales sampled in this study , the correlation of activity with density is a by - product of the density - morphology correlation ( or vice versa ) . this can be further seen by considering the intensity of the two types of activity . we selected the same type of galaxy in different environments and compared the average intensity of the associated activity . star - formation is typically found in galaxies with lower absolute @xmath25 magnitude than the mean absolute @xmath25 magnitude ; we therefore selected two different volume - limited samples at different absolute r magnitudes to maximize the size of the samples . we choose @xmath44 for star - forming galaxies and @xmath45 for agn , defining these as high - luminosity host samples . we compared these with a low - luminosity sample of host for each class ( @xmath46 for sfg and @xmath47 for agn ) . in figure [ vl_ew_trends ] we show the trend for the median equivalent width of h@xmath8 ( ew(h@xmath8 ) ) with varying density in these different samples spectroscopically classified as agn or sfg . no significant variation of the ew(h@xmath8 ) is found with density . for agn this suggests that the relative intensity of their emission is not determined by the frequency of their interactions with other galaxies , in agreement with the results of @xcite . for sfg the ew(h@xmath8 ) measures the ratio of the current to past star - formation activity . since galaxies in denser environments should have undergone more interactions over time , the invariance of ew(h@xmath8 ) with density suggests that the environment plays a relatively small role in shaping their star - formation history , in agreement with other studies ( carter et al . 2001 ; lewis et al . 2002 ; hogg et al . 2006 ) . obviously one might argue that the transformation in morphology and activity may be triggered on much smaller scales by close interactions . however , to recover the correlation of morphology and activity with the _ large - scale _ environmental density sampled in this study requires that , over time , galaxies in denser environments must have undergone more close interactions than their counterparts in low - density environments or that more galaxies are tidally interacting in denser environments . if galaxies in denser environments have undergone more close interactions than their counterparts in low - density environments , there should be a significant difference in the luminosity - concentration properties of the populations in different large - scale environments , because the large and the small scale would be coupled . we do not find such a difference . to assume that more galaxies in denser environments are in close pair / tidally interacting systems than in low - density environments may explain the increase in the fraction of early- to late - type with increasing large - scale density and the invariance of the luminosity - concentration relation with density , but it seems unlikely to account for the _ smooth _ variation of the bulge - to - disk ratio observed in fig . [ bulgy_density ] . we note , however , that in this study we do not divide the extreme environment of the core of clusters from the bulk of the sample . whether gas - gas interactions and harassment play a major role in shaping the properties of galaxies in such rare environments can not be assessed by this study . several scenarios have been proposed to explain the variety of galaxy properties , either by postulating a primordial difference in early- and late - type galaxies or by assuming an evolution of late - type galaxies into early types via mergers , interaction and/or secular evolution . at the moment , no consensus has been reached on which of these scenarios is preferable . certainly all must play a role in shaping the morphology of galaxies and it can not be excluded , _ a priori _ , that they are equally important in creating the trends observed . each of these mechanisms , however , produces a different re - arrangement of gas and angular momentum in galaxies with its own characteristic time scale and induced burst of star formation and , potentially , agn activity . it is therefore likely that understanding the morphological transformation of galaxies is strongly correlated with understanding a galaxy s star - formation history and the evolution of its agn activity . coherent information about galaxy morphology , star - formation , agn activity , and supermassive black hole assembly would allow us to place important constraints on the different mechanisms at play and , possibly , allow us to identify what produces the salient differences in morphology , agn activity and star - formation history along the hubble sequence . the fact that star - formation activity , agn activity and the growth of bulges and smbhs @xcite proceed from bigger to smaller systems suggests that galaxies do form a continuum , with more massive galaxies formed at earlier times and evolving faster than low - mass ones . in agreement with this idea , the _ smooth _ scaling relations between morphology , star - formation and agn activity found in our data seem to confirm a smooth transition in the properties of galaxies with increasing mass or luminosity . in this study we have investigated the transformation of similar types of galaxies in the look back time of the sdss in order to better understand the mechanisms behind the evolution of galaxies . the evolution over the past gyr both in morphology and spectroscopic and radio activity appears more significant for less luminous systems , in agreement with downsizing in the mass assembly and activity observed at higher redshifts @xcite.the observed evolution is therefore likely the tail of the evolution of galaxies and their nuclei observed at higher redshifts . consistent with this notion , we have found that the smooth transition of spectral properties in agn is accompanied by a different history of star - formation : as the nuclear emission lines become less and less prominent , the last star - formation event dates to earlier times . this strongly supports a co - evolution of star - formation and agn activity . furthermore we have shown that mid- and low - luminosity hosts of agn show evidence for a re - arrangement of the stellar component , with the bulge - to - disk ratio increasing within the look - back time of the sdss , while star - forming systems do not show signatures of significant structural evolution , suggesting that the disk phase ( as the elliptical phase ) is stable , while the bulgy phase is not and leads to a progressive growth of the bulge . if this phase is short compared to the red ( early - type ) and blue ( late type ) phases , as our study indicates , a shortage of systems with intermediate colors is to be expected , as suggested by @xcite . from this point of view , the red sequence / blue sequence dichotomy found in the local population of galaxies @xcite can be interpreted as a simple by - product of the different evolutionary timescales for systems of different mass / luminosity . the data presented in this study show that galaxies of different sizes have evolved differently over the past 2 gyrs . for these changes to be environmentally driven , requires that the effect of the environment should be proportionally weaker on smaller galaxies . this seems unlikely , especially considering the fact that bigger galaxies tend to inhabit denser region of the large scale structure where the effects of environment ( harassment , gas - gas interactions ) should be _ more _ significant . in agreement with this we have shown that similar galaxies in different environment share similar morphology and activity properties , leading to the conclusion that for the bulk of galaxies the transformation of both galaxy activity and morphology is mainly a product of the intrinsic properties of the galaxies themselves , not of the environment , which might nonetheless have an important role in the most extreme environments such as the cores of clusters and tidally interacting pairs . from this point of view the most likely factor influencing the evolution of galaxies is the depth of the potential well : in this case more massive galaxies are expected to be more evolved systems . as shown in figure [ hs_fractions ] , star - forming galaxies peak at absolute magnitudes of @xmath48 - 18 , composite systems at @xmath48 -20.5 , agn at @xmath48 -21.5 and passive systems at @xmath48 -23 in agreement with a scenario in which galaxies of different mass / luminosity are at different stages in their evolution from star - forming to passive systems , passing through an intermediate agn phase . since more massive / luminous systems reside in denser environments , the trends in figure [ fraction_rho_final ] would be a consequence of the different rate of evolution for systems of different mass . the simplest way to interpret our data is therefore to posit that galaxies of different sizes underwent evolution at different rates , with smaller systems still in the earlier stages . we note that the infall of baryonic matter in a potential well scales as @xmath49 . therefore bigger primordial over - densities should have faster infall of matter , while small over - densities should have a much slower infall . this may be the key in understanding galaxy formation and evolution . a fast infall / collapse of material in bigger over - densities would more likely resemble a monolithic collapse . these over - densities would form big bulges and big ellipticals ( and supermassive black holes within ) . a slow accretion phase typical of small over - densities would most likely find a thin disk equilibrium configuration , forming smaller bulge - less spiral galaxies over longer timescales . galaxies with intermediate potential wells would have both . while disks and spheroids are mostly stable configurations , the mix of the two may not be . as the gas continues to fall into the potential well , the disk - to - bulge ratio continues to vary , as we have shown in this study . as gas is transferred from the outer regions into the inner regions of the potential well , agn activity is powered . as the reservoir of cold gas gets depleted , the agn strength decreases as we have shown . strong agn in the local universe would be associated with bulgy systems with plenty of gas that at earlier times where mostly star - forming systems , as observed . furthermore , if the gaseous reservoir of galaxies is progressively transferred into the inner regions , one would expect that star - formation would become more and more centrally concentrated as a galaxy ages . we have explored this by considering the ratio of the radio to optical sizes of our galaxies and have found that this ratio decreases for redder systems , consistent with the notion that star - formation becomes more and more centrally concentrated for more evolved types ( figure [ color_size ] ) , in agreement with the early work of hummel ( 1981 ) . such a redistribution of the gas is in agreement with the findings of @xcite , who showed a decreasing ratio of hi to optical radii along the hubble sequence . recently , evidence of gas infall in the inner regions of late - type galaxies produced by secular evolution has been demonstrated by @xcite . this type of scenario seems to favor a transformation of the bulk of galaxies given by the combined effect of the progressive infalling of matter from the outer regions of galaxies @xcite and its re - arrangement by means of disk instabilities such as bars and density waves @xcite . since the properties of the gas infall mostly depends on the potential well of the galaxy , the properties of galaxies must vary smoothly with increasing luminosity and objects of different luminosity should have different signatures of activity , as we find in the data . while this type of scenario would naturally predict downsizing and account for several observational properties , it stands in contrast to cdm models , since it would require the over - densities associated with massive ellipticals to be formed at an early time , and not hierarchically . a merger - driven transformation of galaxies , on the contrary , seems unlikely . first , the evolution appears to be anti - hierarchical , with smaller systems evolving more at later times . secondly , mergers seem unlikely to produce the smooth transformation of both star - formation _ and _ agn activity in hosts of different masses described in this paper . furthermore , since such a mechanism depends on the density of the environment and the relative velocity of the encounters , it is unlikely that it can work equally well in all environments as required by figure [ morpho_lum ] . also , it is not plausible that mergers would affect galaxies of different luminosities in different ways as we find , transforming more low - luminosity systems in the past gyr than high - luminosity ones . the existence of agn activity in non - interacting systems @xcite raises further questions for this mechanism . from this point of view , the bulk of the density - morphology relation might simply be a size - density relation , with bigger disk galaxies preferentially formed in denser environments at earlier epochs and evolving more at earlier times than small size galaxies . late - type galaxies in denser environments are then expected to have more concentrated gas than similar galaxies in less dense environments . interestingly , @xcite compared late - type galaxies in the virgo cluster with galaxies in the field and found that hi is more concentrated in cluster galaxies ( although they interpreted this as a possible effect of the cluster environment on the virgo galaxies , such as stripping ) . the size - density correlation would lead to the correlation of star - formation and agn activity with density shown in figure [ fraction_rho_final ] : smaller galaxies which inhabit less dense environments have longer transformation times and therefore , at the present epoch , are still forming stars , while bigger galaxies have evolved into spheroids and their gaseous reservoir is only used to feed the smbh at their centers . when the trends are normalized respectively to late- and early - type galaxies , the correlation of activity with environment disappears : there is no substantial difference in the ratio of star - forming disk galaxies or agn early - type galaxies across the range in densities across the large scale structure ( figure [ sfg_late_agn_early ] ) . these results are in agreement with earlier work by @xcite . from this point of view , understanding the density - morphology and the density - activity correlation would mean understanding why bigger galaxy - scale over - densities were preferentially born ( or have aggregated ) within denser large - scale environments . in this study based on the sdss and first surveys , we have investigated the evolution of galaxy properties withing the sdss look - back time ( @xmath0 2 gyr ) for a sample of @xmath0 150000 galaxies in the local universe . we have shown a transformation of the activity and morphology of galaxies over the look - back time of the sdss and have further demonstrated that this transformation is more significant for lower luminosity systems , in agreement with the notion of downsizing in star formation , agn activity and bulge formation over cosmic time . the intensity of the h@xmath8 and radio emission has significantly dropped over the last 1 gyr in emission - line agn , a fact in agreement with the progressive fading of active nuclei from high to low redshift traced by other surveys . the hosts of these nuclei have substantially transformed towards more concentrated stellar distributions typical of more bulgy systems . the fact that systems of different luminosity evolve differently over this look - back time supports the claim that , for the bulk of galaxies , transformation is unlikely the product of environment - driven mechanisms , but rather depends on the mass / size of the host itself . this is further supported by the direct study of the galaxy environment presented in this paper . we have shown that galaxies show increasing bulge - to - disk ratio with increasing density and that this increase is accompanied by a transition in the type of activity from star - formation - dominated to composite to agn - dominated and eventually to passive systems . this transition accompanies the well - know density - morphology relation , into regions well beyond the cluster cores where it was originally discovered ( dressler 1980 ) . we argue that these two trends with densities are unlikely driven by environment - driven mechanisms , but rather depend on the mass / size of the host itself and different evolution timescales for galaxies associated to different potential wells . we note that if over - densities of all sizes were formed at early epochs instead of being hierarchically formed over time , the infall time of gas would be much faster for the largest over - densities and correspondingly slower for the smaller ones . a fast infall / collapse of material in bigger over - densities would more likely resemble a monolithic collapse . these over - densities would form big bulges and big ellipticals ( and supermassive black holes within ) . a slow accretion phase typical of small over - densities would most likely find a thin disk equilibrium configuration , forming smaller bulge - less spiral galaxies over longer timescales . galaxies with intermediate potential wells would have both and be a less stable configuration . in agreement with this we have shown that big bulge - dominated galaxies and star - forming disks are stable configurations , while galaxies with intermediate bulges hosting agn tend to evolve towards bulgier systems . this rapid transformation of systems of intermediate luminosity into earlier type systems with more prominent bulges demonstrated here , suggests that the intermediate phase may be short - lived compared to the star - forming disk phase and the red elliptical phase , thus explaining the dichotomy between red- and blue - sequence galaxies . the progressive infall of gas from the outer regions to the inner regions of a galaxy potential well can be traced using the size of radio emission as an indicator of the star - formation region within galaxies . we have shown that a progressive concentration of star - formation activity towards the inner regions of redder star - forming galaxies is found , in agreement with early work of hummels ( 1981 ) , suggesting a picture in which gas is progressively transported by instabilities towards the inner regions . as the gas is funneled towards the center , star - formation is triggered , followed by nuclear activity . as this cold gas reservoir is exhausted the agn activity is progressively brought to a halt , explaining the anti - correlation that we find between line - strength of active nuclei and galaxy color ( and other indicators of star - formation activity ) , which is an extension of the results of @xcite . from this point of view the density - morphology relation and the density - activity relation discussed in this study is the by - product of different formation and evolution timescales for galaxies associated with different primordial over - densities , and the fact that the biggest galaxy - scale over - densities were formed within the denser large - scale structure over - densities . this work greatly benefited from discussion with jacqueline van gorkom , ed spiegel , david schiminovich , david hogg and margaret geller . funding for sloan digital sky survey project has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss website is http://www.sdss.org . the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions ( the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the korean scientist group , los alamos national laboratory , the max - planck - institute for astronomy , the max - planck - institute for astrophysics , new mexico state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington ) . balogh , m. l. , baldry , i. k. , nichol , r. , miller , c. , bower , r. , & glazebrook , k. 2004 , , 615 , l101 barton , e. j. , geller , m. j. , & kenyon , s. j. 2000 , , 530 , 660 becker , r. h. , white , r. l. , & helfand , d. j. 1995 , , 450 , 559 best , p. n. 2004 , , 351 , 70 best , p. n. , kauffmann , g. , heckman , t. m. , & ivezi , . 2005 , , 362 , 9 brinchmann , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , & brinkmann , j. 2004 , , 351 , 1151 bruzual a. , g. 1983 , , 273 , 105 butcher , h. , & oemler , a. , jr . 1978 , , 226 , 559 carter , b.j . , fabricant , d.g . , geller , m.j . , & kurtz , m.j.,2001,,559,606 cassata , p. , et al . 2007 , , 172 , 270 cayatte , v. , kotanyi , c. , balkowski , c. , & van gorkom , j. h. 1994 , , 107 , 1003 christlein , d. 2000 , , 545 , 145 christlein , d. , & zabludoff , a. i. 2005 , , 621 , 201 cimatti , a. , daddi , e. , & renzini , a. 2006 , , 453 , l29 combes , f. 2000 , dynamics of galaxies : from the early universe to the present , 197 , 15 combes , f. 2006 , astrophysics update 2 , 159 condon , j. j. 1992 , , 30 , 575 condon , j. j. , cotton , w. d. , greisen , e. w. , perley , r. a. , yin , q. f. , & broderick , j. j. 1993 , bulletin of the american astronomical society , 25 , 1389 cooper , m. c. , et al . 2008 , , 383 , 1058 cowie , l. l. , songaila , a. , hu , e. m. , & cohen , j. g. 1996 , , 112 , 839 cowie , l. l. , barger , a. j. , bautz , m. w. , brandt , w. n. , & garmire , g. p. 2003 , , 584 , l57 dahari , o. 1984 , , 89 , 966 dressler , a. 1980 , , 236 , 351 dunlop , j. s. , mclure , r. j. , kukula , m. j. , baum , s. a. , odea , c. p. , & hughes , d. h. 2003 , , 340 , 1095 ferrarese , l. , & merritt , d. 2000 , , 539 , l9 fraternali , f. , oosterloo , t. , sancisi , r. , & van moorsel , g. 2001 , , 562 , l47 geller , m. j. , kenyon , s. j. , barton , e. j. , jarrett , t. h. , & kewley , l. j. 2006 , , 132 , 2243 glikman , e. , helfand , d. j. , becker , r. h. , & white , r. l. 2004 , asp conference series 311 : agn physics with the sloan digital sky survey 311 , 351 g ' omez , p. l. , et al . 2003 , , 584 , 210 grimm , h .- j . , gilfanov , m. , & sunyaev , r. 2003 , , 339 , 793 grogin , n. a. , et al . 2005 , , 627 , l97 helou , g. , soifer , b. t. , & rowan - robinson , m. 1985 , , 298 , l7 hogg , d. w. , et al . 2004 , , 601 , l29 hogg , d. w. , masjedi , m. , berlind , a. a. , blanton , m. r. , quintero , a. d. , & brinkmann , j. 2006 , , 650 , 763 hopkins , p. f. , hernquist , l. , cox , t. j. , di matteo , t. , robertson , b. , & springel , v. 2006 , , 163 , 1 hubble , e. p. 1936 , yale university press , kauffmann , g. , et al . 2003 , , 346 , 1055 kauffmann , g. , et al . 2003 , , 341 , 54 kauffmann , g. , white , s. d. m. , heckman , t. m. , m ' enard , b. , brinchmann , j. , charlot , s. , tremonti , c. , & brinkmann , j. 2004 , , 353 , 713 kriek , m. , et al . 2007 , , 669 , 776 559 kormendy , j. , & kennicutt , r. c. , jr . 2004 , , 42 , 603 lewis , i. , et al . 2002 , , 334 , 673 longair , m. s. 1966 , , 133 , 421 lynden - bell , d. , & kalnajs , a. j. 1972 , , 157 , 1 madau , p. , pozzetti , l. , & dickinson , m. 1998 , , 498 , 106 martini , p. , mulchaey , j. s. , & kelson , d. d. 2007 , , 664 , 761 mateus , a. & sodr ' e , l. 2004 , , 349 , 1251 meier , d. l. 2001 , 20th texas symposium on relativistic astrophysics , 586 , 420 merloni , a. , rudnick , g. , & di matteo , t. 2004 , , 354 , l37 miller , c. j. , nichol , r. c. , g ' omez , p. l. , hopkins , a. m. , & bernardi , m. 2003 , , 597 , 142 nandra , k. , et al . 2007 , , 660 , l11 park , c. , choi , y .- y . , vogeley , m. s. , gott , j. r. i. , & blanton , m. r. 2007 , , 658 , 898 pierce , c. m. , et al . 2007 , , 660 , l19 poggianti , b. m. , et al . 2001 , , 563 , 118 postman , m. , & geller , m. j. 1984 , , 281 , 95 ranalli , p. , comastri , a. , & setti , g. 2003 , , 399 , 39 regan , m. w. , et al . 2006 , , 652 , 1112 reviglio p. , `` multiwavelength analysis of star - forming galaxies and active galactic nuclei in a complete redshift survey '' , tesi di dottorato di ricerca , universit di torino , xvi ciclo , 2001 - 2003 reviglio , p. , & helfand , d. j. 2006 , , 650 , 717 reviglio , p. , & helfand , d. j. 2007 , in publication , paper i reviglio , p. , `` active galaxies and their evolution as observed in the first and sloan digital sky surveys '' , ph.d . thesis , columbia university , new york , u.s.a . , 2008 rieke , g. h. , & lebofsky , m. j. 1985 , , 288 , 618 schmitt , h. r. 2001 , , 122 , 2243 sellwood , j. a. , & binney , j. j. 2002 , , 336 , 785 strateva , i. , et al . 2001 , , 122 , 1861 ulvestad , j. s. , & ho , l. c. 2001 , , 558 , 561 veilleux , s. & osterbrock , d. e. 1987 , , 63 , 295 voges , w. , et al . 1999 , , 349 , 389 voges , w. , et al 2000 , vizier online data catalog , 9029 , 0 white , r. l. , helfand , d. j. , becker , r. h. , glikman , e. , & devries , w. 2006 , arxiv astrophysics e - prints , arxiv : astro - ph/0607335 white , n. e. , giommi , p. , & angelini , l. 2000 , vizier online data catalog , 9031 , 0 white , r. l. , helfand , d. j. , becker , r. h. , glikman , e. , & de vries , w. 2007 , , 654 , 99 worthey , g. , & ottaviani , d. l. 1997 , , 111 , 377 york , d. g. , et al . 2000 , , 120 , 1579 or h@xmath22 , misclassified les systems ( mis ) and truly passive galaxies with the x - ray or radio signature of an agn ( abg ) . the bars represent the median absolute deviation in color for each class . _ ] , but for volume - limited sub - samples with narrow host luminosity ranges in r - band ( and therefore small mass ranges in the supermassive black hole powering the agn ) . at all luminosities , galaxies show less remarkable lines for redder colors , suggesting that the trend is not produced by the difference in the supermassive black hole mass , but more likely by the accretion properties of the central engine . when split into these several sub - samples , the abg class with agn confirmed by radio or x - ray emission in each sample becomes quite small ; if fewer than 3 galaxies were present we did not considered the value statistically reliable and did not plot it . _ ] break strength for different classes of agn with increasingly unremarkable lines : seyfert ( sey ) , liners ( lin ) , low - excitation systems ( les ) with marginally detected or undetected [ oiii ] or h@xmath22 , misclassified les systems ( mis ) and spectroscopically passive galaxies with the x - ray or radio signature of an agn ( abg ) . the bars represent the median absolute deviation in color for each class . _ ] for different classes of agn with increasingly unremarkable lines : seyfert ( sey ) , liners ( lin ) , low - excitation systems ( les ) with marginally detected or undetected [ oiii ] or h@xmath22 , misclassified les systems ( mis ) and spectroscopically passive galaxies with the x - ray or radio signature of an agn ( abg ) . the bars represent the median absolute deviation in color for each class . _ ] , but for volume - limited sub - samples with a narrow range of host luminosity in r - band ( and therefore in the mass of the supermassive black hole powering the agn ) . at all luminosities , galaxies which show less remarkable lines also show stronger d4000 breaks , showing that systems with less remarkable lines are hosted by systems with older stellar populations , and therefore are more gas depleted . _ ] , but for volume - limited sub - samples with a narrow range of host luminosity in r - band ( and therefore mass of the supermassive black hole powering the agn ) . at all luminosities , galaxies which show less remarkable lines also show lower @xmath50 , showing that systems with less remarkable lines are hosted by systems with older stellar populations , and therefore are more gas depleted . _ ] 2.22 for early - type ) . in the lower plot , galaxies are classified as early - type if the likelihood of a de - vaucouleur profile for their surface brightness is higher than the likelihood of an exponential fir and late - type otherwise . _ ] ) and a low - luminosity one ( @xmath46 ) . similarly the agn subsamples have been selected from a high - luminosity volume limited sample(@xmath45 ) and a low - luminosity volume - limited sample ( @xmath47 ) . error bars are the absolute median deviation in each bin . _ ] galaxy / mpc@xmath53 ) and low ( @xmath54 galaxy / mpc@xmath53 ) density environments . the emission becomes more centrally concentrated in redder galaxies , and the concentration does not depend on the environment . radio sizes have been determined as the hwhm of a two - dimensional gaussian fit to the first radio maps . _ ]
we present ultraviolet ( uv ) integrated colors of 44 galactic globular clusters ( ggcs ) observed with the galaxy evolution explorer ( galex ) in both and bands . this data - base is the largest homogeneous catalog of uv colors ever published for stellar systems in our galaxy . the proximity of ggcs makes it possible to resolve many individual stars even with the somewhat low spatial resolution of galex . this allows us to determine how the integrated uv colors are driven by hot stellar populations , primarily horizontal branch stars and their progeny . the uv colors are found to be correlated with various parameters commonly used to define the horizontal branch morphology . we also investigate how the uv colors vary with parameters like metallicity , age , helium abundance and concentration . we find for the first time that gcs associated with the sagittarius dwarf galaxy have colors systematically redder than ggcs with the same metallicity . finally , we speculate about the presence of an interesting trend , suggesting that the uv color of gcs may be correlated with the mass of the host galaxy , in the sense that more massive galaxies possess bluer clusters .
1208.5698
the main contributors to the uv emission from any stellar system are the hottest stars . indeed blue horizontal branch ( hb ) stars are well known to be among the hottest stellar populations in globular clusters ( gcs ) and contribute substantially to the uv radiation observed from old stellar systems ( welch & code , 1972 ) . later on , two other sub - classes of post - hb stars were found to be important contributors to far - uv ( @xmath0 , @xmath31600 ) radiation ( e.g. , greggio and renzini 1990 ; dorman et al . 1995 , hereafter dor95 ; han et al . the hottest hb stars ( extreme hb , ehb ) have such a small envelope mass that most of their post - he - core burning phase takes place at high effective temperature ( ) , during the so called `` agb - manqu phase '' , and these stars never return to the asymptotic giant branch ( agb ) . another group of uv - bright stars is that of post - early agb stars , which after a brief return to the agb , spend the bulk of their helium shell burning phase at high . in systems with only red hb a small floor level of @xmath0 is provided by post - agb stars , which evolve to the agb phase with an higher envelope mass where they undergo thermal pulses and eventually lose their envelopes moving at higher temperatures at constant luminosity . the relative contributions of the various types of stars and the factors that might lead to larger or smaller populations of uv - bright stars have remained an open question ( greggio and renzini 1990 ; dor95 ; lee et al . 2002 ; rich et al . 2005 ; sohn et al . 2006 ) . in distant extragalactic systems one can ordinarily observe only the integrated light of unresolved stellar populations , from which the hope is to gain knowledge about the underlying stellar population . galactic globular clusters ( ggcs ) play an important role in understanding the integrated uv colors of extragalactic systems , especially the so called `` uv - upturn '' observed in the spectral energy distributions of elliptical galaxies ( code & welch 1979 ; de boer 1982 ; bertola et al . 1982 ; greggio & renzini 1990 ; oconnell 1999 ) . first of all , gcs are the closest example in nature to a single stellar population ( ssp ) : a system of coeval stars with similar chemical composition dex ) spread in the iron abundance and ( possibly ) age : @xmath4 centauri ( norris et al . 1996 , lee et al . 1999 , ferraro et al . 2004 ) and terzan 5 ( ferraro et al . 2009 ) . ] . moreover ggcs span a large range of metallicities , a small range of ages , and perhaps some range of helium abundance . hence they can be used to test the stellar evolution theory , which in turn is one of the basic ingredients of the models used to interpret the integrated light of distant galaxies . ggcs are relatively nearby objects ( more than @xmath5 are located at distances @xmath6kpc ) , so their populations can be easily resolved . with typically more than 100,000 stars , even relatively short - lived evolutionary stages are sampled . we can directly observe the properties of individual stars and measure the population ratios for objects in different evolutionary stages . in particular we can study the impact of hot and bright populations ( as the agb - manqu stars ) on the integrated uv light of ggcs and then use them as crucial local templates for comparison with integrated properties of distant extragalactic systems . in fact comparing features in the color - magnitude diagrams ( cmds ) of well known and resolved ggcs with integrated quantities can lend important _ model independent _ insights into the nature of extragalactic systems . integrated uv photometry of ggcs has previously been obtained by the _ orbiting astronomical observatory _ ( oao 2 ; welch & code 1980 ) , _ astronomical netherlands satellite _ ( ans ; van albada , de boer & dickens 1981 ) , _ ultraviolet imaging telescope _ ( uit ; hill et al . 1992 ; landsman et al . 1992 ; parise et al . 1994 ; whitney et al . 1994 ) , and _ international ultraviolet explorer _ ( iue ; castellani & cassatella 1987 ) . using a large , but heterogeneous , collection of data obtained with some of these telescopes along with population synthesis models , dor95 showed how the uv colors varied with parameters like metallicity and how they compare with elliptical galaxies . they showed how the uv colors of ggcs could plausibly be produced by hot hb stars and their progeny . at the time of dor95 , uv photometry of individual stars in ggcs was available for only a few clusters . that situation has changed dramatically . our group alone has already published hst uv photometry for a dozen of ggcs ( see ferraro et al . 1997 , ferraro et al . 1998 , ferraro et al . 1999 , ferraro et al . 2001 , ferraro et al . 2003 , lanzoni et al . 2007 , dalessandro et al . 2008 , rood et al . 2008 ) and obtained data for an additional 32 clusters in hst cycle 16s ( go11975 , pi : ferraro ; see contreras et al . 2012 , sanna et al . 2012 and references therein ) . more recently , we have secured observations of 44 ggcs during three observing cycles with the _ galaxy evolution explorer _ ( galex ) . this is the largest homogeneous sample ever collected for ggcs in uv so far . in schiavon et al . ( 2012 ; hereafter paper i ) we presented photometry and cmds for these clusters . here we present integrated uv magnitudes and colors ( 2 ) for each cluster , and we describe ( 3 ) how they are affected by hb class , metallicity , age , and possibly structural parameters ( mass , density , central relaxation time , etc . ) . in 4 we compare our data with observations of gcs in m31 and m87 . images for 38 ggcs were obtained as part of gi1 and gi4 galex programs ( p.i . r. p. schiavon ) and supplemented with somewhat shallower exposures for 8 ggcs obtained as part of program gi3 ( p.i . , s. sohn ) . two clusters ( ngc 6229 and ngc 6864 ) are in common between the two programs . a number of clusters that would have been very interesting targets ( including m13 , m80 , @xmath4centauri , ngc 6388 , and ngc 6441 ) could not be observed because of restrictions on the uv background brightness , due to detector - safety considerations , which prevented us from targeting at low galactic latitudes , or near positions of very bright uv sources . with only one exception ( ngc 6273 ) , images were obtained in both the @xmath0 ( @xmath7 ) and near - uv ( @xmath1 , @xmath8 ) bands . thanks to the wide field of view of galex ( @xmath9 ) , it has been possible to sample the full radial extent of most of the clusters . a few gcs , like ngc 104 ( 47tuc ) and ngc 5272 ( m3 ) , have tidal radii larger than the galex field of view ( mclauglhin & van der marel 2005 , hereafter mvm05 ) , but these missing data do not significantly affect our analysis . each galex image has been pre - processed by the standard pipeline described in detail by morrissey et al . ( 2005 ) . the raw images ( or `` count images '' ) have been converted to flux - calibrated intensity images by applying relative response and flat - field corrections and by scaling the flux by the effective area and exposure time . the integrated magnitudes have been obtained using two different approaches : _ ( i ) _ by fitting the surface brightness profiles ( sbps ) , and _ ( ii ) _ from direct aperture photometry ( ap ) measurement on the images . in both approaches , the integrated photometry will include the effects of objects below the detection threshold for individual stars . comparing the results obtained with these two different approaches we can correct for systematic effects or biases . in particular , ap can be affected by small number statistics when a few very bright stars dominate the flux , especially at @xmath0 wavelengths ( as for example , the case of the bright star in the very central regions of 47 tuc reported by oconnell et al . these very bright objects may be cluster members ( post - hb or post - agb star ) or just foreground stars . on the other hand , while colors obtained by the sbp method are less affected by star count fluctuations , they may suffer from uncertainties arising from the quality of the fit or the determination of the center of the cluster . for these reasons we have calculated integrated magnitudes with both the techniques , and we have compared the results in detail . in both cases , it was first necessary to determine the cluster centers . since previous determinations ( e.g. , harris 1996 , 2010 revision hereafter h10 ; noyola & gebhardt 2006 ; goldsbury et al . 2010 ) either were obtained with quite heterogeneous methods , or only partially overlap our sample , we preferred to perform new estimates . the plate scale of galex ( @xmath10 ) and the `` noise '' due to the relatively small number of stars detected at uv band introduce considerable uncertainty in the measurement of the barycenter of resolved stars ( e.g. , lanzoni et al . we therefore determined the center of the light distribution on @xmath1 images by following the same approach described in bellazzini ( 2007 ) . for any given gc we picked an aperture radius that , based on the size of cluster central region and its density , was large enough to include a good fraction of the brightest sources . we then determined the light density ( i.e. the fraction of observed flux per unit area ) in that aperture and iteratively varied the centering position until maximum light density was found . we do not list the centers of light , since they have been obtained in instrumental coordinates and therefore they could be used only when the images they refer to are available . moreover the poor spatial resolution of the galex detectors would have prevented to obtain a sufficient level of accuracy to make these values useful to the community for other applications . sbps were obtained by using concentric annuli centered on the cluster centers determined as described above . the number and the size of the annuli were chosen for each cluster so as they sampled almost the same light fraction . the flux calculation was performed by using standard iraf tasks ( _ phot _ , in the daophot package , stetson 1987 ) . the flux calculated in each annulus was normalized to the sampled area obtained adopting as pixel size @xmath11 , as reported by morrissey et al . ( 2005 ) for intensity map images . the instrumental surface brightness profiles were then transformed to the abmag galex magnitude system by applying the zero - points reported in the galex on - line user s manual or in section 4.2 of morrissey et al . ( 2005 ) . as an example , the observed sbp of ngc 6341 is shown in figure 1 for both the @xmath0 and @xmath1 channels . in this figure , solid squares are the annular surface brightness measurements at each cluster radius , whereas open circles represent the same values after performing sky subtraction as described below . using sbps to obtain integrated colors requires the subtraction of the background contribution , which is the combination of the sky background emission ( due to unresolved objects ) and the resolved field ( back and foreground ) sources . we determined the background as the average value in the annuli at large radii where the observed sbps show a `` plateau '' ( for @xmath12 in ngc 6341 , see figure [ figprof1 ] ) . the average has been obtained applying a sigma - clipping rejection to take into account possible surface brightness fluctuations , since at low surface brightness levels even a few background stars may cause strong deviations from the general behavior , even in extended areas . we then subtracted the background light density from the observed surface brightnesses in each radial bin thus deriving `` decontaminated '' sbps . as expected , background subtraction does not significantly affect the central regions , but it substantially changes the shape of the profile in the external parts . to further check the robustness of our method , we compared our background subtracted sbps with those computed with standard iraf packages ( _ phot _ and _ fitsky _ ) . the open triangles in the upper panel of figure [ figprof1 ] are obtained with background subtraction and disabling the sigma - clipping routines . they nicely overlap with the open circles thus demonstrating that our approach is essentially equivalent to the automatic procedure . still we preferred to independently fit the observed `` plateau '' in order to have a more direct control of the background level which can be strongly variable case by case . the errors for each bin are defined as the standard deviations of the sampled fluxes . we assumed that counts from both the @xmath1 and @xmath0 detectors have a poisson distribution . to determine the value of the central surface brightness , we have fitted the @xmath0 and @xmath1 sbps with single - mass king models over the entire cluster extension . for each system and each band we adopted the core radius and concentration ( together with their uncertainties ) quoted by mvm05 , and we performed a mono - parametric fit , varying only the central surface brightness : the value yielding the minimum @xmath13 value was adopted as best - fit solution . magnitudes were then calculated by integrating the best - fit king model . the extinction coefficients used to correct the @xmath0 and @xmath1 fluxes are taken from cardelli et al . ( 1989 ) and the @xmath14 values were adopted from h10 . in general the observed sbps are very nicely reproduced by king models with the adopted structural parameters ( see the case of ngc 6341 in figure [ figprof1 ] ) . exceptions to this general trend are the post core collapse clusters in our sample ( ngc 6284 , ngc 6342 , ngc 6397 and ngc 7099 ) . for these systems , which are not studied by mvm05 , it is not possible to fit in a satisfactory fashion the overall shape of the sbps with structural parameters typically adopted for post - core collapse clusters ( @xmath15 and @xmath16 ; see h10 for example ) . in these cases integrated magnitudes have been obtained with aperture photometry only ( see section 2.2 ) . in addition we were not able to obtain a reliable measure in one or both filters for palomar 11 and palomar 12 , because of the low signal to noise ratio of the images . we tested also the impact of using wilson models ( wilson 1975 ) with structural parameters from mvm05 . we find that these models may give up to @xmath17 mag of difference for integrated magnitudes , but they have a null effect on colors . the errors in the integrated magnitudes were obtained by propagating the uncertainties affecting the structural parameters ( as reported by mvm05 ) and the central surface brightness determination . another important source of uncertainty is that affecting the determination of the cluster centers . this can cause variations in central surface brightness estimates that are independent of the other parameters involved in the sbp fit . to quantify the possible errors due to mis - positioning of the cluster centers , we let the centers iteratively vary by up to 5 from our determinations . we found maximum variations of @xmath18 in the @xmath19 color , mainly due to the errors in @xmath0 magnitudes . the colors derived are listed in table 1 . integrated magnitudes in the @xmath20 band were also obtained from integration of king models , adopting the central surface brightnesses reported by h10 . direct measurements of the integrated magnitude were obtained by performing ap using the _ phot _ task under the iraf package . for each cluster we used a single fixed aperture with radius equal to the half - light radius ( @xmath21 ) quoted by mvm05 ( or taken from h10 for the 5 clusters mentioned above ) . by definition , the total magnitude was then computed by adding -0.75 to the value measured within @xmath21 . the adopted center , background flux , and @xmath14 values were the same as above . again the integrated instrumental magnitudes have been converted to the abmag galex magnitude system and corrected for extinction . the photometric errors are defined as the standard deviations of the fluxes . for both approaches , we ignored other sources of errors like those that might affect reddening or distance values . a @xmath22 error in the adopted @xmath14 values may lead to @xmath23 in both filters which in turn gives @xmath24 or @xmath25-v)\sim0.15 $ ] for the most reddened clusters in our sample , such as ngc 6342 or palomar 11 . some of the clusters in our sample may be affected also by differential reddening ( see the case of ngc 6342 ; alonso - garcia et al . the impact of extinction variations was checked by means of synthetic experiments . we simulated a cluster of radius @xmath26 populated by 10,000 stars , all having the same observed magnitude in the three filters ( @xmath0 , @xmath1 and @xmath20 ) ; we also assumed a mean @xmath27 . we then considered 10 circular areas of radius @xmath28 located at randomly extracted positions within the cluster , and we increased by 0.2 the color excess of all the stars falling within these areas ; analogously , we considered 10 additional similar areas and we decreased by 0.2 the color excess of their stars : we therefore simulated a cluster with mean @xmath27 , containing 10 bubbles with @xmath29 and 10 regions with @xmath30 ( this corresponds to a differential reddening of amplitude @xmath31 , as observed in ngc 6342 by alonso - garcia et al . we performed 1000 random extractions of the positions of these 20 areas , and , for each of them , we computed the integrated magnitudes in every band and compared them to the input values . at the end of the experiments we found that the net color variation due to differential reddening is negligible . the largest effects ( @xmath32 mag ) is found for the @xmath2 and @xmath33 colors . a comparison between the integrated magnitudes obtained with the two methods is shown in figure [ figcomp ] . the top panel shows that for most of the clusters there is a very good agreement between the @xmath0 integrated magnitudes obtained from sbp king fitting and from ap . most of the scatter ( @xmath34 ) is due to uncertainties in the adopted half - light radii . three clusters ( marked with black crosses ) are strongly deviant in the @xmath0 panel ( at the top ) . they are 47 tuc , ngc 1851 and ngc 6864 for which @xmath35 and @xmath36 respectively . visual examination of the images reveals that the large @xmath37 is due to the presence of very bright stars at @xmath38 . caution must be paid in these cases , since these objects may be cluster members as for the case of 47 tuc ( dixon et al . 1995 ) or non - members as for ngc 1851 ( wallerstein et al . 2003 ) . in the middle panel of figure [ figcomp ] , the @xmath1 integrated magnitudes are compared . the average difference is essentially zero ( with @xmath39 ) and in this case the outliers are only ngc 1851 and ngc 6864 ( @xmath40 ) . the bottom panel shows the differences in the @xmath41 colors computed with the two approaches as a function of @xmath42 . good agreement is found also in this case ( @xmath43 ) . as discussed above , magnitudes based on sbp fitting are robust against stochastic effects due to the presence of a few very bright stars . this is especially important in the @xmath0 , where uv - bright objects such as post - agb and agb - manqu stars are known to yield an important contribution to integrated light ( e.g. , paper i ) thus significantly affecting the integrated colors of not resolved stellar populations ( see figure 2 ) . however in the present work we focus mainly on clusters global properties which are more likely traced by the bulk of hot stars . therefore , in the following analysis we adopt sbp colors , with the exception of the six clusters that could not be properly fitted by king models . a number of hb morphology classifications have been proposed over the decades . we analyze here the behavior of uv colors as a function of the most important hb parameters . perhaps the most commonly used is @xmath44 introduced by lee , demarque , and zinn ( 1994 ) , where @xmath20 is the number of variables , and @xmath45 and @xmath26 are the numbers of hb stars blue - ward and red - ward of the instability strip . in the leftmost panels of figure [ multi ] the uv colors are shown as a function of @xmath46 . the large number of clusters with @xmath47 simply reflects our target selection as most clusters with red hbs are located at low galactic latitude . the wide range of colors at any given value of @xmath46 arises because there is a large variety of hb morphologies even among the subset of clusters with predominantly blue hbs . the @xmath46 parameter is insensitive to the details of the color distribution of stars bluer than the rr lyrae . the same applies to other hb morphology parameters defined on the basis of optical cmds . some clusters with bimodal hbs , like ngc 2808 ( plotted as a gray triangle ) and ngc 1851 ( plotted as a gray square ) , have an @xmath46 value that ranks them among clusters with reddish hbs , yet they are `` hot '' in the uv . we have also compared our uv colors with the hb parameter @xmath48 defined by buonanno et al . ( 1993 ; 1997 ) , where b2 is the number of stars bluer than @xmath49 ( see also catelan 2009 ) . data have been taken from buonanno et al . ( 1997 ) and preston et al . this parameter is expected to correlate more clearly with uv colors , since it is able to remove the degeneracy which characterizes hbr for clusters with extended blue hbs . in fact we find that both in @xmath41 and @xmath50 there is a clear trend for clusters with @xmath51 , in agreement with findings by catelan 2009 and theoretical expectations ( landsman et al . . clusters with @xmath52 are those in the `` bimodal hb '' zone ( catelan 2009 ) . in this region in fact , we find ngc 1851 and ngc 2808 , as well as other clusters with a clear bimodal hb like ngc 6864 ( m75 ) and ngc 7006 , or with a more populous red hb and a sparsely populated blue hb as ngc 1261 and ngc 362 . in the latter case the extension to the blue of the hb may be actually due to contamination by background stars belonging to the small magellanic clouds . a spearman correlation rank test gives probabilities larger than @xmath53 for correlations with both @xmath41 and @xmath54 . in the @xmath55 the correlation is less clear , in particular if bimodal clusters are not considered . in this case the spearman test gives a probability p@xmath56 . the @xmath48 parameter is more efficient than @xmath46 in characterizing extended blue hbs . thus , in general , it would be preferable . however we stress that @xmath48 may suffer of the same limitations as @xmath46 since it is defined and measured in optical cmds where star counts along the extreme hb blue tails could be significantly incomplete . we therefore encourage the definition of similar quantities based on a proper combination of uv and optical bands . @xcite defined several parameters describing hb blue tails ( bts ) using photometry in the optical bands . in principle , their parameter l@xmath57 describing the length of bts should correlate well with the uv light output of gcs . however , as shown in the third column of panels in figure [ multi ] , there is no obvious correlation . the data used by @xcite were not uniform and in a few cases they were based on cmds dating back to the 1960 s . in many cases the cmds were not deep enough to show bts that we now know to exist . this has also been noticed by @xcite . @xcite included bts in their analysis of hb morphology . they used a homogeneous deep hst survey made in the @xmath58 and @xmath59 ( roughly @xmath45 and @xmath20 ) filters . they measured the length of bts in terms of the maximum effective temperature reached by the hb : the values of were obtained by comparing in the cmds the observed bts with theoretical hb models . however , the optical plane is not ideal for determination of the temperatures of the hottest stars . for this reason , this parameter should be considered as a lower limit of the real hb extension , especially for the most extended bts where the hb may be truncated in optical cmds . indeed , incompleteness may strongly affect clusters with a population of very hot hb stars like the blue hook stars . these objects have been suggested to be stars which experienced the helium - flash at high effective temperatures ( see for example moehler et al . 2004 ; busso et al . 2007 ; rood et al . 2008 ; cassisi et al . 2009 ; brown et al . 2010 ; dalessandro et al . 2011 ; brown et al.2012 ) or stars with an extremely large helium mass fraction ( @xmath60 ; dantona et al . however even in those cases in which incompleteness does not represent a limit , optical photometry still remains a poor measure of for extremely hot stars . in fact , as shown for example in dalessandro et al . ( 2011 ) in the case of ngc 2808 , the value of @xmath61 by @xcite is underestimated by @xmath62k . however even with this limitation , is a useful parameter describing the uv bright population of gcs ( see , e.g. , * ? ? ? the fourth column of panels in figure [ multi ] show the uv colors as a function of . there is an obvious correlation in the case of colors involving @xmath0 magnitudes , not so much in the case of @xmath55 . this is expected , since nuv magnitudes are less sensitive than fuv to the of the hottest stars . a spearman test gives probabilities larger than @xmath63 for correlations with both @xmath41 and @xmath54 , and @xmath64 for the correlation with @xmath55 . all the clusters with @xmath65 ( roughly 1/3 of the sample ) have approximately the same @xmath41 color . there is another group of clusters with @xmath66 with a wide range of @xmath41 . therefore , although the correlation between this parameter and integrated color is statistically robust for colors involving @xmath0 magnitudes , the detailed dependence of integrated color on hb morphology as defined by this parameters seems not to be monotonic . @xcite introduced the parameter to describe hb morphology . it is defined as the difference in the median colors of the hb and the red giant branch ( rgb ) at the level of the hb , and it is derived from a homogeneous , deep hst acs survey in f606w and f814w ( roughly @xmath67 ) . however , since bts are almost vertical in the @xmath68 cmds , we might expect that has problems similar to @xmath46 in characterizing the gc uv light . the rightmost column in figure [ multi ] shows the uv colors as a function of . the correlations are good , in a statistical sense : probabilities larger than @xmath63 are found for @xmath41 and @xmath54 colors , while @xmath69 probability is obtained for a correlation with @xmath55 . however the bulk of the sample is bunched up to the lower right corner of the plots . this is because , unsurprisingly , a parameter based on red optical colors distinguishes clusters with relatively red hbs from their blue counterparts , but provides very poor discrimination between clusters with blue and extremely blue hb morphologies . in order to investigate any possible link between the uv colors and chemical compositions of the milky way gcs , we adopted the values quoted by carretta et al . ( 2009b ) . for terzan 8 , which is not in their sample , we used the equation listed by carretta et al . ( 2009b ) to convert values from zinn & west ( 1984 ) to their metallicity scale . the uv colors derived from sbp fitting are plotted as a function of metallicity in figure [ met ] . first focusing on the top panel of figure [ met ] , one can see that , for the gcs included in this sample , @xmath55 decreases by about 2 magnitudes as decreases from @xmath70 -0.7 to @xmath70 1.5 . for smaller values of , @xmath55 is roughly constant , or perhaps increases slightly as decreases . this clear , although non - monotonic , trend of @xmath55 with metallicity is confirmed by a spearman correlation rank test , according to which the probability of a correlation is 99.99% ( @xmath71 ) . even after removing the most metal - rich clusters ( @xmath72 ) the probability of correlation remains significant at more than @xmath73 . by removing from the sample the most metal - poor clusters ( @xmath74 ) the probability for a correlation actually increases . the more restrictive non - parametric kendall rank correlation test also shows that there is a strong and positive correlation between @xmath55 and with a significance of @xmath75 . the integrated @xmath1 radiation from ggcs is dominated by blue hb stars but it has some contribution also from turnoff and blue stragglers stars ( see for example figure 1 in ferraro et al . the overall trend of @xmath55 as a function of metallicity seen on the top panel of figure [ met ] is therefore in line with expectations from standard stellar evolution theory , as hb and turnoff stars in more metal - rich clusters are expected to be cooler and redder . this expectation was also confirmed by observations collected with ans , oao , and uit , as presented by dor95 . metallicity , however , is only one of the parameters determining the color distribution of hb stars . at least one _ second parameter _ is known to affect the color of the hb in gcs , and it can be recognized in the large spread in color for @xmath76 [ fe / h ] @xmath77 , where clusters with similar can have @xmath55 differing by as much as @xmath70 1.5 mag . the nature of this second parameter has been the subject of debate for several decades now , with candidates such as he abundance , age , mass loss , binarity , rotation , among others , being suggested in the past to explain the effect ( for a review , see catelan 2009 ) . recently dotter et al . ( 2010 ) and gratton et al . ( 2010 ) have particularly emphasized the role of age and they also stressed on the necessity of even a third parameter ( the central luminosity density for the first and the he abundance for the second ) . interpretation of the dependence of colors involving @xmath0 magnitudes as a function of requires a little more care . at first glance , the middle and bottom panels of figure [ met ] show a less clear correlation between color and . the bottom panel in particular looks like a scatter plot . it is true , though , that , @xmath54 varies by almost 5 magnitudes . indeed , a spearman test gives a probability @xmath78 ( corresponding to @xmath79 ) that @xmath54 is correlated with metallicity . the kendall test gives correlation probability of @xmath80 . the probability drops to @xmath81 if the most metal - rich clusters are excluded from the analysis . at face value , therefore , @xmath54 correlates with in a similar way as @xmath55 , although in a less strong or noisier fashion . in summary , then , one finds that the behavior of ggcs in the @xmath54 and @xmath55 planes is essentially the same . on both planes , three sub - families of clusters can be recognized : 1 ) ggcs with @xmath82 , which are predominantly red ; 2 ) ggcs with @xmath83 @xmath84 , the `` second parameter region '' ( see also fusi pecci et al . 1993 ) , where ggcs have a wide range of colors , about @xmath85 mag in @xmath55 and @xmath86 mag in @xmath54 ; and 3 ) ggcs with @xmath87 , which are all blue . it is worth noticing that , intermediate - metallicity ( [ fe / h]@xmath88 ) clusters are the bluest in the three colors combinations . this was also highlighted by dor95 . the extension of their hbs ( see paper i ) is compatible with their integrated colors . on average , the metal - poor ( @xmath89 ) gcs have redder hbs than the intermediate ones . this appears to contradict expectations based on the notion that metallicity is the _ first parameter_. some authors interpreted this discrepancy invoking age differences among clusters . however dotter ( 2008 ) was able to account for this behavior without invoking age differences , by using synthetic hb models and simple assumptions about the mass - loss / metallicity relation . careful inspection of the @xmath41 or @xmath54 _ vs _ plots in figure [ met ] clearly reveals that the color spread at @xmath74 is due to a subset of clusters ( plotted as asterisks ) , which are systematically redder by @xmath70 1.5 and 1.0 mag in @xmath41 and @xmath54 , respectively , than the other gcs in the same metallicity regime . this is also confirmed if colors obtained with ap are used instead of those derived from sbp fitting . likewise , adopting the zinn & west ( 1984 ) metallicity scale or metallicity values by carretta & gratton ( 1997 ) does not affect the general behavior , even though minor differences in a cluster - to - cluster comparison can obviously come out . interestingly these clusters ( ngc 4590 , ngc 5053 , ngc 5466 , arp 2 and terzan 8) are potentially connected with the sagittarius dwarf galaxy stream ( dinescu et al . 1999 , palma et al . 2002 , bellazzini et al . 2003 , law & majewski 2010 ) , and thus may have an extra - galactic origin . we stress that all these clusters have been suggested to be connected to the sagittarius stream by at least two different authors . the other candidate sagittarius gc is the relatively metal - rich ( [ fe / h]=@xmath90 ) palomar 12 , for which we were not able to get @xmath0 magnitude ( see section 3.1 ) . among the clusters considered here , the classification of ngc 4590 is the most uncertain . according to its metallicity , hb morphology and proper motions ( smith et al . 1998 ; dinescu et al . 1999 ; palma et al . 2002 ) it is likely associated with the sgr stream , although forbes & bridges ( 2010 ) suggest that it is connected to the canis major dwarf . in any case , the extragalactic origin seems to be well established . to explore the significance and meaning of this behavior we used the photometric catalogs presented in paper i to understand how the differences in the integrated colors relate to differences in the cluster cmds . we picked two clusters associated with the sagittarius stream , and compared them with ggcs of similar metallicity ( figure [ sgr_cmd ] ) . the photometric catalogs have been corrected by distance modulus and reddening values reported by h10 , as done in paper i. the hbs of the stream clusters are homogeneously populated in color and magnitude . in contrast , the ggcs show an obvious increase of star density towards higher temperatures and bluer colors . in these clusters the median color of hb stars is bluer by @xmath91 . this difference is qualitatively consistent with the measured discrepancy between the integrated colors . the hb star distributions differences in the galex cmds are in agreement with results from optical high - resolution surveys ( see dotter et al . 2010 for example ) . the sagittarius gcs do not show any systematic trend in the @xmath55 _ vs _ [ fe / h ] diagram . more details about the morphology of their hbs will be presented in paper iii ( r. t. rood . 2012 in preparation ) . we checked also for possible differences in relative ages . for this , we used two independent papers ( salaris & weiss 2002 , and dotter et al . 2010 ) containing age estimates for a large sample of gcs including most of our targets . even though some systematic differences are present between the two age scales , within the uncertainties the clusters connected with the sagittarius stream are classified in both cases as old and coeval with genuine ggcs in the same metallicity regime , with the only exception of pal 12 which is @xmath92 gyr younger . in particular , the four sagittarius old clusters in common with salaris & weiss ( 2002 ; terzan 8 has not been analyzed by the authors ) have an average age of @xmath93 gyr which is fully compatible with the estimates for ggcs . in dotter et al . ( 2010 ) the differences between the galactic and the sagittarius gcs are of the order of @xmath94 gyr . in addition , recent high resolution spectroscopic analysis ( carretta et al . 2010 ) showed that , on average , the sagittarius clusters in our sample share the same @xmath95-elements abundances with their galactic twins . we used the _ r-parameter _ reported by gratton et al . ( 2010 ) to highlight possible differences . r-parameter _ is defined as the ratio between the number of hb stars and that of rgbs brighter than the level corresponding to @xmath96 . this quantity is an indirect estimate of @xmath97 ( see cassisi et al . 2003 ; salaris et al . 2004 ) , since the hb luminosity , as well as the rgb and hb lifetimes , depends on the he content . however for this comparison we preferred to use only the _ r-parameter _ in order to avoid uncertainties that may come from the calibrations used to derive helium abundances ( @xmath98 ) from it . three ( ngc 4590 , ngc 5053 and ngc 5466 ) out of the five sagittarius clusters have been studied by gratton et al . it is interesting to note that these clusters have _ r _ values smaller than other clusters with similar metallicity [ fe / h]@xmath99 . given the statistical uncertainties of these measurements , the difference in _ r _ between any two given clusters is somewhat uncertain . therefore we performed a t - test to check the significance of the difference between the mean values of the two distributions . we find that for the clusters potentially connected with the sgr stream @xmath100 while for ggcs @xmath101 . the t - test gives a probability p@xmath102 that they are different . we stress that it is not possible to make a final statement about this point . in fact this simple analysis suffers of low number statistics and incompleteness of our sample . moreover the significance of the result may depend on the clusters considered to be connected with sagittarius . however it emerges that clusters connected with sagittarius share , on average , the same properties as the genuine ggcs , except for the _ r-parameter_. this difference might be an indication that those clusters have lower he abundances than ggcs in the same metallicity regime , and this is likely the main responsible of the differences in fuv integrated colors . further analysis and estimates of the _ r-parameter _ for the other clusters possibly associated with sagittarius are highly desirable and could provide stronger constraints on this problem . a number of authors ( lee et al . 2002 , sohn et al . 2006 , rey et al . 2007 ) made use of uv integrated colors to investigate age trends in the gc systems of the milky way ( by using oao 2 , ans and uit data ) , as well as of m31 and m87 . lee et al . ( 2002 ) and yi et al . ( 2003 ) argued that with a proper modeling of the hb contribution to uv bands , the @xmath2 could be a good age tracer of relatively old stellar populations . the basic assumption is that the hb morphology is driven by metallicity and age . of course a number of uncertainties may arise because of the treatment of mass loss along the upper rgb , or the presence of hot hb stars , like ehb , which are not included in these models . however under this assumption , both lee et al . ( 2002 ) and rey et al . ( 2007 ) were able to reproduce the uv color _ vs _ metallicity distribution of milky way and m31 gcs . in contrast , sohn et al . ( 2006 ) were unable to find an acceptable match between their data for the m87 gcs and the models of lee et al . ( 2002 ) , which required unphysically old ages ( @xmath103 ) to reproduce the observed color distributions . taking advantage of our large and homogeneous sample and the high photometric accuracy of our data , in figure [ met_mod ] we compare the @xmath54-[fe / h ] distribution observed in our gc sample , with the theoretical models of lee et al . ( 2002 ) for integrated colors of ssps with different ages . the color - metallicity distribution of ggcs is consistent with ages @xmath104 gyr , in broad agreement with the results obtained from fitting theoretical isochrones to the turnoff or the white dwarf cooling sequences ( e.g. , salaris & weiss 2002 ; dotter et al . it is worth noticing that the three clusters that appear as the `` youngest '' ( @xmath105 gyr ) , according to the lee et al . 2002 models , are those connected with the sagittarius stream ( ngc 4590 , ngc 5053 and ngc 5466 ) . as discussed above , however , they most likely are old and coeval with those in the same metallicity range , thus indicating that caution must be used to derive ages from this color - metallicity plane ( fusi pecci et al . 1993 ) , since other parameters play a role in distributing clusters in this diagram . the helium content certainly is one of these parameters ( see for example sohn et al . 2006 , kaviraj et al . 2007 , chung et al . 2011 ) . indeed , there is now evidence that at least a few massive gcs could host multiple populations with different he abundance ( see piotto 2009 for a review ) . as recently shown by chung et al . ( 2011 ; see also kaviraj et al . 2007 ) , helium might have a strong impact on the uv emission from an old stellar population . they suggest , in fact , that he - rich sub - populations in gcs could be able to reproduce the `` uv - upturn '' observed in elliptical galaxies . to check the impact of multiple populations on the observed colors we used the helium fractions @xmath106 derived by the _ r-parameter _ by gratton et al . ( 2010 ) for the 36 clusters in common . we split our sample using @xmath107 as threshold , to obtain two roughly equally populated sub - samples , with mean @xmath106 of about 0.27 and 0.23 . may suffer from calibration uncertainties . in this case however , we were forced to use it since _ r _ shows a mild trend with metallicity ( gratton et al . 2010 ) , which would have prevented us to make a comparison for the entire sample . ] from figure [ met_mod ] , it is evident that such differences in he content can have effects on uv colors equivalent to age differences of @xmath108 . while a detailed discussion about the long standing hb second parameter problem is postponed to paper iii , here we briefly consider how uv integrated colors vary with clusters parameters previously suggested to be connected with the presence of long hb bts . @xcite suggested that bts are related to central density and concentration , and that stellar interactions somehow enhanced mass loss . @xcite also suggested that the third parameter acting in modeling the hb morphologies is somehow related to the chance of interactions between stars . figure [ concentration ] shows the galex colors plotted as a function of concentration @xmath109 , where @xmath110 is the king model core radius and @xmath111 is the tidal radius . while the sample as a whole shows no significant correlation , the massive cluster ( @xmath112 ) colors are well correlated with @xmath113 , moving to the redder ( or cooler ) colors as the concentration increases . the spearman test gives @xmath114 probability of correlation between @xmath113 and @xmath41 , larger than @xmath53 with @xmath54 and about @xmath115 with @xmath55 . this is the opposite of what we had expected from fusi pecci et al . ( 1993 ) and dotter et al . ( 2010 ) interpretations . similar results are obtained , although with lower significance , for the relaxation time and central density ( figure [ concentration ] ) . however a direct comparison with results by dotter et al . ( 2010 ) could not be performed , since in their analysis the authors removed the effects of metallicity and age from their parameter . we defer to paper iii for a detailed and comparative analysis . @xcite have used galex to observe gcs in m31 . the authors ( kang et al . 2011 ) have recently published an updated catalog with a larger sample of clusters and reviewed reddening values . in figure [ m31comp ] we compare our results with data for m31 clusters classified as `` old '' ( @xmath116 ) by caldwell et al . ( 2011 ) and we restrict the sample to clusters with @xmath117 , in order to avoid clusters with high reddening uncertainties . + m31 seems to show a lack of red clusters with respect to the galaxy . however , this is likely due to the limited sensitivity of galex to detect relatively red populations in distant systems ( see discussion in rey et . al . 2007 ) . hence for the comparison we focus on the bluest systems , @xmath118 , @xmath119 and @xmath120 , which , for the ggcs , correspond to a metallicity range @xmath121 . in this metallicity regime the distributions in the milky way and in m31 are quite similar . the bluest colors reached are essentially the same , and the distributions show little variations with metallicity . three m31 clusters lie almost in the same region as gcs connected with the sagittarius dwarf in the @xmath54-[fe / h ] plane . they are g 327 , b 366 and interestingly b 009 that is known to be projected against the dwarf spheroidal galaxy ngc 205 . unfortunately cmds are available only for b 366 ( perina et al . this is an old cluster with a red hb and possibly a very mild blue extension , which makes it compatible with its redder uv colors . the case is very different at higher metallicity , @xmath122 . in the milky way sample there are only red ggcs , while in m31 there are many blue gcs . + in order to make the comparison with m31 gcs as complete as possible , we have supplemented our galex sample with 12 additional ggcs from dor95 , not observed by galex because of the target selection limitations discussed in section 2 . first , we converted the ans and oao 2 magnitudes from the stmag to the abmag system and we checked that no systematic color offsets or trends are present for the ( 15 ) gcs in common . the average difference results to be @xmath123 and is fully consistent with the reported errors . then , we adopted the ans and oao 2 magnitudes given in table 1 of dor95 and we corrected these values by using the color excess @xmath14 quoted by h10 . so as not to introduce additional errors through the @xmath20 magnitudes , we consider only the @xmath19 color . the results are shown as open squares in the lower panel of figure [ m31comp ] and demonstrate that ( at least ) two gcs , namely ngc 6388 and ngc 6441 , in the galaxy have colors comparable to those of m31 at @xmath122 . still , m31 appears to have many more hot metal - rich gcs . why would that be the case ? as shown in figure [ m31comp ] roughly half of the blue , metal - rich m31 gcs are indeed quite massive ( @xmath124 ) . hence , the relative paucity of hot , metal - rich gcs in the milky way could be due in part ( but only in part ) to the fact that there are only two massive metal - rich clusters in our supplemented sample . it is also possible that many ggcs with high metallicity and a blue hb are missed because of their location towards highly extinguished regions of the galaxy . + we also compare ggc colors measured with galex with those obtained for the giant elliptical galaxy m87 using hst stis images ( sohn et al . we converted those magnitudes from the stmag to the abmag photometric system . in order to perform a direct comparison , we transformed values to the metallicity indicator @xmath125 using equation a1 in the appendix of sohn et al . as shown in figure [ m87 ] , m87 gcs are on average bluer by @xmath126 mag both in @xmath41 and @xmath54 , while they do not show any appreciable difference in @xmath55 magnitudes , 16 @xmath1 and only 7 have both . ] . these differences are consistent with what observed by sohn et al . ( 2006 ) in comparison with the dor95 sample . on the basis of what we discussed in section [ sec : sagit ] , we may suppose that m87 gcs are on average older or have a higher he content than the milky way objects . as noted above , the age - metallicity grid of theoretical predictions provides realistic ages for the m87 gcs only when the effect of helium is taken into account ( see figure 2 in chung et al . 2011 ; kaviraj et al . 2007 ) . from the comparison between ggcs and those belonging to three other galaxies ( the sagittarius dwarf , m31 and m87 ) , different behaviors emerged . in fact the clusters associated with the sagittarius dwarf are on average redder than the mw ones , the m31 clusters have colors which are comparable to those of the ggcs , while the m87 star systems are bluer . we note that there may be a possible trend between the mass of the host galaxy and the color distribution of its globulars , in the sense that the higher is the galaxy mass , the bluer are the gc uv colors . in fact m87 ( with the bluer systems ) is a super - giant elliptical that is about two orders of magnitude more massive than the milky way , while sagittarius ( @xmath127 ; * ? ? ? * ) with the reddest sample of gcs ( although quite small ) , is a dwarf galaxy , and m31 ( @xmath128 ; * ? ? ? * ) and the galaxy ( @xmath129 ; * ? ? ? * ; * ? ? ? * ) representing intermediate cases . we argued that most of the observed differences between colors involving the @xmath0 band are explainable invoking different helium contents . this would lead us to speculatively think that galaxies with larger masses may have , on average , more he - rich populations . in that case , he abundance differences could be a by - product of chemical evolution differences , in some way connected to the mass of the host galaxy . this could be also connected with the formation and dynamical history of clusters in galaxies with different masses , as suggested by valcarce & catelan ( 2011 ) . in particular they argue that clusters hosted by more massive galaxies are more likely to undergo a more complex history of star formation thus having a larger spread in stellar populations properties . as part of a project aimed at studying the properties of hot stellar populations in the milky way gcs ( see paper i ) , we have presented uv integrated colors obtained with galex for 44 clusters spanning a wide range of metallicities ( @xmath130 ) , hb morphologies , structural and dynamical parameters . this represents the largest homogeneous catalog of uv photometry ever built for ggcs . we compared the behavior of uv colors with several parameters characterizing the morphology of the hb . as expected , there are general correlations , in particular between @xmath54 and the @xmath48 parameter defined by buonanno et al . ( 1993 ; 1997 ) and the hb temperature extension defined by reco blanco et al . there is also a significant correlation with the @xmath131 parameter ( dotter et al . 2010 ) , but , as expected , this parameter is insensitive to hbs with extreme blue extensions . in all color combinations , the bluest clusters are those in the intermediate metallicity regime ( @xmath132 ) . this is in agreement with dor95 and with the l@xmath57 parameters measured by fusi pecci et al . ( 1993 ) . in the @xmath55-@xmath133 plane , clusters more metal rich than @xmath134 show a clear and significant linear correlation with cluster becoming redder as metallicity increases , while there is an opposite trend in the metal - poor regime . the combinations of colors involving the @xmath0 appear more scattered , but a reasonable and similar correlation with metallicity at ( @xmath135 level ) has been also found in these cases . all the clusters suspected to be connected with the sagittarius dwarf spheroidal ( ngc 4590 , ngc 5053 , ngc 5466 , arp 2 and terzan 8) are typically @xmath136mag redder in @xmath0 colors than systems with similar iron content . no appreciable differences are found in @xmath55 . studies from different groups suggest that sagittarius clusters and their galactic counterparts are coeval , while ggcs are on average more he - rich than the sagittarius sub - set . this would tentatively be interpreted as due to a different environment in which they formed . with the aim of showing how sensitive ages derived from uv colors may be to assumptions about helium abundance , we compared our colors with evolutionary models of ssp by lee et al . the color - metallicity distribution of ggcs can be reproduced by assuming an average age of @xmath137 gyr , with a spread of about @xmath138 . alternatively , in the framework in which some gcs have experienced self - enrichment from material ejected from agbs or fast - rotating massive stars ( ventura & dantona 2008 ; decressin et al . 2007 ) , the color spread is consistent with different he content . in particular , we show that an overabundance of helium ( @xmath139 ) can mimic an age difference @xmath140 . the uv colors of ggcs are consistent with those obtained by galex for m31 clusters ( rey et al . 2007 ; kang et al . 2011 ) , at least in the intermediate / low metallicity regime . at @xmath122 , m31 gcs are systematically bluer by 12 mag , behaving like massive mw clusters with bimodal hbs , such as ngc 6388 and ngc 6441 . as already noticed by sohn et al . ( 2006 ) , m87 gcs are on average bluer than ggcs . this might be the signature of different chemical abundances impressed on the `` integrated '' properties of gcs . in particular we speculate that he abundance may be correlated with the mass of the host galaxy , being higher in gcs belonging to higher mass galaxies . + _ the authors dedicate this paper to the memory of co - author bob rood , a pioneer in the theory of the evolution of low mass stars , and a friend , who sadly passed away on 2 november 2011 . _ + we thank the anonymous referee for the useful comments and suggestions . the authors warmly thank m. bellazzini and f. fusi pecci for useful discussions that improved the presentation of the results . we are grateful to m. catelan for providing us informations about hb parameters . this work is based on observations made with the nasa galaxy evolution explorer . galex is operated for nasa by the california institute of technology under nasa contract nas5- 98034 . this research is part of the project cosmic - lab funded by the european research council ( under contract erc-2010-adg-267675 ) . thanks the hospitality and support from gemini observatory where most of this work was developed , during two extended visits in 2009 and 2010 , and a shorter visit in 2011 . r.p.s . acknowledges funding by galex grants @xmath141 nng05ge50 g and nnx08aw42 g and support from gemini observatory , which is operated by the association of universities for research in astronomy , inc . , on behalf of the international gemini partnership of argentina , australia , brazil , canada , chile , the united kingdom , and the united states of america . s.t.s . acknowledges support for this work from the galex guest investigator program under nasa grant nnx07ap07 g . and @xmath1 surface brightness profiles for ngc 6341 . the black squares represent the observed raw profile , while open circles are the sky - subtracted values . the horizontal black lines mark the estimated background level . the background - subtracted profiles obtained by using standard iraf routines are shown with open triangles , for comparison ( see section 3.1 for details ) . the profiles are well reproduced by the king model ( the grey region is defined by the structural parameters uncertainties ) with concentration parameter and core radius quoted by mvm05 and with central surface brightness ( csb ) value providing the minimum @xmath13 ( see labels ) . ] by buonanno et al . ( 1993 ; 1997 ) , horizontal branch length ( l@xmath57 ) by fusi pecci et al . ( 1993 ) , temperature extension log(@xmath142 ) as introduced by reco blanco et al . ( 2006 ) , and median horizontal branch color @xmath143 , by dotter et al . the grey triangle and square are ngc 2808 and ngc 1851 respectively ; these have bimodal horizontal branches ] color as a function of metallicity as in figure [ met ] , with superimposed theoretical models by lee et al . _ bottom panel_. the cluster sample has been split according to the helium content : black squares are `` helium - rich '' clusters ( @xmath144 ) , while the open squares are the `` he - poor '' ones . only clusters in common with gratton et al . ( 2010 ) are plotted . as before , asterisks are clusters potentially associated with the sagittarius stream . ] cccccccc + cluster & & @xmath14 & @xmath145 & @xmath41 & @xmath54 & @xmath55 + ngc 104 & -0.76 & 0.04 & 4.09 & 3.52 & 8.51 & 4.99 + ngc 1261 & -1.27 & 0.01 & 8.63 & 2.69 & 5.91 & 3.22 + ngc 1851 & -1.18 & 0.02 & 7.23 & 2.00 & 6.49 & 4.49 + ngc 1904 & -1.58 & 0.01 & 8.16 & 0.89 & 3.88 & 2.99 + ngc 2298 & -1.96 & 0.14 & 8.89 & 1.62 & 4.91 & 3.28 + ngc 2419 & -2.20 & 0.08 & 10.05 & 1.29 & 4.26 & 2.97 + ngc 288 & -1.32 & 0.03 & 8.13 & 0.91 & 4.27 & 3.36 + ngc 2808 & -1.18 & 0.22 & 5.69 & 1.10 & 4.66 & 3.56 + ngc 362 & -1.30 & 0.05 & 6.58 & 2.73 & 6.90 & 4.17 + ngc 4147 & -1.78 & 0.02 & 10.74 & 1.70 & 4.41 & 2.71 + ngc 4590 & -2.27 & 0.05 & 7.96 & 2.93 & 6.21 & 3.28 + ngc 5024 & -2.06 & 0.02 & 7.79 & 1.90 & 5.05 & 3.15 + ngc 5053 & -2.30 & 0.01 & 9.96 & 2.22 & 5.91 & 3.69 + ngc 5272 & -1.50 & 0.01 & 6.39 & 2.19 & 5.92 & 3.73 + ngc 5466 & -2.31 & 0.00 & 9.70 & 2.48 & 5.60 & 3.12 + ngc 5897 & -1.90 & 0.09 & 8.52 & 1.77 & 5.01 & 3.24 + ngc 5904 & -1.33 & 0.03 & 5.95 & 1.11 & 4.83 & 3.71 + ngc 5986 & -1.63 & 0.28 & 6.92 & 1.57 & 4.45 & 2.89 + ngc 6101 & -1.98 & 0.05 & 10.08 & 1.53 & 4.56 & 3.02 + ngc 6218 & -1.33 & 0.19 & 6.07 & 1.07 & 4.83 & 3.76 + ngc 6229 & -1.43 & 0.01 & 9.86 & 1.29 & 4.47 & 3.18 + ngc 6235 & -1.38 & 0.31 & 7.20 & 1.35 & 5.48 & 4.13 + ngc 6254 & -1.57 & 0.28 & 4.98 & 1.07 & 4.53 & 3.47 + ngc 6273 & -1.76 & 0.38 & 5.57 & & 4.49 & + ngc 6284 & -1.31 & 0.28 & 7.43 & 1.23 & 5.47 & 4.24 + ngc 6341 & -2.35 & 0.02 & 6.52 & 1.10 & 4.46 & 3.36 + ngc 6342 & -0.49 & 0.46 & 10.01 & 2.67 & 7.24 & 4.57 + ngc 6356 & -0.35 & 0.28 & 7.42 & 3.95 & 8.30 & 4.35 + ngc 6397 & -1.99 & 0.18 & 5.17 & 1.60 & 4.39 & 2.79 + ngc 6402 & -1.39 & 0.60 & 5.73 & 1.19 & 3.93 & 2.74 + ngc 6535 & -1.79 & 0.34 & 9.85 & 2.13 & 4.78 & 2.65 + ngc 6584 & -1.50 & 0.10 & 8.17 & 3.28 & 7.29 & 4.01 + ngc 6809 & -1.93 & 0.08 & 6.49 & 1.56 & 4.69 & 3.13 + ngc 6864 & -1.29 & 0.16 & 8.26 & 2.14 & 5.87 & 3.73 + ngc 6981 & -1.48 & 0.05 & 8.96 & 3.03 & 6.48 & 3.45 + ngc 7006 & -1.46 & 0.05 & 10.46 & 2.93 & 6.78 & 3.85 + ngc 7089 & -1.66 & 0.06 & 6.25 & 1.14 & 4.54 & 3.40 + ngc 7099 & -2.33 & 0.03 & 7.10 & 1.21 & 4.54 & 3.33 + ccccccc + cluster & & @xmath14 & @xmath145 & @xmath41 & @xmath54 & @xmath55 + pal 11 & -0.45 & 0.35 & 7.54 & & & + pal 12 & -0.81 & 0.02 & 11.89 & & & 5.05 + ngc 7492 & -1.69 & 0.00 & 10.48 & 1.38 & 5.78 & 4.40 + ic 4499 & -1.62 & 0.23 & 8.56 & 3.01 & 6.80 & 3.78 + terzan 8 & -2.22 & 0.12 & 11.54 & 2.46 & 5.68 & 3.21 + arp 2 & -1.74 & 0.10 & 12.41 & 3.26 & 5.72 & 2.47 + [ tab : masses ]
we study the problem of the optimal execution of a large trade in the presence of nonlinear transient impact . we propose an approach based on homotopy analysis , whereby a well behaved initial strategy is continuously deformed to lower the expected execution cost . we find that the optimal solution is front loaded for concave impact and that its expected cost is significantly lower than that of conventional strategies . we then consider brute force numerical optimization of the cost functional ; we find that the optimal solution for a buy program typically features a few short intense buying periods separated by long periods of weak selling . indeed , in some cases we find negative expected cost . we show that this undesirable characteristic of the nonlinear transient impact model may be mitigated either by introducing a bid - ask spread cost or by imposing convexity of the instantaneous market impact function for large trading rates .
1412.4839
the optimization of trading strategies has long been an important goal for investors in financial markets . as demonstrated in the context of a linear equilibrium model by kyle almost thirty years ago @xcite , the optimal strategy for an investor with insider information on the fundamental price of an asset is to trade incrementally through time . this strategy allows the trader to minimize costs whilst also minimizing the revelation of information to the rest of the market . the precise way in which it is optimal to split the large order ( herein called _ metaorder _ ) @xcite depends on the objective function and on the market impact model , _ i.e. _ the change in price conditioned on signed trade size . in part due to the increasing tendency toward a full automation of exchanges and in part due to the discovery of new statistical regularities of the microstructure of financial markets , the problem of optimal execution is receiving growing attention from the academic and practitioner communities @xcite . as pointed out in gatheral et al . @xcite , a first generation of market impact models @xcite distinguishes between two impact components . the first component is temporary and only affects the individual trade that has triggered it . the second component is permanent and affects all current and future trades equally . these models can be either in discrete or in continuous time and can assume either linear or nonlinear market impact for individual trades . the second generation of market impact models focusses on the _ transient _ nature of market impact @xcite . in such models , market impact is typically assumed to factorize into two components : instantaneous market impact and a decay component . the instantaneous component models the reaction of price to traded volume . the decay component describes how the market price relaxes on average after the execution of an order . in such models , each trade affects future price dynamics with an intensity that decays with time . the problem of optimal execution in the presence of transient impact has been considered in a series of recent studies . in the case of linear instantaneous market impact @xcite , the problem has been completely solved by showing that the cost minimization problem is equivalent to solving an integral equation . in particular gatheral et al . @xcite proved that optimal strategies can be characterized as measure - valued solutions of a fredholm integral equation of the first kind . they show that optimal strategies always exist and are nonalternating between buy and sell trades when price impact decays as a convex function of time . this extends the result of alfonsi et al . @xcite on the non existence of transaction triggered price manipulation , _ i.e _ strategies where the expected execution costs of a sell ( buy ) program can be decreased by intermediate buy ( sell ) trades . however , a series of empirical studies @xcite has clearly shown that the instantaneous market impact is a strongly concave function of the volume , well approximated by a power law function . the resulting optimal execution problem in the presence of nonlinear and transient impact is mathematically much more complicated than the linear case . in this paper we consider this optimization problem and propose several methods to find optimal solutions . some important results in the nonlinear transient case were established by gatheral @xcite who showed that under certain conditions , the model admits price manipulation , _ i.e. _ the existence of round trip strategies with positive expected revenues . this money machine should of course be avoided in the modeling of market impact . in particular gatheral set some necessary conditions for the absence of price manipulation ( see below for details ) . a step toward the solution of the optimal execution problem under nonlinear transient impact has been made recently by dang @xcite . in his paper , dang suggests a way to convert the cost minimization problem into a nonlinear integral equation and proposes a numerical fixed point method on a discretization of the trading time interval to solve this equation . as we discuss in detail below , we find that dang s fixed point method has convergence problems when the degree of nonlinearity of impact is significant and/or when the discretization grid is fine enough . in this paper we propose two methods to solve the optimal execution under nonlinear transient impact . the first is based on the homotopy analysis method ( ham ) @xcite applied to the discretized version of the integral equation proposed by dang @xcite . the method starts from an initial guess and deforms it continuously in order to find better and better approximations of the solution of the integral equation . in doing this , we are implicitly restricting the space of solutions to continuous nonvanishing functions of the trading rate . we find that the optimal solution is a non time - symmetric u - shape ; in the case of concave ( convex ) instantaneous impact , it is optimal to trade more at the beginning ( end ) of the metaorder . a comparative cost analysis shows that our solution outperforms conventional strategies . once again , the ham method explores only a restricted subspace of possible solutions . for this reason , in the second part of the paper we consider a fully numerical cost optimization method on a discrete grid . by using sequential quadratic programming ( sqp ) we minimize directly the cost functional ( _ i.e. _ we do not try to solve the integral equation ) . we find that the cost landscape is rugged , _ i.e. _ composed by a very large number of local minima separated by peaks . a significant number of these minima correspond to strategies with similar costs ; for a buy program the corresponding strategy is an alternation of intense and short bursts of buying periods and long periods of weak selling . in other words , the model admits transaction triggered price manipulation . more important , when the nonlinearity is strong and/or the partition is fine , some such strategies have a negative expected cost , indicating that the model admits price manipulation . we then further extend our analysis by minimizing the cost functional with the extra constraint that all trades should have the same sign , so that for example selling is disallowed during execution of a buy metaorder . this case requires a derivative - free optimization method . by using a direct - search method , namely the generating set search ( gss ) method , we find positive expected execution costs and _ sparse _ optimal strategies , _ i.e. _ it is optimal to trade with a few bursts at a high trading rate interspersed with long periods of no - trading . in order to eliminate negative cost solutions , we propose two ways of regularizing the model , one based on the addition of a spread cost and one based on a modification of the instantaneous impact function . in the latter case the function becomes convex for sufficiently high trading rates . both methods succeed in avoiding solution with negative costs and obviously reflect features of real markets . the paper is organized as follows . in section [ gatheral s model and fredholm integral equations ] , we state the problem and explain why it is difficult to solve . we also briefly summarize our results on the convergence of the dang fixed point method . in section [ sec : homotopy ] , we present the ham approach to the solution of the cost minimization problem and in section [ sec : generalpiecewise ] we present our results on the sqp and direct - search brute force minimization of the cost function . section [ regularize ] presents two proposed regularization methods . in section [ conclusions ] , we summarize and conclude . the model of nonlinear transient market impact for the price @xmath0 of an order execution starting at time @xmath1 when the price is @xmath2 is @xmath3 where @xmath4 is the rate of trading , _ i.e. _ number of shares per unit of time , at time @xmath5 , @xmath6 represents the impact of trading at time @xmath7 , and @xmath8 describes the impact decay . finally , @xmath9 is the volatility and @xmath10 is a wiener process . thus @xmath0 follows an arithmetic random walk with a drift that depends on the accumulated impacts of previous trades . we refer to @xmath11 as the _ instantaneous market impact function _ and to @xmath12 as the _ decay kernel_. in discrete time this is the propagator model originally developed by bouchaud et al . @xcite ; the above continuous time formulation is due to gatheral @xcite . more recently , bacry et al.@xcite have shown how a market impact model of the form may be related to a more fundamental description of order flow using hawkes processes . the optimal execution problem consists in finding the trading strategy @xmath13}$ ] that minimizes the execution cost for a given total amount @xmath14 of shares to be traded . the expected cost @xmath15 $ ] associated with a given strategy @xmath16 is given by @xmath17=\mathbb{e}\left[\int_{0}^{t}\dot x\left(t\right)\left(s\left(t\right)-s_{0}\right)dt\right]=\int_{0}^{t}\dot x\left(t\right)\,\int_{0}^{t}f\left(\dot x\left(s\right)\right)g\left(t - s\right)ds\,dt,\ ] ] and the constraint that all shares should be traded is @xmath18 expression for the expected cost corresponds to expected implementation shortfall . we search for a statically optimal strategy . a statically optimal strategy is also dynamically optimal when the cost depends on the stock price only trough the term @xmath19 , with @xmath0 a martingale @xcite . thus for the model of and the cost function described by , a statically optimal strategy is also dynamically optimal . the @xmath20 shares traded at time @xmath21 are traded at an expected price @xmath22 which represents the cumulative impact of prior trading up to time @xmath21 . the impact model of equation is fully specified by the form of the functions @xmath23 and @xmath24 . a large body of empirical evidences points out two empirical facts on the form of these two functions . first , the instantaneous impact function @xmath25 is strongly concave . for example , based on a large sample of nyse stocks , lillo et al . @xcite observed a concave function of the transaction volume . the concave function is well fitted by a power law with exponent 0.5 for small volumes and 0.2 for large volumes . bouchaud et al . @xcite analyzed stocks traded at the paris bourse and found that a logarithmic form gave the best fit to the data . in addition ( in @xcite for example ) , the decay kernel @xmath26 is found to decay asymptotically as a power law function @xmath27 the presence of these nonlinearities raises the question of the possible presence of price manipulation . there are different forms of manipulation . following @xcite , we define a _ price manipulation _ as a round trip trade whose expected cost is negative . an impact model is free from price manipulation if , for any round trip trade , _ i.e. _ a strategy with @xmath28 , the expected cost is non negative . according to proposition 1 of @xcite , the model of equation admits price manipulation in the nonlinear case unless the decay kernel @xmath29 is singular for @xmath30 . for these reasons we will focus our analysis on decay kernels of the form @xmath31 . moreover , as shown by gatheral @xcite , the requirement of no price manipulation restricts the class of possible joint form for the instantaneous impact and decay kernel . specifically , for power law impact function , @xmath32 , and a power - law kernel , @xmath31 , the following conditions @xmath33 are _ necessary _ for the absence of price manipulations . in this paper we always consider parameters @xmath34 and @xmath35 satisfying the above conditions . however , if these conditions are satisfied , there is no guarantee that the impact model does not admit price manipulation . later in the paper we will show that it is indeed the case that the above conditions are not sufficient to preclude price manipulation . a weaker form of price manipulation , relevant for our paper , is the _ transaction - triggered _ price manipulation defined in @xcite . this occurs when the expected revenues of a buy(sell ) program may be increased by intermediate sell(buy ) trades . the optimization problem of minimizing the expected cost of equation under the constraint of equation in the case of linear impact , @xmath36 , has been solved and widely studied @xcite . in what follows , we use the symbol @xmath37 to indicate the rate of trading @xmath38 . in particular , proposition 22.9 of ref . @xcite states that if @xmath24 is positive definite , then @xmath39 minimizes the expected cost if and only if there is a @xmath40 such that @xmath41 , @xmath39 solves @xmath42 as an important example , relevant for this paper , is the case @xmath31 where the integral equation becomes the abel equation with solution @xmath43^{\frac{1-\gamma}{2}}},\ ] ] where @xmath44 is uniquely determined by the constraint equation as @xmath45 where @xmath46 is euler s gamma function . this solution is u - shaped and symmetric under time reversal , _ i.e. _ @xmath47 , @xmath48 $ ] . in the following we will refer to this solution as the gss solution . in the general nonlinear case , the problem is mathematically much more complicated . a first step in this direction has been presented very recently by dang @xcite and consists in two contributions . the first one is the use of calculus of variations for integrals depending on convolution products @xcite to transform the cost minimization problem into an integral equation generalizing . specifically , given @xmath49 and @xmath50 $ ] , for the class of functions @xmath51 on @xmath52 $ ] satisfying * @xmath51 is absolutely continuous on @xmath53 , * @xmath54 $ ] , the following necessary condition for the stationarity of the functional of equation holds : @xmath55 where again @xmath40 is a constant set by the constraint equation . in the case of a convex impact function , @xmath56 with @xmath57 , equation holds @xmath58 . in contrast , in the concave case , @xmath59 , equation is not defined if the trading velocity @xmath60 vanishes at some time @xmath21 , because the derivative of @xmath23 diverges at zero . this observation restricts the class of trajectories that can be considered . moreover in the concave case there is no guarantee that the necessary condition is also sufficient , because the minimization cost problem could have a large number of extremal points the condition of equation is a stationarity condition for the cost functional only when @xmath61 , _ i.e. _ only for @xmath62 or @xmath63 , because the derivative of @xmath23 diverges in zero . * we can find only strategies described by trading rates of definite sign . non potrebbe saltare ed essere discontinua ? in the concave case we do not have guarantees that the equation describes the global minimum of the cost functional , we can only find a local minimum . chi ci dice che non e vero per convex ? * equation is a weakly singular urysohn equation of the first kind @xcite taking the form @xmath64 where @xmath65 note that there are two sources of nonlinearity in the integral equation : the nonlinear impact function @xmath66 , and the function @xmath67 . in fact , the nonlinearity depends also on the first derivative of the impact function @xmath68 , _ i.e. _ on the response of price to the traded volume per unit time . moreover , the term involving @xmath69 entangles the response of price at time @xmath21 with the future trading rates , _ i.e. _ @xmath70 for @xmath71 , _ i.e. _ a coupling between present and future values of @xmath60 . this means that equation can not be classified as a weakly singular nonlinear fredholm equation , because the function @xmath67 depends both on @xmath21 and on @xmath7 . in the linear impact case , both nonlinearities disappear and one recovers a weakly singular linear fredholm integral equation of the first kind for the trading rate , where there is no such coupling between present and future times . this special form of _ nonlinearity _ makes this integral equation different from standard nonlinear fredholm integral equations of the first kind . the standard form of the nonlinear fredholm integral equations of the first kind @xcite is given by @xmath72 where the kernel @xmath73 and the function @xmath74 are given real - valued functions , and @xmath75 is a nonlinear function of the unknown function @xmath76 . this integral equations of first kind are considered an ill - posed problem @xcite because it does not satisfy the following properties : existence of a solution , uniqueness of a solution , and continuous dependence of the solution @xmath76 from the function @xmath74 . this means that a small change of the function @xmath74 can give a large change in the function @xmath76 . to determine a solution of the equation one usually converts it to a linear fredholm integral equation of the first kind of the form @xmath77 this approach is suitable if the function @xmath78 is an invertible function of its argument . another method is the tikhonov s _ regularization _ , it received a considerable interest especially in solving first kind integral equations @xcite . the regularization consists of replacing the ill - posed problem with a well - posed one . it transforms the integral equation of first kind in a integral equation of second kind dependent from a small parameter @xmath79 @xmath80 and one can find the solution in the limit @xmath81 . the advantage of this method is that integral equations of the second kind are much simpler to solve with respect to that of first kind @xcite . other methods can be applied to find a solution of the linear equation , as the augmented galerkin method @xcite or wavelets method @xcite . it is important to note that @xmath67 in equation is not an invertible function of @xmath60 , because it depends not only on @xmath7 but also on @xmath21 . for this reason , the usual method for solving nonlinear integral equation by setting @xmath82 and solving the linear integral equation for @xmath83 is not applicable here . in section [ sec : homotopy ] , we use the homotopy analysis method to solve the integral equation . the second contribution of dang @xcite is a numerical scheme to solve the integral equation . this is a fixed point iteration scheme to find a numerical solution by quadrature methods which we review in appendix [ sec : dangfixedpoint ] . quadrature relies on a discretization of the time interval @xmath84 $ ] into @xmath85 subintervals , transforming the integral equation into a system of equations for a vector of dimension @xmath85 . a detailed numerical analysis of the convergence properties of dang s fixed point method as a function of the number of subintervals @xmath85 and the deviation of the impact from linearity , as measured by @xmath86 is presented in appendix [ sec : dangfixedpoint ] . we now summarize the results of this analysis . the green area described in the left panel of figure [ fig : conv_region&squared_error ] shows the region where our implementation of dang s algorithm converges , while the right panel shows the squared residual error of solutions ( defined in equation for a generic nonlinear operator ) obtained as fixed points of the map . both panels tell the same story . for small values of the number @xmath85 of subintervals , the method converges , even if the impact function is strongly nonlinear . as @xmath85 increases , _ i.e _ for finer partitions of the interval , we find that the method converges only for very moderately nonlinear impact functions . dang s method obviously always converges when @xmath87 , because in this case , the objective function has only one minimum . that dang s method does not converge is no surprise ; iteration schemes are useful for solving integral equations of the second kind , whilst equation is an integral equation of the first kind . . right panel : squared residual error of solutions obtained as map s fixed points.,scaledwidth=100.0% ] in summary , dang s proposed solution technique in @xcite is problematic . first , there is no guarantee that solving the integral equation leads to a strategy that minimizes the expected execution cost over the class of all admissible strategies . in particular , if the impact function is concave , as observed in reality , solutions are restricted to those for which the trading velocity @xmath37 never vanishes , because @xmath88 diverges at @xmath89 . second , according to our numerical tests , dang s fixed point algorithm seems to converge only for moderate nonlinearity and moderate discretizations of the @xmath84 $ ] interval . thus we need to investigate alternative approaches . in the next section we consider a perturbative approach to equation valid for weak nonlinearity , _ i.e. _ @xmath90 . this enables us to gain some understanding of the effect of nonlinearity on the optimal solution . in section [ sec : homotopy ] we present a new approach to the solution of equation using the homotopy method . finally , in section [ sec : generalpiecewise ] , we will present a brute - force numerical minimization of a discretized version of the cost function , where we will see explicitly that this cost function has many local minima . in this section we present a simple perturbative method to investigate the solution of equation in the weakly nonlinear case . this will provide some intuition for the results we obtain later using the more powerful homotopy method . our perturbative approach is based on two approximations . the first one regards the impact function , while the second one regards the trading rate , _ i.e. _ the unknown solution . let us consider a buy program , @xmath91 and a slightly concave impact function , @xmath92 , with @xmath93 . we then make the approximations @xmath94 substituting into equation , and keeping terms only to order @xmath95 , we obtain @xmath96\,\int_{t}^{t}\,v\left ( s\right)\,g\left ( s - t\right)\,ds\right\rbrace = \lambda . \end{split}\ ] ] as mentioned earlier , the gss solution solves the zeroth order case of @xmath97 in closed form . and @xmath98 . the full line represents the solution @xmath99 . we observe that this solution is not symmetric under time reversal . the dotted line represents the gss solution , _ i.e. _ the solution valid for the linear impact case.,scaledwidth=80.0% ] now write @xmath100 . then , matching terms of order @xmath95 gives the following linear fredholm equation for @xmath101 : @xmath102\,\int_{t}^{t}\,v_0\left ( s\right)\,g\left ( s - t\right)\,ds - \lambda'.\nonumber\\\end{aligned}\ ] ] we solve the above equation for a fixed value of @xmath103 using the constraint @xmath104\,ds = x\ ] ] to determine the correct value of @xmath103 for a fixed value of @xmath95 . we search for @xmath103 iteratively until equation is satisfied to within some given precision , in our case @xmath105 of @xmath14 . an example of this perturbative solution is shown in figure [ fig : weakcuratogatheral ] . we use the quadrature method described in section [ sec : discretehomotopy ] to discretize the kernel @xmath24 and perform matrix inversion to solve a discretized version of equation . we report here the case @xmath106 and @xmath98 using a quadrature grid with @xmath107 points . our first observation is that when market impact is concave , the optimal strategy is no longer symmetric under time reversal , in contrast to the linear case . indeed , it is better to trade faster in the first half of the trading period @xmath108 and more slowly in the second half . repeating the computation in the convex market impact case , one obtains the opposite behavior : it is better to trade faster in the second half , as found also in @xcite with the fixed point method . in section [ sec : homotopy ] , we will observe that the numerical solution of in the strongly nonlinear impact case using the homotopy method has the same properties . it is worth emphasizing again that there is no reason to believe that a solution of the integral equation gives the optimal execution strategy , corresponding to the global minimum of the expected cost . the concept of homotopy can be traced back to henri poincar . in short , a homotopy describes a _ continuous _ variation or deformation . for example , a circle can be continuously deformed into a square or an ellipse , a coffee cup can be continuously deformed into a doughnut but not into a ball . more formally , let us consider the following general nonlinear equation : @xmath109=0,\ ] ] where @xmath110 is a nonlinear operator , @xmath21 denotes the independent variable , and @xmath76 is the unknown function . liao @xcite constructs the so - called zero - order deformation equation @xmath111=p\,\hslash\,h\left(t\right ) \mathcal{n}\left[\phi\left(t;p\right)\right],\ ] ] where @xmath112 $ ] is called the homotopy parameter , or embedding parameter and @xmath113 is a non - zero auxiliary parameter which is called the convergence control parameter @xcite . @xmath114 is an auxiliary function , @xmath115 is an auxiliary linear operator , @xmath116 is an initial guess of @xmath76 , and @xmath117 is an unknown function . there are no particular prescriptions for the choice of auxiliary functions or operators ; often the choice depends on the problem to be solved . when @xmath118 and @xmath119 , we have respectively @xmath120 thus , as @xmath121 increases from @xmath122 to @xmath123 , the solution @xmath117 varies _ continuously _ from the initial guess @xmath116 to the sought solution @xmath76 . expanding @xmath117 in a maclaurin series with respect to @xmath121 , we have @xmath124 where @xmath125 the series representation of @xmath126 in equation is called the homotopy series and @xmath127 in is called the @xmath128th - order homotopy derivative of @xmath126 @xcite . if the auxiliary linear operator , the initial guess , the convergence control parameter @xmath113 , and the auxiliary function are properly chosen , the homotopy series converges at @xmath119 , giving the sought solution of equation . then by using the relationship @xmath129 , one has the so - called homotopy series solution @xmath130 which is one of the solutions of the original nonlinear equation , as proved by liao @xcite . defining the vector @xmath131 and differentiating the zero - order deformation equation @xmath128 times with respect to the homotopy parameter @xmath121 and then setting @xmath118 and finally dividing them by @xmath132 , we have the so - called @xmath128th - order deformation equation @xmath133=\hslash\ , h\left(t\right ) r^{m}\left(\textbf{v}^{m-1}\right),\ ] ] where @xmath134}{\partial p^{m-1}}\big|_{p=0}\ ] ] and @xmath135 the @xmath136th - order approximate solution is then given by @xmath137 and the exact solution by @xmath138 . the two main difficulties of this approach are to compute the derivatives of equation and how to choose an appropriate value of @xmath113 in order to guarantee the convergence of the series solution of equation . the first problem requires us to compute the homotopy derivative of a given nonlinear smooth function @xmath23 of the homotopy maclaurin series of equation , _ i.e. _ to compute the derivative of @xmath139 . we refer the reader to appendix [ apphomo ] for details . the second problem is how to choose an appropriate value of the convergence control parameter @xmath113 in order to guarantee the convergence of the series of equation @xcite . to do this , we will adopt in the following , the so - called _ optimization _ method @xcite , according to which , we define the squared residual of the governing equation as @xmath140\right)^2dt.\ ] ] the optimal value of the convergence control parameter is then obtained by finding the minimum of this squared residual . in fact , if @xmath141 is the solution of the original problem of equation , the residual @xmath142 vanishes . now we apply the homotopy analysis method ( ham ) to the solution of the nonlinear integral equation . thus @xmath143=-\lambda+\int_{0}^{t}g\left(|t - s|\right)\,f\left(v(s),t\right)\,ds.\ ] ] as suggested in @xcite , we choose the linear operator @xmath144 to be the identity , that is @xmath145=\phi\left(t;p\right),\ ] ] and the auxiliary function to be @xmath146 . the zero - order deformation equation is then @xmath147=\hslash\,p\,\mathcal{n}\left[\phi\left(t;p\right)\right].\ ] ] differentiating this zero - order deformation equation @xmath128 times with respect to @xmath121 , and finally dividing by @xmath132 , we obtain the @xmath128th - order deformation equation @xmath148 where for @xmath149 @xmath150}{\partial p^{m-1 } } |_{p=0 } \nonumber\\ \ & = & \int_{0}^{t}g\left(|t - s|\right)\left\lbrace\frac{1}{\left(m-1\right)!}\left.\frac{\partial^{m-1}f\left(\phi\left(s;p\right),t\right)}{\partial p^{m-1}}\right|_{p=0}\right\rbrace\,ds\ , . \nonumber\\\end{aligned}\ ] ] for example , the first - order deformation equation is @xmath151 to compute the higher - order deformation equations we need to write @xmath67 as a function of two homotopy - series , defined respectively at the time points @xmath7 and @xmath21 @xmath152 and then apply the homotopy - derivative of equation , relative to a nonlinear one - dimensional system , for @xmath153 , and the homotopy - derivative of equation , relative to a nonlinear two - dimensional system , for @xmath71 . the expression of the first homotopy - derivative , for @xmath154 , is given by @xmath155 higher orders may be computed using symbolic computation software such as mathematica or maple . obviously , for this algorithm to work , the initial guess needs to satisfy the condition @xmath156 for all @xmath157 $ ] . thus , @xmath158 must have the same sign on the whole interval @xmath84 $ ] . from now on , we consider a buy program . in order to test for possible dependencies on the initial guess , we choose both the vwap strategy , @xmath159 , and the gss solution for the linear case , _ i.e. _ @xmath160 given by equation as initial guesses . to apply ham to our problem , we need to compute the definite integrals , which seem to be analytically intractable . we therefore propose a way to approximate these integrals , and refer to this discretized version of ham as the dham approach . we first discretize equation by splitting the time interval @xmath52 $ ] into @xmath85 subintervals at the times @xmath161 where @xmath162 . this gives the following nonlinear system of @xmath85 equations in the variables @xmath163 where @xmath164 @xmath165 where @xmath166 indicates the time point @xmath167 . the nonlinear function @xmath168 of equation becomes a real @xmath169 matrix @xmath170 the decay kernel @xmath171 becomes a toeplitz real symmetric @xmath172 matrix given by @xmath173 if @xmath174 , for @xmath175 we have @xmath176 and the diagonal terms are given by @xmath177 in this scheme the constraint on the traded volume of equation is given by @xmath178 we are not able to solve the nonlinear system of equation directly . rather , we search an approximate homotopy series solution @xmath179 where the deformation equations for @xmath149 are approximated by @xmath180 and where the @xmath181-th homotopy derivative is evaluated on the grid points @xmath182 @xmath183 the approximate solution of order @xmath136 is given by @xmath184 and we compute the squared residual error of equation as @xmath185 ^ 2.\ ] ] the value of @xmath186 that minimizes this error gives the dham solution @xmath187 of equation . this solution can be considered as a piecewise constant approximation of the exact solution corresponding to a sequence of vwap executions with trading rates @xmath188 . finally , the expected liquidation cost is approximated by @xmath189= \sum_{i=1}^{n}\,\sum_{j=1}^{n } v_{i}^{(n)}\,f\left(v_{j}^{(n)}\,\right ) a_{ij},\ ] ] where the @xmath190 are elements of a toeplitz matrix that describes the decay kernel @xmath8 @xmath191 we present here the optimal strategies obtained with dham in the no - dynamic - arbitrage region given by equation . we analyze in detail the values @xmath192 and @xmath193 in a strong nonlinear regime , i.e. @xmath194 . we consider the case where the volume to be purchased is @xmath98 , which can be interpreted as a metaorder execution where one buys @xmath195 of the available unitary market volume . we discretize using a grid of @xmath196 subintervals and compute the solution up to the @xmath197-th order with a tolerance on the constraint on the total quantity executed of @xmath123 of @xmath14 . as mentioned earlier , we consider two initial guesses , namely one corresponding to a vwap profile and the other to a gss solution . is illustrated on the left panel , the minimum is attained for @xmath198 where we have @xmath199 . the vwap initial guess and the dham solution are reported on the right panel respectively by a full green line with circles and a dashed blue line with circles , are reported also the results of the seven deformation equations.,scaledwidth=100.0% ] is illustrated on the left panel , the minimum is attained for @xmath200 where we have @xmath201 . the gss initial guess and the dham solution are reported on the right panel respectively by a full green line with circles and a dashed blue line with circles , are reported also the results of the seven deformation equations.,scaledwidth=100.0% ] the results for a vwap initial condition are illustrated in figure [ fig.1 ] . the left panel shows that the minimum of squared residual of the @xmath197-th order iteration is attained at @xmath198 . the results for a gss initial condition are illustrated in figure [ fig.2 ] , where the minimum of the squared residual is attained at @xmath200 ( left panel ) . the right panel of both figures shows the trading profile of the different approximating terms @xmath202 , ( @xmath203 , as well as of the @xmath197-th order approximated solution @xmath204 obtained by summing the terms . it is worth noticing that higher order terms become smaller and smaller . in both cases the dham approximated solution is asymmetric with respect to time reversal ; it is optimal to trade faster at the beginning of the trading period and slower at the end . this is consistent with our earlier findings in section [ perturbative approach ] for small deviations from linear impact . finally , it is worth noticing that the two different initial conditions lead to very similar approximated solutions . in fact , if we exclude the last discretization point ( @xmath205 ) , the maximum value of the relative difference between the two profiles is around @xmath206 , but the relative difference of the expected liquidation cost , given by equation , is only @xmath207 . the squared error reported in tab [ table.0.1 ] , as we expected , decreases when the nonlinearity decreases . .minimum of the squared residual @xmath208 on the no - dynamic - arbitrage region for @xmath209 for the dham strategy . the error decreases when the degree of nonlinearity decreases . [ cols="<,<,<",options="header " , ] , _ i.e. _ @xmath195 of a market volume and with @xmath196 subintervals , and a concave - convex impact function with @xmath210 ( top ) and @xmath211 ( bottom ) . we use @xmath212 starting points for each optimization and consider only parameters in the no - dynamic - arbitrage region defined by . we observe that holding @xmath35 fixed , expected cost is not a monotonic function of @xmath34 . for @xmath210 negative costs are observed for any value of @xmath35 , while for @xmath211 we observe no negative costs.,title="fig:",scaledwidth=50.0% ] , _ i.e. _ @xmath195 of a market volume and with @xmath196 subintervals , and a concave - convex impact function with @xmath210 ( top ) and @xmath211 ( bottom ) . we use @xmath212 starting points for each optimization and consider only parameters in the no - dynamic - arbitrage region defined by . we observe that holding @xmath35 fixed , expected cost is not a monotonic function of @xmath34 . for @xmath210 negative costs are observed for any value of @xmath35 , while for @xmath211 we observe no negative costs.,title="fig:",scaledwidth=50.0% ] in this paper , we have studied the problem of finding optimal execution strategies in the nonlinear transient impact model of @xcite . we considered several different approaches , discussing their ranges of validity and the characteristics of the solutions they generate , and compared their associated expected costs . we also focused our attention on the presence of different forms of possible price manipulation . if the class of admissible strategies is constrained so as to eliminate the possibility of zero trading rates , the cost minimization problem may be cast as an urysohn integral equation of the first kind . to solve for the optimal execution strategy with this constraint , we proposed a discretized homotopy analysis technique . the solutions obtained are by construction continuous deformation of initial guesses , such as vwap . we have shown that the optimal solution found in this way is not time - reversal symmetric , but front - loaded in the case of concave market impact . our expected cost analysis shows that such solutions substantially outperform conventional execution strategies such as vwap . the integral equation approach can only find local minima of the cost functional . in order to search for a global minimum we therefore perform a direct numerical minimization of the cost function . from our analysis using an sqp global optimization technique , we found that the discretized cost function exhibits a rugged landscape , with many local minima separated by peaks . the global minimum ( as well as other many other suboptimal minima ) corresponds to a strategy that alternate buys and sells , and thus allows for transaction - triggered price manipulation . in particular , for a buy program the optimal solution consists of short bursts of trading at high rates separated by long periods of selling at low trading rates . even worse , when the nonlinearity is strong and when the discretization is sufficiently fine , we find negative execution costs . coupled with the transient property of market impact in our model , this implies that the nonlinear transient impact model with purely power - law concave impact admits price manipulation , _ i.e. _ round trips with positive expected revenues , eliminating such a model from consideration for practical use . in particular , this shows that the conditions of @xcite for no price manipulation are necessary but not sufficient . we also studied optimal monotonic strategies where selling(buying ) is disallowed in a buy(sell ) program . this optimization is performed by a derivative - free direct search method , specifically the generating set search algorithm . in this case , we eliminate transaction - triggered price manipulation , we no longer see negative expected costs , and the corresponding optimal strategy is sparse . in other words , it is optimal to trade in a few intense bursts separated by long periods of no trading . finally , we propose two ways to regularize the transient impact model , both of them reflecting important features of the market that are neglected in the simple version of the model . in the first approach , we add a spread cost , effectively performing a @xmath213 or lasso regularization , which discourages negative trades for buy programs . in the presence of sufficiently strong transaction costs , the optimal execution strategy no longer exhibits a negative expected cost , and the optimal buy strategy becomes sparse and bursty , with a few moments of strong positive trading separated by periods of zero or slightly negative trading . in the second approach , we add to the market impact function a convex component which makes it very costly to trade at high trading rates . we observe that if the convex component is sufficiently large , negative expected execution costs are again eliminated . our results indicate that a fully satisfactory model of market impact , which is free of price manipulation and consistent with empirical observation is still lacking , leaving a fruitful field for future research @xcite . we thank dario trevisan for useful discussions regarding calculus of variations . this research was partially supported by hsbc holdings plc . [ fixed point method ] in this section we analyze an iteration scheme proposed by dang @xcite to find a numerical solution of equation by quadrature methods . dang considers a power law impact function @xmath214 where @xmath215 . if we want to study the case @xmath216 we should find a way to handle the problem of an infinite first derivative of @xmath217 for @xmath89 . we define a perturbative version of the market impact function @xmath218 @xmath219 where the perturbation is given by the parameter @xmath220 . in this way the trading rate can assume any value on @xmath221 we search a solution @xmath222 of the equation with a perturbation near to zero , _ i.e. _ @xmath223 . dang defines a nonlinear map from equation and searches a fixed point for the map starting from the initial guess @xmath224 . this means that we have a sequence of approximations for the solution @xmath225 , the convergence to a fixed point of the map is a necessary condition in order to have a convergent iteration scheme @xcite . the iterative scheme is defined by a taylor expansion @xmath226 where @xmath227 is the derivative of @xmath228 in equation with respect to @xmath229 . following dang , we solve for each @xmath128-th iteration the following linear system : @xmath230 this defines a nonlinear map @xmath231 @xmath232 where @xmath233 we search a solution that is a fixed point of the map @xmath231 , _ i.e. _ @xmath234 , as in dang @xcite . starting from the initial guess @xmath158 we look for a value @xmath235 for which the @xmath85-dimensional vector @xmath236 is a constant vector for @xmath237 . a simple scalar quantity controlling if the fixed point has been reached is the mean field , defined as @xmath238 . the mean field @xmath239 is also useful in order to implement a procedure that leads to determine the value of @xmath40 that satisfies the constraint on the total quantity traded . it is clear that for a general dynamic system the condition of a constant mean field @xmath239 does not guarantee the existence of a fixed point since there can be also more complex attractors , _ e.g. _ a chaotic attractor , where the mean field is constant . here we are interested in studying only a necessary condition in order to find a fixed point of the nonlinear map @xmath231 . moreover , it is important to notice that the evolution of the map @xmath231 does not satisfy the constraint of equation , _ i.e. _ if the initial point @xmath158 is on the hyper - plane , it could go outside of it when we iterate the map @xmath231 . this is the reason for which we should tune the constant @xmath40 in order to satisfy our constraint within some precision . dang s proposal for the initial guess @xmath240 , @xmath241 , is a constant vector whose components are given by @xmath242 in principle the fixed point could depend on the starting point @xmath158 that we choose as initial guess for the solution . we add to this initial guess a set of initial guesses uniformly distributed on the simplex to investigate the presence of different attractors . the convergence of this method is determined by two parameters : @xmath85 and @xmath34 . we performed an extensive analysis of the convergence of the dang s fixed point method . the convergence criterion is that the relative standard deviation of the mean field is below some threshold , typically @xmath243 . the result of this analysis for @xmath193 and @xmath244 is shown in figure [ fig : conv_region&squared_error ] , but similar results are obtained for small values of @xmath95 . we observe that when @xmath85 increases , the set of values of @xmath34 for which we have the convergence decreases , _ i.e. _ we have a @xmath245 under which the procedure does not converge , _ i.e. _ we do not find a fixed point for the map @xmath231 , see figure [ fig : conv_region&squared_error ] . when the procedure converges , the solution does not depend from the initial guess , i.e the map @xmath231 has only one attractive basin . moreover , the convergence is extremely fast , reaching the fixed point in very few iterations of the map . as @xmath85 increases however , dang s algorithm converges only in the weakly nonlinear case . confining our attention to a regime where dang s algorithm does converge , how does the solution achieved with dang s algorithm compare with the solution achieved using the dham and sqp methods we propose ? in figure [ fig.a.2 ] we report results for the very weakly nonlinear case with @xmath246 . we find that in this case the dang method converges very quickly ( right panel ) and the solution is essentially coincident with the one obtained with the dham method ( left panel ) . the solution of the numerical optimization using the sqp algorithm is oscillating , alternating between positive and negative trading rates . in conclusion , dang s algorithm converges only under weak nonlinearity and/or when discretization of the time interval is very coarse . when the dang method does converge , the solution is very close to the one obtained with the dham method proposed in this paper . however the dham approach is applicable not only in the weakly nonlinear regime where dang s algorithm converges , but also in the strongly nonlinear regime . . the right panel reports the fast convergence of the mean field @xmath247 of the fixed point method with a vwap initial guess and @xmath248 . notice that the initial guess goes outside the constraint at the beginning of the iteration procedure . , scaledwidth=90.0% ] as explained in section [ sec : homotopy ] , the so - called homotopy derivative is used to deduce deformation equations of equation for any order greater than one , _ i.e. _ to compute @xmath236 for @xmath149 . such computations are difficult because they depend on the nonlinear operator @xmath110 . for our problem , we need to compute the homotopy derivative of a power law function . a nonlinear power law function with a integer exponent , _ i.e. _ @xmath249 , was studied by molabahrami and khani @xcite to find an approximate solution of the burgers - huxley equation . the case of a real power - law index was studied by wang et al . @xcite analyzing the flow of a power - law fluid film on an unsteady stretching surface . our case is more complicated than the case of a simple power - law function . recent results given by turkyilmazoglu @xcite and liao @xcite show how to compute the homotopy - derivative of any smooth function . the homotopy - derivative @xmath250=\left(1/m!\right)\partial^{m}f\left(\phi\right)/\partial p^{m}$ ] is given by the recursive relation @xmath251=\sum_{k=0}^{m-1}\left(1-\frac{k}{m}\right)\mathcal{d}^{m - k}\left[\phi\right]\frac{\partial}{\partial \phi}\left\lbrace \mathcal{d}^{k}\left[f\left(\phi\right)\right ] \right\rbrace,\ ] ] evaluated at @xmath118 . the sum consists of two homotopy - derivative terms . the first gives @xmath252 , the second term gives polynomial terms of @xmath253 that multiply @xmath254 , derivatives of @xmath23 evaluated at the initial guess . if the market impact function @xmath23 is of the form @xmath255 with @xmath216 , all such derivatives diverge at @xmath89 . as mentioned in section [ sec : homotopy ] , we avoid this problem by choosing the initial guess to be a strictly positive ( or negative ) function of time . using the @xmath256-th order of equation in equation we can express the @xmath128-th homotopy derivative @xmath236 as a complicated function of the previous @xmath256 derivatives . to handle the nonlinearity of equation in the ham framework we need a further step . the coupling between past and future values of trading rates implies that we have to consider our problem as a two - dimensional system . this means that we use the homotopy - derivative for systems described by two variables @xcite , for example @xmath83 and @xmath257 , in which we have a nonlinear coupling between them given by @xmath258 @xmath259=\sum_{k=0}^{m-1 } \left(1-\frac{k}{m}\right)u^{m - k}\mathcal{d}^{k}\left[\frac{\partial f\left(\phi,\psi\right)}{\partial \phi}\right]+ \left(1-\frac{k}{m}\right)w^{m - k}\mathcal{d}^{k}\left[\frac{\partial f\left(\phi,\psi\right)}{\partial \psi}\right],\ ] ] where @xmath260 are the maclaurin series of @xmath261 respectively . thus , given the expression of equation as @xmath262 , we are able to apply the ham to our optimal execution problem using the equation and choosing a function with a given sign as initial guess . the introduction of lagrange multipliers as additional variables makes it possible to apply to the constrained - extremum problem the same first order condition used in the free - extremum problem . it is tempting to go a step further and borrow the second - order sufficient conditions as well . this , however should not be done . as in the case of free extremum , it is possible to express the second - order sufficient condition in a determinant form . in place of the hessian , in the constrained case we shall encounter what is known as a _ bordered hessian _ this matrix is simply the hessian of the lagrangian function , where the borders are given by the first derivatives of the constraint function . we illustrate it in our specific case , where we have one linear constraint and a cost function defined on @xmath263 by the equation . the second derivatives of the lagrangian are given by @xmath264 where @xmath265 and the borders are given by the second derivative respect to @xmath40 and @xmath266 , _ i.e. _ the derivatives of the constraint with respect to @xmath266 . the sufficient conditions imply the computation of the determinant of @xmath267 minors of the bordered hessian . let @xmath268 be the upper left @xmath269 submatrix of the bordered hessian , we have the following sufficient conditions for a local stationary point 1 . if @xmath270det@xmath271 for @xmath272 , the cost function has a local minimum . if @xmath273det@xmath271 for @xmath272 , the cost function has a local maximum . 3 . if det@xmath274 for @xmath272 , but fall into a different pattern of signs than the above two cases , then the critical point is some point of saddle . 4 . if some of the det@xmath275 , the character of the critical point is indeterminate at the level of second order partial derivatives . abbasbandy , s. , shivanian , e. , vajravelu , k. , mathematical properties of @xmath113-curve in the framework of the homotopy analysis method , _ commun . nonlinear sci . _ * 16 * ( 2011 ) 42684275 . abbasbandy , s. , homotopy analysis method for the kawahara equation , _ nonlinear analysis : real world applications _ * 11 * ( 2010 ) 307312 . abergel f. , bouchaud j .- p . , foucault t. , lehalle c .- a . , rosenbaum m. , _ market microstructure : confronting many viewpoints _ ( the wiley finance series , padstow , cornwall , uk , 2012 ) . akyildiz , f. t. , vajravelu , k. , magnetohydrodynamic flow of a viscoelastic fluid , _ physics letters a _ * 372 * ( 2008 ) 33803384 . alfonsi , a. , and schied , a. , capacitary measures for completely monotone kernels via singular control , _ siam j. control optim . _ * 51 * ( 2013 ) 17581780 . alfonsi , a. , schied , a. , and slynko , a. , order book resilience , price manipulation , and the positive portfolio problem , preprint available at ssrn ( 2009 ) . almgren , r. , and chriss , n. , value under liquidation , _ risk _ * 12 * ( 1999 ) 6163 . almgren , r. , and chriss , n. , optimal execution of portfolio transactions , _ j. risk _ * 3 * ( 2000 ) 539 . almgren , r. , optimal execution with nonlinear impact functions and trading - enhanced risk , _ appl . * 10 * ( 2003 ) 118 . bacry , e. , iuga , a. , lasnier , m. , and lehalle , c .- a . , market impacts and the life cycle of investors orders , http://arxiv.org/abs/1412.0217 ( 2014 ) . berger , m. s. , nonlinearity and functional analysis , lectures on nonlinear problems in mathematical analysis . academic press , inc.(london ) ltd(1977 ) . bertsimas , d. , and lo , a. , optimal control of execution costs , _ journal of financial markets _ * 1 * ( 1998 ) 150 . bouchaud , j .- p . , farmer , j.d . and lillo , f. , how markets slowly digest changes in supply and demand . in _ handbook of financial markets : dynamics and evolution _ , edited by t.hens and k. schenk - hoppe , ( 2008 ) , academic press : new york . bouchaud , j .- p . , kockelkoren , j. , potters , m. , random walks , liquidity molasses and critical response in financial markets , _ quantitative finance _ * 6 * ( 2006 ) 115123 . bouchaud , j .- , gefen , y. , potters , m. and wyart , m. , fluctuations and response in financial markets : the subtle nature of random price changes , _ quantitative finance _ * 4 * ( 2004 ) 176190 . brodie , j. , daubechies , i. , de mol , c. , giannone , d. , and loris , i. , sparse and stable markowitz portfolios , _ pnas _ * 106 * ( 2009 ) 1226712272 . browder , f.e . , construction of fixed points of nonlinear mappings in hilbert space , _ journal of mathematical analysis and applications _ * 20 * ( 1967 ) 197228 . brown , k. s. and sethna , j. p. , statistical mechanical approaches to models with many poorly known parameters , _ physical review e _ * 68 * ( 2003 ) 021904 - 1021904 - 9 . busseti , e. , and lillo , f. , calibration of optimal execution of financial transactions in the presence of transient market impact , _ j. stat . mech . _ * * ( 2012 ) p09010 . chiang , a. c .. fundamental methods of mathematical economics . mcgraw - hill , inc , ( 1984 ) . dang , n. m. , optimal execution with transient impact , _ _ * * ( 2014 ) ssrn - id2183685 . donier , j. , bonart , j. , mastromatteo , i. , and bouchaud , j .- , a fully consistent , minimal model for non - linear market impact , http://arxiv.org/abs/1412.0141v2 ( 2014 ) . gatheral , j. , schied , a. and slynko , a. , transient linear price impact and fredholm integral equations , _ mathematical finance _ * 22 * ( 2012 ) 445474 . gatheral , j. , schied , a. and slynko , a. , exponential resilience and decay of market impact , _ econophysics of order - driven markets _ , springer , berlin , 225236 ( 2011 ) . gatheral , j. , schied , a. , dynamical models of market impact and algorithms for order execution , in _ handbook of systemic risk _ fouque and j. a. langsam editors ) , cambridge university press 2013 . gatheral , j. , no - dynamic - arbitrage and market impact , _ quantitative finance _ * 10 * ( 2010 ) 749759 . hetmaniok , e. , sota , d. , trawiski , t. , witua , r. , usage of the homotopy analysis method for solving the nonlinear and linear integral equations of the second kind , _ numer . * 67 * ( 2014 ) 163185 . kolda , t. g. , lewis , r. m. and torczon , v. , stationarity results for generating set search for linearly constrained optimization , _ siam journal on optimization _ * 17 * ( 2006 ) 943968 . kolda , t. g. , lewis , r. m. and torczon , v. , optimization by direct search : new perspectives on some classical and modern methods , _ siam review _ * 45 * ( 2003 ) 385482 . liao , s. , on the homotopy analysis method for nonlinear problems , _ applied mathematics and computation _ * 147 * ( 2004 ) 499513 . liao , s. , notes on the homotopy analysis method : some definitions and theorems , _ commun . nonlinear sci numer . simulat . _ * 14 * ( 2009 ) 983997 . liao , s. , beyond perturbation : introduction to the homotopy analysis method . boca raton : chapman hall crc / press ; ( 2003 ) . liao , s. , homotopy analysis method in nonlinear differential equations . springer ( 2012 ) . lillo , f. , farmer , j.d . and mantegna , r. n. , master curve for price - impact function , _ nature _ * 421 * ( 2003 ) 129130 . liu , c .- s . , the essence of the generalized taylor theorem as the foundation of the homotopy analysis method , _ commun . nonlinear sci . simulat . _ * 16 * ( 2011 ) 12541262 . magnus , j. r. , neudecker , h .. matrix differential calculus with applications in statistics and econometrics , 3rd edition . john wiley and & sons ltd , ( 2007 ) . optimization toolbox user s guide , 1990 - 2012 by the mathworks , inc . global optimization toolbox user s guide , 2004 - 2012 by the mathworks , inc . parallel computing toolbox user s guide , 2004 - 2012 by the mathworks , inc . molabahrami , a. , khani , f. , the homotopy analysis method to solve the burgers - huxley equation , _ nonlinear analysis : real world applications _ * 10 * ( 2009 ) 589600 . obizhaeva , a. a. , wang , j. , optimal trading strategy and supply / demand dynamics , _ journal of financial markets _ * 16 * ( 2013 ) 132 . polyanin , a. d. , manzhirov , a. , handbook of integral equations , second edition . chapman & hall / crc(2008 ) . predoiu , s. , shaikhet , g. , shreve , s , optimal execution in a general one - sided limit - order book , _ siam journal on finance mathematics _ * 2 * ( 2011 ) 183212 . press , w.h . , teukolsky , s.a . , vetterling , w.t . , flannery , b.p . , numerical recipes 3rd edition : the art of scientific computing , cambridge university press , 3rd edition ( 2007 ) . snchez , d. a. , calculus of variations for integrals depending on a convolution product , _ annali della scuola normale superiore di pisa , classe di scienze _ ( 3 ) * 18 * ( 1964 ) 233254 . snchez , d. a. , some existence theorems in the calculus of variations , _ pacific journal of mathematics _ * 19 * ( 1966 ) 357363 . snchez , d. a. , on extremals for integrals depending on a convolution product , _ journal of mathematical analysis and applications _ * 11 * ( 1965 ) 213216 . turkyilmazoglu , m. , a note on the homotopy analysis method , _ applied mathematics letters _ * 23 * ( 2010 ) 12261230 . verma , a. , schug , a. , lee , k. h. , wenzel , w. , basin hopping simulations for all - atom protein folding , _ the journal of chemical physics _ * 124 * ( 2006 ) 044515 . wang , c. , pop , i. , analysis of the flow of a power law fluid film on an unsteady stretching surface by means of the homotopy analysis method , _ j. non - newtonian fluid mech . _ * 138 * ( 2006 ) 161172 . waterfall , j. j. , casey , f. p. , gutenkunst , r. n. , brown , k. s. , myers , c. r. , brouwer , p. w. , elser , v. , and sethna , j. p. , sloppy - model universality class and the vandermonde matrix , _ physical review letters _ * 97 * ( 2006 ) 150601 - 1150601 - 4 . wazwaz a .- m . , linear and nonlinear integral equations , methods and applications . springer - verlag ( 2011 ) . weinberger , e. , correlated and uncorrelated fitness landscapes and how to tell the difference , _ biol . cybern . _ * 63 * ( 1990 ) 325 - 336 . weinberger , e. d. , local properties of kauffman s @xmath85-@xmath276 model : a tunably rugged energy landscape , _ physical review a _ * 44 * ( 1991 ) 6399 - 6413 . kyle , a. s. , continuous auctions and insider trading , _ econometrica _ ( 1985 ) 13151335 . zarinelli , e. , treccani , m. , farmer , j. d. , and lillo , f. , beyond the square root : evidence for logarithmic dependence of market impact on size and participation rate , http://arxiv.org/abs/1412.2152v1 ( 2014 ) .
ratchet effects can arise for single or collectively interacting brownian particles on an asymmetric substrate when a net dc transport is produced by an externally applied ac driving force or by periodically flashing the substrate . recently , a new class of active ratchet systems has been realized through the use of active matter , which are self - propelled units that can be biological or non - biological in nature . when active materials such as swimming bacteria interact with an asymmetric substrate , a net dc directed motion can arise even without external driving , opening a wealth of possibilities such as sorting , cargo transport , or micromachine construction . we review the current status of active matter ratchets for swimming bacteria , cells , active colloids , and swarming models , focusing on the role of particle - substrate interactions . we describe ratchet reversals produced by collective effects and the use of active ratchets to transport passive particles . we discuss future directions including deformable substrates or particles , the role of different swimming modes , varied particle - particle interactions , and non - dissipative effects . ratchet , active matter , nonequilibrium transport , flocking
1604.01072
when a particle is placed in a symmetric periodic potential and subjected to an ac drive , in general the particle simply rocks back and forth and undergoes no net dc motion . if an external dc drive @xmath0 is imposed on a particle in a substrate , the particle has no net dc motion and remains pinned as long as @xmath1 , where @xmath2 is the critical depinning force for motion over the substrate . once @xmath3 , the particle enters a sliding state @xcite . when the motion of the particle is overdamped , its dynamics can be described by the following equation of motion : @xmath4 where @xmath5 is the damping constant , @xmath6 is the force from the substrate , @xmath7 is the interaction force from the other particles , @xmath8 is an externally applied driving force , and @xmath9 is a stochastic force representing thermal fluctuations . the situation changes if the substrate is asymmetric . an example of a one - dimensional ( 1d ) asymmetric substrate potential with periodicity @xmath10 , @xmath11 , \ ] ] is illustrated in fig . [ fig:1](a ) . in the absence of thermal fluctuations , the maximum pinning force exerted by the substrate on the particle for motion in the positive @xmath12-direction is @xmath13 , while the maximum pinning force for motion in the negative @xmath12-direction has a smaller value of @xmath14 . as a result , if an externally applied ac driving force of the form @xmath15 is introduced , then the particle remains trapped and oscillates within a single potential minimum for @xmath16 . in contrast , when @xmath17 , the particle can overcome the substrate potential barrier for motion in the negative @xmath12-direction but can not overcome the barrier for motion in the positive @xmath12-direction , so that during each ac cycle the particle undergoes a net dc transport in the negative @xmath12-direction . such motion is termed a ratchet effect . the amount of dc translation that occurs per cycle depends on the amplitude and frequency of the ac drive . for fixed ac frequency , the efficiency of the ratchet is generally nonmonotonic , increasing with increasing @xmath18 for @xmath19 and then decreasing with increasing @xmath18 at higher @xmath18 @xcite . ( b ) a quasi-1d version of the same substrate for particles moving in two dimensions . adapted from c. reichhardt , d. ray , and c.j . olson reichhardt , new j. phys . 17 , 070304 ( 2015 ) . ( c ) a line of 2d funnels or barriers ( yellow ) affecting the motion of interacting particles ( green ) . the easy direction for the ratchet effect is the positive @xmath20 direction , toward the top of the panel . , title="fig:",width=384 ] direction . ( b ) a quasi-1d version of the same substrate for particles moving in two dimensions . adapted from c. reichhardt , d. ray , and c.j . olson reichhardt , new j. phys . 17 , 070304 ( 2015 ) . ( c ) a line of 2d funnels or barriers ( yellow ) affecting the motion of interacting particles ( green ) . the easy direction for the ratchet effect is the positive @xmath20 direction , toward the top of the panel . , title="fig:",width=192 ] when the mechanism responsible for producing a ratchet effect is an ac driving force , the system is known as a rocking ratchet . an alternative mechanism operates in a flashing ratchet , where thermally diffusing particles are placed on a substrate that is periodically flashed on and off , and where a maximum ratchet efficiency emerges as a function of flashing frequency @xcite . in the single particle limit for overdamped particles , the ratchet effect causes the particle to translate along the easy direction of the substrate regardless of the nature of the ratchet . when collective particle - particle interactions are introduced , however , it is possible to observe reversals of the ratchet effect , where the motion is along the hard direction of the substrate asymmetry @xcite . these reversals are often produced by commensurability effects between the interacting particles and the substrate , where emergent quasiparticle objects such as kinks or antikinks dominate the transport and behave as if they are moving through an inverted substrate potential . examples of higher - dimensional asymmetric substrates on which ratchet effects can occur appear in fig . [ fig:1](b ) , which shows a quasi-1d asymmetric potential with no structure transverse to the asymmetry direction @xcite , and in fig . [ fig:1](c ) , which shows a linear array of v - shaped barriers @xcite . the asymmetry necessary for generating ratchet effects can also be introduced by placing individual symmetric pinning sites in an array that has a density gradient along one direction @xcite . in geometric or drift ratchets , which require a minimum of two spatial dimensions , particles driven over an asymmetric substrate by a dc drive undergo a net dc drift in the direction transverse to the applied dc drive @xcite . in transverse ratchets , which are a variation of drift ratchets , a net dc ratchet motion arises in the direction transverse to an applied ac driving force @xcite . additional ratchet effects including ratchet reversals can arise when there are non - dissipative terms in the equations of particle motion , such as inertia @xcite or a magnus term @xcite . studies of ratchet effects have been performed for colloids interacting with asymmetric flashing substrates @xcite , vortices in type - ii superconductors interacting with asymmetric periodic 1d @xcite and 2d substrates @xcite , granular matter on sawtooth substrates @xcite , cold atoms @xcite , electron systems @xcite , and domain walls moving over asymmetric substrates @xcite . these systems all require application of some form of external driving in order to create the nonequilibrium conditions necessary for a ratchet effect to occur . in a different class of nonequilibrium systems , the dynamics is governed by internal driving or self - propulsion @xcite . these active matter systems include flocking and swarming particles , pedestrian and traffic flow , swimming bacteria , crawling cells , and actin filaments . active matter has been attracting growing attention due to the increasing availability of non - biological active systems such as artificial swimmers , self - driven colloids , robot swarms , and autonomous agents . such systems exhibit a variety of phenomena including pattern formation @xcite , self - clustering @xcite , and nonequilibrium phase transitions @xcite , and represent a rich and very active field of research . one of the first demonstrations of an active matter ratchet was achieved by p. galajda _ @xcite for a system of _ e. coli _ swimming in a sample divided into two chambers by an array of funnel - shaped barriers similar to those illustrated in fig . [ fig:1](c ) . the entire sample is 400@xmath21 m on a side , while the funnel barriers are each 27@xmath21 m long , form a @xmath22 angle at their tips , and have a minimum separation of 3.8@xmath21 m between adjacent funnels . if brownian particles are placed at a uniform density in such a geometry , the density remains equal in both chambers of the sample . in contrast , when an equal number of swimming bacteria are introduced into each chamber , after roughly an hour there is a buildup of bacteria in the chamber that is on the easy - flow side of the funnel barriers , as illustrated in fig . [ fig:2](a , b ) . in the figure , the easy - flow direction is to the right . in fig . [ fig:2](c ) , the ratio @xmath23 of the bacterial concentration @xmath24 in the right and left chambers , @xmath25 , has an initial value of @xmath26 at time @xmath27 and saturates to a value of @xmath28 at longer times , showing that a ratchet effect is occurring . when the same experiment is performed with chambers separated by symmetric holes instead of asymmetric funnels , then @xmath29 at all times , indicating that the asymmetry of the funnel is a crucial ingredient for the occurrence of rectification . when a mixture of swimming and non - swimming bacteria is placed in the sample , the swimming bacteria are rectified while the non - swimming bacteria are not , showing that swimming is also important in producing the ratchet effect . due to the complexity of bacterial systems , it was not obvious whether this initial observation of an active matter ratchet effect could be understood as being produced solely by the dynamics of the self - propulsion combined with bacteria - wall interactions , or whether more complex bacterial behaviors play a role in the ratcheting motion . ( c ) . the easy - flow direction through the barriers is toward the right chamber . ( a ) the initial uniform distribution of bacteria at time @xmath27 . ( b ) the distribution at @xmath30 minutes showing that the bacteria concentration is significantly higher in the right chamber . ( c ) the ratio @xmath31 of the bacteria density @xmath24 in the right and left chambers versus time , showing that a rectification effect is occurring . reprinted with permission from p. galajda , j. keymer , p. chaikin , and r. austin , j. bacteriol . 189 , 8704 ( 2007 ) . copyright 2007 by the american society for microbiology . , width=288 ] wan _ et al . _ @xcite subsequently studied a simple model of pointlike run - and - tumble particles moving in a 2d container with the same funnel barrier geometry used in the experiments of ref . @xcite , as illustrated in fig . [ fig:3 ] . in this model , particle @xmath32 obeys the following overdamped equation of motion : @xmath33 where @xmath5 is the damping coefficient and @xmath34 is the location of particle @xmath32 . instead of the external driving force used in eqn . 3 contains a motor force @xmath35 representing self - propulsion in a randomly - chosen direction that remains constant during a running time @xmath36 . every @xmath36 simulation time steps , particle @xmath32 undergoes an instantaneous tumbling event during which the motor force is reoriented into a new random direction . the particle then runs in this new direction during the next time period @xmath36 before tumbling again . the running speed @xmath37 is held constant . for an isolated particle in the absence of any barriers , the motor force produces a motion that is ballistic at short times and diffusive at very long times @xcite . if the particle does not collide with any other objects , then during a single running time it translates by one run length @xmath38 , which is a distance of @xmath39 . the stochastic thermal force @xmath40 has the properties @xmath41 and @xmath42 . if @xmath43 so that the particles experience only thermal fluctuations but no motor force , then no rectification occurs . @xmath44 represents particle - particle interactions , while @xmath45 is the particle - barrier interaction term . the barriers and confining walls are modeled as exerting a short - range steric repulsion on particle @xmath32 , and a particle interacting with a barrier or wall runs along the wall at a speed determined by the component of @xmath46 that is parallel to the wall until the particle reaches the end of the wall or undergoes a tumbling event that rotates its running direction away from the wall . between tumbles . ( b ) after 100 tumbling events , the particle density in the upper chamber has increased . ( c ) ratios @xmath47 ( upper red line ) and @xmath48 ( lower blue line ) of the particle density @xmath49 in the upper chamber and @xmath50 in the lower chamber to the initial particle density @xmath51 as a function of time . solid lines are for a sample with a run length of @xmath52 where there is a large rectification effect , and dashed lines are for a sample with a run length of @xmath53 where the rectification effect is negligible . adapted from m. b. wan , c. j. olson reichhardt , z. nussinov , and c. reichhardt , phys . rev . lett . * 101 * , 018102 ( 2008 ) . copyright 2008 by the american physical society . , width=288 ] a simulated system in which particle - particle interactions and thermal fluctuations are both neglected , so that the particles experience only a motor force propulsion and interactions with the barriers and walls , is illustrated in fig . [ fig:3](a , b ) . the sample is of size @xmath54 on a side , where @xmath55 is the unit of distance in the simulation , and contains 12 funnel barriers that each have a tip angle of @xmath22 and an arm length of @xmath56 . the running length is @xmath57 , so a particle can move much further than the length of a funnel arm during a single run time . the particle density is initially uniform , as shown in fig . [ fig:3](a ) , but after a time @xmath58 , fig . [ fig:3](b ) shows that the particle density has increased in the upper chamber . the ratio of the particle density @xmath49 in the upper chamber and @xmath50 in the lower chamber to the initial particle density @xmath51 is plotted as a function of simulation time steps in fig . [ fig:3](c ) . the solid lines show that in a system with @xmath59 , the density in the upper chamber increases over time with a simultaneous decrease in the lower chamber density until the sample reaches a steady state . the dashed lines indicate that when the running length is very short , @xmath60 , the density in both chambers remains nearly constant at a value equal to the initial particle density . when @xmath38 is very small , the particle motion is brownian - like , so no rectification can occur . the steady - state ratio @xmath61 is plotted as a function of @xmath62 in fig . [ fig:4 ] . for small @xmath38 , @xmath63 , while for large @xmath38 , @xmath64 reaches a value of nearly @xmath65 . the inset of fig . [ fig:4 ] illustrates the ratchet mechanism . the left inset shows that under the wall - following rule of motion , when @xmath38 is sufficiently large , a particle approaching the funnel barrier from the top chamber becomes trapped at the funnel tip until a tumbling event allows it to escape back into the top chamber , while a particle approaching the funnel barrier from below is shunted along the barrier wall into the top chamber . in the right inset , a particle with a very short run time spends very little time interacting with the wall , similar to a purely brownian particle , and is unable to sample the barrier asymmetry over a time scale long enough to bias its motion toward the upper chamber . the two ingredients that are required to produce a ratchet effect are the running of particles along the walls and the breaking of detailed balance by the motor force . in the same simulation study , the rectification effect is reduced when a finite temperature is added since this increases the chance that a particle will prematurely move away from a wall . the ratchet effect is also reduced by the inclusion of steric particle - particle interactions , since particle collisions effectively reduce the running length while a buildup of particles in the funnel tip reduces the ability of the funnels to trap particles approaching from the upper chamber . of the particle density in the upper chamber to that in the lower chamber for the system in fig . [ fig:3 ] at steady state vs run length @xmath38 , showing that for small run lengths @xmath66 indicating no rectification , while for large @xmath62 the ratio @xmath64 reaches a value of nearly @xmath65 . inset : schematic illustration of the rectification mechanism . on the left , an individual particle approaching the funnel barrier from above becomes trapped at the funnel tip , while a particle approaching from below is shunted by the funnel wall into the upper chamber . on the right , a brownian particle spends very little time in contact with the funnel walls . adapted from m. b. wan , c. j. olson reichhardt , z. nussinov , and c. reichhardt , phys . * 101 * , 018102 ( 2008 ) . copyright 2008 by the american physical society . , width=288 ] figure [ fig:5](a ) shows a snapshot from a simulation of run - and - tumble particles with steric particle - particle interactions moving between two chambers separated by funnel barriers . when the run length is sufficiently large @xcite , the inclusion of steric interactions produces two additional features : a build up of particles along the walls , and trapping of particles in the funnel tips . both of these effects were observed in the experiments of ref . one question is whether the accumulation of particles along the walls in the experiments is produced by hydrodynamic effects or whether it arises purely due to the self - propulsion of the bacteria . using numerical and analytical methods , tailleur and cates @xcite investigated run - and - tumble particles confined by a spherical trap and found that the dynamics alone produce particle accumulation on the walls . other studies of confined active matter where the motor force rotates diffusively also show a build up of particle density along the walls , and indicate that this effect is enhanced near corners @xcite . the tendency of active matter particles to accumulate in corners is one of the reasons the funnel barriers can produce a ratchet effect , as the increased trapping occurs on only one side of the funnels due to their curvature . the work in ref . @xcite also shows how the nature of the particle - wall interactions leads to rectification . rectification by the funnel geometry is robust against changes in the running length distribution , as is observable by comparing the results of wan _ @xcite , who used only a single running length in each simulation , and of tailleur and cates @xcite , who considered a poisson distribution of running lengths . changes in the nature of the particle - wall interactions , however , can destroy the rectification . four possible interaction rules are shown schematically in fig . [ fig:5](b - e ) . under rule i , the motor force alignment is unchanged by contact with the wall and the particle moves according to the component of the motor force vector that is parallel to the wall . in rule ii , the motor force is realigned to be parallel with the wall upon contact . for rule iii , the particle is reflected by the wall , while in rule iv , the particle collides elastically with the wall . in ref . @xcite , when the motor force realigning rule ii is employed , efficient rectification of the particles occurs ; however , if the elastic collisions of rule iv are substituted , the rectification is lost , as shown in fig . [ fig:5b ] . an examination of all four wall rules in ref . @xcite shows that the running of the particle along the wall , as in rules i and ii , is essential for producing a ratchet effect . when the particle does not remain in contact with the wall , as in rules iii and iv , the rectification is lost . . ( b - e ) different possible particle - wall interaction rules . ( b ) rule i : the particle maintains its original running direction and moves along the wall with the component of its velocity that is parallel to the wall . ( c ) rule ii : the particle realigns its running direction to be parallel with the wall . ( d ) rule iii : the particle reflects from the wall . ( e ) rule iv : the particle undergoes an elastic collision with the wall . ratchet effects occur for rules i and ii , with the strongest ratcheting for rule ii , but rules iii and iv produce no ratcheting . adapted from c.j . olson reichhardt , j. drocco , t. mai , m.b . wan , and c. reichhardt , proc . spie * 8097 * , 80970a ( 2011 ) . , width=288 ] for the particle density in chambers separated by a line of funnel barriers for run - and - tumble particles that align their running direction with any wall they contact , showing increasing amounts of rectification for increasing particle running speed @xmath37 . the purple curve with @xmath67 is for the nonrectifying case where the particles collide elastically with the walls . reprinted with permission from j. tailleur and m. e. cates , europhys . 86 * , 60002 ( 2009 ) . , width=288 ] followup experiments by galajda _ et al . _ @xcite using self - propelled large scale swimming bathtub toys produced a rectification effect similar to that observed for the bacteria , while experiments by hulme _ et al . _ @xcite demonstrate how the ratchet effect can be used to fractionate motile _ e. coli _ using arrays of asymmetric chambers of the type illustrated in fig . [ fig:6](a ) . the arrows in panels 1 to 5 of fig . [ fig:6](b ) show how bacteria moving in the positive @xmath12-direction are guided through the chamber array , while in panels 6 to 10 , bacteria moving in the negative @xmath12 direction are blocked by the chamber geometry . variations on similar channel geometries have also been shown to produce guided motion of bacteria @xcite . use similar microfabricated ratchet channels with asymmetry to concentrate _ e. coli _ at particular locations @xcite . numerous other shapes can produce rectification of bacteria , such as the multiple rows of u - shaped funnel barriers illustrated in fig . [ fig:7](a ) that produce rectification of particles with a motor force that rotates diffusively @xcite . in fig . [ fig:7](b ) , a plot of the rectification ratio @xmath68 , given by the ratio of the density in the right chamber to that in the left chamber , versus the motor force @xmath69 for one , two , and three rows of barriers shows that @xmath68 increases linearly with @xmath69 , and that there is a strong enhancement of the ratchet effect with the addition of more layers of funnels . in the same study , an examination of barriers made from open and closed triangles , u shapes , and half boxes indicates that barriers with an open shape always produce a larger rectification than barriers with closed shapes due to the trapping of particles within the tips of the open barriers . , equal to the ratio of the density in the right chamber to the density in the left chamber , vs motor force @xmath70 for samples containing one ( squares ) , two ( circles ) , and three ( triangles ) layers of funnel barriers . reproduced from y .- f . chen , s. xiao , h .- y . chen , y .- j . sheng , and h .- k . tsao , nanoscale * 7 * , 16451 ( 2015 ) , with permission of the royal society of chemistry . http://dx.doi.org/10.1039/c5nr04124d , width=288 ] theoretical and numerical studies show that rectification effects are a general phenomenon that occurs for self - driven particles in the presence of asymmetric piecewise periodic potentials @xcite . in other numerical studies , the directed transport that occurs for self - propelled particles interacting with arrays of nonsymmetric convex obstacles without cavities was shown to result from the fact that the particles tend to attach to solid surfaces @xcite . in numerical studies of the motion of active particles with a diffusively rotating motor force through funnel arrays , i. berdakin _ _ examined the effect of changing the swimming properties by varying the distribution of run lengths , the magnitude of the rotational diffusion , and the preservation of run orientation memory after a tumbling event @xcite . they observe that there are optimal swimming properties that maximize the rectification as a function of the geometry of the funnels . an illustration of the simulated funnel geometry appears in fig . [ fig:8](a ) along with colored dots indicating the position of three different _ e. coli _ mutants modeled on the basis of characterizations performed by berg and brown @xcite . red particles , or species s1 , represent chec497 with a mean run duration of @xmath71 seconds and a speed of 20 @xmath21 m/s ; blue particles , or species s3 , represent the wild type aw405 with a mean run duration of 0.86 seconds and a speed of 14.2 @xmath21 m/s ; and green particles , or species s4 , represent un602 which is the slowest with a mean run duration of 0.42 seconds and a speed of 14.2 @xmath21 m/s . the particles are initially placed along the leftmost wall , and after 12 minutes the red particles , which have the longest run time and the highest speed , have moved the furthest to the right , followed by the blue particles , which move less far to the right , and finally by the green particles , which have the least motion to the right . figure [ fig:8](b ) shows the number of bacteria as a function of time in the leftmost chamber ( decreasing curves ) and rightmost chamber ( increasing curves ) , indicating that the different species become spatially separated over time . in follow - up studies , the same group observed that the use of multiple arrays of funnels makes it possible to sort species that have only a small variation in a single swimming parameter @xcite . ( b ) the number of bacteria vs time in the leftmost chamber ( solid symbols ) and the rightmost chamber ( open symbols ) , showing that the different bacteria species can be fractionated using the funnel arrays . reprinted with permission from i. berdakin , y. jeyaram , v. v. moshchalkov , l. venken , s. dierckx , s. j. vanderleyden , a. v. silhanek , c. a. condat , and v. i. marconi , phys . e * 87 * , 052702 ( 2013 ) . copyright 2013 by the american physical society . , title="fig:",width=288 ] + ( b ) . ( b ) the number of bacteria vs time in the leftmost chamber ( solid symbols ) and the rightmost chamber ( open symbols ) , showing that the different bacteria species can be fractionated using the funnel arrays . reprinted with permission from i. berdakin , y. jeyaram , v. v. moshchalkov , l. venken , s. dierckx , s. j. vanderleyden , a. v. silhanek , c. a. condat , and v. i. marconi , phys . e * 87 * , 052702 ( 2013 ) . copyright 2013 by the american physical society . , title="fig:",width=288 ] the motion of biological cells through funnel geometries has been explored in several experiments . @xcite considered swimming _ chlamydomonas _ in both simulations and experiments using four chambers separated by funnel barriers , and measured the rectification ratio @xmath72 , where @xmath73 is the steady state number of cells in chamber @xmath74 and @xmath75 is the steady state number of cells in chamber @xmath76 . for funnels with a tip angle of 35@xmath77 , @xmath78 , indicating strong rectification . in ref . @xcite , experiments on human sperm cells in quasi-2d microchambers containing funnel barriers show that rectification occurs when the cells become trapped in the funnel tips . the same study showed that since the sperm are attracted to the boundaries , the ratchet effect can be optimized by using u - shaped barriers . experiments with _ c. elegans _ moving through funnel barriers showed that the wild type , which moves along walls that it contacts , can be rectified by the barriers @xcite . in contrast , mutant strains that lack touch sensory neurons in their body reverse motion upon contacting a wall instead of following the wall , and these mutants are not rectified by the barriers . a variety of other ratchet geometries for active particles have been studied , including 2d corrugated asymmetric substrates @xcite and 2d geometries where entropic effects are important @xcite . other works focus on the effect of changing the active particle shape to swimming ellipses @xcite , or on including an additional external drift force to the active particles to produce an active drift ratchet @xcite . proposed that ratchet effects can occur in a variety of plants where the asymmetry of the roots , seeds or leaves is important for transport in the natural world where fluctuations may be present @xcite . finally , in the case of symmetric substrates , it has been proposed that by periodically modulating the velocity of the active particles and then phase - shifting this velocity with respect to the substrate periodicity , enough symmetries are broken to permit a ratchet effect to occur @xcite . in ratchets produced using funnel barriers , it would be interesting to examine additional modes of motion of the active particles such as particles that avoid walls , particles that reverse their motion if the density is too high , or particles that change swimming strategies as a function of local density or of time . many of the active matter ratchet studies performed so far employ infinitely repulsive barriers or walls ; however , many previous studies of nonactive matter ratchet effects produced by external driving were performed for substrates containing pinning sites with a finite strength , so that a particle experiencing a sufficiently large driving force can pass through the pinning potential or barrier . it would be interesting to examine active matter particles moving through asymmetric finite - strength pinning arrays rather than obstacle arrays . pinning arrays for colloidal particles can be readily created experimentally using optical means , so it should be possible to create optically generated pinning arrays for active colloidal particles or swimming bacteria . additionally , optical traps introduce the possibility of creating flashing substrates , which could significantly enhance active matter ratchet effects . it was recently proposed that by using a light intensity pattern with a funnel or chevron shape to manipulate a dense assembly of sterically interacting active particles that are attracted by light , the patterned regions would become filled with trapped active particles that would then serve as an asymmetric substrate that can produce ratcheting motion of the remaining untrapped active particles @xcite . there has also been work examining how asymmetry in the curvature of the walls can produce different pressures on each side of the wall , producing a ratchet effect @xcite . this suggests that curvature is another route to explore in producing different types of active ratchet devices . other possible systems to explore include crawling insects or animals . studies that reveal boundary following behaviors in animals @xcite suggest that asymmetry in the boundaries could be used to produce ratchet effects . another avenue of study is the introduction of feedback effects to a ratchet system , which could be achieved through optical methods . recent studies of active matter using feedback controls have produced a variety of behavior for very simple feedback rules @xcite . in rocking and flashing ratchet systems , a rich variety of collective effects including ratchet reversals arise when particle - particle interactions are included @xcite . for run - and - tumble active particles moving through funnel barriers , _ observed that the addition of steric particle - particle interactions produced only a monotonic reduction in the rectification effect @xcite . @xcite placed particles obeying a vicsek model , modified to include particle - barrier interactions and steric particle - particle interactions with an exclusion radius @xmath79 , in a funnel barrier geometry . for small @xmath80 , the particles ratchet in the easy flow direction , and the magnitude of the ratchet effect depends on the noise term @xmath5 in the vicsek model . for small @xmath5 the system forms flocks , while for large @xmath5 the system is in a liquidlike state . in the funnel barrier system , the ratchet effect is strongly reduced or completely lost for large @xmath5 , while for small @xmath5 where flocking occurs , a pronounced ratchet effect appears with density ratios in the two chambers approaching @xmath65 . figure [ fig:9](b , c ) highlights the enhancement of the ratchet effect due to the collective rectification of groups of active particles . in fig . [ fig:9](b ) , a flock approaches the funnel barriers from below . the flock becomes compressed and moves through the barrier array as a unit , as illustrated in fig . [ fig:9](c ) . a reversal of the ratchet effect , in which the particles collect in the bottom chamber in the hard flow direction of the funnel barriers , can occur when flocks are present . figure [ fig:9](a ) shows a plot of the particle density in the top chamber @xmath81 versus the particle exclusion radius @xmath79 for a system in which @xmath82 , the flocking radius is 1.0 , the starting uniform density is @xmath83 , and the closest spacing between adjacent funnel barriers is @xmath84 . for @xmath85 , rectification occurs in the easy flow direction of the funnel barriers as illustrated in fig . [ fig:9](b , c ) , but there is a reversal of the rectification into the hard flow direction for @xmath86 . for @xmath87 , the diameter of the particles is larger than the closest spacing between funnels , @xmath88 , so it is no longer possible for particles to pass between the upper and lower chambers , and the rectification effect is lost . the ratchet reversal is produced by a jamming effect that occurs during the compression of a flock of particles as it approaches the opening between adjacent funnel barriers from below . the compression causes the particles at the leading edge of the flock to form a rigid close - packed structure that is wider than the interfunnel spacing @xmath89 , and makes it impossible for the flock to cross the line of funnel barriers . in contrast , when a flock approaches the funnel barriers from above , the barriers cleave the flock into smaller clusters , and during this process one or two particles can escape from the flock and pass between the funnel barriers to enter the lower chamber . this reversed motion process always occurs but can not compete with the forward motion of flocks of particles moving into the top chamber for small @xmath79 . when the forward motion is inhibited by jamming at larger @xmath79 , the reversed motion dominates and a reversed ratchet effect appears . the reverse ratchet effect operates much more slowly than the forward ratchet , so a much longer time is required for the system to reach a steady state in the reverse ratchet regime than in the forward ratchet regime . if two species of particles with different exclusion radii of @xmath90 ( small ) and @xmath91 ( large ) are placed in the container , flocks of the different species ratchet in opposite directions and can be spatially separated , as shown in fig . [ fig:10 ] . of particles in the upper chamber vs the particle exclusion radius @xmath79 . the dashed line indicates the uniform starting density with @xmath92 . the ratchet effect in the easy flow direction is maximized for small @xmath79 . a reversal in the ratchet direction occurs for @xmath93 , and the ratchet effect is lost for @xmath94 when the particles become too large to fit between adjacent funnels . ( b , c ) images of a small portion of the sample illustrating the forward rectification mechanism . ( b ) a flock approaches the funnel barriers from below . ( c ) the flock elongates and passes through the barrier array as a unit . adapted from j.a . drocco , c.j . olson reichhardt , and c. reichhardt , phys . e * 85 * , 056102 ( 2012 ) . copyright 2012 by the american physical society . , title="fig:",width=288 ] of particles in the upper chamber vs the particle exclusion radius @xmath79 . the dashed line indicates the uniform starting density with @xmath92 . the ratchet effect in the easy flow direction is maximized for small @xmath79 . a reversal in the ratchet direction occurs for @xmath93 , and the ratchet effect is lost for @xmath94 when the particles become too large to fit between adjacent funnels . ( b , c ) images of a small portion of the sample illustrating the forward rectification mechanism . ( b ) a flock approaches the funnel barriers from below . ( c ) the flock elongates and passes through the barrier array as a unit . adapted from j.a . drocco , c.j . olson reichhardt , and c. reichhardt , phys . e * 85 * , 056102 ( 2012 ) . copyright 2012 by the american physical society . , title="fig:",width=288 ] ( small , blue ) and @xmath95 ( large , red ) . ( a ) the initial configuration with uniform density in both top and bottom chambers , which are separated by a line of funnel barriers . ( b ) a representative steady state configuration at later time . the larger red particles undergo reversed rectification and collect in the bottom chamber , while the smaller blue particles undergo forward rectification and collect in the top chamber . , width=192 ] ( small , blue ) and @xmath95 ( large , red ) . ( a ) the initial configuration with uniform density in both top and bottom chambers , which are separated by a line of funnel barriers . ( b ) a representative steady state configuration at later time . the larger red particles undergo reversed rectification and collect in the bottom chamber , while the smaller blue particles undergo forward rectification and collect in the top chamber . , width=192 ] ratchet reversals have also been observed for eukaryotic cells moving on asymmetric substrates . @xcite examined cell motility on micropatterned ratchet surfaces . they considered cancerous b16f1 cells , which are a mouse epithelial - like melanoma cell , non - cancerous rat2 fibroblast cells , and cancerous mda - mb-231 human mammary gland cells . for the b16f1 and rat2 cells , directed motion occurs along the easy flow direction of the funnel array . the addition of a protruding spike shape to the funnel shaped chambers does not affect the rat2 cells , which still ratchet in the easy direction , but causes the b16f1 and mda - mb-231 cells to perform a reversed ratchet motion in the hard flow direction . the differences in the direction of the ratchet motion for the different cells were attributed to the different crawling mechanisms of the cells . rat2 cells use long protrusions that can grab the funnel edges or the ends of the spike shapes , while the other cells prefer to spread out in the open spaces as they move . et al . _ also found that it is possible to use arrays of the spike shaped funnels to separate a mixture of different cells over time , and they propose the creation of a bidirectional ratchet , opening the possibility of creating cell traps and ratchets to achieve various types of medical treatments . other studies have shown that unidirectional cell motion can be achieved using asymmetric nanotopography @xcite . self - propelled janus colloids , whose direction of motion slowly diffuses rotationally , represent another type of experimental artificial active matter system . @xcite performed simulations of active janus particles in an asymmetric channel , and observed that rectification occurs in the easy flow direction of the channel and that the magnitude of the ratchet effect increases with increasing correlation time of the particle running direction , eventually saturating for large correlation times . they also considered a mixture of @xmath96 active particles and @xmath97 non - active particles in the same geometry , with the particle - particle interactions modeled as a repulsive harmonic force , and found that the active particles can induce a net rectification of the non - active particles . a particularly interesting effect is that the rectification efficiency for the non - active particles is non - monotonic as a function of particle density . for one choice of parameters , the maximum non - active particle ratchet efficiency occurs for a ratio of @xmath98 , indicating that even a very small fraction of active particles can efficiently rectify the passive particles . in the dense limit of @xmath99 close to the jamming regime , ghosh _ et al . _ find a small reversal of the ratchet effect . in experiments with passive colloids immersed in a bath of swimming bacteria , koumakis _ et al . _ showed that introducing an asymmetric substrate that causes the bacteria to undergo rectification also produces transport of the passive colloids into or out of enclosed regions @xcite . figure [ fig:11](a ) shows the time evolution of the number of passive colloids in each chamber of a sample containing three layers of ratchets for a geometry that concentrates the colloids in the central chamber , while fig . [ fig:11](b ) shows the same quantity for a geometry that ejects the colloids from the box . et al . _ also found that curved ratchet geometries were not as efficient as flat boundaries in their device since the colloids would follow curved trajectories over the curved surfaces . these results show how collective effects in active ratchets could be used to achieve a variety of active pumping techniques . it may also be possible that biological systems such as cells already take advantage of such collective ratchet effects . the field of collective effects in active matter systems is wide open . since a wide variety of flocking models exist , it would be interesting to determine what types of ratchet effects occur for different flocking rules . this would also be an interesting question to examine in experiments on schools of fish or swarming insects which could be placed in funnel like channels , with a comparison between ratchet behaviors for schooling and non - schooling systems . similar experiments could be performed with herding animals . it would also be interesting to explore systems with competing effects , such as a ratchet geometry that biases the motion of the system in one direction combined with some other competing force that biases the motion of the system in the opposite direction . in such a case , flocking or collective effects could be turned on or off to take advantage of or nullify the ratchet effect . there has already been some evidence for competing interactions that can overwhelm the ratchet effect , such as an experiment on bacteria in a funnel barrier ratchet geometry , where the bacteria form fisher wave swarms that move against the ratchet direction @xcite . self - driven colloids have been shown to exhibit a phase separation transition into gaslike and clustered solidlike phases for high enough density or activity @xcite , so it would be interesting to examine how the ratchet efficiency is affected by the onset of self - clustering . introduction of a substrate can modify the density or activity at which the self - clustering begins to occur , as well as the mean distance that individual particles can move @xcite . collective ratchet effects should also depend strongly on the type of interactions between the particles . in the systems studied so far , short - range steric repulsion has been used for the particle - particle interactions ; however , it is also possible to consider active particles with lennard - jones interactions , long - range repulsive interactions , competing long - range repulsion and short - range attraction , or many body interactions . another class of system is active membranes or sheets that could mimic cells or collections of tightly bound organisms . swimming deformable droplets are one example of an extended active object of this type @xcite , and if such droplets were placed on an asymmetric surface , ratchet effects could occur . in this case , varying the substrate periodicity to be larger or smaller than the size scale of the active object could cause changes in the ratchet effect . there has already been a study demonstrating ratchet effects for active polymers in funnel barrier arrays @xcite . other motion rules could also be introduced to model objects that have mechanics similar to cells , such as protrusions that push or pull to mimic the experiments in ref . there could also be ratchet effects for extended active objects that can expand or contract , such as a squirmer model or a system with inertial dynamics . collective ratchet effects can also be examined for groups of interacting rods coupled by molecular motors when an additional asymmetry is introduced , or active gel systems such as active nematics @xcite could be studied in the presence of an asymmetric substrate . active nematics can also exhibit mobile defects @xcite , so it may be possible to produce ratcheting motion of the defects . there are many examples of active biological and artificial systems that exhibit a chirality in their motion , such as biological circle swimmers @xcite and artificial circle swimmers @xcite . in this case , ratchet effects can occur even in the absence of a substrate asymmetry since the chirality of the swimming introduces an asymmetry to the system . one of the first examples of a ratchet effect produced by chirality was observed for circularly moving particles on a 2d periodic substrate @xcite . as the substrate - free particle orbits are varied from circular to more complex lissajous patterns , a series of transitions occurs between localized closed trajectories with no net translation to open trajectories that permit the particles to translate in a particular direction , producing a ratchet effect . additional simulations have shown that directed motion can be achieved for circularly moving particles on periodic substrates , where various types of translating trajectories appear @xcite , while in experiments with circularly moving colloids on 2d substrates , directed or ratchet motion can be produced @xcite . @xcite considered a model of counterclockwise circle swimmers on a 2d periodic substrate for both deterministic and stochastic swimmers . in the deterministic regime they found that the particles can either form closed orbits or translate at fixed angles with respect to the substrate periodicity . in the stochastic regime , the ratchet effect extends over a wider range of parameters since the particles can hop between different rectifying orbits even if they become temporarily trapped in a closed or nonrectifying orbit . it is also possible to obtain ratchet effects for circle swimmers moving over asymmetric substrates @xcite . figure [ fig:12](a ) shows a counterclockwise swimming particle moving with a motor force amplitude of @xmath100 over an array of l - shaped barriers . the particle translates one substrate lattice constant in the positive @xmath20 direction during every swimming cycle . for @xmath101 in fig . [ fig:12](b ) , the particle is confined to a closed non - translating orbit . in fig . [ fig:12](c ) , the translating orbit at @xmath102 includes particle motion in both the positive @xmath12 and positive @xmath20 directions . figure [ fig:12](d ) shows that at @xmath103 the particle translates in the negative @xmath12-direction . the average drift velocity of the particle in the @xmath20 direction , @xmath104 , and in the @xmath12 direction , @xmath105 , is plotted as a function of @xmath23 in fig . . a series of quantized orbits occur in which the particle translates in a fixed direction over a finite interval of @xmath23 . due to the asymmetry of the l - shaped barriers , the locations of the translating or ratcheting regimes differ for clockwise and counterclockwise swimming particles ; however , in general , swimmers with different chirality move in different directions , permitting a mixture of chiral swimmers to be separated by the barrier array . when thermal effects are included , the additional noise expands the range of values of @xmath23 over which rectification occurs ; however , the efficiency of the maximum ratchet effect is reduced compared to the nonthermal or deterministic limit . , translation occurs in the positive @xmath20 direction . ( b ) for @xmath106 a non - ratcheting orbit appears . ( c ) at @xmath107 the particle translates in the positive @xmath12 and @xmath20-directions . ( d ) at @xmath103 the particle translates in the negative @xmath12-direction . adapted from c. reichhardt and c. j. olson reichhardt , phys . e * 88 * , 042306 ( 2013 ) . copyright 2013 by the american physical society . , width=288 ] showing a series of ratchet orbits as a function of the motor force amplitude @xmath23 . ( a ) @xmath108 vs @xmath23 . ( b ) @xmath109 vs @xmath23 . if the particle swimming direction is reversed from counterclockwise to clockwise , a different set of rectification phases occurs due to the asymmetry of the l - shaped barriers . reprinted with permission from c. reichhardt and c. j. olson reichhardt , phys . e * 88 * , 042306 ( 2013 ) . copyright 2013 by the american physical society . , width=288 ] mijalkov and volpe @xcite considered the separation of different species of circle swimmers with different angular velocities by an asymmetric comb channel with a geometry that could be nanofabricated , as illustrated in fig . [ fig:14](a ) . the particles are initially placed in the leftmost chamber , and as time progresses , the particles moving at 2.2 rad / s undergo the largest amount of ratcheting motion along the channel , as shown by the histograms of particle density at different times in fig . [ fig:14](b - e ) . at the longest times of @xmath110s in fig . [ fig:14](e ) the spatial separation of the different particle species is clearly visible . ( black ) , 2.8 ( yellow ) , 2.5 ( blue ) , and @xmath111 ( red ) rad / sec . ( b , c , d , e ) histograms of the particle positions for increasing times . the particles rotating at 2.2 rad / s exhibit the largest amount of rectification . reproduced from m. mijalkov and g. volpe , soft matter * 9 * , 6376 ( 2013 ) , with permission of the royal society of chemistry . , width=288 ] ai _ et al . _ @xcite considered chiral swimmers interacting with an array of half - circle barriers and found that particles with different chirality migrate to opposite sides of the sample due to interactions with the barriers . in a numerical study of chiral swimmers in a geometry containing m - shaped barriers on the bottom of the sample at @xmath112 and periodic boundary conditions in the @xmath12 direction , the chirality of the swimmers leads to a rectification effect and makes it possible to achieve the transport of a larger passive particle @xcite . the direction in which the active particles ratchet depends on the chirality of the swimming , so that counterclockwise and clockwise swimmers move in opposite directions . one of the surprising results of this work is that the direction in which the passive particle drifts can , for some parameters , be opposite to the direction in which the swimmers are moving , and this directional reversal depends strongly on the packing fraction of the system , which again indicates that collective effects can introduce new behaviors . ratchet effects are generally studied in systems with fixed asymmetric substrates ; however , if the substrate is mobile and can respond to the active particle , an active ratchet effect can occur that can be used to transport larger - scale asymmetric objects . @xcite examined a model of run - and - tumble bacteria interacting with large scale passive rotary objects that have asymmetric saw - tooth shapes . the bacteria accumulate in the corners of the sawteeth and generate a net torque that produces a persistent rotation of the object in one direction . such active matter - induced gear rotation was subsequently confirmed in several experiments with swimming bacteria @xcite as well as for gears interacting with small robots @xcite . additionally , if an asymmetric object is placed in a bath of active particles , it is possible to generate enhanced transport of the object . figure [ fig:15](a , b ) shows images of asymmetric objects placed in a bath of rod - like bacteria to create microshuttles @xcite . the bacteria , which swim in the direction of their white tips , accumulate in the corners of each shuttle object and produce a net force @xmath113 that causes the shuttle to translate . figure [ fig:15](c ) shows that the probability distribution of @xmath113 shifts its weight to higher values of @xmath113 as the shuttle length is increased , producing a larger biasing propulsion force . of the time - averaged force @xmath113 exerted on the shuttle by the active particles for shuttles made from 1 ( black ) , 2 ( red ) , 3 ( green ) , or 4 ( blue ) wedges , showing an increase in the effective driving force as the number of wedges increases . reprinted with permission from l. angelani and r. di leonardo , new j. phys . * 12 * , 113017 ( 2010 ) . iop publishing & deutsche physikalishe gesellschaft . cc by - nc - sa . http://dx.doi.org/10.1088/1367-2630/12/11/113017 , width=288 ] + ( a)(b ) + \(c ) of the time - averaged force @xmath113 exerted on the shuttle by the active particles for shuttles made from 1 ( black ) , 2 ( red ) , 3 ( green ) , or 4 ( blue ) wedges , showing an increase in the effective driving force as the number of wedges increases . reprinted with permission from l. angelani and r. di leonardo , new j. phys . * 12 * , 113017 ( 2010 ) . iop publishing & deutsche physikalishe gesellschaft . cc by - nc - sa . http://dx.doi.org/10.1088/1367-2630/12/11/113017 , title="fig:",width=288 ] experiments on asymmetric wedges in bacterial baths reveal enhanced transport of these objects @xcite , and other studies have examined how the shape of the asymmetric passive object affects the transport of the object @xcite . since active particles accumulate in corners or concave regions of a funnel shape , it is possible to use static @xcite or moving wedges @xcite to capture active particles . many additional variants of using active matter to transport passive objects are possible . it would also be interesting to study systems in which the passive objects could change shape in response to the activity . we have highlighted the recent results for a new class of ratchet system called active ratchets , which appear when the directed motion of self - propelled particles , which breaks detailed balance , is combined with an additional asymmetry in the system . unlike rocking and flashing ratchets , active ratchets do not require the application of an external driving force to produce rectification . active ratchet effects and variations upon them will be a growing field of research as the ability to fabricate additional types of artificial swimmers , nanobots , and other self - driven systems continues to improve , along with the possibility of applying simple motion rules to the active objects that can permit them to perform usable work . due to the general nature of the conditions under which active ratcheting can occur , it is likely that nature may already be exploiting active ratchet effects in various biological systems , so it would be interesting to examine motility process in complex biological environments to understand whether active ratchet effects are present . another area for future research is the collective effects that arise in assemblies of interacting active particles or in more complex objects such as cells . it has already been demonstrated that collective effects can produce a variety of novel ratcheting effects including ratchet reversals , which offers the possibility to utilize bidirectional ratchets for various medical technologies and other types of biological applications . the authors are not aware of any affiliations , memberships , funding , or financial holdings that might be perceived as affecting the objectivity of this review . this work was carried out under the auspices of the nnsa of the u.s . doe at lanl under contract no . de - ac52 - 06na25396 . 99 reichhardt c , reichhardt c j o. arxiv:1602.03798 ( unpublished ) reimann p. 2002 . _ 361:57265 derenyi i , vicsek t. 1995 . _ _ 75:374377 de souza silva c c , van de vondel j , morelle m , moshchalkov v v. 2006 . _ nature ( london ) _ 440:651654 lu q , reichhardt c j o , reichhardt c. 2007 . 75:054502 lee c s , jank ' o b , der ' enyi i , barab ' asi a l. 1999 . _ nature ( london ) _ 400:337340 galajda p , keymer j , chaikin p , austin r. 2007 . _ j. bacteriol . _ 189:87048707 reichhardt c , ray d , reichhardt c j o. 2015 . b _ 91:184502 van oudenaarden a , boxer s g. 1999 . _ science _ 285:10461048 ertas d. 1998 . _ 80:15481551 duke t a j , austin r h. 1998 . _ 80:15521555 reichhardt c j o , reichhardt c. 2005 . _ physica c _ 432:125132 dinis l , perez de lara d , gonzalez e m , anguita j v , parrondo j m r , vicent j l. 2009 . _ new j. phys . _ 11:073046 reichhardt c , reichhardt c j o. 2016 . _ phys . rev . 93:064508 lindner b , schimansky - geier l , reimann p , h " anggi p , nagaoka m. 1999 . _ phys . 59:14171424 reichhardt c , ray d , reichhardt c j o. 2015 . _ new j. phys . _ 17:070304 rousselet j , salome l , ajdari a , prost j. 1994 . _ nature ( london ) _ 370:446448 farkas z , tegzes p , vukics a , vicsek t. 1999 . _ 60:70227031 wambaugh j f , reichhardt c , olson c j. 2002 . 65:031308 jones p h , goonasekera m , renzoni f. 2004 . _ 93:073904 salger t , kling s , hecking t , geckeler c , morales - molina l , weitz m. 2009 . _ science _ 326:12411243 linke h , humphrey t e , l " ofgren a , sushkov a o , newbury r , taylor r p , omling p. 1999 . _ science _ 286:23142317 roeling e m , germs w c , smalbrugge b , geluk e j , de vries t , janssen r a j , kemerink m. 2011 . _ nature mater . _ 10:5155 p ' erez - junquera a , marconi v i , kolton a b , ' alvarez - prado l m , souche y , alija a , v ' elez m , anguita j v , alameda j m , mart ' in j i , parrondo j m r. 2008 . _ 100:037203 franken j h , swagten h j m , koopmans b. 2012 . _ nature nanotechnol . _ 7:499503 chat e h , ginelli f , gr ' egoire g , peruani f , raynaud f. 2008 . 64:451456 ramaswamy s. 2010 . rev . condens . matter phys . 1:323345 marchetti m c , joanny j f , ramaswamy s , liverpool t b , prost j , rao m , simha r a. 2013 . _ 85:11431189 bechinger c , di leonardo r , l " owen h , reichhardt c , volpe g , volpe g. arxiv:1602.00081 ( unpublished ) . cates m e , marenduzzo d , pagonabarraga i , tailleur j. 2010 . 107:1171511720 fily y , marchetti m c. 2012 . _ 108:235702 redner g s , hagan m f , baskaran a. 2013 . _ 110:055701 palacci j , sacanna s , steinberg a p , pine d j , chaikin p m. 2013 . _ buttinoni i , bialk ' e j , k " ummel f , l " owen h , bechinger c , speck t. 2013 . 110:238301 reichhardt c , reichhardt c j o. 2014 . _ soft matter _ 10:75027510 wan m b , reichhardt c j o , nussinov z , reichhardt c. 2008 . _ 101:018102 reichhardt c j o , drocco j , mai t , wan m b , reichhardt c. 2011 . 8097:80970a tailleur j , cates m e. 2009 . _ europhys . _ 86:60002 fily y , baskaran a , hagan m f. 2014 . _ soft matter _ 10:56095617 fily y , baskaran a , hagan m f. 2015 . 91:012125 berdakin i , jeyaram y , moshchalkov v v , venken l , dierckx s , vanderleyden s j , silhanek a v , condat c a , marconi v i. 2013 . 87:052702 galajda p , keymer j , dalland j , park s , kou s , austin r. 2008 . 55:34133422 hulme e , diluzio w r , shevkoplyas s s , turner l , mayer m , berg h c , whitesides g m. 2008 . _ lab chip _ 8:188895 kaehr b , shear j b. 2009 . _ lab chip _ 9:26322637 kim s y , lee e s , lee h j , lee s y , lee s k , kim t. 2010 . _ j. micromech . _ 20:085007 chen y - f , xiao s , chen h - y , sheng y - j , tsao h - k . _ nanoscale _ 7:1645116459 angelani l , costanzo a , di leonardo r. 2011 . _ epl _ 96:68002 yariv e , schnitzer o. 2014 . 90:032115 potiguar f q , farias g a , ferreira w p. 2014 . _ phys . 90:012307 berg h c , brown d a. 1972 . _ 239:500504 berdakin i , silhanek a v , cort ' ez h n m , marconi v i , condat c a. 2013 . _ central eur . _ 11:16531661 kantsler v , dunkel j , polin m , goldstein r e. 2013 . 110:11871192 guidobaldi a , jeyaram y , berdakin i , moshchalkov v v , condat c a , marconi v i , giojalas l , silhanek a v. 2014 . _ 89:032720 nam s - w , qian c , kim s h , van noort d , chiam k - h , park s. 2013 . _ _ 3:3247 ai b - q , chen q - y , he y - f , li f - g , zhong w - r . 88:062129 wu j c , chen q , wang r , ai b q. 2014 . _ j. phys . a : math . theor . _ 47:325001 ai b - q , wu j - c . 2014 . _ j. chem . _ 140:094103 reichhardt c , reichhardt c j o. 2013 . _ phys . 88:062310 kulic i m , mani m , mohrbach h , thaokar r , mahadevan l. 2009 . b _ 276:22432247 pototsky a , hahn a m , stark h. 2013 . 87:042124 stenhammar j , wittkowski r , marenduzzo d , cates m e. arxiv:1507.01836 ( unpublished ) nikola n , solon a p , kafri y , kardar m , tailleur j , voituriez r. arxiv:1512.05697 ( unpublished ) weitz s , blanco s , fournier r , gautrais j , jost c , theraulaz g. 2014 . _ phys . 89:052715 wu j - c , chen q , wang r , ai b - q . 2015 . _ physica a _ 428:273278 mijalkov m , mcdaniel a , wehr j , volpe g. 2016 . x _ 6:011008 drocco j a , reichhardt c j o , reichhardt c. 2012 . _ 85:056102 mahmud g , campbell c j , bishop k j m , komarova y a , chaga o , soh s , huda s , kandere - grzyboswka k , grzybowski b a. 2009 . _ nature phys . _ 5:606612 sun x , driscoll m k , guven c , das s , parent c a , fourkas j t , losert w. 2015 . _ proc . 112:1255712562 ghosh p k , misko v r , marchesoni f , nori f. 2013 . 110:268301 koumakis n , lepore a , maggi c , di leonardo r. 2013 . _ nature commun . _ 4:2588 lambert g , liao d , austin r h. 2010 . _ 104:168102 reichhardt c , reichhardt c j o. 2014 . e _ 90 : 012701 maass c c , kr " uger c , herminghaus s , bahr c. 2016 _ annu . rev . matter phys . _ 7:171193 wan m - b , jho y - s . _ soft matter _ 9:32553261 narayan v , ramaswamy s , menon n. 2007 . _ science _ 317:105108 sanchez t , chen d t n , decamp s j , heymann m , dogic z. 2012 . _ 491:431434 giomi l , bowick m j , ma x , marchetti m c. 2013 . _ 110:228101 decamp s j , redner g s , baskaran a , hagan m f , dogic z. 2015 . _ nature mater . _ 14:11101115 diluzio w r , turner l , mayer m , garstecki p , weibel d b , berg h c , whitesides g m. 2005 . 435:12711274 riedel i h , kruse k , howard j. 2005 . _ science _ li g , tam l - k , tang j x. 2008 . 105:1835518359 tierno p , johansen t h , fischer t m. 2007 . _ 99:038303 k " ummel f , ten hagen b , wittkowski r , buttinoni i , eichhorn r , volpe g , l " owen h , bechinger c. 2013 . _ 110:198302 ten hagen b , k " ummel f , wittkowski r , takagi d , l " owen h , bechinger c. 2014 . _ nature commun . 5:4829 reichhardt c , reichhardt c j o. 2003 . 68:046102 speer d , eichhorn r , reimann p. 2009 . _ 102:124101 tierno p , johansen t h , fischer t m. 2007 . _ 99:038303 nourhani a , crespi v h , lammert p e. 2015 . _ _ 115:118101 reichhardt c , reichhardt c j o. 2013 . _ phys . 88:042306 mijalkov m , volpe g. 2013 . _ soft matter _ 9:63766381 ai b , he y , zhong w. 2015 . _ soft matter _ 11:38523859 ai b. 2016 . _ _ 6:18740 angelani l , di leonardo r , ruocco g. 2009 . _ 102:048104 di leonardo r , angelani l , dellarciprete d , ruocco g , iebba v , schlippa s , conte m p , mecarini f , de angelis f , di fabrizio e. 2010 . 107:95419545 sokolov a , apodaca m m , grzyboswki b a , aranson i s. 2010 . 107:969974 kojima m , miyamoto t , nakajima m , homma m , arai t , fukuda t. 2016 . _ sensors actuators b _ 222:12201225 li h , zhang h p. 2013 . 102:50007 angelani l , di leonardo r. 2010 . _ new j. phys . _ 12:113017 kaiser a , peshkov a , sokolov a , ten hagen b , l " owen h , aranson i s. 2014 . _ 112:158101 mallory s a , valeriani c , cacciuto a. 2014 . 90:032309 kaiser a , wensink h h , l " owen h. 2012 . _ phys . rev _ 108:268307 kaiser a , popowa k , wensink h h , l " owen h. 2013 . _ phys . rev .
we study the influence of an imperfect structure of a crystalline undulator on the spectrum of the undulator radiation . the main attention is paid to the undulators in which the periodic bending in the bulk appears as a result of a regular ( periodic ) surface deformations . we demonstrate that this method of preparation of a crystalline undulator inevitably leads to a variation of the bending amplitude over the crystal thickness and to the presence of the subharmonics with smaller bending period . both of these features noticeably influence the monochromatic pattern of the undulator radiation .
0710.1363
in this paper we present the qualitative and quantitative analysis of the influence of imperfect structure of a crystalline undulator ( in particular , the influence of the variation of the bending amplitude over the crystal thickness due to the stress applied to its surface ) on the spectral distribution of the radiation . the parameters of crystalline undulators ( including the types and lengths of crystals , the periods of bending , the positron energies , and the range of photon energies ) as well as the methods of preparation of periodically bent structures discussed below in the paper correspond to those which are available for the experiments to be carried out within the pecu project@xcite . a periodically bent crystal together with a bunch of ultra - relativistic charged particles which undergo planar channeling constitute a crystalline undulator . in such a system there appears , in addition to the well - known channeling radiation , the undulator type radiation which is due to the periodic motion of channeling particles which follow the bending of the crystallographic planes@xcite . the intensity and characteristic frequencies of this radiation can be varied by changing the beam energy and the parameters of bending . in the cited papers as well as in the subsequent publications ( see the review ref . @xcite and the references therein ) a feasibility was proven to create a short - wave crystalline undulator that will emit intensive monochromatic radiation when a pulse of ultra - relativistic positrons is passed through its channels . more recently , it was demonstrated @xcite that the undulator based on the electron channeling is also feasible . a number of corresponding novel numerical results were presented to illustrate the developed theory , including , in particular , the calculation of the spectral and angular characteristics of the new type of radiation . although the operational principle of a crystalline undulator does not depend on the type of a projectile below in this paper we will consider the case of a positron channeling . under certain conditions @xcite , an ultra - relativistic positron , which enters the crystal at the angle smaller than the lindhard critical angle @xcite , will penetrate through the crystal following the bendings of its planes . provided the bending amplitude @xmath0 greatly exceeds the interplanar distance @xmath1 ( see figure [ figure.1 ] left ) one can disregard the oscillations due to the action of the interplanar force , the channeling oscillations@xcite . in this case the trajectory of the particle can be associated with the periodic profile of channel centerline . the undulator radiation appears as a result of this periodic motion of the particle . thus , the operational principle of a crystalline undulator is the same as for a conventional one@xcite , in which the monochromaticity of the radiation is the result of a constructive interference of the photons emitted from similar parts of trajectory . usually , when discussing the properties of a crystalline undulator and of its radiation , one considers the case of _ a perfect _ crystalline undulator . by this term we will understand the crystal whose planes are bent periodically following a perfect harmonic shape , @xmath2 , see figure [ figure.1 ] ( left ) . for clarity , let us stress that we consider the case when the quantities @xmath1 , @xmath0 and @xmath3 satisfy the double inequality @xmath4 . typically , @xmath5 , @xmath6 and @xmath7 . the spectral - angular distribution from the perfect undulator is characterized by a specific pattern which implies that for each value of the emission angle the spectrum consists of a set of narrow , well - separated and powerful peaks corresponding to different harmonics of radiation . in principle , the perfect crystalline undulators can be produced by using the technologies of growing si@xmath8ge@xmath9 structures @xcite . in this case , by varying the ge content @xmath10 one can obtain periodically bent crystalline structure @xcite . the technological restrictions imposed by this method on the crystalline undulator length is @xmath11 m . the periodic bending can also be achieved by making regularly spaced grooves on the crystal surface either by using a diamond blade @xcite or by means of laser - ablation . the latter method was used @xcite to prepare the si - based crystalline undulators for the pecu experiments . in either case , the regular surface deformation results in the periodic pattern of the crystallographic planes bending in the bulk . the question which appears in connection with these methods of preparation concerns the quality of the periodically bending . indeed , for a crystal of a finite thickness it is natural to expect that the surface deformations , regularly spaced with the period @xmath3 , result in the volume deformations of the same period but of a varied amplitude of bending , @xmath12 ( see figure [ figure.1 ] ( right ) ) . the latter has maximum value in the surface layer but decreases with the penetration distance . therefore , it is important to carry out a quantitative analysis ( a ) of the structure of this _ imperfect _ periodic bending in the bulk , and ( b ) of its influence on the spectrum of undulator radiation . both of these problems constitute the subject of the present paper . the paper is organized as follows . in section [ de ] we briefly outline the basic formulae and definitions which refer to a perfect crystalline undulator . section [ averaging ] describes , in general terms , the modifications to be introduced to the formalism due to the imperfectness of the crystalline undulator . in section [ formalism ] we present the formalism and carry out numerical analysis of the periodic deformations in the bulk caused by a regular stress applied to the crystal surface . finally , in section [ averagedspectra ] we carry out quantitative analysis of the differences in the emission spectra formed in a perfect undulator and in the undulator created by means of periodic surface deformations . the spectral distribution of the energy @xmath13 of radiation emitted by an ultra - relativistic ( @xmath14 ) positron in a perfect crystalline undulator can be written in the following form @xcite : @xmath15 where @xmath16 and @xmath17 are the emission angles with respect to the undulator axis chosen along @xmath18 , @xmath19 is the emission solid angle . the factor @xmath20 is explained further in this section . the function @xmath21 is given by @xmath22 \biggr ) , \quad m=0,1\ , . \label{de.3}\end{aligned}\ ] ] here @xmath23 is the fine structure constant , @xmath24 is the relativistic lorentz factor with @xmath25 standing for the positron mass and @xmath26 for its energy , @xmath27 and @xmath28 is the undulator parameter , @xmath29 the parameter @xmath30 is defined as follows @xmath31 a peculiar feature of the undulator radiation is that for each value of the emission angle @xmath32 the spectral distribution consists of a set of narrow and equally spaced peaks ( harmonics ) the frequencies @xmath33 one defines letting @xmath34 : @xmath35 the widths @xmath36 of the peaks satisfy the inequality @xmath37 , so that all peaks are well separated . in an ideal undulator ( i.e. , in which positrons and photons propagate in vacuum ) the peak intensity is proportional to the squared number of periods . formally , it follows from the fact that @xmath38 is proportional to @xmath39 which behaves as @xmath40 for integer @xmath30 ( e.g. , @xcite ) . this factor reflects the constructive interference of radiation emitted from each of the undulator periods . consequently , in an ideal undulator one can increase unrestrictedly the radiated intensity by increasing of the undulator length @xmath41 . the situation is different for a crystalline undulator , where the number of channeling particles and the number of photons which can emerge from the crystal decrease with the growth of @xmath42 . in refs . @xcite the quantitative study of the influence of the dechanneling and the photon attenuation on the spectral - angular distribution was presented . the main result of this analysis is that the peak intensity is no longer proportional to @xmath40 . it was shown , that in a crystalline undulator the factor @xmath43 must be substituted with @xmath20 , which depends not only on @xmath44 and @xmath30 but also on the ratios @xmath45 and @xmath46 . here @xmath47 stands for the dechanneling length which is the mean penetration distance covered by a channeling particle . the quantity @xmath48 , called the attenuation length , defines the scale within which the intensity of a photon flux propagating through a crystal decreases by a factor of @xmath49 due to the processes of absorption and scattering . a convenient formula for @xmath20 , which enters the right - hand side of ( [ de.1 ] ) , is as follows @xcite : @xmath50\ , , \label{de.5 } \ ] ] where @xmath51 ( with @xmath52 being the closest positive integer to @xmath30 ) . despite a cumbersome form of the right - hand side of ( [ de.5 ] ) its main features can be easily understood . the most important is that , as in the case of an ideal undulator ( to which @xmath20 reduces in the limit @xmath53 ) the main maxima of @xmath20 correspond to @xmath30 integers , and , therefore , the harmonic frequencies are still defined by ( [ de.4 ] ) with @xmath54 . for finite @xmath47 and @xmath48 , the maximum value of @xmath20 is smaller than @xmath40 whereas the width of the peak is larger than that in the corresponding ideal undulator . the attenuation length depends on the photon energy , @xmath55 , and can be calculated as the inverse mass attenuation coefficient which are tabulated for all elements and for a wide range of photon frequencies @xcite . the dechanneling effect stands for a gradual increase in the transverse energy of a channeled particle due to inelastic collisions with the crystal constituents @xcite . at some point the particle gains a transverse energy higher than the planar potential barrier and leaves the channel . in a straight crystal dechanneling length @xmath47 of a positron depends on the crystal and on the energy of the projectile . in a bent crystal the potential barrier changes due to the centrifugal force , and , as a result , @xmath47 becomes dependent on the curvature of the channel . a stable channeling of a projectile in a periodically bent channel occurs if the maximal centrifugal force in the channel @xmath56 is less than the maximal interplanar force @xmath57 , i.e. @xmath58 . for an ultra - relativistic particle @xmath59 , where @xmath60 is the minimum curvature radius of the bent channel . in a perfect crystalline undulator @xmath61 , therefore , the condition for a stable channeling reads @xcite : @xmath62 using the diffusion model @xcite one can be demonstrated that dechanneling length in a periodically bent crystal becomes dependent on the parameter @xmath63 . in the case of ultra - relativistic positrons the expression for @xmath64 can be written as follows @xcite : @xmath65 where @xmath66 is the dechanneling length a straight channel . this quantity can be estimated as @xmath67 @xcite , where @xmath68 and @xmath69 are the classical radius of the electron and the thomas - fermi radius of the crystal atom , and @xmath70 ( @xmath71 stands for an average ionization potential of the atom ) . apart from the dechanneling length , the bending parameter @xmath63 also defines another important quantity , _ the acceptance _ of the crystal . the acceptance represents by itself the fraction of the particles trapped in the channeling mode ( or , in other words , it is equal to the ratio of the phase volume corresponding to the channeling regime to the phase volume of the beam particles at the entrance of the crystal ) . in comparison with a linear crystal , the acceptance of the bent crystal gains additional decrease due to the presence of the centrifugal force in the channels @xcite . generalizing the result of ref . @xcite , one writes the following expression for the acceptance @xmath72 of the periodically bent channel and for the case of a parallel beam : @xmath73 where @xmath74 is the acceptance of the linear crystal and @xmath75 . if the centrifugal force exceeds the interplanar force ( i.e. , @xmath76 ) then one uses the identity @xmath77 , which means that the channel does capture the particles . in connection with the radiation from a crystalline undulator , the acceptance comes into play when one calculates the energy emitted per particle of the bunch . to obtain this quantity in a perfect undulator , one multiplies the right - hand side of ( [ de.1 ] ) by the acceptance @xmath72 . the formulae ( [ de.1])-([de.8 ] ) allow one to carry out numerical analysis of the spectral - angular distribution formed in a perfect crystalline undulator . in a recent paper @xcite the results of such analysis were reported for 0.6 and 10 gev positrons channeling through periodically bent si and si@xmath8ge@xmath9 crystals . in this section we analyze the modifications in the formalism of spectral - angular distribution of the radiation from a crystalline undulator which appear due to the imperfectness of periodic bending of the crystallographic planes . for the sake of clarity let us discuss the geometrical conventions which will be used below . figures [ figure.1 ] and [ figure.2 ] help a reader to follow the text . we assume that the crystal has the form of a rectangular box . its length , @xmath42 , width , @xmath78 , and thickness , @xmath79 , are measured along the @xmath18 , @xmath10 and @xmath80 directions , respectively . we chose the frame in which the values @xmath81 denote the @xmath80-coordinates of the upper and lower surfaces of the crystal . hence , it is assumed that @xmath82 labels the central @xmath83-plane of the crystal ( the midplane ) . initially non - deformed crystallographic planes are perpendicular to the @xmath80 axis . we assume that periodic deformation of the crystalline structure occurs only in the @xmath84-plane , so that there is no deformation in the @xmath10-direction . in the deformed crystal the bunch of channeling particles propagates in the @xmath84-plane along @xmath18 direction . with @xmath85 we denote the bunch size along the @xmath80 direction . the width of the bunch ( i.e. , its size in @xmath10 ) is not used in our model . suppose that bending amplitude is not constant over the crystal thickness but changes according to a law @xmath12 , as illustrated by figure [ figure.1 ] . the particular form of this dependence is discussed in section [ formalism ] where we analyze the deformation of the crystal interior due to the periodic stress applied to its surfaces . at the moment , the only restrictions implied on the dependence @xmath86 are : ( a ) the change in amplitude on the scale of interplanar spacing @xmath1 is negligible ; ( b ) a strong inequality @xmath87 is valid for all @xmath80 . particles in the bunch are randomly distributed along the @xmath80-axis . therefore , we may assume that at the entrance a particle can be captured in any channel located within the interval @xmath88 $ ] , where @xmath89 stands for the smallest from @xmath90 and @xmath91 . being captured in the channel at some @xmath80 point , the particle undulates and emits the radiation corresponding to the undulator with the amplitude @xmath86 . for a fixed value of the period @xmath3 , the amplitude defines two important quantities which are the undulator parameter @xmath28 and the bending parameter @xmath63 . the latter , in turn , defines the dechanneling length and the acceptance ( see eqs . ( [ de.7 ] ) and ( [ de.8 ] ) ) . the influence of @xmath28 and @xmath47 on the spectral - angular distribution of radiation is discussed in detail further in this section . here we note that the acceptance @xmath72 determines the probability of a particle to be captured in the channeling more . if @xmath92 , this probability depends on the entrance coordinate @xmath80 . therefore , to obtain the distribution ( per particle ) of radiation formed in the crystal with varied amplitude of bending one should ( a ) multiply eq . ( [ de.1 ] ) by the acceptance ( [ de.8 ] ) , and ( b ) carry out the averaging over the interval @xmath88 $ ] . this leads to the formula : @xmath93 here @xmath94 stands for the the distribution ( [ de.1 ] ) obtained for the amplitude @xmath86 . the right - hand side of this equation represents the spectral - angular distribution per particle averaged over the width of the crystal ( or of the bunch if @xmath95 ) , otherwise @xmath77 , see eq . ( [ de.8 ] ) . ] . if the amplitude does not change within the interval of integration , the formula ( [ averaging.1 ] ) reduces to eq . ( [ de.1 ] ) multiplied by the acceptance corresponding to the fixed @xmath63-value . let us discuss in more detail the influence of the dependence @xmath86 on the characteristics of the crystalline undulator radiation . it is clear that for the @xmath80-varying amplitude the terms , dependent on @xmath28 in the function @xmath21 ( see ( [ de.2 ] ) ) , change with @xmath80 since @xmath96 ( see ( [ no.3 ] ) ) . apart from this influence , the proportionality can lead to the variation of the harmonics frequencies @xmath33 and , consequently , to the loss of the monochromaticity of the radiation . indeed , let @xmath97 and @xmath98 denote the minimum and maximum amplitudes within the interval @xmath99 $ ] . the corresponding extremum values of the undulator parameter , @xmath100 and @xmath101 , having been used in ( [ de.4a ] ) , produce the lower , @xmath102 , and the upper , @xmath103 , bounds on @xmath33 . in the limit @xmath104 there is a weak variation of the harmonics frequencies , @xmath105 . however , in the case @xmath106 the change in the undulator parameter leads to the emission within the band @xmath107 which can greatly exceed not only the peak width @xmath36 but also the interval between the neighbouring harmonics . in the latter case the monochromaticity of the radiation will be smeared out if one carries out the averaging procedure ( [ averaging.1 ] ) . however , the change in the undulator parameter and the harmonics frequencies is not the only impact caused by the amplitude variation . as mentioned above , in a crystalline undulator the peak intensity is defined by the maximum value of the factor @xmath20 , which depends on the ratios @xmath108 and @xmath109 ( see ( [ de.5 ] ) ) . typically , in the crystalline undulators based on @xmath110 gev positrons channeling in crystals ( so far si monocrystals and si - ge mixtures were used @xcite ) the energy of emitted photons lies within the range @xmath111 kev . for these energies of positrons and photons the attenuation length @xmath112 is on the level of several cm @xcite and by far exceeds the positron dechanneling lengths ( lying within @xmath113 cm in straight channels @xcite ) . therefore , assuming @xmath114 in ( [ de.5 ] ) , one finds that the peak value of the factor @xmath20 ( i.e. , at @xmath54 ) in an undulator with fixed number of periods , @xmath115 , is given by @xmath116 the dechanneling length in a channel , which is periodically bent with the amplitude @xmath0 , is defined by equations ( [ de.6 ] ) and ( [ de.7 ] ) . hence , for @xmath12 the right - hand side of ( [ averaging.2 ] ) ( and , generally , of ( [ de.5 ] ) as well ) becomes dependent on @xmath80 since @xmath117 with @xmath118 . in turn , the variation of the factor @xmath20 can strongly influence the averaged spectral - angular distribution ( [ averaging.1 ] ) . in this section we present a formalism which allows one to carry out a quantitative analysis of the parameters of periodic bending in the bulk of a crystal . we consider the case when the bendings in the bulk are due to the periodic deformations on the crystal surfaces . such a situation is illustrated by figure [ figure.2 ] : two parallel opposite surfaces are deformed periodically by means of identical sets of parallel grooves applied to each of the surfaces ( the two sets are shifted by half - period , @xmath119 ) . the deformation of this type can be achieved either by mechanical scratching of the crystal surface @xcite or by means of a more accurate laser - ablation method @xcite . another possible method is in a deposition of si@xmath120n@xmath121 layers onto a si wafer @xcite . in either case the crystallographic planes in the bulk become bent periodically although the shape @xmath122 of bent planes does not follow an ideal harmonic form @xmath123 . the main deviations are : ( a ) the amplitude of bending depends on the distance from the surface , and ( b ) in general case , higher subharmonics ( i.e. the fourier components of @xmath124 with smaller periods , @xmath125 , where @xmath126 ) contribute noticeably into the formation of the periodic shape . however , as we demonstrate below , it is possible to establish the ranges of parameters ( these include @xmath3 , the thickness @xmath79 and elastic constants of the crystal ) within which the deviations of the resulting periodic shape from the ideal form do not affect the spectral - angular distribution of the undulator radiation . direction ) : @xmath3 stands for the period of deformations and @xmath127 denote the width of a groove . the set on the lower surface is shifted by @xmath119 ( along the @xmath18-axis ) with respect to that on the upper surface . the surface stress gives rise to the periodic bending of crystallographic planes in the bulk of crystal . , width=491,height=188 ] to start with , let us formulate the approximations which will be used in our formalism . firstly , we assume that the width of a crystal @xmath78 ( i.e. its size in the @xmath10-direction , see figure [ figure.2 ] ) is much larger than the thickness @xmath79 : @xmath128 . as a result , one can disregard the deformations in the @xmath10-direction and consider the deformations of a solid body in the @xmath84-plane only . additionally , we assume that the period @xmath3 is incomparably smaller than the length @xmath42 of the crystal in the @xmath18 direction . then , taking into account the periodicity of the surface deformations , one represents the displacement vector @xmath129 ( i.e. the vector which characterizes the change in the position vector of a point in the body due to the deformation ) in the form of fourier series : @xmath130 where @xmath131 . the vectors @xmath132 ( with @xmath133 standing for an integer ) are to be defined by solving the equation of equilibrium with proper boundary conditions . in what follows we adopt that the @xmath80 coordinate is measured from the crystal midline , and thus @xmath134 corresponds to the lower surface , and @xmath135 - to the upper one . secondly , we will consider the limit of small deformations only . in this case the _ strain tensor _ @xmath136 is defined as follows ( see , e.g. , @xcite ) : @xmath137 with @xmath138 and @xmath139 . the third approximation concerns the _ stress tensor _ , @xmath140 ( we remind , that physically @xmath140 stands for a stress on the @xmath141-th plane along the @xmath142-th direction ) . in an isotropic media the components of the stress tensor can be related to @xmath136 by means of the two elastic constants : the young s modulus @xmath13 and the poisson s ratio @xmath143 . in the case of a planar deformation the relationship is as follows ( see , e.g. , 5 in @xcite ) : @xmath144 , \\ \sigma_{zz } = { e \over ( 1+\nu)(1 - 2 \nu ) } \bigl[(1-\nu ) u_{zz } + \nu u_{yy}\bigr ] , \\ \sigma_{yz } = \sigma_{zy } = { e \over ( 1+\nu ) } u_{yz}. } \label{formalism1.3 } \ ] ] in anisotropic media ( e.g. , in a crystal ) @xmath13 and @xmath143 depend on the directions of the applied stress and of the deformation . for example , depending on a crystallographic direction the poisson s ratio and the young s modulus in a si crystal varies within the intervals @xmath145 and @xmath146 gpa @xcite . however , to simplify the analysis one can chose some average values . in our numerical analysis we use @xmath147 and @xmath148 gpa which are close to the values used in modeling various deformation processes in silicon @xcite . to check the sensitivity of the results to the choice of the poisson s ratio we also carried out the calculations using the extreme values of @xmath143 in silicon . it turned out that nearly an order of magnitude change in @xmath143 does not noticeably affect the results which are presented below in the paper . the components of the stress tensor satisfy the following equations of equilibrium : @xmath149 using ( [ formalism1.1])([formalism1.3 ] ) in ( [ formalism1.4 ] ) one derives the system of coupled equations for the functions @xmath150 and @xmath151 : @xmath152 to find the unique solution of this system one has to impose the boundary conditions . these can be formulated as follows . each trench acts as a source of a normal ( ` @xmath153 ' ) and a tangential , or a shear ( @xmath154 ) tension which are characterized by the average pressures @xmath155 and @xmath156 applied to the crystal surface in vicinity of the trench . in the case when the period @xmath3 the width @xmath127 ( see figure [ figure.2 ] ) , the trenches are equivalent to the sets of concentrated normal and shear forces applied along the equally spaced lines on the upper and lower surfaces . is well fulfilled in the crystalline undulators which have been manufactured . typical values are : @xmath157 @xmath158 m and @xmath159 @xmath158 m @xcite . ] within this model the pressures @xmath155 and @xmath156 can be related to the components of the stress tensor calculated at the upper ( @xmath135 ) and the lower ( @xmath134 ) surfaces . let us first formulate the boundary conditions due to the normal tension . the pressure @xmath155 is applied inward the crystal along straight lines , parallel to the @xmath10 axis , passing through the equally - spaced points in the @xmath18 direction . therefore , recalling that @xmath160 represents the @xmath141-th component of the force per unit area ( with @xmath161 standing for the @xmath142-th component of the outward - pointing normal ) one derives the boundary conditions : @xmath162 here the coefficient @xmath3 ensures that the period - averaged pressure , equals to @xmath155 . the arguments of the delta functions fix the @xmath18 coordinates of the lines parallel to the @xmath80 axis . as mentioned , the sets of trenches on the upper and lower surfaces are shifted by @xmath119 , and this explains the difference of the arguments of the delta functions for @xmath135 and @xmath134 . in the case of shear stress , supposing that the trench profile is symmetric , one notices that the tangential forces created at the opposite edges of a trench are equal in magnitude but are antiparallel . therefore , the component @xmath163 as a function of @xmath18 must change the sign when crossing a trench . it means that in the limit of infinitesimal width @xmath164 this component becomes proportional not to the delta functions as in ( [ formalism1.6 ] ) but to their derivatives . as a result , one writes the boundary conditions in the form : @xmath165 where @xmath166 stands for the tangential force associated with the period - averaged pressure @xmath156 . with the help of ( [ formalism1.1])([formalism1.3 ] ) the equations ( [ formalism1.6 ] ) and ( [ formalism1.7 ] ) can be re - written in terms of the functions @xmath150 and @xmath151 taken at @xmath167 . the obtained formulae suffice to determine completely the solutions @xmath168 ( @xmath169 ) for the normal or the shear stress . to obtain the solution in the case when both types of stress act simultaneously one constructs the corresponding linear combinations of the functions @xmath170 and @xmath171 . using these functions further in ( [ formalism1.1 ] ) one determines the displacement vector @xmath172 . the @xmath80 component of this vector is of a special interest in connection of the crystalline undulator problem since it determines the profile of the periodically bent channel in the bulk . let us first analyze the periodic deformation in the bulk due to the normal and the shear stresses separately . resolving the system ( [ formalism1.5 ] ) with the boundary conditions ( [ formalism1.6 ] ) ( or ( [ formalism1.7 ] ) ) one finds the functions @xmath174 ( or @xmath175 ) for all @xmath133 . using these in ( [ formalism1.1 ] ) one represents the @xmath80-component of the displacement vector in the following form : @xmath176 here @xmath177 can be associated with the amplitude of the @xmath133th harmonic of teh periodic bending , i.e. , the one with the period @xmath125 . a perfect crystalline undulator ( see figure [ figure.1 ] , left panel ) is characterized only by the term @xmath178 whose amplitude is independent on @xmath179 $ ] . however , if a crystalline undulator is prepared by applying concentrated periodic normal or shear stress then : ( a ) higher amplitude harmonics ( with @xmath180 ) appear , and ( b ) the homogeneity of the bending amplitudes is lost since they become @xmath80-dependent . to stress these features of the amplitudes one can use the formulae which conveniently expresses @xmath177 via @xmath181 - the first harmonic amplitudes is the midplane of the crystal ( i.e. in the @xmath182-plane with @xmath82 ) . in the case of normal stress alone the formula is as follows : @xmath183 for a shear stress the relationship reads : @xmath184 in ( [ formalism2.2 ] ) and ( [ formalism2.3 ] ) the following notations are used : @xmath185 and @xmath186 the amplitudes @xmath181 are proportional to the applied stresses and are given by : @xmath187 where @xmath188 let us note that due to the half - period relative displacement of the deformations on the upper and lower surfaces of the crystal the amplitudes @xmath177 with odd values of @xmath133 are even functions of @xmath80 and vice versa ( see ( [ formalism2.2 ] ) and ( [ formalism2.3 ] ) ) . therefore , it is sufficient to analyze the amplitudes in the upper half of the crystal , i.e. for @xmath189 $ ] . the behaviour of the ratios @xmath192 and @xmath193 as functions of the scaled distance @xmath194 from the midplane is illustrated by figures [ figure.any2a10_perp ] and [ figure.any2a10_par ] . the dependences were calculated for several values of the crystal thickness @xmath79 ( measured in the periods @xmath3 as specified by the ratio @xmath195 ) and several even and odd values of @xmath133 . the data refer to a si crystal and the poisson s ratio was chosen as @xmath147 . ( see ( [ formalism2.2 ] ) and ( [ formalism2.8 ] ) ) versus @xmath194 ( for @xmath196 $ ] ) calculated for several values of the si crystal thickness as indicated by the parameter @xmath195 . the solid curves correspond to @xmath133 odd ( which are indicated explicitly in the top left graph ) , the dashed curves - to @xmath133 even ( see the bottom left graph ) . ] but for the ratio @xmath193 ( see ( [ formalism2.3 ] ) and ( [ formalism2.8 ] ) ) . ] it is clearly seen from the figures that for either type of stress the inhomogeneity of the bending amplitudes along the @xmath80 direction is much more pronounced for a thick crystal ( i.e. , when @xmath197 ) than for a thin one with @xmath198 . indeed , in the limit @xmath199 the amplitudes corresponding to odd values of @xmath200 do not vary noticeably over the crystal thickness . being even functions of @xmath80 , the odd-@xmath133 amplitudes @xmath177 depend quadratically on @xmath80 in the vicinity of the midplane , @xmath201 , and have minimum at @xmath82 . in a thin crystal the quadratic term is small for all @xmath80 provided @xmath133 satisfies the condition written above . as a result , @xmath202 for these @xmath133 . the main reason for the even-@xmath133 amplitudes to vary even in the case of a thin crystal is that they are odd functions of @xmath80 . therefore , being non - zero at the surfaces @xmath167 these amplitudes attain zero in the midplane . ( a non - monotonous behaviour of @xmath191 for particular @xmath133 and @xmath195 , which is most pronounced for the @xmath203 curves on the top left graph of figure [ figure.any2a10_par ] , just reflects the fact that the right - hand side of ( [ formalism2.3 ] ) is not , generally , a monotonous function of @xmath80 . ) another important feature of a thin crystal is that over the whole thickness the amplitude with @xmath178 greatly exceeds those with higher @xmath133 . this dominance is more pronounced for a normal stress than for a shear one as reflected by additional factor @xmath204 on the right - hand side of ( [ formalism2.2 ] ) . as a result , the terms with @xmath205 prevail in the series from ( [ formalism1.1 ] ) , so that the periodic bending in a thin crystal is of nearly harmonic shape : @xmath206 . as the crystal thickness increases the variation of the amplitudes over half - width evolves dramatically . in the case of a thick crystal this variation reaches orders of magnitudes . such a behaviour one understands analyzing the right - hand sides of ( [ formalism2.2 ] ) and ( [ formalism2.3 ] ) . assuming @xmath207 one derives the following expressions for the amplitudes on the surface and in the midplane : @xmath208 where @xmath209 in the case of a normal stress and @xmath210 for a shear stress . for even @xmath133 the identity @xmath211 is valid . equations from ( [ formalism3.1 ] ) demonstrate that for all @xmath133 the amplitudes decrease exponentially with the penetration distance into the crystal . for a fixed odd @xmath133 the decrease rate can be characterized by the ratio @xmath212 , which is , basically , independent on the type of applied stress ( the difference manifests itself only in the pre - factor @xmath213 ) . on the other hand , the first line in ( [ formalism3.1 ] ) indicates that in the limit of a very thick crystal the amplitudes weakly depend on @xmath133 in a surface layer of the width @xmath214 ( this dependence is more pronounced in the case of a normal stress due to the additional factor @xmath133 ) . these two features of the amplitude suggest that deviation of the periodic bending from the harmonic shape is very strong in the outer layers of the crystal whereas in the central layer the terms with @xmath180 are negligibly small and the profile of bending @xmath215 is nearly perfect . the deviation of the amplitudes @xmath216 from their values at @xmath82 is described by the following formulae : @xmath217 to conclude this section we note that figures [ figure.any2a10_perp ] and [ figure.any2a10_par ] suggest that the use of crystalline undulators manufactured by means of periodic surface deformation can be justified in two cases ( irrelevantly to the type of the stress applied to crystal ) . firstly , it is the limit of a thin crystal @xmath198 which ensures ( a ) the ( nearly ) constant value of the amplitude @xmath218 , and , ( b ) a small contribution of higher-@xmath133 terms to sum on the right - hand side of ( [ formalism2.1 ] ) . secondly , for @xmath219 it is meaningful to use only the central part of the crystal , i.e. , where @xmath220 , as an undulator . in this case it is necessary to use a narrow beam ( along the @xmath80-direction ) which is accurately aligned with the crystal midplane . equation ( [ formalism2.1 ] ) shows that in general case the profile of bending contains contributions of the terms with various @xmath133 thus deviating from a pure harmonic shape @xmath221 . hence , it is meaningful to analyze the influence of this deviation on the undulator parameter @xmath28 and the bending parameter @xmath63 which , as it was discussed at the end of section [ averaging ] , are of importance for the calculation of the averaged spectral - angular distribution ( [ averaging.1 ] ) of radiation . generally , the parameter @xmath28 characterizes the mean - square velocity , @xmath222 , of the periodic transverse motion of a particle moving in an undulator ( e.g. , @xcite ) . for an ultra - relativistic projectile the relationship is as follows : @xmath223 . in a perfect undulator this formula leads to equation ( [ no.3 ] ) . in the imperfect undulator discussed here , a particle moves along the trajectory defined by the right - hand side of ( [ formalism2.1 ] ) with @xmath224 . calculating the transverse velocity @xmath225 and averaging over the period @xmath226 , one derives the expression for the undulator parameter as a function of @xmath80 : @xmath227 where @xmath228 stands for the partial undulator parameters corresponding to the @xmath133-th subharmonics of the trajectory . in figure [ figure.py ] we present the dependences of the ratio @xmath229 on @xmath194 calculated for several values of the crystal thickness and for the two types of stress , as indicated in the caption . as it was discussed in section [ averaging ] the dependence of the undulator parameter on @xmath80 leads to a variation in the harmonic frequencies @xmath33 ( [ de.4a ] ) over the crystal thickness . if this variation becomes large enough ( e.g. , when the shift in the harmonic position becomes comparable with the separation @xmath230 of the neighbouring harmonics ) then the spectral - angular distribution will loose the peak - like pattern typical for the undulator radiation . from this viewpoint the graphs from figure [ figure.py ] allow one to estimate the influence of the change in the undulator parameter across the crystal cross section on the shape of spectral - angular distribution . for both types of stress the dependences @xmath231 exhibit the common trend : the undulator parameter slowly varies with @xmath80 in the limit of a thin crystal but becomes a rapidly ( exponentially ) increasing function for a thick crystal . this is a direct consequence of the definition ( [ formalism4.1 ] ) and the behaviour of the amplitudes @xmath177 ( see figures [ figure.any2a10_perp ] and [ figure.any2a10_par ] ) . another peculiarity , which one notices comparing the two panels in figure [ figure.py ] , is that for the same value of @xmath195 the ratio @xmath232 grows faster than @xmath233 . such a feature , which is more pronounced for higher @xmath195 values , can be understood if one recalls the differences in the @xmath133-dependences of the bending amplitudes in the case of normal and shear stresses , and the behaviour of @xmath234 as functions of @xmath80 . indeed , for moderate and large @xmath195 ratios and for both types of the stress , the amplitudes @xmath234 with @xmath180 , being rapidly increasing functions , satisfy the relations @xmath235 and @xmath236 ( see ( [ formalism3.1 ] ) ) . hence , for @xmath82 the sum on the right - hand side of ( [ formalism4.1 ] ) is defined , basically , by the term with @xmath178 , whereas for larger @xmath80 also the terms with higher @xmath133 contribute noticeably to the sum . on the other hand , it follows from ( [ formalism2.2 ] ) and ( [ formalism2.3 ] ) that @xmath237 , resulting in a similar estimate for the ratio of the partial undulator parameters : @xmath238 . therefore , in the case of the shear stress the undulator parameter varies with @xmath80 more rapidly than for the normal stress . typically , in crystalline undulators prepared by means of surface deformation and based on @xmath110 gev positron channeling @xcite ) , the values of parameter @xmath28 lie within the range @xmath239 . the curves in figure [ figure.py ] allow us to estimate the degree of consistency of using the undulators with various @xmath240 and @xmath195 values . it was pointed out in section [ averaging ] , that there will be no dramatic change in the spectral - angular distribution of the radiation due to the imperfectness of the undulator provided the variation of the undulator parameter does not lead to a noticeable change in the position of the peaks located at the frequencies @xmath33 ( see ( [ de.4a ] ) ) . to analyze this condition let us consider , for the sake of clarity , emission in the forward direction ( @xmath241 ) . in this case the frequency of the @xmath142-th harmonics , radiated from the the channel located at the distance @xmath80 from the midplane , is given by @xmath242 it is clear that in the limit @xmath243 the position of the peak will be practically unchanged for those @xmath80 where @xmath244 . these two inequalities , depending on the absolute value of @xmath245 , allow for a wide - range variation of the ratio @xmath229 , up to the order of magnitude . it follows from figure [ figure.py ] that for a thin crystal such a situation can be realizes over the full thickness @xmath79 . in a thick crystal ( @xmath219 ) only the central part ensures the needed variation of the undulator parameter . in the opposite limit of large undulator parameters , when @xmath246 , the stability of the peak position can be achieved only in that part of the crystal where @xmath247 . taking into account rapid variation of the ratio @xmath229 in the case of a thick crystal , one deduces that this inequality can be ensured only by using very thin crystals with @xmath248 . the channeling process in a bent crystal takes place if the centrifugal force , @xmath249 , due to the channel bending is less than the interplanar force @xmath57 @xcite . in a perfect crystalline undulator this condition , applied to the points of maximum curvature , results in the inequality ( [ de.6 ] ) which relates @xmath57 with the period and the amplitude of periodic bending @xcite . in the case when the profile of periodic bending contains a number of subharmonics , e.g. , as in ( [ formalism2.1 ] ) , is it more constructive to relate the parameter @xmath63 to the mean - square curvature @xmath250 where the averaging is carried out over the period @xmath3 . recalling that the curvature @xmath251 is proportional to the modulus of the second derivative of the bending profile , @xmath252 , and carrying out the averaging , one derives the following expression for average bending parameter as a function of @xmath80 : @xmath253 where @xmath254 is the mean - square partial bending parameter corresponding to the @xmath133-th subharmonics of the trajectory ( [ formalism2.1 ] ) . the parameter @xmath255 , being used in ( [ de.7 ] ) , defines the dechanneling length @xmath256 as a function of @xmath80 . as mentioned in section [ averaging ] , the variation of @xmath256 over the crystal thickness may influence destructively the averaged spectral - angular distribution ( [ averaging.1 ] ) . figure [ figure.cy ] presents the dependences of @xmath255 , scaled by its value at the crystal center , versus @xmath194 calculated for several values of the crystal thickness and for the two types of stress , as indicated in the caption . qualitatively , the behaviour of the @xmath257 curves is similar to those of the ratio @xmath258 discussed above . nevertheless , there is a quantitative difference : for a fixed value of @xmath195 the curves @xmath257 increase faster with @xmath80 than the corresponding ratio of the undulator parameters . this feature is due to the difference in the @xmath133-dependence of the partial terms in ( [ formalism2.1 ] ) and in ( [ formalism2.2 ] ) . indeed , the partial bending parameters behave as @xmath259 ( see ( [ formalism2.3 ] ) ) instead of the proportionality to @xmath260 of the partial undulator parameters . it was discussed in connection with figure [ figure.py ] that for moderate and large @xmath195 values the terms with @xmath180 provide the increase of @xmath231 with @xmath80 . this increase is more pronounced for @xmath255 since its partial terms contain an extra factor @xmath133 . a crystalline undulator can operate only in the regime when @xmath261 . if otherwise , then the centrifugal force will drive the particles out of the channel . more detailed analysis @xcite indicated that the reasonable range for the bending parameter is @xmath262 . it means that if for @xmath82 the bending parameter is of the order of @xmath263 , then the variation of @xmath255 within the order of magnitude are acceptable . from figure [ figure.cy ] it follows that for a thin crystal such a situation can be realizes over the full thickness . in a thick crystal ( @xmath219 ) only the central part ensures the needed variation of the bending parameter . the numerical data , discussed above in sections [ formalism3 ] and [ formalism4 ] , represent the @xmath80-dependence of the amplitudes , undulator parameters and bending parameters _ scaled _ by their values at @xmath82 . the latter , in turn , can be expressed in terms of the amplitudes @xmath264 which can be calculated from ( [ formalism2.8 ] ) and ( [ formalism2.8a ] ) . these equations relate the amplitudes to the bending period @xmath3 . in connection with a perfect crystalline undulator it was established ( for a detailed discussion see @xcite ) that the operation of the undulator should be considered in the large - amplitude regime , i.e. when the bending amplitude is much larger than the interplanar distance @xmath1 . in this limit the characteristic frequencies of undulator and channeling radiation ( see , e.g. , ref . @xcite ) are well separated . as a result , the channeling radiation does not affect the parameters of the undulator radiation , whereas the intensity of undulator radiation becomes comparable or higher than that of the channeling one @xcite . to apply this approach to a crystalline undulator with varied amplitude one assumes that the large - amplitude regime is applicable to the amplitudes at the crystal midplane : @xmath265 . for the convenience of further consideration let us introduce the quantity @xmath266 which explicitly measures the amplitude in the units of interplanar separation . other quantities , which enter equations ( [ formalism2.8 ] ) and ( [ formalism2.8a ] ) include : + ( a ) the crystalline medium dependent parameters , the poisson s ratio @xmath143 and the young s modulus @xmath13 . ( as already mentioned , for a si crystal we use the average values @xmath147 and @xmath148 gpa . ) + ( b ) the crystal thickness @xmath79 , which enters via the ratio @xmath195 . + ( c ) the applied stress , @xmath155 or @xmath156 . to estimate the stress one takes into account that we are interested in elastic deformations of the crystalline structure . therefore , @xmath267 must not exceed the plastic yield strength , @xmath268 , which stands for the stress at which material strain changes from elastic deformation to the plastic one . for a silicon crystal one can adopt @xmath269 gpa @xcite . for further use let us introduce the quantity , which stands for the stress measured in the units of @xmath268 : @xmath270 using ( [ formalism5.1 ] ) and ( [ formalism5.2 ] ) one re - writes equation ( [ formalism2.8 ] ) as follows ( to simplify the notations the superscripts ` @xmath153 ' and/or ` @xmath154 ' are omitted ) : @xmath271 explicit forms of the functions @xmath272 for the two types of stress are given in ( [ formalism2.8a ] ) . relationship ( [ formalism5.3 ] ) allows one to find the values of @xmath3 and @xmath79 which ensure , for a given crystal ( the parameters @xmath13 , @xmath143 and @xmath1 ) and for a relative stress ( the parameter @xmath273 ) , a desired value of the relative amplitude @xmath274 in the center of crystal . to illustrate the relationship we present figure [ figure.lambda ] , where the dependence of @xmath3 on @xmath195 is plotted for several values of @xmath273 ( as indicated ) and for two values of the relative amplitude , @xmath275 and @xmath276 the two panels in the figure correspond to different types of the applied stress . it is seen that the @xmath277 and @xmath278 curves obtained for the same values of @xmath273 and @xmath274 look much alike although there is a distinguishable quantitative difference ( note the log scale of the vertical axis ) . in this section we present the results of numerical analysis of the influence of the periodic bending imperfectness on the spectral - angular distribution . the calculations were performed for two energies , @xmath279 gev and @xmath280 ev , of a positron channeling along periodically bent ( 110 ) crystallographic planes in si ( the interplanar distance @xmath281 , the maximal interplanar force @xmath282 gev / cm ) . the data presented below refer to the emission in the forward direction . ( i.e. , @xmath283 with respect to the @xmath18 axis , see figure [ figure.2 ] ) . in this case the integrals ( [ de.3 ] ) can be evaluated analytically and the spectral distribution ( [ de.1 ] ) from an undulator with fixed parameters is expressed in terms of the appel function @xcite . these features , being not too important physically , simplify the numerical analysis . the parameters of the crystalline undulators used in the calculations were chosen as follows . firstly , for both mentioned positron energies the amplitude @xmath284 of the @xmath178 subharmonic in the crystal center was fixed as @xmath285 , thus satisfying the large - amplitude condition ( [ formalism5.1 ] ) . secondly , for each @xmath26 the length @xmath42 of the crystal was chosen to be equal to the dechanneling length in the straight channel ( see ( [ de.7 ] ) ) : @xmath286 cm and @xmath287 cm for @xmath279 and @xmath280 gev , correspondingly . the bending period is @xmath288 m for @xmath279 gev ( resulting in @xmath289 undulator periods ) , and @xmath290 m for @xmath280 gev ( with @xmath291 ) . using ( [ de.4b ] ) with @xmath292 one calculates the energies of the first harmonics emitted in the perfect undulators ( i.e. , with the cited @xmath3 and @xmath274 values ) : @xmath293 kev and @xmath294 mev for @xmath279 and @xmath280 gev , correspondingly . let us note that the mentioned values of positron energies , the crystal lengths and the parameters of periodically bent channels are close to those discussed recently in connection with the experiments on crystalline undulators @xcite . the results of calculations of the spectral distributions are presented in figures [ figure.de ] and [ figure.part_of_h ] . two graphs in figure [ figure.de ] correspond to two energies @xmath26 , as indicated in the caption . in each graph , the solid curve represents the profile of the first harmonic peak ( in the forward direction ) calculated using ( [ de.1 ] ) for the perfect undulators with the parameters @xmath274 and @xmath3 given above . other three curves in each graph correspond to the averaged spectra calculated for the same values of @xmath274 and @xmath3 but for different crystal thicknesses @xmath79 . these spectra were obtained from ( [ averaging.1 ] ) by setting @xmath295 ( the effective range of integration was restricted by the condition @xmath296 , see the footnote comment below eq . ( [ averaging.1 ] ) ) . as it was mentioned in section [ formalism5 ] , one can vary the crystal thickness together with the relative stress @xmath273 to achieve fixed values of @xmath274 ( see ( [ formalism5.3 ] ) ) . the short - dashed , long - dashed and chained curves in the figure were obtained for the relative stress @xmath297 , @xmath298 and @xmath299 , respectively . the corresponding values of @xmath79 one finds from figure [ figure.lambda ] ( left ) , where the filled circles mark the ratios @xmath195 for @xmath279 gev and the filled squares for @xmath280 gev . comparing different curves in the figure and recalling the discussion presented in section [ averaging ] , one sees the extent to which the imperfectness of the undulator structure over the crystal thickness can influence the emission spectrum . the increase in @xmath79 ( or , more generally , in the interval over which the averaging in ( [ averaging.1 ] ) is carried out ) leads to a more pronounced variation of the amplitudes @xmath234 ( see figures [ figure.any2a10_perp ] and [ figure.any2a10_par ] ) . this , in turn , trigger strong variations of the effective undulator parameter @xmath28 and bending parameter @xmath63 , figures [ figure.py ] and [ figure.cy ] . the increase in @xmath28 with @xmath80 results in the decrease of the first harmonic energy ( see ( [ de.4b ] ) ) . in the figure this feature is reflected by the ( relative ) enhancement of the photons with the energy smaller than the values of @xmath300 cited above . thus , in the peaks in the averaged spectrum become wider and the width increases with @xmath79 . another tendency , clearly seen in the figure , is the decrease in the peak intensity with @xmath79 . this is mainly due to the variation of the bending parameter @xmath255 . indeed , @xmath63 varies from its minimum value @xmath301 at the center up to the @xmath302 in the surface layer , figure [ figure.cy ] . for a given @xmath80 the value @xmath255 defines the dechanneling length @xmath303 , which , in turn , influences the peak value of the spectrum via the factor @xmath20 : smaller @xmath47 result in smaller peak intensities ( see ( [ de.5 ] ) and section [ averaging ] ) . additionally , the channel acceptance ( [ de.8 ] ) decreases with @xmath255 increasing . therefore , the relative contribution of the trajectories with larger @xmath255-values to the integral from ( [ averaging.1 ] ) increases with @xmath79 leading to the decrease in the peak intensity of the averaged spectrum . the two features mentioned above are more pronounced for a positron energy @xmath280 gev than for @xmath279 gev ( compare right and left graphs in figure [ figure.de ] ) . the reason is a s follows . for a 5 gev positron , the undulator parameter at the center is @xmath304 which is noticeably larger than the value @xmath305 for a @xmath279 gev positron . hence , the influence on the position of the first harmonic peak due to the variation of the undulator parameter with @xmath80 is smaller for lower @xmath26 . as a result , the averaged peaks for @xmath280 gev are ( relatively ) wider than those for @xmath279 gev . the more pronounced decrease in the intensities for a @xmath306 gev positron is due to the larger contribution of the trajectories with higher @xmath255-values . this , in turns , happens because for the same value of the relative stress @xmath273 the relative thickness @xmath195 is higher for a @xmath280 gev positron ( compare the positions of filled squares and filled circles which correspond to the same @xmath273 values in figure ( [ figure.lambda ] ) ) . figure [ figure.de ] demonstrates that the pattern of spectral - angular distribution can change dramatically . the well - separated peak - like structure , typical for the emission spectrum from a perfect undulator , can be completely smeared out in the imperfect undulator due to the variation of the parameters of periodic bending . the degree to which the peaks are destroyed depends on the crystal thickness and on the values of the undulator parameter @xmath245 and the bending parameter @xmath307 at the center of the crystal . in particular , for @xmath308 and in the case of a thick crystal ( @xmath207 ) the averaged spectrum has a nearly uniform distribution which is typical for the noise rather than for the undulator - type radiation . however , even in the case of a comparatively thick crystals one can restore the coherence of radiation . to achieve this , it is necessary to avoid using the layers of the crystal located far off the centerline , i.e. to use not the whole thickness of the crystal but its central part of the width lower than @xmath3 . in the central layer , the amplitudes @xmath234 and the related quantities do not deviate noticeably from their values at @xmath82 ( see ( [ formalism3.2 ] ) ) , and , therefore , the averaging procedure will not radically influence the peak profile . gev positron channeling along periodically bent si(110 ) . the undulator period , @xmath290 m , and the crystal thickness , @xmath309 , correspond to the relative normal stress @xmath298 ( these parameters are indicated in figure [ figure.lambda ] by a central filled square ) . the the solid curve stands for the intensity from a perfect undulator with the fixed amplitude @xmath310 . other curves represent the intensities ( [ averaging.1 ] ) averaged over different intervals @xmath311 $ ] : the short - dashed curve corresponds to @xmath295 , the long - dashed curve to @xmath312 , the short - dashed chained curve to @xmath313 , the long - dashed chained curve to @xmath314 . further explanations see in the text . ] to illustrate this statement , we present figure [ figure.part_of_h ] which contains the spectral intensity in the vicinity of the first harmonic for the perfect undulator ( the solid line ) along with the averaged spectra calculated for different @xmath315 . the data refer to @xmath280 gev positron channeling in a si crystal with @xmath316 , the relative normal stress @xmath298 , the parameters @xmath274 and @xmath3 as in figure [ figure.de ] right . the solid curves in figures [ figure.part_of_h ] and [ figure.de ] ( right ) are identical . the short - dashed curve ( the lowest one ) in figure [ figure.part_of_h ] stands for the spectrum averaged over the whole thickness , i.e. @xmath317 . other curves corresponds to smaller @xmath318 values , as indicated . it is clearly seen that by narrowing the averaging interval the profile of the line can be made closer to that for the perfect undulator . in our paper we discussed the influence of imperfect structure of a crystalline undulator on the spectral distribution of the radiation . we mainly analyzed the undulators in which the periodic bending in the bulk appear as a result of regular surface deformations . we demonstrated that this method inevitably leads to two main deviations from the perfect harmonic shape @xmath319 . thsese are : ( a ) the dependence of the bending amplitude on the distance @xmath80 from crystal midplane , and ( b ) the presence of subharmonics with smaller bending periods . as a result , the quantities which characterize the crystalline undulator , the undulator parameter @xmath28 and the bending parameter @xmath63 , vary over the crystal thickness @xmath79 . in turn , this leads to the loss of the monochromaticity of the radiation formed in the undulator . typical scale , within which the parameters vary noticeably , is equal to the period @xmath3 of the surface deformations . as we demonstrated in the paper , one can choose the following two strategies to partly restore the monochromaticity of radiation . firstly , one can use thin crystals , @xmath320 . in this case , the variation of the amplitude over the width as well as the contribution of higher subharmonics do not lead to dramatic changes in the spectrum . however , this limit corresponds to very thin crystals , if one takes into account that the period of surface deformations lies within the range @xmath321 @xmath158 m @xcite . the second approach prescribes the use of a thick crystal but in combination with a narrow ( @xmath322 ) positron beam along the midplane . this limit seems to be achievable by using existing positron beams in the gev range and with the size ( along one direction ) of several microns ( see @xcite ) . to minimize the destructive role of the imperfect structure , one can also consider alternative schemes of the surface deformations . namely , if instead of just a periodic surface deformation one applies a _ harmonic _ surface deformation with a period @xmath3 , then the only imperfectness of the periodic bending in the bulk will be associated with the variation of the amplitude @xmath323 since for the amplitudes of subharmonics with @xmath324 will be identically equal to zero . this , in turn , will result in a much smaller variation of the undulator parameter and the bending parameter over the crystal width ( see eqs . ( [ formalism4.1 ] ) and ( [ formalism4.2 ] ) , and figures [ figure.py ] and [ figure.cy ] ) since this variation is to a great extent due to the contribution of higher subharmonics . to achieve the harmonic shape of the surface deformation one can either place a crystal between two press molds of the harmonic profile ( shifted by @xmath119 with respect to each other ) or apply a modulated pressure by means of two piezoelectric layers . the authors are grateful to kim kirsebom and ulrik uggerhj for constructive discussions . this work has been supported by the european commission ( the pecu project , contract no . 99 s. bellucci et al , phys . rev . ( 2003 ) 034801 . v. guidi , a. antonioni , s. baricordi , f. logallo , c. malag , e. milan , a. ronzoni , m. stefancich , g. martinelli , a. vomiero . inst . and meth . b 234 ( 2005 ) 40 . p. balling , k. kirsebom , d.q . le , u.i . uggerhj , s. ballestrero , s. connell , j. hrtwig , f. masiello . ` producing and investigating periodic deformations in silicon crystals ' . talk at the pecu meeting ( frankfurt am main , 2007 ) . korol , a.v . solovyov , w. greiner , j. phys . g 27 ( 2001 ) 95 . korol , a.v . solovyov , w. greiner , proc . spie - int . soc . opt . eng . 5974 ( 2005 ) 597405 . kaplin , s.a . vorobiev , phys . 67 a ( 1978 ) 135 . korol , a.v . solovyov , w. greiner , proc . spie - int . opt . eng . 6634 ( 2007 ) 66340p . baranov et al , nucl . inst . and meth . b 252 ( 2006 ) 32 . chung - jen lu , d.b . bogy , int . j. solid . structures 32 ( 1995 ) 1759 . v.n . baier , v.m . katkov and v.m . strakhovenko , high energy electromagnetic processes in oriented single crystals , world scientific , singapore,1998 . tsyganov , fermilab preprint tm-682 , tm-684 , fermilab , batavia , 1976 . kumakhov , f.f . komarov , radiation from charged particles in solids , aip , new york , 1989 . youn , c.g . kang materials science and engineering a 384 ( 2004 ) 275 ( erratum : ibid . 393 ( 2005 ) 398 ) .
the nuclear mean - field potentials obtained in the hartree - fock method with different choices of the in - medium nucleon - nucleon ( nn ) interaction have been used to study the equation of state ( eos ) of the neutron star ( ns ) matter . the eos of the uniform ns core has been calculated for the np composition in the-equilibrium at zero temperature , using version sly4 of the skyrme interaction as well as two density - dependent versions of the finite - range m3y interaction ( cdm3y and m3y - p ) , and versions d1s and d1n of the gogny interaction . although the considered effective nn interactions were proven to be quite realistic in numerous nuclear structure and/or reaction studies , they give quite different behaviors of the symmetry energy of nuclear matter at supranuclear densities that lead to the _ soft _ and _ stiff _ scenarios discussed recently in the literature . different eos s of the ns core and the eos of the ns crust given by the compressible liquid drop model have been used as input of the tolman - oppenheimer - volkov equations to study how the nuclear symmetry energy affects the model prediction of different ns properties , like the cooling process as well as the gravitational mass , radius , and moment of inertia .
1105.5222
the determination of the equation of state ( eos ) of asymmetric nuclear matter ( nm ) has been the main object of numerous nuclear structure and reaction studies involving unstable nuclei lying close to the neutron or proton driplines @xcite . the knowledge about the eos of asymmetric nm is vital for any model of neutron star @xcite , and the nuclear mean - field potential is the most important input for the determination of the nuclear eos . many microscopic studies of the eos have been done based on the nuclear mean field given by both nonrelativistic and relativistic nuclear many - body approaches , using realistic two - body and three - body nucleon - nucleon ( nn ) forces or interaction lagrangians ( see recent reviews @xcite ) . these many - body studies have shown the important role played by the pauli blocking effects as well as higher - order nn correlations at different nm densities . these medium effects are normally considered as the physics origin of the density dependence that has been introduced into various versions of the effective nn interactions used in the modern mean - field approaches . among them , very popular is the so - called m3y interaction which was originally constructed to reproduce the g - matrix elements of the reid @xcite and paris @xcite nn potentials in an oscillator basis . several realistic density dependences have been added later on to the m3y interactions @xcite to properly account for the nm saturation properties as well as the ground - state structure of finite nuclei @xcite . these density dependent versions of the m3y interaction have been used in the nonrelativistic hartree - fock ( hf ) studies of symmetric and asymmetric nm . some of them have been successfully used in the folding model studies of the nucleon - nucleus and nucleus - nucleus scattering @xcite . in attempt to find a realistic version of the effective nn interaction for consistent use in the mean - field studies of nm and finite nuclei as well as in the nuclear reaction calculations , we have performed recently a systematic hf study of nm @xcite using the cdm3y@xmath2 interactions , which have been used mainly in the folding model studies of the nuclear scattering @xcite , and the m3y - p@xmath2 interactions carefully parametrized by nakada @xcite for use in the hf studies of nuclear structure . for comparison , the same hf study has also been done with the d1s and d1n versions of the gogny interaction @xcite and sly4 version of the skyrme interaction @xcite . while these effective nn interactions give more or less the same description of the saturation properties of the symmetric nm , the hf results for the asymmetric nm @xcite show that they are divided into two families , which are associated with two different ( _ soft _ and _ stiff _ ) behaviors of the nm symmetry energy at high nucleon densities . as a result , these two families predict very different behaviors of the proton - to - neutron ratio in the @xmath1-equilibrium that can imply two drastically different mechanisms for the neutron star cooling ( with or without the direct urca process ) @xcite . as a further step in this direction , we try to find out in the present work how such a difference in the nm symmetry energy can affect the eos of the @xmath1-stable neutron star ( ns ) matter as well as the main ns properties like the maximum mass , radius , central density and moment of inertia . for this purpose , the tolman - oppenheimer - volkov ( tov ) equations have been solved using different eos s of the ns matter that are associated with the nuclear mean - field potentials given by different in - medium nn interactions under study . given the complex , inhomogeneous structure of the ns crust , it is a tremendous task to develop a consistent structure model for the inner and outer ns crusts using all versions of the in - medium nn interaction considered here . therefore , we have used the eos of the ns crust given by the compressible liquid drop model ( cldm ) @xcite with the model parameters determined by the sly4 interaction @xcite . different eos s of the uniform ns core are then calculated for the np@xmath0 composition in the @xmath1-equilibrium at zero temperature and extended to the supranuclear densities , using the mean - field potentials given by different density - dependent nn interactions . in this way , any difference found in the solutions of the tov equations is entirely due to the choice of the eos of the ns core , i.e. , to the choice of the in - medium nn interaction . the main ns properties obtained in each case are compared with the empirical data given by the recent astronomical observation of neutron stars . we recall here the main features of our hf study @xcite of the uniform ( spin - saturated ) nm at zero temperature that is characterized by given values of the neutron and proton densities , @xmath3 and @xmath4 , or equivalently by the total density @xmath5 ( hereafter referred to as the baryon density ) and the neutron - proton asymmetry @xmath6 . with the direct ( @xmath7 ) and exchange ( @xmath8 ) parts of the interaction determined from the singlet- and triplet - even ( and odd ) components of the central nn force , the total energy density of the nm is determined as @xmath9 , \label{e2}\ ] ] where @xmath10 are the ordinary plane waves . dividing @xmath11 over the total baryon number density @xmath12 , we obtain the total nm energy per particle @xmath13 that can be expressed as @xmath14 the nm pressure @xmath15 and incompressibility @xmath16 are then calculated as @xmath17 the contribution of @xmath18 and higher - order terms in eq . ( [ e3 ] ) was proven to be quite small @xcite and is often neglected in the so - called _ parabolic _ approximation , where the nm _ symmetry energy _ @xmath19 equals the energy required per particle to change the symmetric nm into the pure neutron matter . the value of @xmath19 at the symmetric nm saturation density , @xmath20 @xmath21 , is known as the symmetry energy coefficient @xmath22 that has been predicted by numerous many - body calculations to be around 30 mev @xcite . the knowledge about the density dependence of @xmath19 is extremely important for the construction of nuclear eos and it has been , therefore , a longstanding goal of many nuclear structure and reaction studies . the main method to probe @xmath19 associated with a given in - medium nn interaction is to test this interaction in the simulation of heavy - ion ( hi ) collisions using transport and/or statistical models @xcite or in the structure studies of nuclei with large neutron excess @xcite . based on the physics constraints implied by such studies , extrapolation is often made to draw conclusion on the low- and high - density behavior of @xmath19 . however , such conclusions still remain quite divergent in some cases @xcite . one of the most intriguing issues discussed recently in the literature is whether the soft " or stiff " density dependence of the nm symmetry energy is more realistic . these two scenarios are well illustrated in fig . [ f1 ] where the hf results obtained with the two groups of the considered in - medium nn interactions are plotted . given by the density - dependent nn interactions under study . the shaded ( magenta ) region marks the empirical boundaries deduced from the analysis of the isospin diffusion data and double ratio of neutron and proton spectra data of hi collisions @xcite . the square and triangle are the constraints deduced from the consistent structure studies of the gdr @xcite and neutron skin @xcite , respectively . the circles and crosses are results of the ab - initio calculation by akmal , pandharipande and ravenhall ( apr ) @xcite and microscopic monte carlo ( mmc ) calculation by gandolfi _ @xcite , respectively.,scaledwidth=70.0% ] for comparison , we have also plotted in fig . [ f1 ] results of the ab - initio variational chain - summation calculation using the a18+@xmath23+uix * version of the argonne nn potential by akmal , pandharipande and ravenhall ( apr ) @xcite and recent microscopic monte carlo calculation of neutron star structure by gandolfi _ et al . _ @xcite , using the argonne av6 potential added by an empirical density dependence . around the saturation density @xmath24 of the symmetric nm all the models predict the symmetry coefficient @xmath25 mev , in a reasonable agreement with the empirical values deduced recently from the structure studies of of neutron skin @xcite . in the low - density region ( @xmath26 ) there exist empirical boundaries for the symmetry energy deduced from the analysis of the isospin diffusion data and double ratio of neutron and proton spectra data of hi collisions @xcite , which enclose the hf results given by both groups of the in - medium nn interactions . at the baryon density @xmath27 @xmath21 , all the hf results also agree quite well with the empirical value deduced from a consistent structure study of the isovector giant dipole resonance ( gdr ) in heavy nuclei @xcite . so far there are no firm empirical constraints on the nm symmetry energy at supranuclear densities , and the behavior of @xmath19 at high densities remains uncertain . the two different behaviors of @xmath19 shown in fig . [ f1 ] have been observed earlier @xcite and often discussed in the literature as the _ asy - stiff _ ( with symmetry energy steadily increasing with density ) and _ asy - soft _ ( with symmetry energy reaching saturation and then decreasing to negative values ) behaviors . ) - ( [ e3 ] ) using the in - medium nn interactions that give a _ stiff _ behavior of @xmath19 as shown in upper panel of fig . [ f1 ] . the circles and crosses are results of the ab - initio calculation by akmal , pandharipande and ravenhall ( apr ) @xcite and microscopic monte carlo ( mmc ) calculation by gandolfi _ @xcite , respectively.,scaledwidth=70.0% ] the main characteristics of the eos s obtained with these two groups of density - dependent nn interactions have been discussed in details in ref . we note here that most of the microscopic calculations of nm like the ab - initio apr results @xcite or the monte carlo calculation by gandolfi _ @xcite do predict a stiff behavior of the nuclear symmetry energy , excepting perhaps the microscopic study by wiringa _ _ @xcite that predicted a soft behavior of @xmath19 using the argonne or urbana nn interaction plus a three - nucleon interaction term . the stiff behavior is also predicted by the recent microscopic brueckner - hartree - fock ( bhf ) or dirac - brueckner - hartree - fock calculations of nm that include the higher - order many - body correlations and/or three - body forces @xcite as well as by the latest relativistic mean - field studies @xcite . but obtained with the in - medium nn interactions that give a _ soft _ behavior of @xmath19 as shown in lower panel of fig . [ f1].,scaledwidth=70.0% ] because the isovector density dependence of the cdm3y@xmath2 interactions has been parametrized @xcite to reproduce simultaneously the bhf results for the isospin- and density dependent nucleon optical potential by the jlm group @xcite and the charge exchange data for the isobaric analog excitation @xcite , the _ stiff _ behavior of @xmath19 given by the cdm3y@xmath2 interactions is quite close to that given by the bhf calculation . if we simply assume the density dependence of the isovector part of cdm3y@xmath2 interactions to be the same as that of the isoscalar part , then @xmath19 has a _ soft _ behavior that has been discussed in our earlier hf study @xcite . on the other hand , the isovector density dependence of the m3y - p@xmath2 , d1s and d1n interactions were carefully fine tuned against the structure data observed for a wide range of the neutron ( and proton- ) dripline nuclei and the low - density tail of @xmath19 predicted by these interactions should be quite realistic . however , there is no physics ground to confirm the validity of the high - density behavior of @xmath19 predicted by the _ soft _ m3y - p@xmath2 , d1s and d1n interactions . for the total nm energy ( [ e3 ] ) that is sometimes referred to as the nuclear eos , the hf results given by all considered interactions for the symmetric nm agree well with the microscopic apr or monte carlo predictions . however , the difference in the symmetry energy lead to very different behaviors of the energy of pure neutron matter given by the stiff and soft groups of interactions ( see lower panels of figs . [ f2 ] and [ f3 ] ) . in terms of the nm pressure ( [ e4 ] ) the soft - type interactions have been shown @xcite unable to reproduce the empirical pressure @xmath28 of the pure neutron matter deduced from the hi flow data @xcite . in the present work we make such a comparison more accurately , based on the mean - field prediction for the eos of the ns matter of the np@xmath0 composition in the @xmath1-equilibrium and the recent empirical data deduced from astrophysical measurements of neutron stars by zel _ _ @xcite and steiner _ et al . nuclei in the ns crust are described by the compressible liquid drop model by douchin _ _ ( see ref . @xcite and references therein ) , using the model parameters determined with the version sly4 of the skyrme interaction @xcite . within the cldm , electrons inside the inhomogeneous ns crust are assumed to form a relativistic fermi gas . the structure of the ns crust has been given by the cldm for the baryon densities up to the _ edge _ density @xmath29 @xmath21 , where a weak first - order phase transition between the ns crust and liquid ( uniform ) core takes place @xcite . at baryon densities @xmath30 the ns core is described as a homogeneous matter of neutrons , protons , electrons and negative muons ( @xmath31 appear at @xmath12 above the muon threshold density , where electron chemical potential @xmath32 mev ) . such a np@xmath0 composition of the ns core is a realistic assumption up to the high densities of @xmath33 . although the appearance of hyperons can be expected at higher densities , douchin and haensel have extrapolated their np@xmath0 model for the eos of the ns matter up to the maximum central density ( approach used earlier in the ab - initio study by akmal _ thus , the total energy density @xmath11 of the np@xmath0 matter ( including the rest energy of baryons and leptons ) is determined in the present study as @xmath34 where @xmath35 is the hartree - fock energy density of nucleons ( [ e2 ] ) ; @xmath36 and @xmath37 are the energy densities of electrons and muons , respectively , which are evaluated in the relativistic fermi gas model , neglecting electrostatic interaction @xcite . the number densities of leptons , @xmath38 and @xmath39 , are determined from the charge neutrality condition ( @xmath40 ) and the relation for the chemical potentials , implied by the @xmath1-equilibrium of the ( neutrino - free ) ns matter @xmath41 as a result , we can determine uniquely all fractions of the constituent particles @xmath42 at the given baryon density @xmath5 . below the muon threshold density ( @xmath43 mev ) the charge neutrality condition leads to the following relation @xcite @xmath44 using the total nm energy @xmath13 given by the hf calculation ( [ e3 ] ) , the density dependence of the proton fraction at the @xmath1-equilibrium , @xmath45 , is readily obtained from the solution of eq . ( [ e7 ] ) . if we assume the parabolic approximation and neglect the contribution from higher - order terms in ( [ e3 ] ) , then @xmath45 is given by the solution of the well - known equation @xcite @xmath46 ^ 3=0 , \label { e8}\ ] ] which shows the crucial role of the nm symmetry energy in the determination of the proton abundance in the ns matter . as the baryon density exceeds the muon threshold density , where @xmath47 mev , it is energetically favorable for electrons to convert to negative muons and the charge neutrality condition leads now to the relation @xcite @xmath48^{3/2}\theta(\hat\mu - m_\mu c^2)=0 , \label { e9}\ ] ] where @xmath49 is the heaviside step function . based on the solutions of eqs . ( [ e7 ] ) and ( [ e9 ] ) , the eos of the np@xmath0 matter is fully determined by the mass density @xmath50 and total pressure @xmath28 inside the neutron star @xmath51+p_e+p_\mu . \label { e10}\ ] ] of constituent particles of the ns matter obtained from the solutions of eqs . ( [ e7 ] ) and ( [ e9 ] ) using the mean - field potentials given by the m3y - p5 and d1n interactions.,scaledwidth=110.0% ] in the present study , we have first solved eqs . ( [ e7 ] ) and ( [ e9 ] ) to determine all fractions @xmath42 and the eos of the np@xmath0 matter using the total baryon energy @xmath52 given by the sly4 interaction . the accuracy of the numerical procedure was double checked against the published results for @xmath53 and @xmath28 , obtained with the sly4 interaction by douchin and haensel ( tabulated in refs . different eos s of the ns core have been then calculated for the np@xmath0 matter , using the hf mean - field energies given by different in - medium nn interactions considered in our study . but using the mean - field potentials given by the cdm3y4 and sly4 interactions . the circles are @xmath54 values calculated at the maximum central densities @xmath55 given by the solution of the tov equations.,scaledwidth=110.0% ] our results for the fractions @xmath42 obtained with different interactions are plotted in figs . [ f4 ] and [ f5 ] . for the typical soft - type interactions ( like the results obtained for the m3y - p5 and d1n interactions shown here ) , the proton and lepton fractions are quite small and reach their maxima of around 4% at @xmath56 @xmath21 ( see lower panels of fig . [ f4 ] ) , at exactly the same baryon density where the symmetry energy @xmath19 goes through its maximum value . the fast decrease of @xmath19 to zero at @xmath57 @xmath21 leads also to a drastic decrease of the proton and lepton components in the ns matter that then becomes _ @xmath1-unstable _ , pure neutron matter at @xmath58 @xmath21 ( see upper panels of fig . [ f4 ] ) . the fractions @xmath42 obtained with the stiff - type interactions ( see the results obtained for the cdm3y4 and sly4 interactions shown in fig . [ f5 ] ) are substantially different from those given by the soft - type interactions . namely , the proton and lepton fractions increase steadily with the baryon density like the corresponding symmetry energy @xmath19 . for the cdm3y@xmath2 interactions , the proton fraction @xmath59 is above 30% at the maximum central density @xmath60 @xmath21 , and the matter at the ns center becomes less neutron rich ( with @xmath61 ) . in this case , the @xmath1-equilibrium of the charge neutral ns matter is kept throughout the ns core , with the lepton fractions reaching more than 30% at @xmath55 ( see left - lower panel of fig . it is interesting to note that about the same behavior of @xmath59 is also predicted by the recent complete eos of nuclear matter by shen _ @xcite , constructed for use in astrophysical simulations . for the sly4 interaction , with a less stiff increase of @xmath19 at large densities ( see upper panel of fig . [ f1 ] ) , the maximum proton fraction at @xmath62 @xmath21 is only about 12% and the ns matter is , therefore , more neutron rich compared to the case of cdm3y@xmath2 interactions . nevertheless , in the case of sly4 interaction the ns matter remains always in the @xmath1-equilibrium @xcite . of the @xmath1-stable ns matter obtained from the solutions of eqs . ( [ e7 ] ) and ( [ e9 ] ) using the mean - field potentials given by the stiff - type cdm3y@xmath2 and sly4 interactions . the circles are @xmath4 values calculated at the maximum central densities @xmath55 given by the tov equations . the thin lines are the corresponding du thresholds ( [ e11 ] ) . , scaledwidth=88.0% ] as already discussed in ref . @xcite , the behavior of the density dependence of the proton fraction @xmath45 plays a very important role in the determination of the ns cooling rate . in particular , the powerful direct urca ( du ) process of neutrino emission is allowed only if the fermi momenta of the constituent particles in the np@xmath0 matter satisfy the triangle conditions @xcite that lead to the existence of a du threshold @xmath63 for the proton fraction that can be estimated @xcite as @xmath64 where @xmath65 is the leptonic electron fraction . @xmath63 has its lowest value of 11.1% at @xmath66 that corresponds to the muon - free threshold for du process @xcite . it can be concluded immediately from lower panel fig . [ f4 ] that the du process is _ not _ possible for the ns matter generated with the soft - type interactions when @xmath59 can reach at most 4% and then decreases quickly to zero at @xmath58 @xmath21 . such a ( mean - field ) challenge to the soft - type in - medium nn interactions has been pointed out and discussed in ref . [ f6 ] shows that for the stiff - type cdm3y@xmath2 interactions , the proton fraction becomes larger than the du threshold at a rather modest threshold density of @xmath67 @xmath21 that is far below the corresponding maximum central densities of @xmath68 @xmath21 . therefore , the du process should be a realistic scenario for the ns cooling if the mean - field potential is generated with the cdm3y@xmath2 interactions . for the sly4 interaction , the @xmath59 value remains well below the du threshold up to the maximum central density determined by the tov equations and the du process is thus not likely for the ns matter generated with the sly4 interaction @xcite . , in comparison with the empirical data points deduced from the astronomical observation of neutron stars @xcite . the shaded band shows the uncertainties associated with the data determination . the circles are @xmath15 values calculated at the corresponding maximum central densities given by the tov equations , and the vertical arrows indicate the baryon densities above which the ns matter predicted by the m3y - p3 and m3y - p4 interactions becomes superluminal ( see fig . [ f10 ] below).,scaledwidth=70.0% ] we discuss now our results obtained for the total pressure @xmath28 inside the neutron star determined by relation ( [ e10 ] ) , which defines the eos of the ns matter to be used in the tov equations . the ns pressure obtained with the hf mean - field energies @xmath52 given by different in - medium nn interactions are plotted in fig . [ f7 ] . since the lepton pressure ( @xmath69 ) inside the ns is about one order of magnitude weaker than the baryon pressure @xmath70 , the results shown is fig . [ f7 ] are determined predominantly by @xmath70 . one of the main constraints for the ns matter is that the pressure must satisfy relation @xmath71 to ensure the ns matter stability @xcite . this ( microscopic ) stability condition also is known as le chatelier s principle @xcite . one can see in lower panel of fig . [ f7 ] that @xmath15 obtained with the d1s version of gogny interaction does not comply with such a constraint and this interaction should not be used to generate the eos of the ns matter at densities @xmath72 . this results stresses again that there could be a plethora of systematic uncertainties in different models of in - medium nn interaction that are not visible at low nuclear densities , and the success of any interaction in the nuclear structure study is not sufficient to ensure its extrapolation to supranuclear densities . with the advance in both the astrophysical techniques and modeling of the ns structure , it became recently feasible to empirically deduce the pressure of the ns matter at supranuclear densities @xcite . the empirical pressure determined from the masses and radii observed for the binaries 4u 1608 - 248 , exo 1745 -248 , and 4u 1820 - 30 are plotted in fig . [ f7 ] as the three data points spanned by a shadow region of the uncertainties associated with the data determination @xcite . from fig . [ f7 ] one can see that both groups of in - medium nn interaction agree more or less with the data for ns pressure , while a similar comparison @xcite of the pressure calculated for pure neutron matter with the empirical value deduced from the hi flow data @xcite seemed to favor the stiff - type interactions . in order to distinguish more clearly the eos s given by these two groups of interaction , we have further used them in the input of the tolman - oppenheimer - volkov equations to study the main ns properties different sets of predicted mass density @xmath73 and total pressure @xmath15 inside the neutron star ( [ e10 ] ) have been further used to solve the well - known tolman - oppenheimer - volkov equations @xmath74 where @xmath75 is the universal gravitational constant , @xmath76 is the radial coordinate in the schwarzschild metric , and @xmath77 is the gravitational mass enclosed within the sphere of radius @xmath76 . the tov equations ( [ e12 ] ) are supplemented with the following equation determining the number of baryons @xmath78 inside this sphere @xcite @xmath79 eqs . ( [ e12 ] ) , ( [ e13 ] ) have been integrated from the ns center , with the boundary conditions at @xmath80 , and @xmath81 . the stellar surface at @xmath82 is determined from the boundary condition @xmath83 . the total gravitational mass and total number of baryons are then determined as @xmath84 , respectively . as a result , with different inputs for the nm pressure , the corresponding solutions of the tov equations give different ns models in terms of _ one - parameter _ families @xcite that can be labeled by the central pressure @xmath85 or equivalently by the central density @xmath86 of the neutron star . thus , at each central density we can uniquely determine the corresponding gravitational mass @xmath87 and radius @xmath88 , and the behavior of @xmath87 versus @xmath88 is often used to compare with the measured masses and radii of neutron stars . our results obtained with different eos s are plotted in fig . the recently measured masses and radii for the binaries 4u 1608 - 248 , exo 1745 -248 , and 4u 1820 - 30 @xcite are plotted in fig . [ f8 ] as the shaded contours . one can see that all mass - radius curves lie well below the limit allowed by the general relativity @xcite and go through or very closely nearby the empirical contours spanned by the data , excepting the curve given by d1n version of the gogny interaction that lies well below the data . , but in comparison with the empirical data ( shaded contours ) deduced by steiner _ @xcite from the observation of the x - ray burster 4u 1608 - 52.,scaledwidth=70.0% ] we note that the same eos for the ns crust ( generated in the cldm using model parameters determined by the sly4 interaction @xcite ) has been used in the input of the tov equations ( [ e12 ] ) . therefore , the difference found between the mass - radius curves shown in fig . [ f8 ] is entirely due to the different choices of the in - medium nn interaction used to generate the eos of the ns core . the maximum gravitational masses @xmath89 given by the eos s under study are plotted in fig . [ f8 ] as solid circles and they agree more or less with the recent data . while all the stiff - type interactions give the corresponding ns radius @xmath90 quite close to the empirical range around 10 km , the @xmath89 values given by the sly4 and cdm3y6 interactions are slightly higher than the observed masses , close to about twice the solar mass ( @xmath91 ) . we note , however , that the ns matter in the present study has been assumed to consist only of baryons , electrons and muons . at high baryon densities ( @xmath92 ) the hyperons are expected to appear , and the maximum ns mass becomes then smaller @xcite . in this case , the nucleon matter generated with the sly4 or cdm3y6 interactions seems well suitable for the description of the baryon component of the ns matter , while the inclusion of hyperons might pull the @xmath89 values given by the soft - type m3y - p@xmath2 interaction to values lying below the empirical boundaries shown in fig . [ f8 ] for the maximum ns mass . moreover , the @xmath89 values around @xmath93 are still allowed by a broad systematics of the measured ns masses @xcite . for the d1n interaction , the @xmath89 value is still within the broad range of the measured ns masses @xcite but the radius @xmath90 is rather small , almost 1 km smaller than the minimum limit for @xmath88 versus @xmath87 : @xmath94 km @xcite . we stress again that the existing data for the ns mass still allow a wide range for the realistic @xmath89 value , from the lowest value of 1.25 @xmath91 @xcite up to around 2 @xmath91 @xcite , and the comparison of the predicted results _ for both mass and radius _ is , therefore , vital in testing different eos s of the ns matter . as another example , we have also compared in fig . [ f8a ] the results predicted by different eos s with the mass - radius data deduced by steiner _ @xcite from the observation of the type - i x - ray burster 4u 1608 - 52 . one can see that results given by the eos s obtained with the stiff - type interactions agree nicely with the empirical data , while those given by the soft - type interactions clearly disagree with the considered mass - radius data . because the tov equations are deduced from the einstein s general relativistic equations for a _ gravitationally _ bound star in the hydrostatic equilibrium , the results shown in fig . [ f8 ] indicate that the eos s given by both groups of the in - medium nn interactions ) . the circles are values calculated at the maximum central densities.,scaledwidth=70.0% ] , but in comparison with the empirical data ( shaded contours ) deduced from the mass - radius data by steiner _ et al _ @xcite for the x - ray burster 4u 1608 - 52 using eq . ( [ e14]).,scaledwidth=70.0% ] ( excepting perhaps the d1n interaction ) can describe reasonably the empirical mass - radius data deduced by zel _ et al _ @xcite , despite drastically different proton fractions predicted for the ns matter at supranuclear densities that lead to very different scenarios for the ns cooling as discussed in the previous section . nonetheless , the comparison of our results with the mass - radius data deduced by steiner _ @xcite seems to prefer the stiff - type interactions . the moment of inertia @xmath95 of slowly rotating neutron star has been shown @xcite as a good constraint for the eos of neutron stars and the physics of their interiors . in general , @xmath95 can be determined from the einstein field equations for a compact star @xcite . it has been shown @xcite that for slowly rotating neutron stars the expression for @xmath95 determined from the einstein field equations can be approximated quite well by the following empirical relation of the ns mass and radius @xmath96 . \label { e14}\ ] ] assuming the validity of relation ( [ e14 ] ) , we have transformed the recent mass - radius data by zel _ et al _ @xcite and steiner _ et al _ @xcite into the empirical boundaries for realistic values of the moment of inertia . these new empirical data " for the ns moment of inertia @xmath95 are plotted in figs . [ f9 ] and [ f9a ] by similar shaded contours as in figs . [ f8 ] and [ f8a ] , and compared with the results obtained with different eos s using relation ( [ e14 ] ) . in this way , it is natural to see that the eos giving the best agreement with the mass - radius data also give the corresponding @xmath95 curves agreeing well with the empirical data for the moment of inertia . we conclude from the results shown in figs . [ f8]-[f9a ] that the stiff - type nn interactions give consistently good description to both sets of the empirical data for the ns masses , radii and moments of inertia . the main characteristics of the ns configuration determined from the tov equations ( [ e12 ] ) , using different eos s , are given in table [ t1 ] . as noted above , the eos determined by d1s version of the gogny interaction gives _ negative _ pressure at high baryon densities and violates , therefore , the main constraint for a gravitationally bound star @xcite in the hydrostatic equilibrium . as a result , the eos given by the d1s interaction can not be used in the tov equation ( [ e12 ] ) . from table [ t1 ] one can see that the stiff - type cdm3y@xmath2 and sly4 interactions give the maximum gravitational mass @xmath97 and radius @xmath98 km that are well within the established empirical boundaries as shown in figs . [ f8 ] and [ f8a ] . we note that both the ab - initio apr calculation @xcite and microscopic monte carlo study @xcite have obtained @xmath99 and the corresponding moment of inertia @xmath100 is , therefore , also somewhat larger than @xmath100 values given by the stiff - type interactions considered here . the soft - type m3y - p@xmath2 and , especially , d1n interactions give the @xmath89 and @xmath90 values significantly lower than those given by the stiff - type interactions . therefore , if hyperons are included at high baryon densities , the @xmath89 and @xmath90 values given by the soft - type interactions could be well below all the existing empirical estimates . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] given by different eos s of the ns matter plotted versus the corresponding total baryon mass @xmath101 . the shaded rectangle is the empirical value inferred from observations of the double pulsar psr j0737 - 3059 by podsiadlowski _ _ @xcite.,scaledwidth=88.0% ] from solutions of the tov equations at a given radius @xmath88 , the total baryon mass enclosed within the sphere of radius @xmath88 can be determined as @xmath102 , where @xmath103 is the nucleon mass and the total baryon number @xmath104 is given by eq . ( [ e13 ] ) . as a result , the @xmath105 values are closely correlated with the corresponding gravitational mass @xmath106 , and difference between these two masses depends upon the compactness of the neutron star . furthermore , the low - mass part of the dependence of gravitational mass on the total baryon mass , @xmath107 , can be directly compared with the constraint suggested by podsiadlowski _ namely , in a likely scenario that the massive component of the double pulsar psr j0737 - 3059 ( the lightest ns observed to date ) has been formed by an electron - capture supernova , the total pre - collapse baryon number of the stellar core ( rather precisely known from the model calculations ) has a very small loss of material in the subsequent collapse . therefore , the accurate gravitational mass @xmath108 estimated from the pulsar timing can be used to extract a stringent constraint on the low - mass part of @xmath107 , which is plotted as the shaded rectangle in fig . [ f9b ] . as discussed also in ref . @xcite , a realistic eos of the ns matter should give @xmath107 curve going through or nearby the shaded box in fig . [ f9b ] in order to be consistent with the observation of the double pulsar psr j0737 - 3059 . one can see that the stiff - type nn interactions give @xmath107 curves passing very close to the left corner of the shaded box , while those given by the soft - type interactions clearly underestimate the empirical data . at @xmath109 , the baryon density ( @xmath110 ) is well above the hyperon threshold and the gravitational mass should become smaller if the hyperons are included @xcite . this would probably push the @xmath107 curves given by the stiff - type interactions right into the shaded box in fig . in contrast , there is no physics mechanism that can push the @xmath107 curves given by the soft - type interactions higher up , to be close to the empirical shaded region . another important ns observable is the surface redshift of photons emitted from the ns photosphere that is determined as @xcite @xmath111 it is obvious from eq . ( [ e15 ] ) that the measurement of @xmath112 is vital for the determination of the mass / radius ratio . at the typical masses around @xmath113 , the stiff - type cdm3y@xmath2 interactions give @xmath114 while the soft - type m3y - p@xmath2 interactions give @xmath115 , which agree reasonably with the empirical value @xmath116 deduced from the x - ray spectra of the burster exo 0748 - 676 @xcite . although the maximum @xmath112 value given by the d1n interaction is close to that empirical data , the maximum ns mass is only around @xmath117 , and this effect is also showing up in a poor agreement of the d1n results with the mass - radius data as illustrated in figs . [ f8 ] and [ f9 ] . for the ns binding energy , we have used the standard definition of @xmath118 as the mass defect with respect to an unbound configuration consisting of the same baryon number @xcite . namely , the mass defect with respect to a dispersed configuration of a pressureless cloud of @xmath119fe dust , with mass per nucleon @xmath120 = mass of @xmath119fe atom divided by 56 @xmath121 at the typical masses around @xmath122 , all considered interactions give @xmath118 values within the range of @xmath123 mev . exception again is the d1n interaction that give a weaker binding energy of @xmath124 mev with the maximum gravitational mass @xmath125 . ) , and the vertical arrows indicate the baryon densities above which the ns matter predicted by the m3y - p3 and m3y - p4 interactions becomes superluminal ( see details in the text).,scaledwidth=70.0% ] concerning the maximum central pressure , the tov equations using the eos s based on the soft - type m3y - p@xmath2 interactions give much too high @xmath85 values ( see table [ t1 ] ) . the behavior of the central pressure is directly correlated with the central density . in this context , it is of interest to consider the _ causality _ condition @xcite that implies the adiabatic sound velocity in the stellar medium to be _ subluminal _ , i.e. , @xmath126 where @xmath127 is the total pressure of the ns matter as function of the total mass density @xmath73 . we have estimated the sound velocity @xmath128 using the eos s given by the two groups of the in - medium nn interactions and the results are plotted in fig . while the soft - type m3y - p@xmath2 interactions give more or less the same @xmath128 values at low densities as those given by the stiff - type cdm3y@xmath2 interactions , eos given by the m3y - p3 and m3y - p4 interactions begins to violate the causality condition at @xmath129 and 1.64 @xmath21 , respectively . for further use in the tov equations , we have assumed in these two cases a causal eos @xcite given by @xmath130 at the @xmath12 larger than 1.6 and 1.64 @xmath21 , respectively , with constant @xmath131 chosen to ensure the continuity of the energy density across the critical densities . the m3y - p5 interaction is doing better and gives @xmath132 at the maximum central density . for the stiff - type sly4 interaction , we obtained @xmath133 at the maximum @xmath12 , and this value indicate a sound velocity closely comparable to velocity of light in the dense ns core , as discussed earlier in ref . @xcite . in conclusion , we have tested two sets of the in - medium nn interactions in the description of the main properties of neutron star based on the tov equations . we found that all stiff - type interactions are quite realistic in describing the latest empirical constraints for the eos of the ns matter , pressure and mass - radius data . in particular , the cdm3y6 and sly4 interactions should be the appropriate choice for the future ns studies , when hyperon presence is taken into account at supranuclear densities . concerning the soft - type interactions , the eos s given the m3y - p3 and m3y - p4 interactions have been modified at high densities to avoid the violation of the causality condition . moreover , the overall agreement of the results given by the soft - type interactions with the same empirical data is not as good as that obtained with the stiff - type interactions . it is clear that such effects are caused not only by a drastic difference in the nuclear symmetry energy alone , but also by quite different functional structures of the considered interactions , like , e.g. , the _ zero - range _ density - dependent form of the m3y - p@xmath2 and gogny interactions versus the _ finite - range _ form of the cdm3y@xmath2 interactions . to explore explicitly the effects caused by the nuclear symmetry energy to the ns properties in this kind of study , one needs to make a similar analysis with the hf energy densities ( [ e2 ] ) obtained essentially with the same in - medium nn interaction but using different ansatzs for its isospin dependence so that different behaviors of the symmetry energy @xmath19 can be tested . we present here the results of such a test using the cdm3y@xmath2 interactions that have the isoscalar ( is ) and isovector ( iv ) parts determined as @xmath134 where @xmath135 is the internucleon distance . the radial strengths of the is and iv interactions @xmath136 were kept unchanged , as derived from the m3y - paris interaction @xcite , in terms of three yukawas ( see detailed formulas in ref . the isoscalar density dependence @xmath137 has been parametrized @xcite to reproduce the saturation properties of symmetric nm in the hf calculation , while the isovector density dependence @xmath138 has been parametrized @xcite to reproduce the bhf results for the isospin- and density dependent nucleon optical potential in the nuclear matter limit @xcite . thus , the stiff behavior of @xmath19 given by the cdm3y@xmath2 interactions is actually associated with that given by the bhf calculation . in our earlier hf study @xcite we have simply assumed the iv density dependence to be the same as that of the is part , and @xmath19 has then a typical soft behavior that is illustrated in fig . [ f11 ] . the same assumption has also been used in the nm studies by basu _ _ @xcite using the density dependent version ddm3y of the m3y - reid interaction @xcite . but @xmath19 curves in the lower panel were obtained with the soft cdm3y@xmath2s interactions.,scaledwidth=80.0% ] to study the effects caused by the stiffness of the symmetry energy , we have used in the present work also a soft version of the cdm3y@xmath2 interactions with the iv density dependence taken as @xmath139 . the factor of 1.1 has been found @xcite to give the best fit of the charge exchange @xmath140he@xmath141li data as well as realistic value of the symmetry coefficient @xmath142 mev . these soft cdm3y@xmath2 interactions ( denoted hereafter as the cdm3y@xmath2s interactions ) give a more moderate _ soft _ behavior of the symmetry energy compared to the m3y - p@xmath2 interactions ( compare figs . [ f1 ] and [ f11 ] ) . but using the mean - field potentials given by the cdm3y6 and cdm3y6s interactions.,scaledwidth=110.0% ] the soft cdm3y@xmath2s interactions have been further used to determine the composition of the @xmath1-stable ns matter by solving eqs . ( [ e7 ] ) and ( [ e9 ] ) , and the fractions @xmath42 obtained with the soft cdm3y6s interaction are compared with those given by the stiff cdm3y6 version in fig . as expected , the proton and lepton fractions obtained with the cdm3y6s interaction are much smaller than those obtained with the stiff cdm3y6 interaction . the fast decrease of @xmath19 to zero at @xmath143 @xmath21 leads also to the disappearance of protons and leptons in the ns matter that soon becomes the @xmath1-unstable , pure neutron matter at @xmath144 @xmath21 . like the results obtained with the soft - type m3y - p@xmath2 interactions , the du scenario of the ns cooling should also be excluded for the ns matter generated with the soft cdm3y@xmath2s interactions , because @xmath59 can reach only around 5 - 6% and then decreases quickly to zero at @xmath145 @xmath21 . but @xmath28 curves in the lower panel were obtained with the soft cdm3y@xmath2s interactions.,scaledwidth=80.0% ] the pressures of the ns matter ( [ e10 ] ) obtained with the two sets of the cdm3y@xmath2 interaction are compared with the empirical data in fig . [ f13 ] , and one can see that the soft cdm3y@xmath2s interactions fail to account for the empirical ns pressure at high densities . this effect is well expected because @xmath28 is determined from the first derivative of the nm energy and the decrease of the nm symmetry energy @xmath19 at high densities leads to a negative contribution of the symmetry term of the nm pressure to the total @xmath28 value . in other words , the eos s given by the two sets of the cdm3y@xmath2 interaction are substantially different at high baryon densities , and this effect is entirely due to the different behaviors of the symmetry energy at high baryon densities . but mass - radius curves in the lower panel were obtained with the soft cdm3y@xmath2s interactions.,scaledwidth=80.0% ] but mass - radius curves in the lower panel were obtained with the soft cdm3y@xmath2s interactions.,scaledwidth=80.0% ] the mass density @xmath73 and total pressure @xmath15 of the ns matter ( [ e10 ] ) obtained with the _ soft _ cdm3y@xmath2s interactions have been further used as input of the tov equations ( [ e12 ] ) , and the obtained ns properties are given in table [ t1 ] and illustrated in figs . [ f14 ] and [ f15 ] . the most obvious effect caused by changing slope of the symmetry energy from _ stiff _ to _ soft _ is the reduction of the maximum gravitational mass @xmath89 and radius @xmath90 as illustrated in fig . the @xmath89 value is changing from @xmath146 to a significantly lower range of @xmath147 , with a much worse description of the empirical mass - radius data @xcite as shown in figs . [ f14 ] and [ f15 ] . then , if the hyperons are included at high baryon densities , the @xmath89 values given by the soft cdm3y@xmath2s interactions can be driven to values lying well below the mass of the lightest ns observed so far ( @xmath148 ) @xcite . together with the maximum central pressure @xmath85 , the total baryon number , surface redshift , binding energy and moment of inertia also become smaller when the eos of the ns core is obtained in the hf calculation using the soft cdm3y@xmath2s interactions . it is interesting to note that the eos s obtained with the soft cdm3y3s and cdm3y4s interactions not only give the ns mass and radius values lying outside the empirical boundaries but also the surface redshifts @xmath112 clearly in disagreement with the data ( @xmath116 ) deduced from the x - ray burst spectra of neutron stars @xcite . these results together with those obtained with the soft - type m3y - p@xmath2 and d1n interactions discussed above indicate that a _ soft _ density dependence of the nm symmetry energy could cause strong deviation of the calculated ns properties from the empirical estimates . although one still can not completely rule out the eos of the ns matter with a _ soft _ behavior of the symmetry energy based on this discussion , the results shown in figs . [ f13]- [ f15 ] clearly favor the _ stiff _ behavior and confirm again the vital role of the nuclear symmetry energy in the theoretical modeling of neutron star . the eos of the np@xmath0 matter of neutron star in @xmath1-equilibrium has been studied in details using the nuclear mean - field potentials obtained in the hf method with different choices of the effective ( in - medium ) nn interactions that give two different behaviors of the nm symmetry energy at supranuclear densities ( the _ soft _ and _ stiff _ scenarios ) . the fast decrease of the soft nm symmetry energy to zero at @xmath57 @xmath21 results on a drastic decrease of the proton and lepton components in the uniform ns core that then becomes the _ @xmath1-unstable _ , pure neutron matter at @xmath58 @xmath21 . very small proton fraction in the ns matter given by the soft - type interactions excludes the direct urca process in the ns cooling , whereas the du process is well possible for the _ @xmath1-equilibrated _ ns matter predicted by the stiff - type cdm3y@xmath2 interactions . the ns pressure obtained with different in - medium nn interactions are compared with the empirical ns pressure deduced from the recent astronomical observation . in general , the eos given by the soft - type interactions tend to give pressure lower the empirical values at high densities . in particular , the d1s version of the gogny interaction gives negative pressure at baryon densities @xmath72 and violates , therefore , le chatelier s principle that ensures the ns stability . the adiabatic sound velocity estimated from the ns pressure given by two versions of the soft - type m3y - p@xmath2 interaction becomes superluminal at high baryon densities and violates , therefore , the causality condition , and the eos has been corrected by hand in this case for further use in the tov equations . it seems , therefore , likely that there could be uncertainties in a well parametrized effective ( density dependent ) nn interaction that are not visible at low nuclear densities , and its success in the nuclear structure study is not sufficient to validate its extrapolation to supranuclear densities . different eos s of the ns core supplemented by the sly4 eos of the ns crust given by the compressible liquid drop model have been used for the input of the tolman - oppenheimer - volkov equations to study how different behaviors of the symmetry energy affect the model prediction of the ns properties . the eos s obtained with the stiff - type interactions were found to give consistently reasonable description of the empirical data for the ns mass and radius , and to comply well with the causality condition . in comparison with the same empirical ns data , the soft - type interactions were found less successful , especially , the two versions of the famous gogny interaction certainly need an appropriate modification before they can be used in the tov equations to study structure of neutron star . the vital role of the nm symmetry energy has been demonstrated in our specific test of the cdm3y@xmath2 interactions , where we found a significant reduction of the maximum gravitational mass @xmath89 and radius @xmath90 away from the empirical boundaries when the slope of the nm symmetry energy is changed from the _ stiff _ behavior to the _ soft _ one . it is natural to expect that if hyperons ( and other hypothetical constituents like kaons or quark matter ) are included at high baryon densities , the @xmath89 and @xmath90values given by the soft - type interactions could be driven to region lying well below all existing empirical estimates . the present research has been supported by the national foundation for scientific and technological development ( nafosted ) under project nr . the first two authors gratefully acknowledge the financial support from ipn orsay and the lia fvppl programme for their short research stays at ipn orsay in 2010 . dtk thanks betty tsang for her helpful comments and discussions . chen , and c.m . ko , phys . rep . * 464 * , 113 ( 2008 ) . bethe , rev . phys . * 62 * , 801 ( 1990 ) . k. summiyoshi and h. toki , astrophys . j. * 422 * , 700 ( 1994 ) . k. summiyoshi , k. oyamatsu , and h. toki , nucl . phys . * a595 * , 327 ( 1995 ) . f. douchin and p. haensel , astronomy & astrophys . , * 380 * , 151 ( 2001 ) . lattimer and m. prakash , science * 304 * , 536 ( 2004 ) ; j.m . lattimer and m. prakash , phys . * 442 * , 109 ( 2007 ) . et al . _ , c * 74 * , 035802 ( 2006 ) . m. baldo and c. maieron , j. phys . g * 34 * , r243 ( 2007 ) . g. bertsch , j. borysowicz , h. mcmanus , and w.g . love , nucl . a284 * , 399 ( 1977 ) . n. anantaraman , h. toki , and g.f . bertsch , nucl . phys . * a398 * , 269 ( 1983 ) . d.t . khoa and w. von oertzen , phys . lett . * b304 * , 8 ( 1993 ) . d.t . khoa and w. von oertzen , phys . lett . * b342 * , 6 ( 1995 ) . khoa , w. von oertzen , and a.a . ogloblin , nucl . phys . * a602 * , 98 ( 1996 ) . khoa , g.r . satchler , and w. von oertzen , phys . c * 56 * ( 1997 ) 954 . khoa , h.s . than , and d.c . cuong , phys . c * 76 * , 014603 ( 2007 ) . basu , p. roy chowdlhury , c. samanta , nucl . phys . * a811 * , 140 ( 2008 ) ; p. roy chowdlhury , c. samanta , and d.n . basu , phys . c * 80 * , 011305(r ) ( 2009 ) . h. nakada and m. sato , nucl a699 * , 511 ( 2002 ) . h. nakada , phys . c * 68 * , 014316 ( 2003 ) . h. nakada , phys . c * 78 * , 054301 ( 2008 ) ; h. nakada , phys . rev . c * 82 * , 029902 ( 2010 ) . d.t . khoa and h.s . than , phys . c * 71 * , 044601 ( 2005 ) . khoa , w. von oertzen , h.g . bohlen , and s. ohkubo , j. phys . * g34 * , r111 ( 2007 ) . n.d . chien and d.t . khoa , phys . c * 79 * , 034314 ( 2009 ) . than , d.t . khoa , and n.v . giai , phys . rev . c * 80 * , 064312 ( 2009 ) . berger , m. girod , and d. gogny , comp . . comm . * 63 * , 365 ( 1991 ) . f. chappert , m. girod , and s. hilaire , phys . b * 668 * , 420 ( 2008 ) . e. chabanat , p. bonche , p. haensel , j. meyer , and r. schaeffer , nucl . phys . * a635 * , 231 ( 1998 ) . lattimer , c.j . pethick , m. prakash , and p. haensel , phys . lett . * 66 * , 2701 ( 1991 ) . lattimer , k.a . van riper , m. prakash , and m. prakash , astrophys . j. * 425 * , 802 ( 1994 ) . d. page , j.m . lattimer , m. prakash , and a.w . steiner , astrophys . series * 155 * , 623 ( 2004 ) . f. douchin , p. haensel , and j. meyer , nucl . a665 * , 419 ( 2000 ) . w. zuo , i. bombaci , and u. lombardo , phys . c * 60 * , 024605 ( 1999 ) . m. brack , c. guet , and h.b . hkansson , phys . rep . * 123 * , 276 ( 1985 ) . pearson and r.c . nayak , nucl . a668 * , 163 ( 2000 ) . tsang , y. zhang , p. danielewicz , m. famiano , z. li , w.g . lynch , and a.w . steiner , phys . * 102 * , 122701 ( 2009 ) ; m.b . tsang , z. chajecki , d. coupland , p. danielewicz , f. famiano , r. hodges , m. kilburn , f. lu , w.g . lynch , j. winkelbauer , m. youngs , and y.x . zhang , prog . part . nucl . phys . * 66 * , 400 ( 2011 ) . p. danielewicz , r. lacey and w.g . lynch , science * 298 * , 1592 ( 2002 ) . a. ono , p. danielewicz , w.a . friedman , w.g . lynch , and m.b . tsang , phys . c * 68 * , 051601(r ) ( 2003 ) . z. xiao , b.a . chen , g.c . yong , and m. zhang , phys . lett . * 102 * , 062502 ( 2009 ) . shetty , s.j . yennello , and g.a . souliotis , phys . c * 76 * , 024606 ( 2007 ) . shetty , s.j . yennello , and g.a . souliotis , nucl . inst . and meth . b * 261 * , 990 ( 2007 ) . l.w . chen , c.m . ko , and b.a . li , phys . lett . * 94 * , 032701 ( 2005 ) p. danielewicz , nucl . phys . * a727 * , 233 ( 2003 ) . p. arumugam , b.k . sharma , p.k . sahu , s.k . patra , tapas sil , m. centelles , and x. vias , phys . b * 601 * , 51 ( 2004 ) . b. g. todd - rutel and j. piekarewicz , phys . 95 * , 122501 ( 2005 ) . j. piekarewicz and m. centelles , phys . c * 79 * , 054311 ( 2009 ) . j. piekarewicz , phys . c * 76 * , 064310 ( 2007 ) . m. centelles , x. roca - maza , x. vinas , and m. warda , phys . * 102 * , 122502 ( 2009 ) . l. trippa , g. col , and e. vigezzi , phys . c * 77 * , 061304(r ) ( 2008 ) . brown , phys . lett . * 85 * , 5296 ( 2000 ) . furnstahl , nucl . phys . * a706 * , 85 ( 2002 ) . a. akmal , v.r . pandharipande , and d.g . ravenhall , phys . c * 58 * , 1804 ( 1998 ) . s. gandolfi , a.yu . illarionov , s. fantoni , j.c . miller , f. pederiva , and k.e . schmidt , mon . not . . soc . * 404 * , l35 ( 2010 ) . v. baran , m. colonna , v. greco , and m. di toro , phys . rep . * 410 * , 335 ( 2005 ) . stone , j.c . miller , r. koncewicz , p.d . stevenson , and m.r . strayer , phys . c * 68 * , 034324 ( 2003 ) . wiringa , v. fiks , and a. fabrocini , phys . c * 38 * , 1010 ( 1988 ) . van dalen , c. fuchs , and a. faessler , eur . j. a * 31 * , 29 ( 2007 ) . li , u. lombardo , h.j . schulze , w. zuo , l.w . chen , and h.r . ma , phys . c * 74 * , 047304 ( 2006 ) . jeukenne , a. lejeune , and c. mahaux , phys . c * 16 * , 80 ( 1977 ) . f. zel , g. baym , and t. gver , phys . d * 82 * , 101301(r ) ( 2010 ) . steiner , j.m . lattimer , and e.f . brown , astrophys . j. * 722 * , 33 ( 2010 ) . i. bombaci , _ isospin physics in heavy ion collisions at intermediate energies _ , edited by b.a . li and w.u . schrder ( nova science , new york , 2001 ) , p. 35 . g. shen , c.j . horowitz , and s. teige , phys . c * 83 * , 035802 ( 2011 ) . n.k . glendenning , _ compact stars : nuclear physics , particle physics and general relativity _ ( springer : springer - verlag new york , inc . lattimer and b.f . schutz , astrophys . j. * 629 * , 979 ( 2005 ) . a. worley , p.g . krastev , and b.a . li , astrophys . j. * 685 * , 390 ( 2008 ) . j. cottam , f. paerls , and m. mendez , nature * 420 * , 51 ( 2002 ) . rhoades , jr . , and r. ruffini , phys . * 32 * , 324 ( 1974 ) . podsiadlowski , j.d.m . dewi , p. lesaffre , j.c . miller , w.g . newton , and j.r . stone , mon . not . r. astron . soc . * 361 * , 1243 ( 2005 ) .
we study in this paper logarithmic derivatives associated to derivations on graded complete lie algebra , as well as the existence of inverses . these logarithmic derivatives , when invertible , generalize the correspondence between a lie algebra and its lie group . such correspondences occur naturally in the study of dynamical systems when dealing with the linearization of vector fields and the non linearizability of a resonant vector fields corresponds to the non invertibility of a logarithmic derivative and to the existence of normal forms . these concepts , stemming from the theory of dynamical systems , can be rephrased in the abstract setting of lie algebra and the same difficulties as in perturbative quantum field theory ( pqft ) arise here . surprisingly , one can adopt the same ideas as in pqft with fruitful results such as new constructions of normal forms with the help of the birkhoff decomposition . the analogy goes even further ( locality of counter terms , choice of a renormalization scheme ) and shall lead to more interactions between dynamical systems and quantum field theory .
1302.6037
since the work of a. connes and d.k kreimer ( see @xcite , @xcite ) in perturbative quantum field theory ( pqft ) , it has been possible to have a purely algebraic interpretation of some renormalization schemes , as the birkhoff decomposition of regularized characters ( with values in a commutative algebra of laurent series ) , that is of elements of the group of algebra morphisms from a graded commutative hopf algebra to the the algebra of laurent series ( see @xcite , @xcite ) . many graded commutative hopf algebras ( shuffle , quasi shuffle , tree ) do appear in this framework and , surprisingly , the same objects ( groups of characters on hopf algebras ) arise in the study of analytic dynamical systems ( see for example @xcite ) , where changes of coordinates can be computed as elements either of the group of formal identity - tangent diffeomorphisms ( namely characters over the fa di bruno hopf algebra ) or of some subgroup that corresponds to characters over a combinatorial hopf algebra of trees or words ( using mould calculus , see @xcite , @xcite ) . such interactions should allow to enrich each domain with the ideas of the other and some steps have already been done in this direction ( see @xcite or @xcite ) . the origin of this paper comes from the observation that , in pqft , the need for renormalization comes from the ill - definedness of some character on a hopf algebra , which , after dimensional regularization , is replaced by a character with values in an algebra of laurent series . the attempted character is then obtained with the help of the birkhoff decomposition . the same phenomenon occur in dynamical systems , in the study of normal forms of vector fields ( when the the change of coordinates that should linearize the vector field is ill - defined ) . apart from some remarks in the literature ( see for example @xcite ) , there does not seem that renormalization schemes have been used to compute , for example , normal forms . we shall explain in this paper in an abstract algebraic context a common framework to pqft and dynamical systems how the ideas developed in pqft ( dimensional regularization , birkhoff decomposition ... ) provide a way to compute normal forms as well as other objects that play a crude role in dynamics . in the next section , we briefly recall the definition of complete graded lie algebras and their link with commutative hopf algebras . we define then in section [ secder ] graded derivations on lie algebras and their associated logarithmic derivative . under some invertibilty condition on such derivations , we define the inverse of such logarithmic derivative ( see also section [ secovin ] ) that generalizes the @xmath1@xmath0 correspondence between a lie algebra and its lie group . whenever such a derivation is not invertible , the theory of dynamical systems ( see appendix [ secds ] ) , provide , in the framework of lie algebra , a way to study these logarithmic derivatives , using @xmath2-conjugacy and `` normal '' forms ( see section [ secnoninv ] ) . there are surprising similarities with the need for renormalization in pqft , the same ideas regularization and birkhoff decomposition provide a new method to compute normal forms ( section [ norenor ] ) and the analogy goes even further , see section [ secmore ] , since such notions as the locality of counter terms or the choice among several `` renormalization schemes '' also appear here . in the sequel we consider complete graded lie algebras @xmath3 ( as well as complete vector spaces or algebras ) over a field @xmath4 of characteristic 0 . in other words , @xmath3 is the completion @xmath5 of a graded lie algebra @xmath6 ( with finite dimensional graded components @xmath7 and @xmath8 \subset l^k_{n + m}$ ] ) with respect to the graduation . we write for short @xmath9 ( @xmath10 ) and the lie bracket extends to the completion @xmath3 : if @xmath11 and @xmath12 then @xmath13 = \sum_{n \geqslant 1 } z_n$ ] with @xmath14\ ] ] under these hypotheses , the universal enveloping algebra @xmath15 of @xmath3 is a unital , graded associative algebra with a graduation @xmath16 component @xmath17 . this algebra turns into a graded co - commutative hopf algebra for the graded co - product defined on @xmath3 by @xmath18 that extends , thanks to the universal property of lie algebras , to @xmath15 ( see @xcite ) . using once again the graduation , we note @xmath19 the completion of @xmath15 and , as a graded map the co - product extends to a map from @xmath20 to the completion of @xmath21 ( that may only contain @xmath22 ) . @xmath20 is not necessarily a hopf algebra but it contains @xmath3 as the lie algebra of _ primitive elements _ of @xmath20 , that is the elements @xmath23 of @xmath20 such that @xmath24 , but also the lie group @xmath25 of group - like elements of @xmath20 , that is the elements @xmath23 of @xmath20 such that @xmath26 . let us also remind that there is a one to one correspondence between @xmath3 and @xmath25 , using the well - defined exponential an logarithm : @xmath27 where @xmath28 . for a given commutative @xmath4algebra @xmath29 , one can also consider the lie algebra @xmath30 , which inherits grading of @xmath3 and is defined as the completion , for the same graduation as in @xmath3 , of the lie algebra @xmath31 with the lie bracket @xmath32 = ( ab)\otimes[x , y]\ ] ] note that @xmath30 is both a vector space over @xmath4 ( whose graded component may be of infinite dimension ) and a module over @xmath29 ( with graded components of finite dimension ) and one can as well define @xmath33 and @xmath34 . there is no difficulty to extend iteratively this construction . for example , if @xmath35 , then @xmath36 and the same inclusions hold for the completed enveloping algebras as well as for the groups . on the same way , we could define @xmath37 as the completion of @xmath38 . in the sequel , we will adopt this lie algebra point of view but there is no difficulty to identify such lie algebra @xmath39 ( resp . lie groups @xmath40 ) with the lie algebra ( resp . group ) of infinitesimal characters ( resp . characters ) on a graded commutative algebra , with values in @xmath29 . the reader can refer to @xcite on this matter . this simply means that the ideas stemming from the study of characters on hopf algebras can be easily translated in the context of such lie algebras . let us give , as an example , a `` toy model '' stemming from the theory of dynamical systems . when dealing with the simple differential equation : @xmath41 the lie algebra ( over @xmath42 ) of differential operator ( or vector fields ) , @xmath43 appears with a natural action on series of @xmath44 $ ] and the elements of the completion of @xmath45 are formal operators @xmath46 with @xmath47 $ ] . for short we write @xmath48\partial_y$ ] this lie algebra . note that it coincides , for a different graduation , with the lie algebra @xmath49 with @xmath50 $ ] and @xmath51 . as another example related to dynamical systems , consider the lie algebra @xmath52}$ ] , whose elements can be written : @xmath53 the element @xmath54}$ ] ( @xmath55 ) is the unique solution of the differential equation @xmath56 with the initial condition @xmath57 . this suggests to explore similar `` differential '' equations that may lead to new correspondences between lie algebras and lie groups , since such equations appear naturally in the framework of differential equations ( see section [ secds ] ) . in the case of our toy model , the differential operator @xmath58 is a substitution automorphism : @xmath59,\quad \exp(a(x)y^2\partial_y ) .f(x , y)=f(x,\exp(a(x)y^2\partial_y).y)\ ] ] with @xmath60 and the map @xmath61 is an anti isomorphism between the lie group of @xmath62\partial_y$ ] and the group ( for the composition ) of formal identity - tangent diffeomorphisms @xmath63 the reader can check that @xmath64 is the flow of the differential equation @xmath65 ( see section [ secds ] for details ) . for short , in the sequel @xmath45 ( resp . @xmath66 , @xmath67 ) will denote @xmath39 ( resp . @xmath68 , @xmath40 ) . note also that any @xmath4linear endomorphism @xmath69 of @xmath70 ( resp . graded @xmath4linear endomorphism of @xmath45 ) can be extended to a @xmath4linear graded endomorphism of @xmath71 , still noted @xmath69 when there is no ambiguity , setting @xmath72 ( resp . @xmath73 ) . in order to avoid ambiguities , whenever such an endomorphism is defined on possibly many @xmath4vector spaces ( or algebras ) @xmath74 , we will write the image @xmath75 and the corresponding kernel @xmath76 a derivation @xmath2 is a graded @xmath4linear endomorphism on @xmath45 ( @xmath3 or @xmath39 ) such that @xmath77 ) = [ d ( x ) , y ] + [ x , d ( y)]\ ] ] from the universal property of lie algebras , as @xmath2 is graded ( @xmath78 ) it extends to a graded derivation on @xmath66 : @xmath79 , @xmath80 and @xmath81 . for such a derivation , we can define and study the `` differential '' equation @xmath82 any such derivation defines a map : @xmath83 that sends @xmath67 on @xmath45 . the map is well defined since any @xmath84 is invertible for the product in @xmath66 with an inverse @xmath85 and , thanks to the completion , @xmath86 is well - defined in the complete algebra @xmath66 . it remains to prove that it maps @xmath67 on @xmath45 and this results comes from the magnus formula ( see @xcite ) : [ dexp ] let @xmath87 ( @xmath88 ) , then @xmath89 where @xmath90 = \alpha x - x \alpha$ ] is the adjoint action of @xmath91 . since @xmath66 is complete , the above series converges in @xmath92 when @xmath91 is in @xmath92 . moreover , since @xmath91 is in @xmath45 ( @xmath93 ) , @xmath94 and @xmath95 belong to @xmath45 thus @xmath96 is in @xmath45 . note that we would have similar formulas for @xmath97 ( @xmath98 ) and , since @xmath99 , @xmath100 this logarithmic derivative is a kind of generalization of the logarithm and such @xmath2-logarithm do appear in the study of dynamical systems ( see section [ secds ] ) . for a given @xmath2-logarithm from @xmath67 to @xmath45 it is natural to ask if there exists an inverse . whenever the derivation @xmath2 has a graded right inverse on @xmath45 ( @xmath3 or @xmath30 ) , there is a positive answer to the previous question : [ rightinv ] if @xmath2 admits a graded right inverse @xmath101 on @xmath102 then for any @xmath103 there exists @xmath104 such that @xmath105 . let @xmath103 . if we look for @xmath106 and set , thanks to the graduation , @xmath107 thanks to lemma [ dexp ] , @xmath108 in @xmath109 we have , @xmath110 and we define @xmath111 . for @xmath112 , @xmath113=u_2\ ] ] and we have the solution @xmath114)\ ] ] and the proof follows recursively : @xmath115 is in @xmath45 and @xmath106 is a solution of @xmath116 . if @xmath2 is invertible on @xmath102 then the solution @xmath117 of @xmath118 is unique and noted @xmath119 the reader can notice that in the case of @xmath3 , since we assumed that the graded components @xmath7 are of finite dimension , a derivation has a right inverse if and only if it is invertible . otherwise , for any derivation @xmath2 , @xmath120 is a sub - lie algebra of @xmath45 ( @xmath121 when @xmath2 is invertible ) that gives rise to a subgroup of @xmath67 : @xmath122 and @xmath123 and @xmath124 are two solutions of @xmath125 if and only if @xmath126 consider @xmath127 , that is @xmath128 , then @xmath129 thus @xmath130 and @xmath131 . note that , using once again lemma [ dexp ] , we get : @xmath132 a classical ( and universal ) example of invertible derivation in graded lie algebras is the derivation associated to the graduation : @xmath133 in this case , @xmath134 is the dynkin operator , which is a bijection between @xmath45 and @xmath67 ( see @xcite ) . let us just point out another example that relates the dynkin operator and the usual logarithm . for a given complete graded lie algebra @xmath135 over a commutative @xmath4algebra @xmath29 , consider the lie algebra @xmath136}$ ] . this algebra inherits the graduation of @xmath45 and the graded derivation @xmath137 is invertible . if @xmath138 , then @xmath139 and @xmath140 are in @xmath136}$ ] and : @xmath141 so that the classical exponential as well as @xmath142 are related to the same differential equation in @xmath136}$ ] : @xmath143 we will come back to the interaction between the logarithm and the dynkin operator in the following section but let us end by the definition of a wide class of derivations that appear in dynamical systems . suppose that @xmath45 is a sub - algebra of a lie algebra @xmath144 where @xmath145 is a lie algebra of graduation @xmath16 , then for any @xmath146 , the adjoint action of @xmath147 on @xmath45 defines a graded derivation on @xmath45 . whenever this adjoint action is invertible , we have a unique solution to the equation @xmath148=\varphi.x\ ] ] which is fundamental is the study of vector fields ( see section [ secds ] ) . but conversely , the study of vector fields provides many ideas to deal with the case of non invertible derivations . let us illustrate this on the toy model @xmath149\partial_y$ ] with @xmath150 . the associated derivation @xmath151 ( with @xmath152 ) gives for @xmath153 @xmath154=(\alpha n+\beta)x^n y^2\partial_y \\ d(a(x)y^2 \partial_y ) & = & ( \alpha xa'(x)+\beta a(x))y^2\partial_y \end{array}\ ] ] and the derivation is invertible if @xmath155 does not vanish for @xmath153 or , equivalently , when , for any @xmath156 $ ] , there exists a unique formal solution @xmath157 to the equation @xmath158 more precisely , for a fixed @xmath159 and @xmath160 , if we look for an element @xmath161 in the lie group such that @xmath162 then , since any lie bracket in @xmath163\partial_y$ ] vanishes , theorem [ rightinv ] ensures that @xmath164 is given by the solution of @xmath165 from the dynamical systems point of view , the reader can check that , when identifying @xmath161 to the formal diffeomorphism @xmath166 the ( formal ) solutions of the differential system @xmath167 are in one to one correspondence with the solutions of @xmath168 since @xmath169 depending on the choice of @xmath91 and @xmath170 , the associated derivation could , in the best case , be invertible on the complete enveloping algebra and , in the worst case ( for example @xmath171 ) , be non - invertible on the lie algebra . if @xmath2 is invertible as well on @xmath45 as on @xmath172 , we will say that @xmath2 is `` overall '' invertible and it is natural to transform the equation @xmath173 into the so - called atkinson s recursion : @xmath174 where @xmath101 is the inverse of @xmath2 on @xmath172 . in this case , @xmath175 namely , @xmath101 is a rota - baxter operator of weight @xmath16 : let @xmath176 an associative algebra and @xmath177 a linear operator on @xmath176 . @xmath177 is a rota - baxter operator of weight @xmath178 if:@xmath179 in this case ( see @xcite ) , the equation ( with perturbative parameter @xmath180 ) : @xmath181 admits a perturbative solution : @xmath182 } ( x)\ ] ] with @xmath183 } ( x ) = i ( x)$ ] and @xmath184 } ( x ) = i(r^{[n ] } ( x ) .x)$ ] . this solution is in @xmath67 and its logarithm , as well as its image by the dynkin map @xmath134 , are given by combinatorial formulas ( see @xcite ) . we proved in @xcite : [ thstrigen ] let @xmath185 a graded derivation that commutes with @xmath2 ( thus with @xmath101 ) , the above perturbative solution satisfies : @xmath186 } ( x)\ ] ] with @xmath187}_{\delta } ( x ) = i ( \delta ( x))$ ] et @xmath188}_{\delta } ( x ) = i ( [ \mathcal{l}_{\delta}^{[n ] } ( x ) , x])$ ] . the proof of this theorem is based on the fact that @xmath2 is overall invertible . this is necessary to get the perturbative expansion of the solution @xmath189 and then the formula [ strigen ] . but this formula only involves the inverse of @xmath2 on @xmath45 and shall still hold if @xmath2 is only invertible on @xmath45 . this idea could certainly lead to a new proof of theorem [ rightinv ] , when @xmath2 is invertible on @xmath45 but not necessary on @xmath172 . for example , in the case @xmath190 , since the vector spaces ( over @xmath4 ) @xmath191 and thus @xmath192 are finite dimensional , for a given parameter @xmath193 , the regularized operator @xmath194 is a derivation on the lie algebra @xmath195[{\varepsilon}^{-1}]}$ ] where @xmath196[{\varepsilon}^{-1}]$ ] is the algebra of laurent series in @xmath193 with coefficients in @xmath4 . on one hand , on each @xmath197[{\varepsilon}^{-1}]}_n$ ] the restriction of @xmath194 is invertible : @xmath198[{\varepsilon}^{-1}]}$ ] is a finite dimensional @xmath196[{\varepsilon}^{-1}]$]module and the determinant of the restriction of @xmath194 to this module is a non zero laurent series . the derivation @xmath194 is thus invertible , of inverse @xmath199 , on @xmath200[{\varepsilon}^{-1}]}$ ] . on the other hand , on each @xmath201[{\varepsilon}^{-1}]}_n$ ] , since @xmath2 is invertible , @xmath199 still makes sense when @xmath202 and coincides , for this value , with the inverse of @xmath2 ( say @xmath203 ) . the perturbative solution of @xmath189 makes no sense since @xmath2 is not necessarily overall invertible but , after the regularization , @xmath194 is overall invertible and the perturbative expansion of the solution of @xmath204 makes sense , especially when @xmath23 is in @xmath3 . the formula [ strigen ] , with @xmath185 commuting with @xmath194 still hold . if we take @xmath205 then the right part of this formula make sense when @xmath202 thus @xmath206 and then @xmath207 can be evaluated at @xmath202 and gives the solution in the invertible case ! there is no contradiction here , this simply means that the expansion giving the perturbative solution may have poles , but once rewritten , for example in a basis of @xmath66 , poles in @xmath193 cancel out . this `` @xmath193regularization '' turn a derivation in an overall invertible regularized derivation . such regularizations appear in quantum field theory and will provide a way to deal with non invertible derivations , see section [ norenor ] . inspired by the situation in dynamical systems , whether @xmath2 is invertible or not , it is relevant to study @xmath2-conjugate elements in @xmath45 two elements in @xmath102 are @xmath2-conjugate if there exists @xmath208 such that@xmath209 we say that @xmath117 d - conjugates @xmath69 to @xmath210 . this is an equivalence relation and we note @xmath211 . if @xmath117 @xmath2-conjugates @xmath69 to @xmath210 , then @xmath212 conjugates @xmath210 to @xmath69 since : @xmath213 on the same way , if @xmath117 @xmath2-conjugates @xmath69 to @xmath210 and @xmath214 @xmath2-conjugates @xmath210 to @xmath215 , then @xmath216 @xmath2-conjugates @xmath69 to @xmath215 : @xmath217 the idea of @xmath2-conjugacy comes from dynamical systems and is motivated by the following fact : on one hand , if , for example @xmath2 is invertible on @xmath102 , then there is only one @xmath2-conjugacy class since any @xmath69 in @xmath102 is conjugated to @xmath16 by @xmath218 . on the other hand , whenever @xmath2 is not invertible , these equivalence classes are not trivial and their classification or description is of great interest , especially in the framework of dynamical systems . the complete classification , up to @xmath2-conjugacy , shall , in many cases , be out of reach . for example , if @xmath219 . the equation reads @xmath220 using the graduation and @xmath221 ( @xmath222 ) , in the first graded components we get : @xmath223 thus we must find @xmath224 such that @xmath225=v_2 -u_2\ ] ] and the existence of a solution highly depends on the structure and the relations in @xmath45 . this last equation suggest , in order to prove that two elements @xmath69 and @xmath210 are @xmath2conjugate , to look for a solution @xmath221 of the equation : @xmath226 if @xmath227 and @xmath228 , let us recall that we already have ( see proposition [ dexp ] and equation [ adjgrad ] ) : @xmath229 where @xmath230 is a lie polynomial in @xmath231 and @xmath232 with @xmath233 . on the same way , [ uexp ] if @xmath210 and @xmath91 are in @xmath45 , then @xmath234 the straightforward proof follows the same lines as for lemma [ adj ] and if @xmath227 , @xmath235 and @xmath236 : @xmath237 where @xmath238 is a lie polynomial in @xmath231 and @xmath239 with @xmath233 . thanks to equations [ loggrad ] and [ conjgrad ] , one can already give a particular element in each conjugacy class : [ fnor ] let @xmath2 a graded derivation on @xmath45 and @xmath240 a supplementary vector space of @xmath241 in @xmath45 . for any @xmath242 , there exists an element @xmath243 which is a @xmath2-conjugate of @xmath69 . if @xmath244 we must find @xmath245 and @xmath246 ( @xmath247 ) such that @xmath248 that is to say , for all @xmath249 : @xmath250 for @xmath251 , this read @xmath252 . let @xmath253 the projection on @xmath254 , parallel to @xmath241 . this is a graded linear operator and if @xmath255 then @xmath256 is in @xmath257 and one can find a solution to the equation @xmath252 . suppose for a given @xmath249 , that we have found @xmath258 in @xmath45 and @xmath259 in @xmath254 that solve equation [ fnorgrad ] . the polynomials @xmath260 and @xmath261 are well defined and , at order @xmath262 , we must find @xmath263 and @xmath264 such that @xmath265 as in the case @xmath251 , we take @xmath266 and we can then find an element @xmath267 that solves the equation . this ends the recursive construction of @xmath91 and @xmath268 . note that the element @xmath269 is not unique since there is a latitude in the determination of @xmath91 . there would be more to say if one could chose @xmath254 as a sub lie algebra , especially @xmath120 . this is the case if we assume that @xmath270 this hypothesis is natural dynamical systems and , in the sequel , we will deal with derivations satisfying equation ( [ dnormal ] ) . under this assumption , the restriction of @xmath2 from @xmath241 to itself is invertible and we note in the sequel @xmath101 this inverse , defined from @xmath241 to itself . from now on , we can choose @xmath271 in the previous theorem and we note @xmath272 the projector on @xmath120 , parallel to @xmath241 . this choice allow to define and study `` normal forms '' . following the example of @xmath149 \partial_y$ ] if @xmath273 then we have @xmath274 \partial_y,\quad \ker_l d=\text{vect}\{y^2 \partial_y\}\ ] ] so that @xmath275 . in this case , with @xmath276 , one can easily see , using equation [ fnorgrad ] and the very simple lie bracket in @xmath48 \partial_y$ ] , that @xmath277 is @xmath2conjugate to @xmath278 . as we shall see now , the choice @xmath271 provides @xmath2normal forms that play a crude role in dynamical systems . let us look closer to the classification of @xmath102 , up to @xmath2-conjugacy , when @xmath275 . [ thnor ] for any @xmath279 , there exists @xmath280 such that @xmath281 moreover , @xmath282 are @xmath2-conjugated to @xmath69 if and only if there exists @xmath283 such that @xmath284 such elements are called @xmath2-normal forms of @xmath69 . the existence of such normal forms follows from theorem [ fnor ] with @xmath271 . for the second part of this theorem , let us assume that @xmath210 is a @xmath2normal form of @xmath69 and let @xmath285 . if @xmath286 , that is to say @xmath287 , then , thanks to lemma [ uexp ] , @xmath215 is in the sub lie algebra @xmath120 and , since @xmath288 , @xmath289 thus @xmath215 is @xmath2-conjugated to @xmath210 and to @xmath69 : @xmath215 is a @xmath2normal form . conversely , if @xmath210 and @xmath215 are two normal forms , then they are @xmath2-conjugated : there exists @xmath93 such that @xmath290 and it remains to prove that @xmath285 . as in the proof of theorem [ fnor ] , if @xmath291 , we have for @xmath292 : @xmath293 in graduation 1 this gives @xmath294 or @xmath295 . thus @xmath296 and @xmath297 . in graduation 2 the equation @xmath298 + w_2 -[{\alpha}_1 , w_1]=v_2\ ] ] reduces to @xmath299 \in \ker_l d \cap d({l})\ ] ] since @xmath300\in \ker_l d$ ] . once again @xmath301 and @xmath302 . we get recursively that @xmath303 , using the fact that the polynomials @xmath304 are lie polynomials of elements of the sub - lie algebra @xmath120 . this theorem gives a description of the @xmath2-conjugacy classes : the @xmath2-conjugacy classes are exactly the classes of @xmath120 for the classic conjugacy by the group @xmath305 . as this will be used in section [ seccor ] , this implies that if @xmath69 is @xmath2-conjugated to @xmath16 , then @xmath16 is its unique normal form because @xmath306 ! the second part of this theorem implies that there can be more than one normal form . let us mention that , in the framework of dynamical systems , j. ecalle and b. vallet ( see @xcite , @xcite ) , provide a way to define for each @xmath242 a unique normal form , with the help of a supplementary condition : let @xmath307 be an invertible derivation such that @xmath308 , there exist a unique solution to the system : @xmath309 let us give an idea of the proof . once a solution @xmath310 of @xmath311 is given the other solutions are given by @xmath312 with @xmath313 . in order to prove this proposition we have to show that the condition @xmath314 determines a unique @xmath313 . but , @xmath315 once again , using the graduation , if @xmath316 and @xmath317 , we get for @xmath251 , @xmath318 thus @xmath319 that has a unique solution in @xmath120 since @xmath307 is invertible and @xmath120 is stable by @xmath307 . the remainder of the proof would follow recursively , as in the previous theorems . for any such @xmath307 we can define a map @xmath320 from @xmath102 to @xmath321 ( note that the conjugating map is also unique ) . the reader can notice that in the case of @xmath322 on @xmath163 \partial_y$ ] there is exactly one normal form of @xmath46 , namely @xmath278 . this is mainly due to the extreme simplicity of this toy model but it does not reflect the complexity of the general case , for which the choice of a normal form and its computation are not so easy to deal with . the aim of this section is to point out that similar procedures as those used in perturbative quantum field theory can be used to compute normal forms in the case of non - invertible derivations . in the connes - kreimer picture of renormalization ( see @xcite and @xcite ) , the computation of feynman integrals shall deliver a character @xmath117 on the the graded hopf algebra of 1pi feynman graphs , that is an element of the lie group associated to the complete graded lie algebra of infinitesimal characters on the hopf algebra of feynman graphs @xmath323 ( with @xmath324 ) . this character is defined by feynman integrals that could be divergent but can be renormalized , using dimensional regularization and birkhoff decomposition . more precisely , after dimensional regularization , the feynman integrals define a character @xmath325 with values in the algebra of laurent series @xmath326[{\varepsilon}^{-1}]$ ] , that is to say an element of the lie group @xmath327 of the lie algebra @xmath49 . the presence of poles at @xmath16 reflects the divergence of the initial character . the usual renormalization of such a group - like element , is then based on the birkhoff decomposition : since @xmath328 where @xmath329 $ ] and @xmath330 $ ] are two sub - algebras , it is possible to factorize @xmath325 , that is @xmath331 where @xmath332 ( resp . @xmath333 ) is in @xmath334 ( resp . @xmath335 ) and the evaluation of @xmath333 at @xmath336 gives the renormalized character @xmath337 . we refer to @xcite for further details on the birkhoff decomposition . such ideas can be applied in order to find normal forms . as in perturbative quantum field theory , the solutions of @xmath338 are ill - defined whenever the derivation @xmath2 is not invertible . one can then chose to `` regularize '' @xmath2 as follows . let @xmath2 a derivation and @xmath307 a derivation on @xmath339 such that @xmath120 is stable by @xmath307 and the restriction of @xmath185 from @xmath120 to itself is invertible . these operators @xmath2 and @xmath185 can be extended to @xmath49 by @xmath340linearity and we get a `` @xmath185-renormalization '' scheme that delivers a normal form : [ noren ] let @xmath2 and @xmath185 two derivations on @xmath190 such that : 1 . @xmath341 , 2 . @xmath120 is stable by @xmath307 , 3 . the restriction of @xmath185 from @xmath120 to @xmath120 is invertible . for @xmath342 ( @xmath326[{\varepsilon}^{-1}]$ ] ) , the equation @xmath343 has a solution @xmath344 and , after birkhoff decomposition , @xmath345 the group - like element @xmath346 conjugates @xmath69 to a normal form @xmath347 . moreover , we have @xmath348 . note that this result still hold on a lie algebra @xmath349 by considering laurent series in @xmath350[{\varepsilon}^{-1}]$ ] . as we shall see in section [ locres ] , the normal form @xmath170 is an analog to the @xmath170 function that appears in perturbative quantum field theory ( see @xcite , @xcite ) . in the case @xmath273 on @xmath351\partial_y$ ] , the reader can check that one can choose @xmath352 , that coincides with the grading @xmath353 . in this case , @xmath354 and for a given element @xmath355 a solution of @xmath356 is given by @xmath357 with @xmath358 where fractions in @xmath359 identify with laurent series and , thanks to the simplicity of the lie algebra , @xmath360 so that a normal form of @xmath361 is @xmath362 let @xmath2 a graded derivation on @xmath339 such that @xmath363 and @xmath364 , @xmath365 the respective projections on the kernel of @xmath2 on @xmath45 and its image . it is clear that @xmath2 , @xmath272 and @xmath366 extend to @xmath30 with @xmath367 . for example , if @xmath368 , @xmath369 and @xmath370 with @xmath371 the identity @xmath367 implies that the restriction of @xmath2 from @xmath372 ( or @xmath373 ) to itself is invertible and we note @xmath101 this graded inverse . on the same way @xmath185 can be extended to @xmath30 , its restriction from @xmath374 ( or @xmath375 ) to itself is invertible and we note abusively @xmath376 its inverse on @xmath374 ( or @xmath375 ) . note that all these operators are graded . the first step of the proof is based on the following lemma : the endomorphism @xmath377 is an invertible derivation on @xmath30 and its ( graded ) inverse is the linear map @xmath378 defined for @xmath379 by @xmath380 it is a matter of fact to check that @xmath377 and @xmath378 are well - defined : as all the operators do respect the graduation , we can work on graded components @xmath381 with elements as in equation ( [ lacomp ] ) . it remains to prove that @xmath382 . since @xmath383 is stable by @xmath307 , @xmath384 thus @xmath385 and , when restricted to @xmath373 , @xmath386 if we apply to @xmath387 it is now clear that @xmath388 thus @xmath389 . in order to prove that @xmath390 , let us notice that@xmath391 @xmath392 and @xmath393 . we get @xmath394 and finally @xmath395 note that the expression of this inverse can be simplified if we assume that @xmath2 and @xmath185 commute . since @xmath396 is invertible on @xmath30 , thanks to theorem [ rightinv ] , the equation @xmath397 has a solution @xmath398 for any @xmath69 in @xmath30 or @xmath45 . thanks to the birkhoff decomposition ( see @xcite ) , we have @xmath399 if @xmath242 , since @xmath397 we get @xmath400 and then @xmath401 but @xmath402 is in @xmath403}$ ] whereas the other terms are in @xmath404}$ ] thus these two parts of the identity do not depend on @xmath193 . let @xmath405 ( not in @xmath30 ! ) , then @xmath406 and for @xmath202 the element @xmath407 conjugates @xmath69 to @xmath170 . it remains to prove that @xmath170 is a normal form , that is @xmath408 . note that if @xmath409 with @xmath410 , then @xmath411 with @xmath408 ( see lemma [ dexp ] ) . the proof of this last property is based on the following lemma : [ lemconst ] let @xmath412 . if @xmath413 then @xmath414 the proof of the lemma is based on the following observation : if , for a given @xmath292 , @xmath415 is such that @xmath416 with @xmath417 and @xmath418 } } d\ ] ] then @xmath419 and @xmath420 . one can check that @xmath421 and @xmath422 on the other hand , if @xmath417 , @xmath423 it follows that necessarily @xmath424 ( thus @xmath419 ) and @xmath420 . this is the key to prove the lemma . if @xmath425 with @xmath426 ( @xmath427 ) then , using lemma [ dexp ] and the graduation ( see equation ( [ loggrad ] ) ) , we get : @xmath428 where @xmath230 is a lie polynomial in @xmath231 and @xmath429 with @xmath233 . in graduation 1 : @xmath430 and , thanks to the previous result ( with @xmath431 and @xmath432 } } d$ ] ) , @xmath433 and finally @xmath434 , @xmath435 and @xmath436 . let us suppose now that @xmath437 we have @xmath438 with @xmath439 since , for any @xmath4 , @xmath440 , it is clear that we can set @xmath441 } } d\\ y_n & = & q(\beta_n ) \in d(l^k_n ) \\ \end{array}\ ] ] and we already proved that @xmath419 ( thus @xmath442 ) and @xmath443 it follow recursively that @xmath444 , @xmath445 for @xmath446 using this lemma , together with the birkhoff decomposition , @xmath447 is a normal form related to @xmath448 by @xmath449 we enclose in this section many ideas , developments that are suggested by the strong similarity between our results , stemming from dynamical systems , and the algebraic machinery developed in the context of perturbative quantum field theory . the last identity ( [ resid ] ) has to be related to residues . if we write @xmath450 then , the residue @xmath451 is such that @xmath452 and it looks very close to the beta function in perturbative quantum field theory , see @xcite , @xcite and @xcite . this concept is related to the `` locality '' of counter terms ( that is @xmath448 ) . in our framework one can define on @xmath453}$ ] ( and then on @xmath454}$ ] ) the automorphism @xmath455 : @xmath456 following section 7 in @xcite ( in the case @xmath457 ) , @xmath458 is @xmath307local if for any @xmath459 , the birkhoff decomposition of @xmath460 is such that @xmath461 that is the counter terms do not depend on @xmath459 . under the assumption that @xmath462=0 $ ] ( thus @xmath463=0 $ ] ) , the same results as in @xcite could be obtained here and one can check that our renormalization procedure delivers a @xmath307local element . in fact , we have @xmath464 and @xmath465 and after birkhoff decomposition ( as in equation ( [ eqbir ] ) ) : @xmath466 but as in the proof of theorem [ noren ] , this identity shows that @xmath467 ^ -}\cap l^{\mathcal{a}[[\tau]]^+}=\lbrace 0 \rbrace$ ] and @xmath468 . if we expand in powers of @xmath459 , since @xmath469 with @xmath470 , we get @xmath461 and this proves the @xmath307locality of @xmath117 . as we can see here , this renormalization is very similar to the one used in physics . note also that it could be of some interest to consider `` perturbations '' of @xmath2 such as @xmath471 . on the same way , the following section suggests that other renormalization procedure appear in the framework of dynamical systems . as pointed out in @xcite , in pqft , the attempted ( but ill - defined ) group - like element @xmath117 is associated to a lagrangian @xmath472 and the renormalization procedure can be interpreted as an iterative process , based on the graduation of the considered hopf algebra , that consists in modifying the lagrangian : @xmath473 so that the renormalized group - like element @xmath474 corresponds to the modified lagrangian @xmath475 . the same principle appear in the framework of dynamical systems but the final object differs from a normal form , this is the correction : let @xmath2 a derivation such that @xmath476 . for any @xmath477 there exists a unique @xmath478 such that @xmath479 is in the @xmath2-conjugacy class of @xmath16 . @xmath480 is called the correction of @xmath69 . moreover , if @xmath481 , then @xmath482 . as in the lagrangian interpretation of renormalization in pqft , the idea of the proof is to modify @xmath69 , graded component by graded component , so that we can solve the equation : @xmath483 whenever there exists @xmath117 such that @xmath105 , then we can take @xmath484 , but , when @xmath2 is not invertible , such @xmath117 does not necessarily exists . let us consider @xmath485 the equation @xmath486 reads ( see lemma [ dexp ] and equation ( [ loggrad ] ) ) : @xmath487 where @xmath488 depends only on @xmath489 ( with @xmath490 ) . for @xmath251 , we must solve : @xmath491 thus @xmath492 ( unique ) , @xmath493 and then we can chose @xmath494 up to an element of @xmath495 : @xmath496 , @xmath497 . now suppose that for @xmath498 we can find @xmath499 so that there exists solutions ( say @xmath500 ) to the equations : @xmath501 for the next equation : @xmath502 as @xmath476 , we must define @xmath503 and this ends the ( recursive ) proof of the existence of @xmath504 . but in this recursive definition @xmath504 seems to depend on the choice of @xmath180 . indeed @xmath504 is unique : suppose we have @xmath505 and @xmath506 in @xmath495 such that @xmath507 , then there exists @xmath508 , @xmath509 such that : @xmath510 in @xmath511 , let @xmath512 , then @xmath513 thus @xmath514 or rather @xmath515 using lemmas [ dexp ] and [ uexp ] with @xmath516 and @xmath517 : @xmath518 thanks to the graduation we get first @xmath519 : since it is in the kernel and the image of @xmath2 , we get @xmath520 and @xmath521 . one can then prove recursively on @xmath522 that , for @xmath523 : @xmath524 if this is true then the equation reduces , in graduation @xmath262 to @xmath525 thus @xmath526 and this ends the recursion . note that , @xmath117 is unique if we assume that @xmath527 with @xmath528 , using the inverse @xmath101 of the restriction of @xmath2 from @xmath241 to @xmath241 . we do nt know if this correction can be interpreted in the framework pqft but let us just end by two remarks on the link between normal forms obtained by the `` renormalization scheme '' described in theorem [ noren ] and the correction . for a given @xmath307 as in theorem [ noren ] , we get a function @xmath529 from @xmath45 to @xmath120 such that , for all @xmath242 , @xmath530 and the correction associates to any @xmath242 a unique @xmath531 , independent of @xmath307 , such that @xmath532 . as pointed out in section [ secnor ] , there is no other normal form than @xmath16 which is conjugated to @xmath16 : @xmath533 thus @xmath534 is a fixed point for the operator @xmath535 . one could even go further : if we introduce a parameter @xmath536 then , @xmath537 so that , in a first approximation , the normal form and the correction coincide . this looks like two renormalizations , with different prescriptions but with a common `` minimal substraction '' @xmath538 . under the same hypothesis as in theorem [ noren ] , since @xmath539 is invertible on @xmath30 with @xmath326[{\varepsilon}^{-1}]$ ] , the equation @xmath540 has a unique solution @xmath541 . as for the birkhoff decomposition , the reader can check that there exist @xmath542 such that , if @xmath543 , @xmath544 as before , the recursive proof is based on equation ( [ loggrad ] ) : for @xmath292 , @xmath545 thus @xmath546 but @xmath547}$ ] , @xmath548}$ ] and their sum is @xmath549 . this means that @xmath550 and , thanks to lemma [ lemconst ] , @xmath551 thus @xmath552and this identity makes sense at @xmath202 . we recover the correction @xmath553 by this `` additive '' birkhoff decomposition . this use of `` renormalization schemes '' is new in the framework of dynamical systems and shall open new perspectives in this area . it shall interacts with j. ecalle s mould calculus that provides a way to `` lift '' the computation of diffeomorphisms to computations in the lie group of the lie algebra of infinitesimal characters on a shuffle hopf algebra ( see @xcite , @xcite ) . if mould calculus is sufficient to lead to explicit computations , this does not tackle the difficult and crucial question of the analyticity of the computed diffeomorphisms . surprisingly , the analytic regularity of such diffeomorphism can sometimes be easily obtained by computing in a hopf algebras of trees , which is strongly related to the prelie structure of vector fields ( see @xcite ) . there shall be no major difficulty to adapt the results of this paper to the case of derivations on prelie algebras . this may even be the right theoretical framework to deal with the similar questions in `` discrete dynamical systems '' that are related to group equations @xmath554 where @xmath178 is a graded automorphism on @xmath45 : @xmath555)=[\theta(x),\theta(y)]$ ] , with an `` invertibility '' condition on @xmath556 . this is the right framework to deal with hurwitz multizetas ( see @xcite ) and their regularization . let us end this paper with a remark on the choice of an @xmath193-regularization . in section [ secmt ] , we illustrated theorem [ noren ] in the case @xmath273 on @xmath351\partial_y$ ] with @xmath352 and the solution of @xmath557 was @xmath357 where @xmath558 theorem [ noren ] gave then the normal form @xmath278 . following @xcite , we could have chosen the regularization @xmath559 , assuming that one can now work on @xmath560 $ ] . after expansion @xmath561 and birkhoff decomposition , it appear that there exists solutions to the equation @xmath562 but with a series @xmath563 in @xmath23 * and * @xmath564 . on one hand , this result is very close to the results of j. ecalle on `` ramified linearization '' ( see @xcite ) . on the other hand , the apparition of logarithmic terms shall be familiar to the experts of pqft and shall be related to the toy model described in @xcite ( section 4.2 ) . the research leading these results was partially supported by the french national research agency under the reference anr-12-bs01 - 0017 . let us first recall some results on autonomous analytic local differential equations ( see @xcite for proofs and details ) , that is a system ( in dimension @xmath565 ) @xmath566 where @xmath567 and @xmath568 is an analytic map defined in a neighborhood of the origin of @xmath569 , with values in @xmath569 such that @xmath570 . there are several strategies to study such a local dynamical system . the first one is to compute the flow ( whose existence , uniqueness and analyticity is due to the cauchy - lipschitz theorem ) , that is the local biholomorphism from @xmath571 to @xmath571 : @xmath572 that describes the local solution , of the system @xmath573 such that @xmath574 . note that , since the system is autonomous ( @xmath575 does not depend on @xmath536 ) , for small values @xmath536 and @xmath576 , we have @xmath577 . although this flow encodes all the properties of the solutions , its computation does not give many information on the geometry of the solutions . since poincar , in order to get more geometrical information , a better approach is to study this solutions up to a change of coordinates . let @xmath214 be a local ( @xmath578 ) biholomorphism of @xmath569 , then if @xmath579 where @xmath23 is a solution of the initial system , then : @xmath580 and the vector field @xmath575 is transformed into the vector field @xmath581 by the equation @xmath582 this defines an equivalence relation of vector field up to local biholomorphism and , in order to grab the geometry of the solutions , it seem reasonable to try to find the most simple vector field @xmath581 which is conjugated to @xmath575 . for the sake of simplicity we will suppose in the sequel that the _ linear part _ of @xmath575 is diagonal : @xmath583 where @xmath584 is a diagonal matrix whose diagonal @xmath585 is called the _ spectrum _ , and the components of @xmath586 are of valuation at least 2 ( for the usual valuation in @xmath587 ) . in this context ( as in the context of perturbative quantum field theory ) , it seem reasonable to consider @xmath575 as a _ perturbation _ of its linear part and to ask if it is conjugated to its linear part : does there exist a local biholomorphism of @xmath569 , tangent to the identity ( in order to preserve the linear part ) such that : @xmath588 this equation is called the homological equation and it provides recursive equations on the coefficients of @xmath214 . unfortunately ( see for example @xcite or @xcite ) many difficulties arise in the resolution of such equation ( see also below ) : if , for any @xmath589 and any sequence of non - negative integers @xmath590 with @xmath591 , we have : @xmath592 then the vector field is _ non - resonant _ and one can solve recursively the homological equation . unfortunately , the series @xmath214 is not necessarily convergent ( i.e. analytic ) , unless the spectrum satisfies some _ diophantine condition _ ( see @xcite ) . otherwise , when there are some cancellations of the values @xmath593 , then the homological equation can not _ even formally _ be solved . in this case , the case of resonant vector fields , solving recursively the homological equation involves some divisions by @xmath16 . this also means that , in this case , the vector field can not be _ linearized _ : at best , one can try to conjugate @xmath575 to a simpler vector field ( but not its linear part ) which is called a normal ( or pre - normal ) vector field @xmath594 such that the components @xmath595 of @xmath210 contains only monomials @xmath596 such that @xmath597 . this is indeed possible but nothing ensures once again that we get an analytic solution , either for @xmath214 nor for @xmath581 . this small survey of the situation calls for some remarks . we shall leave aside the analytic difficulties ( for which their still exists open problems ) and focus on the algebraic difficulties , that is computing formal series rather than analytic series . even formally , there are some obstruction to the resolution of the homological equation : we started with the naive guess that , since @xmath575 can be seen as perturbation of its linear part , it shall be conjugated to it . this look like the situation in perturbative quantum field theory ( pqft ) , where we see the action as a perturbation of the free action , and as in pqft , this gives a ill - defined solution ( here the map @xmath214 ) . the analogy between resonant vector field and pqft goes even further : in the resonant case , the computation of a normal form can be obtained by a renormalization procedure ( as in pqft ) . as pointed out in @xcite and @xcite , pqft leads to the computation of a character ( maybe ill - defined ) on a graded commutative hopf algebra , that is to say to the computation of an element of the lie group of a complete graded lie algebra . the situation is indeed the same in the study of _ formal _ differential equation , when considering vector fields as _ operators _ : let @xmath575 is a formal vector field without constant term ( @xmath598^{\nu } = \mathbbm{c}_{\geqslant 1 } [ [ x]]^{\nu}$ ] ) , and @xmath599 $ ] a formal series , if @xmath600 and @xmath601 then : @xmath602 so that we can identify , in a unique way , a vector field @xmath575 with a differential operator ( even a derivation ) @xmath603 _ acting _ on @xmath604 $ ] : @xmath605 on the same way , if @xmath214 is a local formal , invertible diffeomorphism of @xmath569 , that is to say @xmath606^{\nu } = \mathbbm{c}_{\geqslant 1 } [ [ x]]^{\nu}\ ] ] with an invertible linear part @xmath607 , then one can associate an invertible operator ( called a substitution automorphism ) @xmath608 acting on @xmath604 $ ] : @xmath609 moreover , it is a matter of fact to check that a linear operator @xmath254 on @xmath604 $ ] such that : let @xmath613 ( resp . @xmath614 ) the set of derivations @xmath615 such that @xmath570 ( resp . @xmath575 has no linear part ) . @xmath613 ( resp . @xmath614 ) is a complete graded lie algebra ( for the lie bracket associated to the composition of operators ) . moreover , if @xmath616 ( resp . @xmath617 ) is the group of substitution automorphisms associated to local formal , invertible diffeomorphisms ( resp . identity - tangent formal diffeomorphisms ) , then the exponential map : @xmath618 } + \sum_{s \geqslant 1 } \frac{1}{s ! } x^s_{}\ ] ] defines an injective ( resp . bijective ) map from @xmath613 ( resp . @xmath614 ) to @xmath616 ( resp . @xmath619 ) whose inverse on @xmath617 is given by the logarithm : @xmath620})\ ] ] a proof can be found in @xcite and we refer to @xcite and @xcite for some applications of this operator formalism in dimension @xmath621 . there is no difficulty in proving such a proposition . let us just notice that the graduation is related to the action of such derivation on monomials : a vector field in @xmath613 ( resp . @xmath622 ) is given by a series of operators such as @xmath623 with @xmath624 ( resp . @xmath625 ) that acts on monomials @xmath626 : @xmath627 so that the total degree goes from @xmath628 to @xmath629 and the graduation for such an operator is then @xmath630 . note also that this proposition implies that any substitution automorphism in @xmath617 is a differential operator . this could have been proved directly , using the taylor formula : if @xmath631 , then @xmath632 if we look back at the ( formal ) flow of a vector field ( or rather derivation ) @xmath633 in @xmath613 ( resp . @xmath614 ) , the reader can check that the associated one parameter family of substitution automorphisms @xmath634 in @xmath616 ( resp . @xmath617 ) is given by @xmath635 and satisfies the operator equation @xmath636 on the same way the homological equation can be formulated as follows : let @xmath637 be a vector field of @xmath638 , with linear part @xmath639 and @xmath640 , if this vector field is conjugated by @xmath214 to its linear part and @xmath641 , then : @xmath642 and if @xmath67 is the inverse of @xmath254 , this reads @xmath643 = x_0 .g - g.x_0 = g.b\ ] ] so that the homological equation is an equation relating an element of @xmath614 and an element of @xmath617 , through the adjoint action of @xmath639 : @xmath644 is a logarithmic derivative of @xmath67 for the derivation @xmath645 ( that respects the graduation ) . this is precisely the kind of equation we studied in the abstract context of complete graded lie algebras , equipped with a _ graded derivation _ @xmath2 that has essentially the same properties as in the case of the adjoint action for vector fields . for example , if @xmath646 , then @xmath647 so that a vector field @xmath648 is resonant if @xmath2 is not invertible on @xmath614 . one can also notice in this case that , on restriction to @xmath614 , we have @xmath649 and @xmath495 is a sub - lie algebra . alain connes and dirk kreimer . renormalization in quantum field theory and the riemann - hilbert problem . the @xmath170-function , diffeomorphisms and the renormalization group . , 216(1):215241 , 2001 . jean ecalle and bruno vallet . passive and active resonance . non - linear resurgence and isoresurgent deformations . in _ the stokes phenomenon and hilbert s 16th problem ( groningen , 1995 ) _ , pages 103138 . world sci . publ . , river edge , nj , 1996 . giovanni gallavotti . a criterion of integrability for perturbed nonresonant harmonic oscillators . `` wick ordering '' of the perturbations in classical mechanics and invariance of the frequency spectrum . , 87(3):365383 , 1982/83 . jean martinet . normalisation des champs de vecteurs holomorphes ( daprs a .- d . brjuno ) . in _ bourbaki seminar , vol . 1980/81 _ , volume 901 of _ lecture notes in math . _ , pages 5570 . springer , berlin , 1981 . frdric menous . formal differential equations and renormalization . in _ renormalization and galois theories _ , volume 15 of _ irma lect . math . _ , pages 229246 . soc . , zrich , 2009 .
we develop a time - dependent theory to investigate electron dynamics and photoionization processes of diatomic molecules interacting with strong laser fields including electron - electron correlation effects . we combine the recently formulated time - dependent generalized - active - space configuration interaction theory [ d. hochstuhl and m. bonitz , phys . rev . a * 86 * , 053424 ( 2012 ) ; s. bauch et al . , phys . rev . a * 90 * , 062508 ( 2014 ) ] with a prolate spheroidal basis set including localized orbitals and continuum states to describe the bound electrons and the outgoing photoelectron . as an example , we study the strong - field ionization of the two - center four - electron lithium hydride molecule in different intensity regimes . by using single - cycle pulses , two orientations of the asymmetric heteronuclear molecule are investigated : , with the electrical field pointing from h to li , and the opposite case of . the preferred orientation for ionization is determined and we find a transition from , for low intensity , to , for high intensity . the influence of electron correlations is studied at different levels of approximation , and we find a significant change in the preferred orientation . for certain intensity regimes , even an interchange of the preferred configuration is observed , relative to the uncorrelated simulations . further insight is provided by detailed comparisons of photoelectron angular distributions with and without correlation effects taken into account .
1507.04107
the rapid progress in experimentally observing and even controlling electron dynamics in atoms and molecules demands powerful theoretical approaches ; see , e.g. , @xcite for reviews on this subject . one of the most challenging and , therefore , interesting tasks is the accurate description and understanding of the ultrafast and complex behavior arising from the electron - electron interaction . it can be expected that mean field , i.e. , hartree - fock - type approaches are insufficient , and that electronic correlations are important . these are especially difficult to treat in time - dependent theories with more than two active electrons , due to the complexity of the multi - electron wave function and even more so if the continuum is included for photoionization ; see @xcite for an overview . of particular interest in the context of strong field physics are molecular systems due to their much more complex dynamics and degrees of freedom owing to their geometrical structure . with the development of alignment and even orientation techniques @xcite measurements in the molecular - fixed frame of reference become accessible , which allows for an investigation beyond orientation - averaged quantities . one question is the preferred direction of electron emission with respect to the electrical field direction of a linearly polarized laser and the influence of correlation effects in strong - field excitation scenarios of heteronuclear molecules . in a first approximation , the tunnel ionization maps the highest - occupied molecular orbital ( homo ) to the continuum @xcite , and by recollision strong - field ionization was even used to illustrate the homo experimentally @xcite . however , experimental evidence using co molecules @xcite showed that this simplified one - electron picture needs to be adjusted , as effects such as inner - shell polarizations @xcite , stark shifts and orbital distortions @xcite have impact on the ionization dynamics . the question to what extent electronic correlations are important remains open . in ref . @xcite this topic has been addressed within a one - dimensional model of the four - electron lih molecule , and a shift of the preferred direction of emission is observed when electronic correlations are included . the immediate question of whether these effects are present in a full three - dimensional analysis shall be answered by this work and completed by angle - resolved investigations . all of these above - discussed issues call for a general , time - dependent theory including external ( possibly strong ) fields beyond a perturbative approach . the fundamental equation describing the physics of these quantum systems is the ( non - relativistic ) time - dependent schrdinger equation ( tdse ) . however , its direct numerical solution , even by means of supercomputers , is limited to systems consisting of only one or two electrons , e.g. helium @xcite or molecular hydrogen @xcite . semi - analytical theories , such as the strong - field approximation and tunneling theories @xcite provide physical insight but often draw on a simplified picture of the electron - electron interactions . in order to solve the time - dependent schrdinger equation for more than two active electrons including the electrons interactions , approximate numerical techniques need to be employed . these include the time - dependent configuration interaction singles ( td - cis ) method @xcite , multi - configuration time - dependent hartree - fock ( mc - tdhf ) @xcite or its generalizations time - dependent restricted or complete active space self - consistent - field ( td - ras / cas - scf ) @xcite and the state - specific - expansion approach @xcite ( see also ref . @xcite for an overview ) . further , time - dependent density - functional theory ( td - dft ) @xcite and time - dependent close - coupling solutions of the tdse by using pseudo - potentials for the description of more than two electrons @xcite have been applied to photoionization of molecules . especially the mctdhf family suffers from complicated non - linear numerics , and its applicability to photoionization is not yet fully understood . td - dft and the pseudo - potential approaches , on the other hand , rely strongly on the chosen functionals or potentials with unknown accuracy and lack tunable parameters to achieve convergence to the fully correlated solution . one of the most successful methods which bears some similarities to our present approach , is the time - dependent r - matrix method @xcite . the aims of the present work are ( i ) to provide a fully _ ab - initio _ time - dependent approach to electron dynamics in diatomic molecules exposed to strong laser fields including a systematic ( i.e. controllable ) approach to electron correlation without relying on pseudo - potentials and ( ii ) to demonstrate the method by shining light onto the question of whether electronic correlation decides from which end an electron leaves a heteronuclear molecule which is exposed to a strong electric single - cycle pulse . our approach is based on the time - dependent generalized - active - space configuration interaction ( td - gas - ci ) formalism which we apply within a prolate spheroidal single - particle basis set in combination with the well - established partition - in - space concept to tackle the scattering part of the hamiltonian . the paper is organized as follows . after a brief introduction into the theory of td - gas - ci , we give a detailed overview on the used basis set and details of our implementation , in section [ sec : theory ] . technical aspects and the explicit formulas and strategies of their efficient numerical handling are presented in the corresponding appendices . in sections [ sec : results ] and [ sec : sfi - lih ] , we show illustrative numerical examples and demonstrate the abilities of the present approach . we focus on the lih molecule in strong single - cycle infrared ( ir ) pulses and explore the influence of electronic correlations on the molecular photoelectron angular distributions ( pads ) and the preferred direction of electron emission as a function of the geometrical set - up . the paper closes with conclusions and a discussion of future applications of the present theory . let us consider @xmath0 electrons moving in the potential of two nuclei at positions @xmath1 and @xmath2 with charge numbers @xmath3 and @xmath4 . throughout , we employ the born - oppenheimer approximation @xcite , which decouples the nuclear and electronic degrees of freedom , and use atomic units ( @xmath5 ) . the ( electronic ) hamiltonian is given by @xmath6 with the one - electron contribution of the @xmath7-th electron @xmath8 consisting of the kinetic and potential energies with @xmath9 and the coulombic electron - electron interaction . the time - dependent external laser field is denoted by @xmath10 and is included in dipole approximation using the length gauge via the position operator @xmath11 . we solve the tdse for the @xmath12 electrons , @xmath13 within the td - gas - ci framework @xcite with the hamiltonian . thereby , we expand the many - particle wave function into a basis of time - independent slater determinants @xmath14 , @xmath15 with time - dependent complex coefficients @xmath16 , which results in the matrix representation of the tdse , @xmath17 the slater determinants @xmath14 are constructed from single - particle spin orbitals @xmath18 with the spatial coordinate @xmath19 and the spin coordinate @xmath20 and @xmath21 , where @xmath22 is the dimension of the spatial basis set . details of the orbitals are given in sec . [ ssec : spbasis ] . the matrix elements of the gas hamiltonian , @xmath23 , can be evaluated either by directly using slater - condon rules @xcite or by efficient techniques from ( time - independent ) quantum chemistry @xcite . the most demanding task , besides the time propagation of eq . , remains the evaluation of the one- and two - electron integrals in @xmath24 . details of our strategy are given in sec . [ ssec : rot - basis ] and app . [ app : integral_trafo ] . the included determinants @xmath25 in sums and are chosen according to the gas concept described in detail in ref . thereby , the method ranges from single - active electron ( sae ) @xcite to ( exact ) full ci . we propagate eq . using a short - iterative arnoldi - lanczos algorithm @xcite which results in the repeated application of large - scale matrix - vector multiplications ( up to @xmath26 in one simulation ) where the high degree of sparsity of the gas hamiltonian can be efficiently exploited . the algorithm is applied with an adaptive dimension of the krylov space @xcite for the propagation of the wave packet with a time - dependent hamiltonian and with an adaptive time - step @xcite for the propagation with a time - independent hamiltonian after excitation with a laser pulse . for the efficient solution of eq . , a proper single - particle spin - orbital basis @xmath18 with an associated spatial orbital basis @xmath27 is required to construct the determinantal basis @xmath14 . for quantum - chemistry calculations , typical basis sets are founded on expansions in localized functions , such as gaussian- or slater - type orbitals ( see , e.g. , @xcite ) . these sets achieve a high precision for bound - state properties but lack an efficient description of the scattering part of the hamiltonian . for atomic targets , i.e. , single - center potentials , typically mixed basis sets with radial grids in combination with spherical harmonics are used @xcite . however , for multi - center geometries , the convergence exhibits unfavorable scaling properties for our case with the required expansion of the angular coordinates @xcite . to overcome this problem for diatomic molecules , we use two - center prolate spheroidal coordinates , in the following . alternatively , also for larger molecules with more complicated geometries , a combined gaussian and discrete variable representation ( dvr ) can be applied @xcite . the two - center problem of the diatomic system can be handled efficiently by prolate spheroidal ( confocal elliptic ) coordinates . therein , we define @xmath28 as half the distance between the two centers ( nuclei ) , see fig . [ fig : prol_coords ] : @xmath29 and @xmath30 are the distances between center @xmath31 and center @xmath32 and the electron , @xmath33 the vectors @xmath34 and @xmath35 ( @xmath36 ) are given in cartesian coordinates . the prolate spheroidal coordinates @xmath37 and @xmath38 are then defined as @xmath39,\\ \phi & = \arctan\left(\frac{r_y}{r_x}\right),\ ; \;\ ; \phi \in [ 0,2\pi ] . \end{aligned}\ ] ] with this definition , center a is located at @xmath40 , and the binding potential for the electrons , eq . , takes the form @xmath41 . \label{eq : pot_diat}\ ] ] explicit expressions needed for the implementation are comprised in the appendix . for the set - up of the spatial basis , we follow closely refs . @xcite . we use a direct product basis where @xmath37 is represented by a finite - element dvr(fedvr ) basis whose flexibility regarding the density of grid points avoids complicated coordinate scalings @xcite . coordinate @xmath38 is handled by a usual gauss - legendre dvr@xcite , which is well suited for this problem because spheroidal wave functions are represented by legendre polynomials . the spheroidal wave equation is similar to the one - particle schrdinger equation ( see , e.g. , @xcite ) . although @xmath42 ( @xmath43 is the electronic orbital angular momentum ) does not commute with the hamiltonian , @xmath44 ( component of the electronic orbital angular momentum along the internuclear axis ) does , and the associated quantum number @xmath45 is a `` good '' one @xcite . we , therefore , expand the @xmath46-dependent part of @xmath47 into the eigenvectors of @xmath44 , which , in our case , outperformed a fourier - grid - hamiltonian basis in the @xmath46 coordinate @xcite : @xmath48 note that in refs . @xcite spherical harmonics @xmath49 $ ] for the @xmath38 and @xmath46 coordinates are used . is then the same basis used in this work . ] this basis shows slightly better convergence than a dvrin @xmath38 . however , the resulting electron integrals are less sparse and the basis is non - orthogonal , which complicates the determinantal basis . to fulfill proper boundary conditions , we use gauss - radau quadrature for the first finite element , which ensures that there is no grid point at the singularity @xmath50 . gauss - lobatto quadrature is used for the remaining elements , as usual in fedvr @xcite . to avoid singularity of the kinetic energy matrix , i.e. , to render the matrix invertible for the calculation of the two - electron repulsion integrals ( see app . [ app : matrix_elements ] ) , the very last dvrpoint of the last element is not included in our grid . thereby an infinite potential barrier at the grid end is created , which forces the wave function to vanish asymptotically ( dirichlet boundary condition ) . if the grid is large enough ( i.e. , reflections are avoided ) , effects due to this procedure are negligible . the one - electron primitive functions used in this work are @xmath51 where a multi - index @xmath52 has been defined for convenience . the form of the functions @xmath53 and the matrix elements of the kinetic , potential and interaction energies are given in app . [ app : matrix_elements ] along with details of their derivation . in analogy to refs . @xcite , we use a partition - in - space concept to allow for an efficient description of the photoionization process . similar strategies are also applied in time - dependent @xmath54-matrix theory , e.g. @xcite , and in ref . @xcite . here the basis set is split at @xmath55 into two parts : an inner region , @xmath56 , and an outer part , @xmath57 . the splitting point @xmath58 is chosen such that it coincides with an element boundary of the fedvr expansion . this assures the continuity of the wave function across the grid and avoids the evaluation of connection conditions , see ref . @xcite for a detailed investigation in one spatial dimension . the basis in the inner region , @xmath59 , is constructed from hartree - fock - like rotated orbitals , @xmath60 where @xmath61 is the orbital coefficient matrix . this rotation is needed because the energy of a truncated ci wave function changes under a unitary transformation of the underlying single - particle basis ; a good single - particle basis drastically enhances convergence with respect to the size of the truncated ci space , but comes at the cost of expensive integral transformations which destroy the desired ( partial ) diagonality of the integral matrices @xcite . since we are interested in one - electron photoionization with the simultaneous excitation of the ion , we follow the detailed investigations in @xcite and use pseudo - orbitals based on the @xmath62 electron hartree - fock problem for the virtual orbitals in the rotated part of the basis . here , in contrast to the procedure shown in ref . @xcite , the hartree - fock problem for the @xmath62 electronic problem is solved with the exchange - potential included , which is appropriate for obtaining localized virtual orbitals . the outer part , @xmath57 , of the basis consists of non - rotated , `` raw '' functions which describe the wave packet in the continuum accurately . one block of the coefficient matrix @xmath61 is hence diagonal , see appendix [ subsec : integral_trafo_part_rot_bas ] . an exploitation of the properties of this basis is inevitable for a fast and memory - friendly code [ recall that the two - electron integrals scale , for an arbitrary basis , as @xmath63 whereas for the dvrbasis set they scale as @xmath64 . appendix [ app : integral_trafo ] gives details for the efficient transformations of the integrals . in this part , we discuss the extraction of the relevant observables from the gas wave function in the mixed - prolate basis set . more basis - independent details can also be found in ref . @xcite . angular distributions of photoelectrons contain a wealth of information ( see , e.g. , @xcite ) , and , especially in strong fields , dynamical properties of the rescattering process lead to rich structures @xcite . the molecular angular - resolved photoionization yield ( photoelectron angular distribution , pad ) is defined as @xmath65 @xmath66 is the charge density in spherical coordinates ( @xmath67 is the azimuthal angle in the @xmath68-@xmath69-plane ) , @xmath70 with the spin - summed single - particle density matrix @xmath71 and @xmath72 are the creation and annihilation operators , respectively , of a spin orbital with spatial index @xmath52 and spin @xmath20 . the critical radius @xmath73 is chosen to be sufficiently large such that only the `` ionized '' part of the charge density is used for integration . this neglects the long - range character of the coulomb potential and is strictly valid only for @xmath74 . therefore , several @xmath73 are used to check convergence . @xmath73 lies usually in the outer region of the partially rotated basis . the orbitals sampled at grid points in spherical coordinates , @xmath75 , cf . . , can be efficiently stored in a sparse vector format by exploiting the locality of the fedvrfunctions . this decreases the memory requirements and the computation of the charge density in typical computations by more than three orders of magnitude . the photoionization yield at time @xmath76 can be retrieved from the integrated pad : @xmath77 the momentum distribution of the photoelectron is obtained by using the fourier - transformed basis functions @xcite . only basis functions outside @xmath73 are used , ignoring the central region . this approach is exact for sufficiently large @xmath73 @xcite . the fourier transform of function @xmath78 is defined as @xcite @xmath79 where @xmath80 and @xmath81 are vectors in prolate spheroidal coordinates and @xmath82 is the inner product in these coordinates . if the @xmath46-component of @xmath81 is either @xmath83 or @xmath84 , analytical expressions of eq . can be obtained . however , the integral kernel is nonanalytic which prohibits the usage of the dvr properties ( gauss quadrature ) of the basis functions for the integration . therefore , we employ the fast fourier transform of the basis functions in cartesian coordinates . an application of eq . with the fourier - transformed basis functions gives then the momentum distribution . let us consider the diatomic molecule lih , i.e. , @xmath85 and @xmath86 . it is the smallest ( @xmath87 ) possible heteronuclear molecule and exhibits a spatial asymmetry with respect to its geometrical center . it is a frequently chosen theoretical model to test correlation methods ( e.g. , in one spatial dimension in refs . @xcite ) . ) . the iboview program was used for the generation of the orbitals using an isosurface with a density - threshold of @xcite.,scaledwidth=30.0% ] we use an internuclear distance of @xmath88 ( @xmath89{\textup{\aa}}$ ] ) , which is the equilibrium geometry at ccsd(t)/cc - pcv5z level ( without the frozen core approximation ) and the experimental value @xcite . the hf electronic structure consists of a valence ( homo ) and a core orbital ( 1s of lithium ) , which are given in fig . [ fig : orbitals ] . the electric field is linearly polarized parallel to the internuclear axis . we consider envelopes of gaussian shape , @xmath90 \cos\left[\omega(t - t_0)+ \varphi_{\textup{cep}}\right ] , \label{eq : field_gauss}\end{aligned}\ ] ] and of @xmath91 shape , @xmath92 with the photon energy @xmath93 , the amplitude @xmath94 , and the carrier - envelope phase ( cep ) , @xmath95 . the simulations are carried out within the fixed - nuclei approximation , which is well - justified since the dynamics of the nuclei are on a much longer timescale than the considered pulse durations . all data are retrieved by using the length gauge , cf . eq . , which is preferable over the velocity gauge in the case of tunnel - ionization dynamics with few - cycle pulses @xcite . the td - gas - ci method is well suited for photoionization problems @xcite , as it can be tailored to the problem at hand . for constructing the gas , we assume that multiple ionization is negligible due to the much larger ionization potential of . we consider three types of gas partitions which are sketched in fig . [ fig : gas_cas ] : the single - active electron ( sae ) approximation @xcite , the ci singles ( cis ) approximation and a complete - active - space ( cas ) with single excitations from this subspace to the remaining orbitals in the outer region . following ref . @xcite , we denote this type of gas with cas@xmath96 , with @xmath97 describing the number of electrons and @xmath98 the number of spatial orbitals in the cas . the star indicates the single excitations out of the cas . for the lih molecule , cas@xmath99(2,@xmath98 ) describes a cas with frozen core , and for cas@xmath99(4,@xmath98 ) , all electrons are active . as a sidemark , we mention that this type of gas is equivalent to a multi - reference cis description . the size of the active space , i.e. , @xmath98 in cas@xmath100 , remains to be chosen adequately and to be checked carefully for convergence . for , the accurate description of correlation effects requires orbitals with higher quantum numbers @xmath101 in the @xmath46 coordinate , i.e. , @xmath102 . in contrast to time - independent cas - scf simulations , where typically a few additional orbitals , such that open shells are filled in the cas , give satisfactory results for the stark - shifted ground - state energies , time - dependent calculations are more involved . here , in order to describe intermediate states accurately , a larger size of the cas is crucial . reasonable cas configurations contain closed subshells in the number of active orbitals , @xmath98 , i.e. , all orbitals with a certain symmetry . this leads , for example , to a reasonable space of cas@xmath99(2,5 ) for orbitals with @xmath103 and @xmath104-orbitals for two active electrons [ cas@xmath99(4,6 ) for four active electrons ] . typically , we use cas spaces with orbitals of up to @xmath105 , cas@xmath99(2,8 ) and cas@xmath99(2,12 ) , and @xmath106 symmetry , cas@xmath99(2,10 ) . { w\ , cm^{-2}}$ ] with @xmath107 . data is shown for single - cycle excitation of configuration , see sec . [ sec : sfi - lih ] for details . the pad for cas@xmath99(2,10 ) with @xmath108 ( cas@xmath99(4,4 ) ) coincides with that of cas@xmath99(2,8 ) ( cas@xmath99(2,3 ) ) and is therefore not shown . ] the convergence of the method is illustrated with an example of the photoelectron angular distributions ( pads ) for the strong - field ionization of lih using single - cycle pulses in fig . [ fig : pad_convergence ] , see sec . [ sec : sfi - lih ] for the field parameters . all gas approximations predict the dominant electron emission in direction of the field polarization , @xmath109 . however , the sae and cis approximations drastically underestimate the total yield and fail to predict the correct positions of the side maxima ( see inset of fig . [ fig : pad_convergence ] ) . the strongest difference is observed in direction of @xmath110 . by increasing the active space , a successive convergence is achieved and for a cas@xmath99(2,8 ) only small differences appear in comparison to larger spaces ( blue dashed vs. solid blue line ) . therefore , we typically use the cas@xmath99(2,8 ) model in the following . the convergence was checked for different field parameters additionally . to demonstrate the method , we first consider the case of one - photon absorption in . the parameters for the electric field , eq . , are @xmath111 ( @xmath112{ev}$ ] or @xmath113{nm}$ ] ) , @xmath114 ( @xmath115{w\ , cm^{-2}}$ ] ) , @xmath116 [ @xmath117{fs}$ ] full width at half maximum ( fwhm ) of intensity ] , @xmath118 ( @xmath119{fs}$ ] ) , and @xmath120 . the tdse in gas approximation is propagated until @xmath121 ( @xmath122{fs}$ ] ) . we use up to three functions in @xmath46 ( @xmath107 ) , ten in @xmath38 , and 1386 in @xmath37 . the inner region in @xmath37 consists of two elements with 10 and 18 basis functions each and ranges @xmath123 $ ] and @xmath124 , respectively . the non - rotated basis is formed by 80 equidistantly distributed elements with 18 basis functions in each element within @xmath125 . { fs}$ ] ) pulse with @xmath112{ev}$ ] and an intensity of @xmath115{w\ , cm^{-2}}$ ] within sae and cas@xmath99(2,8 ) ( converged ) approximation . ] the kinetic energy spectrum of the photoelectron is shown in fig . [ fig : distr_resonant ] for the sae approximation and the converged cas@xmath99(2,8 ) . the spectrum shows a strong peak at the expected position of @xmath126 , where @xmath127 is the ionization potential according to koopman s theorem , i.e. , the negative hf energy of the valence orbital , @xmath128 . this peak is rarely shifted by correlations ( solid line ) but a series of additional peaks appears at lower kinetic energies ( see inset of fig . [ fig : distr_resonant ] , @xmath129 ) . these can be attributed to a correlation - induced sharing of the photon s energy between the photoelectron and a second electron still bound in the ion ( `` shake - up '' process ) . thereby , the photoelectron energy is reduced and the ion remains in an excited state . the origin of this process is purely correlation - induced and can neither be described within neither the sae approach , the cis approach @xcite , nor td - hf simulations @xcite . the population dynamics of these states can be measured , e.g. , by strong - field tunneling @xcite . the corresponding angle - resolved momentum spectrum depicted in fig . [ fig : momentum_distr_resonant ] , contains additional information . the photoelectron shows characteristic angular distributions for the different peaks with distinct locations of the maxima . the outer circle with a radius of about @xmath130 corresponds to the main photoelectron peak at an energy around @xmath131 in fig . [ fig : distr_resonant ] . the inner ( fainter ) circles stem from the shake - up state population and exhibit a significantly different angular dependence than the main peak . this is caused by the different selection rules for the simultaneous excitation of two electrons and , therefore , the changed angular momentum of the escaping electron in comparison to the dominant ionization channel with in its ground state . an integration cutoff of @xmath132{a_0}$ ] ( radial coordinates ) was used . see fig . [ fig : distr_resonant ] and text for parameters.,scaledwidth=45.0% ] we now turn our attention to the case of strong and short pulses , for which a time - dependent theory is indispensable . let us consider single - cycle pulses of the form of eq . with @xmath133 ( @xmath134{nm}$ ] ) which corresponds to a duration of . the electrical field exhibits a strong cep dependence with a predominating orientation at the maximum field strength , see fig . [ fig : scenarios ] ( b ) . for the configuration ( ) . the black horizontal line denotes the energy of the active orbital . panel ( b ) shows the time dependence of the single - cycle pulse.,scaledwidth=48.0% ] this dependence corresponds to two different ionization scenarios , i.e. , orientations of the molecule with respect to the field at the maximum intensity of the linearly polarized pulse . we will refer to these situations as , if the field points from the h to the li end [ panel ( a ) in fig . [ fig : scenarios ] ] and , if the field points from the li to the h end [ panel ( c ) ] . the single - particle basis is similar to sec . [ sec : one - photon_ionization ] , but with 16 functions in @xmath38 and 920 functions in @xmath37 : 8 and 14 functions in the inner region and 100 elements with 10 functions each in the outer region within @xmath135 . the total number of basis functions in our simulation is @xmath136 ( @xmath137 ) , @xmath138 ( @xmath107 ) and @xmath139 ( @xmath108 ) . we further note by comparing the ionization potentials of ( @xmath140 ) and ( @xmath141 ) that the keldysh parameter @xcite @xmath142 is @xmath143 times larger for the ion . therefore , we conclude that single excitations into the outer region is a valid gas approximation and double ionization is negligible . convergence aspects of the size of the cas were discussed in sec . [ subsec : results_ci_spaces ] . { w\ , cm^{-2}}$ ] for both scenarios , ( left ) and ( right ) , cf . [ fig : scenarios ] . the classical force @xmath144 points downwards.,scaledwidth=48.0% ] the charge densities at different times during the simulation for a field strength of @xmath145 ( @xmath146{w\ , cm^{-2}}$ ] ) and the two scenarios and are given in fig . [ fig : charge - density ] for the converged cas@xmath99(2,8 ) ( @xmath107 ) . the main dynamics happens after about @xmath147{fs}$ ] when the maximum amplitude of the pulse is reached and the tunneling ionization sets in . the ejected part of the wave packet exhibits a characteristic angular distribution , visible in the logarithmic density plot . after the ionization , the molecular ion remains in rydberg states and performs coherent oscillations between the electronic states . a closer inspection of the released wave packet after the pulse is over ( @xmath148{fs}$ ] ) reveals a significant difference for scenarios and in both , the angular distribution and the absolute yield . the results presented in fig . [ fig : charge - density ] confirm also the observation in ref . @xcite utilizing a one - dimensional model for that ionization for field strength @xmath149 is preferred from the -end , i.e. , using the configuration . this finding will be quantified and discussed in detail in the remaining part of the paper . to address the question of whether the ionization yield is larger if the linearly polarized light is pointing from the - to the -end ( configuration ) or _ vice versa _ ( ) , the ratio @xmath150 is defined @xcite . @xmath151 is smaller ( larger ) than one if the electron is ejected mostly in the direction from the - to the -end ( - to -end ) ; see also fig . [ fig : scenarios ] . in ref . @xcite a similar parameter was defined for the co molecule and a change of the direction for single - active - orbital calculations in comparison to ( uncorrelated ) td - hf calculations including inner orbitals was found . let us first consider a field strength of @xmath145 ( @xmath146{w\ , cm^{-2}}$ ] ) . from the previous discussion , we expect @xmath152 , since ionization is preferred for configuration , cf . [ fig : charge - density ] . this can be understood due to the increased total density at the li nucleus and the direction of the classical force @xmath144 acting on the electronic density . for the opposite configuration , , large portions of the electronic density have to `` pass '' an additional potential well , cf . [ fig : scenarios ] . this simple picture is confirmed by sae and cis calculations in tab . [ tab : conv_eta ] with values of @xmath153 . the total ionization yield is , therefore , about a factor of three larger for than for at field strength @xmath149 . additionally , tab . [ tab : conv_eta ] demonstrates the behavior of @xmath151 with respect to electron correlations by using different sizes of the cas . similar convergence behavior is also found for other field strengths . by successively increasing the cas , first by including only two active electrons , cas@xmath99(2,@xmath154 ) , @xmath151 increases , as it was observed in one - dimensional @xcite . most of the correlation contributions are captured by a cas@xmath99(2,8 ) with a value of @xmath155 and only less than 4% change is found by further increasing the active space . sae and the cis approximations , however , predict values between @xmath156 and @xmath157 . therefore , correlations shift the preferred end of ionization from the li to the h end significantly . we note that too small cas , e.g. cas@xmath99(2,3 ) , give inaccurate results because of a bias due to an improper selection of additional important configurations , which is a general pitfall of multi - reference methods @xcite . however , to test the frozen - core approximation ( correlations arising from the two core electrons are not taken into account ) , we also performed calculations with all four electrons active for a small active space [ cis and cas@xmath99(4,4 ) in tab . [ tab : conv_eta ] ] . by comparing to cis ( frozen core ) or cas@xmath99(2,3 ) , respectively , we find that the observable does not change substantially . for larger cas , a similar behavior is expected , as was demonstrated for one - dimensional systems in ref . also an increase of @xmath158 from one to two does not change the result . for both , the sae approximation and cis , even @xmath137 is sufficient , because the used orbitals in @xmath46 are the corresponding eigenvectors of the one - electron problem . .ionization asymmetry parameter @xmath151 , eq . , for the single cycle pulse with field strength @xmath149 ( @xmath146{w\ , cm^{-2}}$ ] ) calculated in different gas approximations with one [ sae / cis ] , two [ cas@xmath99(2,@xmath154 ) ] and all four [ cas@xmath99(4,@xmath154 ) ] electrons active . @xmath152 corresponds to a preferred configuration for ionization . [ cols="<,<,>",options="header " , ] , cf . eq . , of the photoionization yields for the two configurations [ see fig . [ fig : scenarios ] ] as a function of the field intensity for the relevant gas approximations . @xmath152 ( @xmath159 ) corresponds to ionization from the - ( ) -end . see text for all other parameters.,scaledwidth=48.0% ] in the previous paragraph , we found a dominating release for configuration at a rather high intensity of @xmath146{w\ , cm^{-2}}$ ] , which is well above the barrier . this is understandable by the higher electron density at the li nucleus . the homo , however , is more localized at the h end ( green orbital to the right in fig . [ fig : orbitals ] ) , in contrast to the full single - particle density . thus , entering the tunneling regime by reducing the intensity , the mapping mechanism of the homo to the continuum becomes important . therefore , we expect larger values of @xmath151 for low intensities , and thus by increasing the intensity , a decrease of @xmath151 ; in other words , we expect a shift from the preferred to the configuration . this expectation is readily verified by the simple sae approximation in fig . [ fig : meta ] ( red line with squares ) . for small intensities , @xmath159 , exhibiting values of about @xmath160 indicating a slightly more favorable ionization of the configuration . for this situation , the cis approximation ( bright yellow line with triangles ) fails to describe the change of preferred configuration . a @xmath151 of maximal @xmath161 is predicted by cis . the better qualitative description of the physics by the sae approximation compared to cis is probably due to error cancellation . by increasing the intensity , a monotonic transition to the previously discussed strong - field over - the barrier regime with @xmath152 is observed . the change of the preferred configuration for ionization from to ( @xmath162 ) occurs for the sae approximation at around @xmath163{w\,cm^{-2}}$ ] . for larger intensities , sae and cis calculations approximately coincide and the curves approach a ratio of about @xmath164 , i.e. , a factor of three higher yield for . for very high field strengths , the value increases again , which results from the fact that for extremely intense pulses , the asymmetry of the binding potential is negligible in comparison to the excitation potential and , therefore , a ratio of @xmath165 in the limit @xmath166 is expectable . electronic correlations [ cas@xmath99(2,8 ) , blue line with circles ] change this picture qualitatively , in particular for smaller intensities : at a specific intensity range from around @xmath167{w\,cm^{-2}}$ ] to @xmath168{w\,cm^{-2}}$ ] , the correlation contributions interchange the dominant direction of emission , @xmath159 for cas@xmath99(2,8 ) and @xmath152 for sae and cis . at @xmath168{w\,cm^{-2}}$ ] , @xmath151 is much larger ( @xmath169 ) for cas@xmath99(2,8 ) compared to sae ( @xmath160 ) , indicating that , during the slow tunneling process of the electron from the -end to the continuum , much electron correlation can be built up . for large intensities , a similar trend as for the uncorrelated calculation is found , but with larger absolute values of @xmath151 , similar to the case of field strength @xmath149 , which is discussed in detail above ; see tab . [ tab : conv_eta ] . we now turn our attention to the fully resolved pads which contain more information than the integral quantity @xmath151 . the correlated cas@xmath99(2,8 ) pads for the two molecular orientations with respect to the electric field at different field strengths are depicted in fig . [ fig : pads_field_strengths ] in polar plots . as expected , the dominant ionization occurs along the field polarization axis for all considered intensity regimes . however , for all field intensities shown , the most favorable ionization direction is the opposite direction of the field , see sketch at lower right of fig . [ fig : pads_field_strengths ] . further , the ejection direction of electrons differs from the preferred configuration for ionization measured by the ratio of the integral quantity @xmath151 . for example , at the highest field strength ( top left in fig . [ fig : pads_field_strengths ] ) the pad shows a preference of over , whereas the value of @xmath152 prefers @xmath170 [ see eq . and fig . [ fig : meta ] ] . this can be attributed to the propagation of the released electrons in the field : whereas @xmath151 depends mainly on the orientation of the molecule with respect to the main peak of the field , the pads are modified by all cycles of the pulse and the cycles following the field maximum at which the tunneling - release of the electrons predominantly occur can change the direction of electron emission significantly . this is similar to the rescattering mechanism for higher - harmonics generation and above - threshold ionization . this picture is verified by the time - dependent electron densities in fig . [ fig : charge - density ] where after the main peak of the pulse at @xmath171{fs}$ ] , the electron density gets accelerated by the smaller side - extremum in opposite direction ( compare with @xmath172{fs}$ ] in fig . [ fig : charge - density ] ) . a similar effect is , e.g. , observed in strong - field ionization of atoms , where rescattering effects can drastically modify the angular distributions of the photoelectrons @xcite . with decreasing field strength ( top left to bottom right ) , the shape of the curves along the main maximum become more oblate and the smaller maxima first increase and then decrease again . for the highest field strengths , the pads for the configuration ( bright yellow curves ) show also ionization contributions to the opposite direction pointing to the end , which is not the case for ( black curves ) . however , with decreasing field strength , the maximum pointing to the other direction is growing for configuration , and gets even larger than that of . remarkably , the positions of the secondary maxima at this intensity are the same regardless of the position of the nuclei , but at intermediate field strengths , an additional maximum for the configuration at the site of the nucleus is visible . this indicates a higher angular momentum for the ejected electron . to single out the influence of electron - electron correlations , the pads from correlated cas@xmath99(2,8 ) calculation and those from a sae calculation are shown in fig . [ fig : pad - sae - cas ] . for the configuration ( left two columns in fig . [ fig : pad - sae - cas ] ) , sae ( bright curve ) underestimates the size of the side maxima , especially at intermediate field strengths ( central panels ) . at smaller field strengths ( @xmath173{w\,cm^{-2}}$ ] and @xmath174{w\,cm^{-2}}$ ] ) , the maximum pointing away from the nucleus is either over- or underestimated , showing that no general pattern can be reasoned from correlation effects in pads at different intensities . for the configuration and for all field intensities ( right two columns in fig . [ fig : pad - sae - cas ] ) , sae drastically underestimates the side maxima at the site of the nucleus . on the other hand , at intermediate intensities , the other side maxima are overestimated by the sae approximation . thus , the favored direction of emission of electrons is decided by the electron - electron correlation for intermediate field strengths ( see fig . [ fig : meta ] ) . in this paper we presented a time - dependent approach to correlated electron dynamics following the excitation of diatomic molecules with strong electromagnetic fields . the method is based on the td - gas - ci approach using a prolate - spheroidal representation of the single - particle orbitals within a partition - in - space concept to allow for good convergence of the truncated ci expansion . thereby , parts of the multi - particle wave function close - by the nuclei are represented within a hartree - fock - like orbital basis and the ejected part is represented in a grid - like fe - dvr basis set . we illustrated the method by its application to the calculation of angle - resolved photoelectron spectra of the four - electron heteronuclear lih molecule with and without taking electron - electron correlation contributions into account . to demonstrate the capabilities of the present approach , we then concentrated on the strong - field ionization of lih using single - cycle pulses . the ionization yield for the two opposite orientations of the molecule along the linearly polarized electric field was calculated and an intensity - dependent shift of the preferred configuration was observed : while for low intensities in the tunneling regime , ionization for is larger , for high intensities well above the barrier , shows higher yields . in between both regimes , a smooth transition is found . by turning on electronic correlations in the simulation , we find that especially yields in the tunneling regime are affected whereas the high - intensity regime is well described using the sae or cis approximations . correlations shift the preferred configuration from to for low intensities and vice versa for high intensities . in a certain intermediate intensity regime even an interchange of the preferred configuration in comparison to uncorrelated calculations is observed . additionally , angle - resolved photoionization distributions were presented and discussed , and the correlation effects were singled out by comparison to sae calculations . our results demonstrate the importance of electron - electron correlations in strong - field excitation scenarios of diatomic molecules . we expect the td - gas - ci approach in combination with the prolate spheroidal basis set to be applicable to larger systems such as the co molecule and to arbitrary polarization of the exciting pulse in the near future , where experimental data is available @xcite . further , the application to two - color excitation scenarios , such as streaking and xuv - xuv pump - probe , and the exploration of correlation effects in molecular systems on ultrashort time scales , e.g. , post - collision interaction effects @xcite or the time - delay in photoemission is within reach . the authors thank c. hinz for indispensable optimizations of the td - gas - ci code . the authors gratefully acknowledge discussions with l. b. madsen . h. r. larsson acknowledges financial support by the `` studienstiftung des deutschen volkes '' and the `` fonds der chemischen industrie '' . this work was supported by the bmbf in the frame of the `` verbundprojekt fsp 302 '' and computing time at the hlrn via grants ` shp00006 ` and ` shp00013 ` . because , to our knowledge , the formulas of all needed matrix elements have not been stated in one single publication or exhibit some misprints , we briefly summarize the equations to simplify their implementation . the volume element and the laplacian are @xmath175 , \end{aligned}\ ] ] the action of @xmath176 on @xmath177 becomes @xmath178 \psi^m.\nonumber\end{aligned}\ ] ] the factor @xmath179 [ e.g. eq . ] is canceled by the volume element , which avoids numerical problems due to the singularities at @xmath180 , @xmath181 . still , for odd @xmath45 values , the exact eigenfunctions show non - polynomial behavior at these points and contain factors of @xmath182 @xcite . non - polynomial functions are poorly represented by dvr , in which gauss quadrature is used . therefore , for odd @xmath45 , we multiply the basis functions @xmath183 by @xmath184 or @xmath185 to avoid a non - polynomial integrand @xcite : @xmath186 @xmath187 ( @xmath188 ) is the dvr grid point of the corresponding basis function . these definitions differ in case of a bridge function in the fedvr basis @xcite . the potential , eq . , is independent of @xmath46 . hence , the potential matrix elements are evaluated using the dvrproperties : @xmath189}{\xi_i^2 - \eta_a^2}. \end{aligned}\ ] ] the matrix elements of the dipole operator in @xmath68-direction are @xmath190 the term stemming from the volume element , eq . , is canceled by the normalization factor of the basis functions , eq . . the matrix elements for the kinetic energy are evaluated using integration by parts and the dvrquadrature @xcite : @xmath191 the coulomb interaction of the electrons in eq . can be decomposed using the well - known neumann series in prolate spheroidal coordinates @xcite : @xmath192 with @xmath193 and @xmath194 . @xmath195 are the spherical harmonics and @xmath196 ( @xmath197 ) are the ( irregular ) associated legendre functions @xcite . for their computation for @xmath198 , the code from ref . @xcite was used . the @xmath38- and @xmath46-dependent parts are evaluated straight - forwardly using the properties of the basis functions . because the legendre functions exhibit a singularity for @xmath199 , the @xmath37-dependent parts are not evaluated by dvrquadrature but by solving the corresponding differential equation of the green s function @xmath200 @xcite . the final expression for the integrals is then @xmath201^{-1}}{\sqrt{\omega_{i_1}\omega_{i_2 } } } \biggr],\nonumber\\ t_{ij}^{lm } = & -\delta_{ij } \left(\frac{m^2}{\xi_i^2 - 1 } + l(l+1)\right ) - \int { \ensuremath{\mathrm{d}}}\xi ( \xi^2 - 1 ) { \ensuremath { \frac{\partial { \theta_i^m } } { \partial { \xi } } } } { \ensuremath { \frac{\partial { \theta_j^m } } { \partial { \xi } } } } .\label{eq : w12_tmat } \end{aligned}\ ] ] @xmath202 is the value of the last ( excluded ) fedvrgrid point , and @xmath203 are the quadrature weights of the fedvrgrid points . in the implementation , the legendre polynomials in @xmath38 and the term in brackets are precomputed and stored in arrays so that the actual computation of the matrix elements is just summing up the product of three array values . the matrix elements are symmetric in @xmath37 and @xmath38 , which can be exploited as well . since the integration over @xmath38 is still handled by usual quadrature , the maximum used value of @xmath204 in the series expansion , @xmath205 , should not be too large such that the integration kernel for the @xmath38-dependent part is not a polynomial of degree @xmath206 any more ( @xmath207 is the number of functions in @xmath38 ) . numerically , however , the results are not very sensitive to the choice of @xmath205 and setting it to the number of used basis functions in @xmath38 leads to good results @xcite . the bottleneck in usual hartree - fock calculations is the generation of the coulomb and the exchange matrix as parts of the fock matrix . the elements of the coulomb matrix are generated by @xmath209 @xmath210 is the density matrix , eq . . the scaling with the number of basis functions @xmath22 is @xmath211 if no additional techniques like density fitting are used . for the exchange matrix , the summation runs over the integral @xmath212 and the optimization procedure is similar . resolving the indices using eqs . and , @xmath213 the last two @xmath214-symbols can be resolved in the overall loop creating the coulomb matrix . therefore , an overall scaling of less than @xmath215 is achieved for the construction of @xmath216 , where @xmath217 is the number of basis functions in @xmath46 . the diagonalization of the fock matrix is then the bottleneck of the scf procedure . in the following , orbitals indexed with @xmath221 or @xmath222 denote rotated orbitals ( from the inner spatial region ) and those indexed with @xmath223 or @xmath224 denote nonrotated orbitals from the inner region that have to be rotated . the rotated orbitals are unitarily transformed by the real - valued coefficient matrix @xmath61 , eq . . the integrals in the rotated molecular orbital frame are then @xcite @xmath225 because of the structure of the coefficient matrix , eq . , the sum runs only over all rotated orbitals , if @xmath226 are all themselves rotated orbitals . it is well known that the formally @xmath227-scaling transformation can be massively reduced by employing partial transformations @xcite , i.e. , first transforming and storing the fourth orbital to @xmath228 , then , transforming the third orbital to @xmath229 and so on . this scales as @xmath230 but twice the memory is needed for storing the intermediate transformed integrals . doing only two two - index transformation steps saves a considerable amount of memory but the scaling becomes worse . however , this is sometimes favorable @xcite . a similar procedure can be applied for the one - electron integrals , eq . . in our case however , the transformation can again be sped up massively , as exemplified by the calculation of the coulomb matrix , see eq . . because of the diagonality in @xmath37 and @xmath46 , it is beneficial to first transform the last two indices like in eq . , where @xmath231 has to be replaced with @xmath232 . note that , although the integrals are real - valued , the orbitals for @xmath233 are not : @xmath234 hence , only the following symmetries hold : @xmath235 because the two - index transformed integrals have even less symmetry , @xmath236 elements need to be stored for them . the transformation of the remaining two indices are done using partial transformations , as usual in quantum chemistry ( see above ) . for large non - rotated bases , the fully transformed integral tensor requires too much memory and the latter transformation is done on the fly : @xmath237 the index @xmath238 on the right - hand - side is constructed by @xmath239 and the indices for the functions in @xmath37 and @xmath38 of the index @xmath240 . the product @xmath241 can be precalculated which simplifies and accelerates the summation . transforming the last two indices on the fly requires @xmath242 operations , which is still better than the usual requirement of @xmath243 operations for a non - dvr basis such that the overall scaling for the complete two index transformation is then @xmath244 instead of the best achievable @xmath245 scaling . if all basis functions are nonrotated dvrfunctions , the formulas for the one- and two - electron integrals , eq . , can be applied directly . the crucial point is the efficient implementation of the integrals between both nonrotated and rotated basis functions . a lot of simplifications come from the structure of the coefficient matrix , eq . , which is diagonal if both basis functions are nonrotated functions and zero if one function is a rotated and another a nonrotated function . in the following , nonrotated basis functions are underlined . if @xmath248 , eq . reduces to @xmath249 because @xmath250 , but @xmath251 , all integrals @xmath252 are zero . this originates from the diagonality of the @xmath37-functions . for symmetry reasons , this applies as well to @xmath253 and so on , see eq . . if @xmath254 and @xmath255 are nonrotated functions , the integrals are also zero : @xmath256 this changes if @xmath255 and @xmath222 are nonrotated functions : @xmath257 this sum is computed very efficiently , see section [ sec : hf_integrals ] thus , we do not store these integrals but compute them on the fly . to summarize , only if two nonrotated basis functions are used for the same electron , the mixed integrals are nonzero but can be computed in a very efficient manner exploiting the diagonality inherent to the underlying dvrbasis in @xmath37 and @xmath38 . therefore , the number of nonrotated basis functions for the gas - ci - code do not influence the computational costs of the integrals considerably ( but the number of configurations in the ci expansion ) . the computation of the two - electron integrals in the `` raw '' dvrbasis is negligible . 98ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/revmodphys.72.545 [ * * , ( ) ] http://www.annualreviews.org/doi/abs/10.1146/annurev.physchem.59.032607.093532 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.81.163 [ * * , ( ) ] link:\doibase 10.1140/epjst / e2014 - 02092 - 3 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.75.543 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.153002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.023001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.023001 [ * * , ( ) ] link:\doibase 10.1103/physreva.71.013418 [ * * , ( ) ] link:\doibase 10.1126/science.1157980 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.043429 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.183001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.163001 [ * * , ( ) ] link:\doibase 10.1103/physreva.91.043406 [ * * , ( ) ] link:\doibase 10.1103/physreva.90.062508 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.133001 [ * * , ( ) ] link:\doibase 10.1016/j.elspec.2005.01.270 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.063402 [ * * , ( ) ] link:\doibase 10.1103/physreva.68.013409 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.043420 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.073001 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/37/23/l01 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.85.2280 [ * * , ( ) ] link:\doibase 10.1103/physreva.66.033402 [ * * , ( ) ] link:\doibase 10.1070/pu2004v047n09abeh001812 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.053423 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.053404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.68.245421 [ * * , ( ) ] link:\doibase 10.1063/1.1999636 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.223902 [ * * , ( ) ] link:\doibase 10.1016/0009 - 2614(90)87014-i [ * * , ( ) ] link:\doibase 10.1063/1.1862243 [ * * , ( ) ] link:\doibase 10.1103/physreva.71.012712 [ * * , ( ) ] link:\doibase 10.1016/j.cplett.2004.05.106 [ * * , ( ) ] \doibase doi:10.1063/1.3553176 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.062511 [ * * , ( ) ] link:\doibase 10.1063/1.4872005 [ * * , ( ) ] link:\doibase 10.1103/physreva.91.012509 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.023402 [ * * , ( ) ] http://www.sciencedirect.com/science/article/pii/s0065327610600068 [ * * , ( ) ] link:\doibase 10.1103/physreva.80.043412 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.013409 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/40/7/r01 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/46/3/035201 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.043405 [ * * , ( ) ] link:\doibase 10.1103/physreva.79.053411 [ * * , ( ) ] link:\doibase 10.1103/physreva.89.053407 [ * * , ( ) ] @noop _ _ , ed . ( , ) link:\doibase 10.1103/physreva.86.053424 [ * * , ( ) ] link:\doibase 10.1063/1.455063 [ * * , ( ) ] @noop _ _ , ed . ( , ) @noop _ _ ( , , ) link:\doibase 10.1103/physrevlett.70.1599 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.063403 [ * * , ( ) ] link:\doibase 10.1016/s0370 - 1573(99)00047 - 2 [ * * , ( ) ] \doibase doi:10.1063/1.451548 [ * * , ( ) ] link:\doibase 10.1007/s004600050342 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.461021 [ * * , ( ) ] link:\doibase 10.1103/physreva.74.052702 [ * * , ( ) ] link:\doibase 10.1103/physreva.82.023423 [ * * , ( ) ] link:\doibase 10.1103/physreva.72.052709 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.023405 [ * * , ( ) ] link:\doibase 10.1103/physreva.90.063421 [ * * , ( ) ] link:\doibase 10.1063/1.1726326 [ * * , ( ) ] link:\doibase 10.1103/physreva.79.012719 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.063416 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.043403 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.022510 [ * * , ( ) ] link:\doibase 10.1103/physreva.62.032706 [ * * , ( ) ] @noop _ _ , ed . ( , ) @noop _ _ , ed . ( , ) link:\doibase 10.1103/physreva.81.053407 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.093006 [ * * , ( ) ] link:\doibase 10.1103/physreva.78.043403 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.063407 [ * * , ( ) ] @noop _ _ , ed . ( , ) link:\doibase 10.1103/physreva.82.033427 [ * * , ( ) ] http://arxiv.org/abs/1508.01629 [ ( ) ] , link:\doibase 10.1002/anie.201410637 [ * * , ( ) ] link:\doibase 10.1103/physreva.65.053417 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/91/53001 [ * * , ( ) ] link:\doibase 10.1038/nature05648 [ * * , ( ) ] @noop _ _ , ed . ( , ) link:\doibase 10.1103/physrevlett.108.253003 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.053416 [ * * , ( ) ] link:\doibase 10.1016/0021 - 9991(76)90063 - 2 [ * * , ( ) ] http://www.digizeitschriften.de/dms/img/?ppn=gdzppn002146398 [ * * , ( ) ] link:\doibase 10.1063/1.1748101 [ * * , ( ) ] @noop _ _ , ed . ( , ) link:\doibase 10.1016/s0010 - 4655(97)00126 - 4 [ * * , ( ) ] link:\doibase 10.1016/0021 - 9991(72)90010 - 1 [ * * , ( ) ] link:\doibase 10.1016/0021 - 9991(74)90054 - 0 [ * * , ( ) ] link:\doibase 10.1002/qua.560310322 [ * * , ( ) ] \doibase 10.1002/(sici)1096 - 987x(199808)19:11<1241::aid - jcc4>3.0.co;2-k [ * * , ( ) ] link:\doibase 10.1002/wcms.82 [ * * , ( ) ]
the relation between the ratio of infrared ( ir ) and ultraviolet ( uv ) flux densities ( the infrared excess : irx ) and the slope of the uv spectrum ( ) of galaxies plays a fundamental role in the evaluation of the dust attenuation of star forming galaxies especially at high redshifts . many authors , however , pointed out that there is a significant dispersion and/or deviation from the originally proposed irx- relation depending on sample selection . we reexamined the irx- relation by measuring the far- and near - uv flux densities of the original sample galaxies with and imaging data , and constructed a revised formula . we found that the newly obtained irx values were lower than the original relation because of the significant underestimation of the uv flux densities of the galaxies , caused by the small aperture of , further , since the original relation was based on data which covered a wavelength range of m , using the data from which has wider wavelength coverage toward longer wavelengths , we obtained an appropriate irx- relation with total dust emission ( tir ) : + 0.22 $ ] . this new relation is consistent with most of the preceding results for samples selected at optical and uv , though there is a significant scatter around it . we also found that even the quiescent class of ir galaxies follows this new relation , though luminous and ultraluminous ir galaxies distribute completely differently as well known before .
1206.3905
the true star formation rate ( sfr ) of galaxies at various redshifts is one of the most fundamental physical quantities to understand the formation and evolution of galaxies in the universe . especially , the sfr density of galaxies in a cosmic volume is a very convenient tool to explore the galaxy evolution . the concept of the cosmic sfr density was originally introduced by @xcite , and subsequently it has been measured observationally at various redshifts up to @xmath3 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * among others ) . since the sfr of galaxies is essentially determined by measuring observables related to the ultraviolet ( uv ) flux from ob stars , like hydrogen recombination lines or forbidden lines through ionizing uv photons , or uv continuum itself . however , the largest uncertainty to determine the sfr from these observables is the dust extinction . since active star formation ( sf ) is always accompanied by dust production because of various dust grain formation processes related to the final stage of stellar evolution ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , it is always necessary to deal with the energy `` hidden '' by dust . the absorbed energy by dust is reradiated by dust grains at far - infrared ( fir ) wavelengths ( e.g. , * ? * ; * ? ? ? now it is shown that the fraction of the hidden star formation is even dominant at @xmath4 @xcite . the most straightforward way to remedy the dust extinction problem would be to sum up the sfr calculated from observed ( uncorrected ) uv and fir luminosities @xcite . however , this is not always possible to obtain both restframe uv and fir flux , especially at high redshifts to date . in general for local galaxies , when dust grains in galaxies absorb uv light , shorter wavelength flux is more strongly attenuated than longer one , which causes _ reddening _ of galaxy spectra . then , the strength of reddening and absorption fraction of uv are expected to be related in some way . evaluating this process , @xcite ( hereafter m99 ) proposed a formula to estimate dust attenuation from the slope of the ultraviolet ( uv ) spectrum , @xmath0 . this is based on the tight relation between the infrared ( ir ) to ultraviolet flux ratio ( traditionally referred to as the infrared excess : irx ) and @xmath0 . since this method does not require any other wavelength data to obtain the value of attenuation , like ir emission , it was very much appreciated by researchers of high - redshift galaxies ( @xmath5 or more ) . this is because the dust emission of such high-@xmath6 galaxies is usually much more difficult to observe , and their uv spectra are observed at optical wavelengths thanks to the redshifts ( e.g. , * ? ? ? * ; * ? ? ? thus , the irx-@xmath0 relation is very popularly used in these studies up to now . however , recently , subsequent studies have shown that the original formula of m99 does not always fit to local galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? especially , rather quiescently star - forming galaxies tend to distribute significantly below the irx-@xmath0 relation of m99 . in addition , another extreme categories of starburst galaxies , luminous and ultraluminous ir galaxies ( lirgs and ulirgs , respectively ) , occupy a region well above the m99 relation @xcite . also , it is reported that high-@xmath6 star - forming galaxies also tend to deviate from the m99 relation ( e.g. , * ? ? ? * ; * ? ? ? since the m99 s irx-@xmath0 relation was proposed for uv - luminous starburst galaxies , one may think it is not surprising that the m99 sample galaxies have different property from ir - selected galaxies . indeed , some attempts have been done to explain _ physically _ the discrepancy and dispersion of the irx-@xmath0 relation for various categories of galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? however , since this relation is often applied to high-@xmath6 galaxies whose nature is quite unknown , it is reasonable to examine the m99 relation according to their original recipe for its firm application in future studies . the original irx-@xmath0 relation of m99 was obtained based on and data . though the latter was an epoch - making facility for the uv astronomical studies , it had one observational limitation : the aperture of is @xmath7 ellipse with effective circle - equivalent diameter of @xmath8 . as m99 discussed in their original work , the sample of starburst galaxies in m99 sometimes have large angular diameters ( see section [ subsec : m99 ] ) . in contrast , the beam size of is quite large , and most of the part of galaxies is covered by the instrument . this difference of apertures of and would cause a problem in the estimation of the irx-@xmath0 relation since it is based on the ratio of the two . today , we have a much larger dataset open to public at uv bands provided by . ( galaxy evolution explorer ) is a uv astronomical satellite launched in 2003 by nasa @xcite . has far - uv ( fuv ) and near - uv ( nuv ) bands , and is performing all - sky survey as well as deep surveys in some sky areas . in this work , we reexamine the m99 relation by using the same galaxy sample as m99 , with paying a particular attention to the aperture problem . for this purpose , we use nuv and fuv images for the m99 s original sample galaxies . another important aspect of using filters for estimating @xmath0 of the m99 s sample is that since the main target of the dust correction by the irx-@xmath0 relation is high-@xmath6 galaxies , for which usually only broadband photometric data are available . since data are spectroscopic , it is crucial to recalibrate photometrically estimated uv slope with directly measured one from spectroscopic data ( see section [ subsubsec : filter ] ) . it is also interesting and important to examine the irx-@xmath0 relation from the fir side . for this , we also have a large all - sky ir data obtained by . is an infrared astronomical satellite launched in 2006 by jaxa ( japan aerospace exploration agency ) . is equipped with two imaging instruments , a near- and mid - ir camera irc @xcite and fir surveyor fis @xcite , and a fourier spectrograph @xcite . in this study , we make use of the data obtained by fis . the fis has four photometric bands , _ ( @xmath9 m ) , _ wide - s _ ( @xmath10 m ) , _ wide - l _ ( @xmath11 m ) , and _ n160 _ ( @xmath12 m ) . since has longer wavelength bands than that of , we can estimate the total fir emission from dust directly without any complicated conversion formula ( e.g. , * ? ? ? * ; * ? ? ? thus , with and data , we can obtain an unbiased irx-@xmath0 relation for the m99 sample . this paper is organized as follows : in section [ sec : sample ] , we describe the data of m99 and measurements newly done in this study . section [ sec : results ] , we first show the result of the - irx-@xmath0 relation . then we examine this new relation , and discuss the origin of the discrepancy between the m99 relation and recent observations from uv side . then , we examine the ir side with fis bright source catalog ( fisbsc ) . after considering all the measurement - related effects with diffuse map , we propose a new irx-@xmath0 relation for the uv - selected starbursts with and . in section [ sec : discussion ] , we further examine some properties related to the irx-@xmath0 relation we obtained in this work . first we estimate the bolometric correction for fir by and flux densities . then , we examined the behavior of the irx as a function of aperture radius but both on the and images . by this analysis , we can obtain the irx only from the central starburst regions of the original m99 sample . after checking these properties , we finally compare our new relation to various previous results . section [ sec : conclusion ] is devoted to our conclusions . we review the formulation of m99 in appendix [ sec : meurer ] . in appendix [ sec : akari_diffuse ] , we evaluate the aperture effect in fisbsc flux densities with fis diffuse map . first , we describe the original data of m99 . @xcite has constructed a uv sed atlas of galaxy sample of 143 spiral , irregular , blue compact dwarfs , seyfert 2 , and starburst galaxies with spectrum archive . m99 selected a subsample of very actively star - forming galaxies from this database which consisted of starburst nucleus , blue compact , and blue compact dwarfs . further , in order to reduce the effect of small aperture of , m99 have selected galaxies whose surface brightness distribution is strongly concentrated in the central small region . to construct this sample , they used a criterion for an optical diameter @xmath13 ( @xmath14 is measured at the @xmath15 isophote ) . we will revisit this issue in the following . this further selection resulted in 57 starbursts in the sample . m99 used 44 galaxies which were observed at @xmath16 m and @xmath17 m bands to make the irx@xmath18 plot and obtained the meurer relation . the sample and its property are summarized in table [ table : meurer ] . in order to examine the possible aperture effect of , we remeasured the uv flux densities from images . as already mentioned , has two uv bands , fuv ( @xmath19 ) and nuv ( @xmath20 ) . the images are available through mast . we have made a fuv and nuv photometry of the sample galaxies as follows : 1 . cut out a @xmath21 square subimage from ais images around each galaxy 2 . select a subimage with the largest exposure time when multiple observations were available . 3 . measure fuv and nuv flux densities . the nuv image is taken as the reference . in this study , it is important to mention that almost all of the sources are resolved by . then , they very often appear as an assembly of small bright patchy regions , and the pipeline misidentifies these fragments as individual objects . this is referred to as shredding . we must deal with the shredding to obtain sensible flux density measurements for nearby extended galaxies . for this purpose , we have used an idl software package developed by ourselves . this software performs aperture photometry in the nuv sub - image using a set of elliptical apertures . total flux density is calculated within the aperture corresponding to the convergence of the growth curve . the sky background is measured by combining several individual regions around the source . nuv and fuv flux densities are corrected for galactic extinction using the schlegel map @xcite and the galactic extinction curve of @xcite . a detailed description of the photometry process can be found in @xcite . this was already used for previous studies @xcite , and its performance is carefully checked and established . the new measurements are tabulated as @xmath22 and @xmath22 in table [ table : meurer ] . [ 0.71 ] lccccccccccc + name&@xmath0&@xmath23 & @xmath24 & @xmath25 & @xmath22 & @xmath26&@xmath27 & major axis & minor axis & @xmath6&type + & & & [ @xmath28&[@xmath28&[@xmath28&[@xmath28&[@xmath28&[arcsec]&[arcsec ] & & + & & & & & & & & & & & + ngc4861 & @xmath292.46 & @xmath291.89 & 8.76 & 9.13 & 8.97 & 8.68 & 8.50 & 180.00 & 46.15 & 0.0028 & sb + izw18 & @xmath292.43 & @xmath292.09 & ... & 8.05 & 7.86 & ... & ... & 31.50 & 14.70 & 0.0025 & bcg + ngc1705 & @xmath292.42 & @xmath292.09 & 8.88 & 8.95 & 8.75 & 8.12 & 8.50 & 76.50 & 73.44 & 0.0021 & sa0 + mrk153 & @xmath292.41 & @xmath291.78 & 9.20 & 9.38 & 9.24 & 8.65 & ... & 27.00 & 18.69 & 0.0080 & scp + tol1924@xmath29416 & @xmath292.12 & @xmath291.80 & 9.66 & 9.86 & 9.72 & 9.54 & 9.53 & 30.00 & 20.00 & 0.0095 & pec + ugc9560 & @xmath292.02 & @xmath291.78 & 8.57 & 8.77 & 8.63 & 8.54 & 8.77 & 42.00 & 27.00 & 0.0039 & pec + mrk66 & @xmath291.94 & @xmath291.41 & 9.60 & 9.86 & 9.79 & 9.85 & ... & 22.50 & 16.36 & 0.0210 & bcg + ngc3991 & @xmath291.91 & @xmath291.67 & 9.68 & 9.96 & 9.84 & 10.02 & 10.11 & 61.50 & 28.38 & 0.0106 & pec + ngc3738 & @xmath291.89 & @xmath291.49 & ... & 7.94 & 7.85 & ... & 7.86 & 76.50 & 62.16 & 0.0008 & irr + ugca410 & ... & ... & ... & ... & ... & ... & ... & ... & ... & 0.0022 & bcdg + mrk357 & @xmath291.80 & @xmath291.44 & 10.66 & 10.81 & 10.73 & 10.87 & 11.03 & 25.50 & 19.83 & 0.0529 & pair + ngc3353 & @xmath291.79 & @xmath291.52 & 8.50 & 8.73 & 8.63 & 9.15 & 9.23 & 61.50 & 45.32 & 0.0031 & irr + mrk54 & @xmath291.78 & @xmath291.41 & 10.54 & 10.71 & 10.64 & 10.82 & 10.77 & 43.50 & 23.20 & 0.0449 & sc + ngc1140 & @xmath291.78 & @xmath291.51 & 9.07 & 9.46 & 9.37 & 9.39 & 9.72 & 81.00 & 47.03 & 0.0050 & pec + mrk36 & @xmath291.72 & @xmath291.32 & 7.75 & 8.47 & 8.41 & 7.49 & 8.40 & 150.00 & 28.68 & 0.0022 & bcg + mcg6@xmath2928@xmath2944 & @xmath291.77 & @xmath291.40 & 10.51 & 10.71 & 10.64 & 10.82 & 10.77 & 43.50 & 23.20 & 0.0449 & sc + ngc1510 & @xmath291.71 & @xmath291.56 & 8.27 & 8.54 & 8.44 & 8.37 & 8.49 & 63.00 & 51.88 & 0.0030 & sa0 + mrk19 & ... & ... & ... & ... & ... & ... & 9.74 & ... & ... & 0.0138 & sa + ngc4214 & @xmath291.69 & @xmath291.56 & ... & 9.01 & 8.91 & ... & 9.05 & 337.50 & 225.00 & 0.0010 & iab + ngc4670 & @xmath291.65 & @xmath291.53 & 8.84 & 9.08 & 8.98 & 9.01 & 9.36 & 69.00 & 53.32 & 0.0036 & sb0 + ngc1800 & @xmath291.65 & @xmath291.63 & ... & 8.62 & 8.51 & ... & 8.31 & 67.50 & 35.10 & 0.0027 & ib + ugc5720 & @xmath291.62 & @xmath291.42 & 8.93 & 9.05 & 8.97 & 9.15 & 9.66 & 34.50 & 31.85 & 0.0048 & i m + ugc5408 & @xmath291.60 & @xmath291.06 & 8.98 & 9.01 & 9.00 & 9.54 & 9.78 & 34.50 & 26.83 & 0.0100 & e / s0 + ngc7673 & @xmath291.50 & @xmath291.27 & ... & 9.98 & 9.93 & ... & 10.41 & 51.00 & 45.00 & 0.0114 & sac + ngc3125 & @xmath291.49 & @xmath291.05 & 8.71 & 8.99 & 8.99 & 9.26 & 9.52 & 40.50 & 29.25 & 0.0037 & bcdg + haro15 & @xmath291.48 & @xmath291.63 & 9.96 & 10.33 & 10.22 & 10.27 & 10.44 & 52.50 & 43.24 & 0.0214 & sb0 + ngc2537 & @xmath291.44 & @xmath291.48 & 7.29 & 8.36 & 8.27 & 8.34 & 8.53 & 66.00 & 58.38 & 0.0014 & sb(s)m + ugc3838 & @xmath291.41 & @xmath291.46 & 8.97 & 9.41 & 9.33 & 9.23 & 8.30 & 42.00 & 24.71 & 0.0102 & sd + ngc7793 & @xmath291.34 & @xmath291.58 & ... & 8.94 & 8.84 & ... & 9.08 & 354.00 & 200.26 & 0.0008 & sa(s)d + ngc5253 & @xmath291.33 & @xmath291.18 & ... & 8.95 & 8.92 & ... & 9.26 & 123.00 & 69.62 & 0.0014 & i m + ngc7250 & @xmath291.33 & @xmath291.14 & 8.71 & 9.20 & 9.17 & 9.17 & 9.34 & 49.50 & 19.04 & 0.0039 & sdm + mrk542 & @xmath291.32 & @xmath290.99 & ... & 9.64 & 9.64 & ... & 10.27 & 27.00 & 23.14 & 0.0245 & i m + ngc7714 & @xmath291.23 & @xmath291.34 & 9.54 & 9.90 & 9.84 & 10.38 & 10.48 & 40.50 & 24.92 & 0.0093 & sb(s ) + mrk487 & ... & ... & ... & ... & ... & ... & ... & ... & ... & 0.0022 & bcdg + ngc3049 & @xmath291.14 & @xmath291.04 & 8.44 & 8.93 & 8.92 & 9.29 & 9.38 & 90.00 & 45.00 & 0.0049 & sb(rs ) + ugc6456 & @xmath291.10 & @xmath291.87 & 5.98 & 6.64 & 6.49 & 6.09 & ... & 37.50 & 26.47 & 0.0003 & pec + ngc3310 & ... & ... & ... & ... & ... & ... & 9.41 & ... & ... & 0.0033 & sab(r ) + ngc5996 & @xmath291.04 & @xmath291.12 & 9.23 & 9.90 & 9.88 & 10.21 & 10.38 & 76.50 & 42.08 & 0.0110 & sbc + ngc4385 & @xmath291.02 & @xmath291.06 & 8.85 & 9.20 & 9.19 & 9.82 & 10.01 & 73.50 & 36.75 & 0.0071 & sb(rs ) + mrk499 & @xmath291.02 & @xmath290.82 & ... & 9.88 & 9.92 & ... & 10.35 & 31.50 & 27.56 & 0.0260 & i m + ngc5860 & @xmath290.91 & @xmath290.78 & 9.32 & 9.51 & 9.55 & 10.24 & 10.38 & 43.50 & 35.59 & 0.0180 & e / s0 + ic1586 & @xmath290.91 & @xmath291.15 & 9.36 & 9.65 & 9.63 & 10.07 & 10.20 & 37.50 & 33.33 & 0.0194 & bcg + ngc2782 & @xmath290.90 & @xmath290.95 & 9.12 & 9.55 & 9.56 & 10.29 & 10.49 & 82.50 & 61.21 & 0.0085 & sab(rs ) + eso383@xmath2944 & ... & ... & ... & ... & ... & ... & 10.12 & ... & ... & 0.0126 & sa(s ) + ngc5236 & @xmath290.83 & @xmath290.80 & ... & 9.98 & 10.01 & ... & 10.68 & 352.50 & 320.11 & 0.0017 & sab(s ) + ngc2415 & @xmath290.80 & @xmath290.66 & 9.64 & 10.01 & 10.08 & 10.61 & 10.44 & 34.50 & 34.50 & 0.0126 & i m + mcg@xmath2901@xmath2930@xmath2933 & @xmath290.77 & @xmath290.93 & ... & ... & ... & ... & ... & 57.00 & 21.92 & ... 00 & sb(s)b + ngc1614 & ... & ... & ... & ... & ... & ... & 11.42 & ... & ... & 0.0159 & sb(s)c + ngc6217 & ... & ... & ... & ... & ... & ... & 10.03 & ... & ... & 0.0045 & sb(rs ) + ngc6052 & @xmath290.72 & @xmath290.76 & 9.55 & 10.11 & 10.15 & 10.69 & 10.89 & 55.50 & 45.71 & 0.0158 & sc + ngc4500 & ... & ... & ... & ... & ... & ... & 10.37 & ... & ... & 0.0104 & sb(s)a + ic214 & @xmath290.61 & @xmath290.92 & 9.83 & 10.18 & 10.20 & 11.18 & 11.30 & 43.50 & 31.90 & 0.0302 & irr + ngc3504 & @xmath290.56 & @xmath290.61 & 8.77 & 9.31 & 9.38 & 10.22 & 10.44 & 102.00 & 63.41 & 0.0051 & sab(s ) + ngc4194 & ... & ... & ... & ... & ... & ... & 10.67 & ... & ... & 0.0083 & bcg + ngc2798 & 0.03 & @xmath290.25 & 8.22 & 8.48 & 8.62 & 10.27 & 10.26 & 28.50 & 23.75 & 0.0058 & sb(s ) + ngc3256 & ... & ... & ... & ... & ... & ... & 11.29 & ... & ... & 0.0094 & pec + ngc7552 & 0.48 & @xmath290.37 & 8.66 & 9.35 & 9.47 & 10.78 & 10.78 & 111.00 & 62.83 & 0.0054 & sb(s ) + we show the result of new measurements for m99 s sample galaxies to examine the irx-@xmath0 relation . we quantitatively define the slope of uv spectrum @xmath0 as the exponent of an approximated spectrum ( in flux density per unit wavelength interval ) @xmath30 the relation between the flux at wavelength @xmath31 , @xmath32 , and the flux density per unit wavelength interval at @xmath31 , @xmath33 is @xmath34 thus the 1600 flux @xmath35 is defined as the average flux around the central wavelength 1600 , @xmath36 the fir flux @xmath37 is defined by using flux densities @xmath38 at @xmath39 m and @xmath17 m as @xmath40 \ ; , \end{aligned}\ ] ] where the unit of @xmath41 , @xmath42 , and @xmath37 is @xmath43 $ ] , and the unit of @xmath44 is @xmath45 $ ] @xcite . in this work , we denote the flux density per unit frequency interval at a wavelength @xmath31 as @xmath46 . we also use flux density in unit wavelength interval , @xmath33 . the unit of @xmath33 is @xmath47 $ ] throughout this paper . in m99 , the ir and the uv fluxes were estimated from the and data , respectively . hence , we first use the original values of m99 for ir flux , and focus on the new uv measurements with _ galex_. we should note that the fir flux evaluated by eq . ( [ eq : helou ] ) is only the integrated flux over a wavelength range @xmath48 m , and not the total ir flux often defined as integrated flux in a range @xmath49 . we evaluate the ir flux with later in section [ subsec : akari ] . with these observables and some empirical assumptions , m99 have formulated the relation between @xmath50 and @xmath51 as @xmath52 where bc stands for the bolometric correction . m99 adopted @xmath53 for 1600 and bc(fir)@xmath54 for fir , which leads @xmath55 . we revisit this issue in section [ subsec : dust_bc ] . thus , we obtain @xmath56 the derivation of eqs . ( [ eq : irx_extinction_general ] ) and ( [ eq : irx_extinction ] ) is shown in appendix [ sec : meurer ] . here , the value @xmath57 is for and , and if we use different data , it will also change . since we use fuv , we should recalculate @xmath57 for and . however , @xcite obtained @xmath58 , which is very close to that of , yielding practically the same value for @xmath57 . assuming a linear relation between @xmath51 and @xmath0 as @xmath59 namely the general relation between irx and @xmath0 is expressed as @xmath60 + 0.076 \;.\ ] ] performing a least square fit with observed values , m99 obtained a result @xmath61 combining eqs . ( [ eq : ext_beta_meurer ] ) with ( [ eq : irx_beta_meurer_general ] ) , we obtain @xmath62 + 0.076 \;.\ ] ] this is the irx-@xmath0 relation proposed by m99 . is a survey - oriented astronomical satellite , and its data are mainly images ( though is also equipped with a spectrograph ) . hence , we need to determine the uv spectral slope @xmath0 from fuv and nuv images , different from the case of which is equipped with a spectrograph . @xcite proposed a method to estimate irx and @xmath0 by using a flux density per unit wavelength @xmath63 and @xmath64 as follows . @xmath65 here the effective wavelength of fuv and nuv bands are @xmath66 and @xmath67 , respectively . the irx-@xmath0 relation newly obtained by these quantities is shown in fig . [ fig : irx_beta_galex_iras ] . we also show the original values of m99 for comparison . same galaxies are connected with dotted lines in fig . [ fig : irx_beta_galex_iras ] . the new values distribute clearly below the original ones of m99 for almost all galaxies . we found that since also the new values for @xmath0 are different , the distribution is shifted toward lower - right direction . fitting eq . ( [ eq : irx_beta_meurer_general ] ) to the new - irx-@xmath0 measurements , we obtain @xmath68 + 0.076 \;,\ ] ] which is different from eq . ( [ eq : irx_beta_meurer ] ) ( see the dashed line in fig . [ fig : irx_beta_galex_iras ] ) . the newly obtained relation is very close to that of normal star - forming galaxies proposed by previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * , among others ) . this implies that the starbursts in m99 sample are hosted in a normal galaxies with redder colors . we will revisit this issue in section [ sec : discussion ] . as we mentioned above , data are basically images , then the photometric data would be affected by the filter response function . we should stress that the uv spectral slopes @xmath0 were obtained directly from the spectrum , with avoiding the wavelength range corresponding to the 2170 bump ( see e.g. * ? ? ? * ) of the galactic extinction curve @xcite . in contrast , the nuv filter covers the bump . this would cause a systematic effect on the measurement of irx and @xmath0 . since most of the high-@xmath6 galaxy surveys use photometric data to evaluate @xmath0 for applying the irx-@xmath0 relation fort dust attenuation , it is crucial to calibrate the related quantities carefully . we first examine this effect . for this , we convolve the response function with the energy spectra of the m99 sample galaxies ( see fig . [ fig : ngc4861 ] as an example ) . let @xmath69 and @xmath70 be the fuv and nuv filter response functions , respectively , and denote the raw spectra as @xmath71 and the flux density at fuv ( nuv ) as @xmath72 ( @xmath73 ) @xmath74 $ ] , then we have @xmath75 with eqs . ( [ eq : flux_fil_galex_fuv ] ) and ( [ eq : flux_fil_galex_nuv ] ) , irx and @xmath0 after transmitting the filters , @xmath76 and @xmath77 , are expressed as @xmath78 we show the comparison of the uv spectral slope @xmath0 and the 1600 flux density before and after transmitting the filter in fig . [ fig : iue_galex_filter ] . for this analysis , we restricted to galaxies only with spectra in the full range of wavelength coverage ( @xmath79 ) . in the m99 sample , there are samples only with spectra at sw spectral coverage , and they are not ideal for the purpose of this analysis here . as we see in the left panel of fig . [ fig : iue_galex_filter ] , @xmath0 directly measured by and that defined by show a very good agreement . the linear fit gives @xmath80 with the pearson s correlation coefficient of @xmath81 and the standard deviation is @xmath82 . the systematic deviation is smaller than @xmath83 . as for the flux density , as we see on the right panel in fig . [ fig : iue_galex_filter ] , the 1600 flux density measured by , @xmath35 , tends to be about 10 % larger than that transmitted through the fuv filter , @xmath84 . the logarithmic linear fit yields @xmath85 with the pearson s correlation coefficient of @xmath86 and the standard deviation of flux density @xmath87 in logarithmic scale . again , the systematic deviation is smaller than @xmath88 . considering the intrinsic scatter in the irx-@xmath0 plot in m99 , we can safely neglect the systematic differences in @xmath0 and @xmath89 in the following analysis . in order to examine the effect of aperture photometry on images , we observe the behavior of flux with varying the aperture size . in this work , the aperture radius @xmath90 of the photometry is defined as an ellipse with semi - major axis size @xmath90 . this ellipse is determined automatically by fitting to the isophotal ellipse of a galaxy by our photometry software . then , the flux within a radius of @xmath91 means the integrated flux in an ellipse with semi - major axis size of @xmath91 . figure [ fig : irx_beta_all ] shows the change of measured values on the irx@xmath0 plot when we change the aperture radius as @xmath92 , @xmath91 , @xmath93 , @xmath94 , and the convergence radius ( depends on the sample ) . to see the change more clearly , we also show the change of the measured values with increasing aperture size with a step of @xmath91 in fig . [ fig : five_aperture ] . in fig . [ fig : frac_galex_1600 ] , we compare fuv and nuv flux densities and @xmath95 flux . here , the circle - equivalent `` effective '' aperture of @xmath8 is shown as vertical lines . this is an approximate of the aperture size which was an ellipse of @xmath96 . we set the ellipse shape and position angle to be the same as that used in actual observation of the sample galaxies . this effective aperture is depicted by vertical lines in both panels of fig . [ fig : frac_galex_1600 ] . as shown in this figure , fuv fluxes do not start to converge around a radius of @xmath91@xmath94 but continue to increase , and finally converge to values approximately 2.5 times larger than the 1600 fluxes . the nuv flux also show a similar behavior to that of fuv . we should also note that the behavior of fuv and nuv growth curves is not exactly the same . this causes an aperture dependence of @xmath0 . to see this effect , we examined the difference of @xmath0 between and . figure [ fig : beta_radius ] depicts the behavior of @xmath97 as a function of aperture radius . we observe that the behavior of @xmath0 is not monotonic : some increase and others decrease after exceeding @xmath8 aperture , and finally start to converge to certain values at much larger radius . within the effective aperture radius , @xmath98 changes significantly for most of the sample and is still far from the convergence values . we see a weak tendency that the converged values of @xmath0 are larger than that listed in m99 , i.e. , the global uv color of the sample turned out to be slightly redder when we measure the integrated flux . this can be explained as follows : since the m99 sample consists of central starbursts , it may be natural that their outer part of the disk has redder colors than the central part . hence , the colors of central part are bluer than the global ones . thus , the values of @xmath0 determined at a radius similar to that of are systematically bluer than the uv slopes when larger apertures are used . we , therefore , conclude that the systematic deviation of the irx-@xmath0 relation from the original one is due to the effect of uv flux because of the small aperture . especially , we see there is a galaxy that shows a very prominent behavior in fig . [ fig : frac_galex_1600 ] . both at fuv and nuv , this galaxy converges to exceptional values much brighter than the fluxes at a radius of @xmath91 . this galaxy is ngc 2537 , which has very strong contribution of uv flux from the peripheral regions . because of this , for ngc 2537 , the flux is more than an order of magnitude underestimated . in the previous sections , we have shown that the uv flux measurement of m99 was significantly underestimated because of the small aperture size of . however , since the photometry software we used for this work has been developed originally by ourselves ( cf . * ) , we should examine if there would be any systematic effect caused by the difference in photometry softwares when we compare our results with other ones . @xcite also used original software to perform a photometry of images . we checked the sample of @xcite and found 18 galaxies common with m99 , among which we have 11 galaxies with measurements needed for this study ( * ? ? ? * see their table 4 ) . figure [ fig : gil_de_paz ] shows that the results of this work and that of @xcite agree with each other very well . the correlation coefficient and the standard deviation for fuv data are @xmath99 and @xmath100 in logarithmic scale , and the same for nuv data are @xmath99 and @xmath101 , respectively . thus , we can safely conclude that both measurements are in excellent agreement . this means that the aperture effect on the original irx-@xmath0 relation was independently proved by @xcite . now we can safely conclude that the reason of the systematic shift of the irx-@xmath0 relation of recent studies and the original one is dominantly caused by the small aperture of . now we have the - based irx-@xmath0 relation which is free from the aperture effect [ eq . ( [ eq : irx_beta_galex_iras ] ) ] . up to now , we concentrated on the examination of uv flux measurement for the irx-@xmath0 relation , and for this purpose we kept the m99 s original values @xmath37 from . here we discuss the ir flux . the m99 relation was constructed to relate the uv slope and irx @xmath102 , and they used the fir luminosity @xcite @xmath103 \ ; [ \rm{l}_{\odot } ] \ ; .\ ] ] however , as mentioned above , @xmath104 contains only the luminosity in a wavelength range of @xmath105 m , and energy emitted at mir ( @xmath106 m ) nor submillimeter ( @xmath107 m ) is not considered . hence , the fir luminosity is proved to be a significant underestimation as a representative of the total ir luminosity ( @xmath108 ) ( e.g. , * ? ? ? @xcite and @xcite proposed formulae to convert fir into tir luminosity by making use of flux densities . the fis all - sky survey data are publicly available . with fluxes , @xcite proposed a new simple formula to estimate @xmath108 precisely as follows by making use of _ n60 _ , _ wide - s _ , and _ wide - l _ flux densities . however , since _ n60 _ flux density is more noisy than the other wide bands , @xcite found that the formula using two wide bands [ _ wide - s _ and _ wide - l _ : eq . ( [ eq : tir_akari ] ) ] gives a smaller scatter . hence , we use this two - band based formula in this work . ] @xmath109 \ ; , \label{eq : l_tir_wides}\\ & & \delta\nu(\mbox{\it wide - l\ , } ) = 0.831\times 10^{12}~[\rm{hz } ] \ ; , \label{eq : l_tir_widel}\\ & & \log l_{\it akari } ^ { 2 \rm{band } } = \delta\nu ( \mbox{\it wide - s\ , } ) l_{\nu}(90~\mu \mbox{m } ) \ ; , \nonumber \\ & & \qquad\qquad\qquad+\delta\nu ( \mbox{\it wide - l\,})l_{\nu}(140~\mu \mbox{m } ) \label{eq : l_tir_2band}\\ & & \log l_{\rm tir } = 0.964\log l_{\it akari}^{2\rm{band}}+0.814 \label{eq : l_tir } \;.\end{aligned}\ ] ] here @xmath110 is the bandwidth of each fis filter expressed in units of frequency @xcite . combining the -based uv flux free from the aperture effect and the tir flux [ eq . ( [ eq : l_tir ] ) ] covering the whole ir wavelength range , we construct a new irx-@xmath0 relation applicable for recent studies with , _ spitzer _ , and _ we cross - matched the m99 sample with the database of fis bsc . among the m99 sample , 48 are detected by fis . among these 48 galaxies , 40 have fuv and nuv ( see table [ table : meurer ] ) . with these 40 subsample galaxies of m99 , we estimated the new irx-@xmath0 relation . again the uv slope of the sed , @xmath0 , was obtained by eq . ( [ eq : beta_galex ] ) , and the irx was estimated from fuv flux and tir flux as follows . @xmath111 the result is presented in fig . [ fig : irx_akari_galex ] . since the fir is only a part of the tir , the - values distribute in a lower region than the - measurements on the irx-@xmath0 plots . however , we should carefully check one issue here . since we tried to be careful about the aperture effect at uv , we should also examine if the bsc fluxes are not affected by the aperture effect . importantly , @xcite examined a sample of local galaxies extracted from that of @xcite , since they are mostly nearby resolved galaxies by using diffuse map . they remeasured the ir fluxes with diffuse images and found that the flux of extended galaxies tends to be underestimated in the fisbsc , which adopted a complicated point source extraction procedure . then , according to @xcite , we measured the m99 sample with fis diffuse map and compared the obtained flux with that in the fisbsc ( appendix [ sec : akari_diffuse ] ) . since m99 tried to remove very extended galaxies in the primary construction of the sample , the current sample was found to be less affected by this effect , but still for some galaxies the underestimation was significant . we thus used the fluxes from the diffuse map and used them for the final determination of the irx-@xmath0 relation . this is shown in fig . [ fig : irx_akari_galex ] . for fitting eq . ( [ eq : irx_beta_meurer_general ] ) to the data , we should recall that this is a formula constructed for measurement , reflected in the term @xmath112 . since the bolometric correction for tir , @xmath113 is unity by definition , the last constant term in eq . ( [ eq : irx_beta_meurer_general ] ) [ i.e. , eq . ( [ eq : bc_ratio ] ) ] simplifies to the logarithm of the bolometric correction at uv , @xmath114 . fitting the eq . ( [ eq : irx_beta_meurer_general ] ) to the - data with this modification yields the following formula . @xmath115 + 0.22 \ ; .\ ] ] this is the new unbiased irx-@xmath0 relation for uv - selected starburst galaxies free from any aperture effect in uv or ir , applicable for present - day observational datasets with uv images ( like ) and total integrated ir flux ( like that obtained by ) . in the above discussion , we have rather focused on the aperture effect of the uv photometry on the irx-@xmath0 relation . since we have the diffuse image map , it would be also interesting to explore the radial dependence of the irx values . here we estimate the irx-@xmath0 relation as a function of aperture radius both on the and images . for this purpose , we have put the same elliptical aperture homologous to the original one , both on these images for each galaxy . since the angular resolution of the diffuse image map is @xmath116 at _ wide - s _ band , smaller aperture radii than @xmath94 are not meaningful . we started an aperture radius from @xmath94 which is similar to the aperture radius , and then estimated the irx with @xmath117 , @xmath118 , and @xmath119 radii . we should also take into account the different of the psfs at fir and fuv . it is pointed out that the synthesized psfs of the diffuse map are more extended than the previous estimates , with a significant flux leak to the sidelobes ( arimatsu et al . , in preparation ) . since it is still difficult to evaluate this effect precisely at this moment , we instead convolved the fuv images with the difference of the angular resolutions of and , we have convolved the fuv images with the synthesized psf of fis 140 @xmath120 m band and measured the irx within the same aperture at fir and fuv bands . the result is shown in figs . [ fig : galex_akari_2band ] and [ fig : five_aperture_2band ] . it is very clearly seen that the irx values of central regions of the sample are much lower than the relation obtained from the whole galaxies . this is mainly because of the difference of the surface brightness profiles at fuv and fir wavelengths . in general , the fuv light of galaxies is much more concentrated than that of fir . then , the irx values are well below the convergence values within @xmath93 aperture , and then moves up the diagram with increasing aperture radius , as clearly seen in fig . [ fig : five_aperture_2band ] . even with the @xmath118 aperture radius , the irx values are still far from the convergence values for some galaxies , even if we take into account the difference of the psfs . this implies that the starburst regions of the m99 s original sample galaxies emit less energy at fir compared to the whole galaxy scale . to see this more clearly , we show the fraction of tir flux as a function of aperture radius in fig . [ fig : tir_frac ] . though the values are largely dispersed , on average the fraction contained in @xmath94 aperture radius is @xmath121 % . namely , though we should keep in mind that the correction factor of the flux leak could be large at small aperture radii , at least we can claim that less than @xmath122 % of the tir light is associated to the central starburst regions in the sample , but comes from outer part of galaxies . it will be interesting to simulate the irx as a function of redshift , taking into account a realistic beam size and detection limit for current / future observational facilities to interpret the high-@xmath6 irx-@xmath0 relation . since the original irx-@xmath0 relation was based on observation , we should convert the formula to that of tir and fuv . the assumption of m99 was , as stated above , the ratio of tir / fir [ the bolometric correction of dust emission , @xmath123 : see eq . ( [ eq : irx_extinction_general ] ) ] is @xmath124 . many authors used this value to rescale the relation with this factor to obtain the irx defined by @xmath125 ( e.g. * ? ? ? * ) . then , it is important to examine the value of @xmath126 by new ir data . by making use of the fis diffuse map , we examine the ratio of tir and -based fir . we calculated a ratio of tir / fir ( @xmath127 ) , shown in fig . [ fig : tir_fir_radial ] . in fig . [ fig : tir_fir_radial ] , the tir / firs among the sample galaxies and its convergence values are presented . in fig . [ fig : tir_fir_radial ] , dotted lines represent the tir / fir ratio , and the symbols represent convergence values of tir / fir for each sample . the thick solid broken line shows the average value . the horizontal thick line shows the convergence value of tir / fir at large aperture limit . from this analysis , we found that the tir / fir ratio converges to @xmath128 , which is very close to the ratio originally assumed by m99 . by using a formula of @xcite , @xcite discussed that tir / fir is @xmath129 , which is also quite close to our value . we now compare our new formula with previous studies on the irx-@xmath0 relation with various sample selection . we summarize the comparison in fig . [ fig : irx_beta_comp ] . @xcite compiled a sample of optically selected galaxies observed by and analyzed the irx-@xmath0 relation , as well as the -selected uv starbursts . they have converted the fir flux to the tir one by the formula of @xcite . since the formula of @xcite is defined for a wavelength range of @xmath130 m , it gives a slightly larger value than our formula , but the difference is small enough compared with the intrinsic scatter and measurement errors @xcite . to describe the average trend , they fit a simple linear function instead of the form of eq . ( [ eq : irx_beta_meurer_general ] ) . in their analysis , @xcite found that the optically selected sample distribute systematically lower than starbursts on the irx-@xmath0 plane . they suggested qualitatively that this is partially due to the aperture , as well as some physical reason related to the star formation activity . our formula passes through the peak of the distribution of their optical sample , and agrees very well with their linear fit at @xmath131 . @xcite selected 43 nearby late - type spiral galaxies optically and performed a detailed analysis on the dust attenuation . the sample was cross - matched with atlas of nearby galaxies @xcite for uv and hires imagesover.html . ] for ir . the same as @xcite , @xcite also used the formula of @xcite to compute the tir flux from flux . for the fitting of the irx-@xmath0 relation , they used a form @xmath132 \ ; , \end{aligned}\ ] ] which has one more free parameter than eq . ( [ eq : irx_beta_akari ] ) . their sample also distributed below the m99 relation ( their fig . 1 ) . their irx-@xmath0 relation has a higher irx at smaller @xmath0 and lower irx at larger @xmath0 , but since there are very few or no data at @xmath133 , the discrepancy on this side is not crucial . at the larger @xmath0 side , actually there is a large dispersion in optically selected spiral galaxy sample in @xcite . this may be the reason of their flat slope . indeed , @xcite used the sings galaxy sample @xcite for a similar analysis and obtained an irx-@xmath0 relation with steeper slope at large @xmath0 side . the new relation of @xcite is in a very good agreement with our result at @xmath134 . thus , we can safely say that our new irx-@xmath0 formula determined by the original uv starburst sample of m99 , also describes the attenuation of optical sample very well . however , we should keep in mind that there is a considerable scatter around the relation obtained by fitting , as these authors also cautioned . next , we also examine if eq . ( [ eq : irx_beta_akari ] ) would apply to the ir - selected samples . as we have already seen , the behavior of ir - selected galaxies on the irx-@xmath0 plane is quite complicated . these galaxies have two branches of distribution , one locates below the m99 relation and the other spreads well above the line ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite first discovered that the population distributing above the m99 line is lirgs or ulirgs . @xcite examined this aspect intensively and found that the population distributing lower than the m99 line is quiescent star - forming galaxies , and confirmed that the one above the m99 line is lirgs or ulirgs , by evaluating the 60-@xmath120 m luminosity of fisbsc sample . @xcite presented an impressive irx-@xmath0 plots for both nuv- and 60 @xmath120m - selected samples . the quiescent branch of ir - selected galaxies distribute in almost the same way as the uv - selected sample on this plane , i.e. , below the m99 s irx-@xmath0 relation but in parallel with it . more active branch galaxies are spread over the m99 relation . we see that the former is well described by eq . ( [ eq : irx_beta_akari ] ) without significant bias . however , since the distribution of ( u)lirgs are very different from that of quiescent galaxies , our new relation can not apply to these active galaxies . as for high-@xmath6 galaxies , still it is not very easy to obtain both uv and ir data , but gradually direct measurements are reported recently . @xcite showed the irx-@xmath0 measurements for two such objects , ms1512-cb58 ( @xmath135 ) and the cosmic eye ( @xmath136 ) , both of which are lyman - break galaxies ( lbgs ) , i.e. , uv - luminous starbursts . they found that they locate significantly below the original m99 relation . @xcite also discussed the irx-@xmath0 relation for both uv- and ir - luminous galaxies at @xmath137 . they found that their `` typical '' @xmath137 galaxies do follow the m99 relation ( but with a large scatter ) , and ulirgs at the same redshift locate well above it similarly to their local counterparts . also they argue that young galaxies at these redshifts distribute below the m99 line . the young galaxies seem to follow the irx-@xmath0 relation expected for the smc extinction curve @xcite , also suggested by @xcite . comparing their results with our new irx-@xmath0 relation ( see fig . [ fig : irx_beta_comp ] ) , we would say the @xmath137 typical galaxies of @xcite locate above the new irx-@xmath0 relation , while the young galaxies follow it . as for lensed @xmath138 galaxies in @xcite , cb58 is approximately on the new relation , while the cosmic eye may be still below the relation . it would be very interesting to make a statistical analysis when a large sample of galaxies at these redshifts will be available , to discuss whether evolutionary effect really would take place . we should note that the irx-@xmath0 relation of the central starburst regions of our sample locates well below the m99 s original relation and our new result , but more similar to the pettini s relation ( see top - left panel of fig . [ fig : five_aperture_2band ] ) . since m99 originally selected the sample to have a central starbursts , we might speculate , for example , that the properties of high-@xmath6 star - forming galaxies could be similar to that of a pure starburst region . further investigation in this direction will be interesting . @xcite suggested a systematic sequence of the irx-@xmath0 relation from lower to higher along with the birthrate parameter @xmath139 @xmath140 where @xmath141 is a supposed age of a galaxy @xcite . they claimed that @xmath139 is related to the attenuation as @xmath142 in @xcite , the original m99 relation corresponds to the extremely bursting group , i.e. , @xmath143 , not very consistent with the properties of m99 s uv - luminous starburst sample . we see that our new irx-@xmath0 relation corresponds to the intermediate @xmath139 galaxies similar to normally star forming galaxies , now consistent with the original m99 sample . @xcite independently made essentially the same analysis with this study but with their own photometry and -based ir flux estimation with a conversion formula of @xcite , with _ measurement for a subsample of 12 galaxies their formula is , as we expect , not very different from our result [ eq . ( [ eq : irx_beta_akari ] ) ] , but slightly deviated upward than ours at large @xmath0 and high irx regime . they have measured the uv flux up to @xmath144 times kron radius of galaxies , their converged values are on average twice larger than the original values of m99 . as we have shown , the flux is underestimated with a factor of 2.5 on average , and sometimes the underestimation is of an order of magnitude . hence , though they discuss that the quiescent ir galaxies in @xcite locate lower than their line , for example , this might be due to the uv flux still missed . the relation between the ratio of infrared ( ir ) and ultraviolet ( uv ) flux densities ( the infrared excess : irx ) and the slope of the uv spectrum ( @xmath0 ) of galaxies plays a fundamental role in the evaluation of the dust attenuation of star forming galaxies especially at high redshifts . m99 first introduced a very useful method to estimate the dust attenuation by using the relation between the fir - to - fuv flux ratio and the uv slope , the irx-@xmath0 relation . however , subsequent studies revealed a dispersion and/or systematic deviation from the original irx-@xmath0 relation of m99 , depending on sample selection . we reexamined the original irx-@xmath0 relation proposed by m99 by measuring the far- and near - uv flux densities of the original sample galaxies used by m99 with imaging data . we summarize our conclusion as follows : 1 . the uv flux densities of the galaxies used in m99 were significantly underestimated because of the small aperture of which was @xmath7 . consequently , the irx values were lower than that in m99 , which lead the irx-@xmath0 relation shifted downward on the plot . 2 . the aperture effect affected not only the irx but also @xmath0 , because the curve of growth of the flux densities is a function of aperture which depends strongly on the wavelength . correcting the aperture effect , we obtained a new irx-@xmath0 relation for uv - selected starburst galaxies for the same sample with m99 . the original relation was obtained based on data . since bands cover only a wavelength range of @xmath1 m , the fir flux calculated from flux does not represent the total ir emission from dust . we used data from which has a much wider wavelength coverage especially toward longer wavelengths , and obtained an appropriate irx-@xmath0 relation for the data with total dust emission . we also estimated the irx-@xmath0 relation of the sample as a function both on the and images . we found that the irx-@xmath0 relation of the central starburst regions of the sample locates significantly below the relation obtained for the whole galaxy regions . this means that the extended dust emission is important to evaluate various global properties of galaxies , though we need to know the behavior of the psfs of akari diffuse map more precisely to evaluate more quantitatively . 6 . in many previous studies , authors use a conventional conversion factor ( bolometric correction : bc ) of @xmath124 to obtain the total ir flux from dust . we examined the bc of dust emission and found that indeed the value was @xmath128 on average . our new relation is consistent with most of the preceding results on the irx-@xmath0 relation for samples selected at optical and uv , though there is a significant scatter around it . we also found that even the quiescent class of ir galaxies follow this new relation , though luminous and ultraluminous ir galaxies distribute completely differently . the irx-@xmath0 relation for the central starburst regions of m99 s sample galaxies is very similar to the relation proposed by @xcite , based on the smc attenuation curve . this may imply interesting property of high-@xmath6 star - forming galaxies . in this work , we constructed a proper formula for a category of galaxies which are luminous in uv and actively forming stars . we found that this formula also applies to various class of moderately star - forming galaxies , like optically selected or non - luminous ir selected ones . this will be a more firm basis than before when one needs to correct dust extinction of high-@xmath6 galaxies . however , we should note the large intrinsic scatter on the irx-@xmath0 plane , especially the ( u)lirg population which does not obey the irx-@xmath0 relation at all . many issues and problems still remain to be solved for understanding this relation . first of all , we thank the anonymous referee for her / his extremely careful and constructive comments and suggestions that improved this article significantly . this work is based on observations with , a jaxa project with the participation of esa . we thank yasuo doi for preparing the fis diffuse map for our sample galaxies , and ko arimatsu for providing the new synthesized psf of the map . we also thank vronique buat , denis burgarella , agnieszka pollo , ryosuke asano , fumiko nagaya , mai fujiwara , daisuke yamasawa , and takashi kozasa for fruitful discussions and comments . we are grateful to takako t. ishii for developing the basis of the photometric software with idl . ttt has been supported by program for improvement of research environment for young researchers from special coordination funds for promoting science and technology . ttt and aki have been also supported by the grant - in - aid for the scientific research fund ( 20740105 , 23340046 , 24111707 : ttt , 19740108 : aki ) commissioned by the ministry of education , culture , sports , science and technology ( mext ) of japan . ttt , fty , ai , and klm have been partially supported from the grand - in - aid for the global coe program `` quest for fundamental principles in the universe : from particles to the solar system and the cosmos '' from the mext . facilities : , , , here , we review the original formulation of the irx-@xmath0 relation by m99 . @xmath145 where @xmath146 is the unattenuated flux density of the emitted spectrum , @xmath147 is the net absorption in magnitudes by dust as a function of wavelength , and @xmath148 is the ly@xmath149 flux . in eq . ( [ eq : irx_def ] ) , it is assumed that the ionizing photons does not affect dust heating significantly , namely they are not absorbed directly by dust grains .. % of the bolometric luminosity of starbursts is emitted below 912 @xcite , and the fraction of ionizing photons that is directly absorbed by dust grains is estimated to be @xmath150 % @xcite . however , the direct absorption of lyman photons by dust grains in ionizing regions has been first discussed by @xcite and extensively by @xcite , who found that the fraction of ionizing photons directly absorbed by dust is significantly larger than these values . however , even if all the ionizing photons were absorbed by dust , they would still not drastically affect the conclusions . hence , the formulation of m99 will remain valid . ] the ly@xmath149 flux @xmath148 was produced by cascade of photons with wavelengths shorter than 912 , which finally reach the transition @xmath151 . if the ionized gas in case b , ly@xmath149 is resonantly scattered repeatedly by hydrogen atoms and finally absorbed by dust . thus , the numerator of eq . ( [ eq : irx_def ] ) corresponds to the total flux absorbed by dust , consisting of two terms , the first is the heating by ly@xmath149 and the second is by nonionizing photons . the denominator in eq . ( [ eq : irx_def ] ) represents the observed 1600 flux density which is attenuated by dust . namely , @xmath152 most of the photons absorbed by dust are contribution from massive stars . then , by assuming that most of the radiation from massive stars is uv , we can substitute @xmath153 with @xmath51 in eq . ( [ eq : irx_def ] ) as @xmath154 further , we rearrange eq . ( [ eq : irx_ap1600 ] ) as follows . @xmath155 the subscript asterisk means it is related to stellar emission , and subscript dust means it is for dust emission . here , @xmath156 we assume @xmath157 on one hand , @xmath158 can be approximated as a constant if we fix the initial mass function ( imf ) and a constant star formation rate . if we adopt the salpeter imf @xcite with an upper mass limit of @xmath159 , we have @xmath53 ( m99 ) . the uncertainty is calculated based on the variation of @xmath160 depending on the burst duration . on the other hand , @xmath161 is the ratio between the total ir flux and the fir flux . empirically from the observation @xcite , m99 adopted a value of bc(fir)@xmath54 , which leads @xmath162 . putting them together with eq . ( [ eq : irx_bc ] ) , we obtain @xmath163 which is eq . ( [ eq : irx_extinction ] ) in the main text , the basic relation between @xmath50 and @xmath51 . the fisbsc was produced by a dedicated pipeline for point source extraction @xcite . however , for very extended nearby galaxies , the point source assumption is not valid anymore . to examine this `` aperture '' effect in fis flux density , we measured ir flux density from fis diffuse map . we performed this procedure by using both sextractor @xcite and our own software , and both yielded more or less the same result . since @xcite used sextractor , we also present flux density given by sextractor for the diffuse map . we show the comparison between flux densities from fisbsc and diffuse map for m99 sample in fig . [ fig : akari_diffuse ] . basically these flux densities agree well , because m99 put a constraint to have galaxies with a small angular diameter in their sample selection . however , even so , the bsc flux density turned out to be underestimated for non - negligible number of galaxies . the horizontal dashed lines represent the average difference between bsc and diffuse - map flux densities . for both _ wide - s _ and _ wide - l _ , the average is positive . @xcite performed a careful comparison of bsc flux density with other well - understood database and concluded that this is due to the underestimation of the bsc and not the overestimation in diffuse map photometry . thus , we use the diffuse - map based ir flux densities to have the final irx-@xmath0 relation fitting .
we analyze the possibility of probing non - standard neutrino interactions ( nsi , for short ) through the detection of neutrinos produced in a future galactic supernova ( sn ) . we consider the effect of nsi on the neutrino propagation through the sn envelope within a three - neutrino framework , paying special attention to the inclusion of nsi - induced resonant conversions , which may take place in the most deleptonised inner layers . we study the possibility of detecting nsi effects in a megaton water cherenkov detector , either through modulation effects in the spectrum due to ( i ) the passage of shock waves through the sn envelope , ( ii ) the time dependence of the electron fraction and ( iii ) the earth matter effects ; or , finally , through the possible detectability of the neutronization burst . we find that the spectrum can exhibit dramatic features due to the internal nsi - induced resonant conversion . this occurs for non - universal nsi strengths of a few % , and for very small flavor - changing nsi above a few .
0704.0032
the very first data of the kamland collaboration @xcite have been enough to isolate neutrino oscillations as the correct mechanism explaining the solar neutrino problem @xcite , indicating also that large mixing angle ( lma ) was the right solution . the 766.3 ton - yr kamland data sample further strengthens the validity of the lma oscillation interpretation of the data @xcite . current data imply that neutrino have mass . for an updated review of the current status of neutrino oscillations see @xcite . theories of neutrino mass @xcite typically require that neutrinos have non - standard properties such as neutrino electromagnetic transition moments @xcite or non - standard four - fermi interactions ( nsi , for short ) @xcite . the expected magnitude of the nsi effects is rather model - dependent . seesaw - type models lead to a non - trivial structure of the lepton mixing matrix characterizing the charged and neutral current weak interactions @xcite . the nsi which are induced by the charged and neutral current gauge interactions may be sizeable @xcite . alternatively , non - standard neutrino interactions may arise in models where neutrinos masses are radiatively `` calculable '' @xcite . finally , in some supersymmetric unified models , the strength of non - standard neutrino interactions may arise from renormalization and/or threshold effects @xcite . we stress that non - standard interactions strengths are highly model - dependent . in some models nsi strengths are too small to be relevant for neutrino propagation , because they are either suppressed by some large mass scale or restricted by limits on neutrino masses , or both . however , this need not be the case , and there are many theoretically attractive scenarios where moderately large nsi strengths are possible and consistent with the smallness of neutrino masses . in fact one can show that nsi may exist even in the limit of massless neutrinos @xcite . such may also occur in the context of fully unified models like @xmath3 @xcite . we argue that , in addition to the precision determination of the oscillation parameters , it is necessary to test for sub - leading non - oscillation effects that could arise from non - standard neutrino interactions . these are natural outcome of many neutrino mass models and can be of two types : flavor - changing ( fc ) and non - universal ( nu ) . these are constrained by existing experiments ( see below ) and , with neutrino experiments now entering a precision phase @xcite , an improved determination of neutrino parameters and their theoretical impact constitute an important goal in astroparticle and high energy physics @xcite . here we concentrate on the impact of non - standard neutrino interactions on supernova physics . we show how complementary information on the nsi parameters could be inferred from the detection of core - collapse supernova neutrinos . the motivation for the study is twofold . first , if a future sn event takes place in our galaxy the number of neutrino events expected in the current or planned neutrino detectors would be enormous , @xmath4 @xcite . moreover , the extreme conditions under which neutrinos have to travel since they are created in the sn core , in strongly deleptonised regions at nuclear densities , until they reach the earth , lead to strong matter effects . in particular the effect of small values of the nsi parameters can be dramatically enhanced , possibly leading to observable consequences . this paper is planned as follows . in sec . [ sec : preliminaries ] we summarize the current observational bounds on the parameters describing the nsi , including previous works on nsi in sne . in sec . [ sec : neutrino - evolution ] we describe the neutrino propagation formalism as well as the sn profiles which will be used . in sec . [ sec : two - regimes ] we analyze the effect of nsi on the @xmath5 propagation in the inner regions near the neutrinosphere and in the outer regions of the sn envelope . in sec . [ sec : observables ] we discuss the possibility of using various observables to probe the presence of nsi in the neutrino signal of a future galactic sn . finally in sec . [ sec : summary ] we present our conclusions . a large class of non - standard interactions may be parametrized with the effective low - energy four - fermion operator : @xmath6 where @xmath7 and @xmath8 is a first generation fermion : @xmath9 . the coefficients @xmath10 denote the strength of the nsi between the neutrinos of flavors @xmath11 and @xmath12 and the @xmath13handed component of the fermion @xmath8 . current constraints on @xmath10 come from a variety of different sources , which we now briefly list . neutrino scattering experiments @xcite provide the following bounds , @xmath14 cross section measured at lep ii leads to a bound on @xmath15 @xcite . future prospects to improve the current limits imply the measurement of @xmath16 leptonically in the scattering off electrons in the target , as well as in neutrino deep inelastic scattering in a future neutrino factory . the main improvement would be in the case of @xmath17 and @xmath18 , where values as small as @xmath19 and @xmath20 , respectively , could be reached @xcite . the search for flavor violating processes involving charged leptons is expected to restrict corresponding neutrino interactions , to the extent that the @xmath21 gauge symmetry is assumed . however , this can at most give indicative order - of - magnitude restrictions , since we know @xmath21 is not a good symmetry of nature . using radiative corrections it has been argued that , for example , @xmath22 conversion on nuclei like in the case of @xmath23 also constrains @xmath24 @xcite . non - standard interactions can also affect neutrino propagation through matter , probed in current neutrino oscillation experiments . the bounds so obtained apply to the vector coupling constant of the nsi , @xmath25 , since only this appears in neutrino propagation in matter . the role of neutrino nsis as subleading effects on the solar neutrino oscillations and kamland has been recently considered in ref . @xcite with the following bounds at 90 % cl for @xmath26 with the allowed range @xmath27 , while for the diagonal term @xmath28 , the only forbidden region is @xmath29 $ ] @xcite . only in the ideal case of infinitely precise solar neutrino oscillation parameters determination , the allowed range would `` close from the left '' for negative nsi parameter values , at @xmath30 for @xmath31 and @xmath32 for @xmath33 . non - standard interactions involving muon neutrinos can be constrained by atmospheric neutrino experiments as well as accelerator neutrino oscillation searches at k2k and minos . in ref . @xcite super - kamiokande and macro observations of atmospheric neutrinos were considered in the framework of two neutrinos . the limits obtained were @xmath34 and @xmath35 at 99 % cl . the same data set together with k2k were recently considered in refs . @xcite to study the nonstandard neutrino interactions in a three generation scheme under the assumption @xmath36 . the allowed region of @xmath37 obtained for values of @xmath38 smaller than @xmath39 becomes @xmath40 @xcite , where @xmath41 stands for the fermion number density . if non - standard interactions with electrons were large they might also lead to important cosmological and astrophysical implications . for instance , neutrinos could be kept in thermal contact with electrons and positrons longer than in the standard case , hence they would share a larger fraction of the entropy release from @xmath42 annihilations . this would affect the predicted features of the cosmic background of neutrinos . as recently pointed out in ref @xcite required couplings are , though , larger than the current laboratory bounds . according to the currently accepted supernova ( sn ) paradigm , neutrinos are expected to play a crucial role in sn dynamics . as a result , sn physics provides a laboratory to probe neutrino properties . moreover , many future large neutrino detectors are currently being discussed @xcite . the enormous number of events , @xmath4 that would be `` seen '' in these detectors indicates that a future sn in our galaxy would provide a very sensitive probe of non - standard neutrino interaction effects . the presence of nsi can lead to important consequences for the sn neutrino physics both in the highly dense core as well as in the envelope where neutrinos basically freely stream . the role of non - forward neutrino scattering processes on heavy nuclei and free nucleons giving rise to flavor change within the sn core has been recently analyzed in ref . the main effect found was a reduction in the core electron fraction @xmath43 during core collapse . a lower @xmath43 would lead to a lower homologous core mass , a lower shock energy , and a greater nuclear photon - disintegration burden for the shock wave . by allowing a maximum @xmath44 it has been claimed that @xmath45 , where @xmath46 @xcite . on the other hand it has been noted since long ago that the existence of nsi plays an important role in the propagation of sn neutrinos through the envelope leading to the possibility of a new resonant conversion . in contrast to the well known msw effect @xcite it would take place even for massless neutrinos @xcite . two basic ingredients are necessary : universal and flavor changing nsi . in the original scheme neutrinos were mixed in the leptonic charged current and universality was violated thanks to the effect of mixing with heavy gauge singlet leptons @xcite . such resonance would induce strong neutrino flavor conversion both for neutrinos and antineutrinos simultaneously , possibly affecting the neutrino signal of the sn1987a as well as the possibility of having @xmath47process nucleosynthesis . this was first quantitatively considered within a two - flavor @xmath48 scheme , and bounds on the relevant nsi parameters were obtained using both arguments @xcite . one of the main features of the such `` internal '' or `` massless '' resonant conversion mechanism is that it requires the violation of universality , its position being determined only by the matter chemical composition , namely the value of the electron fraction @xmath43 , and not by the density . in view of the experimental upper bounds on the nsi parameters such new resonance can only take place in the inner layers of the supernova , near the neutrinosphere , where @xmath43 takes its minimum values . in this region the values of @xmath43 are small enough to allow for resonance conversions to take place in agreement with existing bounds on the strengths of non - universal nsi parameters . the sn physics implications of another type of nsi present in supersymmetric r - parity violating models have also been studied in ref . @xcite , again for a system of two neutrinos . for definiteness nsi on @xmath49quarks were considered , in two cases : ( i ) massless neutrinos without mixing in the presence of flavor - changing ( fc ) and non - universal ( nu ) nsis , and ( ii ) neutrinos with ev masses and fc nsi . different arguments have been used in order to constrain the parameters describing the nsi , namely , the sn1987a signal , the possibility to get successful @xmath47process nucleosynthesis , and the possible enhancement of the energy deposition behind the shock wave to reactivate it . on the other hand several subsequent articles @xcite considered the effects of nsi on the neutrino propagation in a three neutrino mixing scenario for the case @xmath50 , typical for the outer sn envelope . together with the assumption that @xmath51 this prevents the appearance of internal resonances in contrast to previous references . motivated by supersymmetric theories without r parity , in ref . @xcite the authors considered the effects of small - strength nsi with @xmath49quarks . following the formalism developed in refs . @xcite they studied the corrections that such nsi would have on the expressions for the survival probabilities in the standard resonances msw - h and msw - l . a similar analysis was performed in ref . @xcite assuming z - induced nsi interactions originated by additional heavy neutrinos . a phenomenological generalization of these results was carried out in ref . the authors found an analytical compact expression for the survival probabilities in which the main effects of the nsi can be embedded through shifts of the mixing angles @xmath52 and @xmath53 . in contrast to similar expressions found previously these directly apply to all mixing angles , and in the case with earth matter effects . the main phenomenological consequence was the identification of a degeneracy between @xmath53 and @xmath54 , similar to the analogous `` confusion '' between @xmath53 and the corresponding nsi parameter noted to exist in the context of long - baseline neutrino oscillations @xcite . we have now re - considered the general three neutrino mixing scenario with nsi . in contrast to previous work @xcite , we have not restricted ourselves to large values of @xmath43 , discussing also small values present in the inner layers . this way our generalized description includes both the possibility of neutrinos having the `` massless '' nsi - induced resonant conversions in the inner layers of the sn envelope @xcite , as well as the `` outer '' oscillation - induced conversions @xcite small enough not to lead to drastic consequences during the core collapse . ] . in this section we describe the main ingredients of our analysis . our emphasis will be on the use of astrophysically realistic sn matter and @xmath43 profiles , characterizing its density and the matter composition . their details , in particular their time dependence , are crucial in determining the way the non - standard neutrino interactions affect the propagation of neutrinos in the sn medium . as discussed in sec . [ sec : preliminaries ] in an unpolarized medium the neutrino propagation in matter will be affected by the vector coupling constant of the nsi , @xmath55 . ] . the way the neutral current nsi modifies the neutrino evolution will be parametrized phenomenologically through the effective low - energy four - fermion operator described in eq . ( [ eq : lnsi ] ) . we also assume @xmath56 , neglecting possible @xmath57 violation in the new interactions . under these assumptions the hamiltonian describing the sn neutrino evolution in the presence of nsi can be cast in the following form @xmath58 where @xmath59 stands for the kinetic term @xmath60 with @xmath61 , and @xmath62 the three - neutrino lepton mixing matrix @xcite in the pdg convention @xcite and with no @xmath57 phases . the second term of the hamiltonian accounts for the interaction of neutrinos with matter and can be split into two pieces , @xmath63 the first term , @xmath64 describes the standard interaction with matter and can be written as @xmath65 = diag @xmath66 up to one loop corrections due to different masses of the muon and tau leptons @xcite . the standard matter potential for neutrinos is given by @xmath67 where @xmath68 ev , the density is given in @xmath69 , and @xmath43 stands for the relative number of electrons with respect to baryons . for antineutrinos the potential is identical but with the sign changed . the term in the hamiltonian describing the non - standard neutrino interactions with a fermion @xmath8 can be expressed as , @xmath70 with @xmath71 . for definiteness and motivated by actual models , for example , those with broken r parity supersymmetry we take for @xmath8 the down - type quark . however , an analogous treatment would apply to the case of nsi on up - type quarks , the existence of nsi with electrons brings no drastic qualitative differences with respect to the pure oscillation case ( see below ) . therefore the nsi potential can be expressed as follows , @xmath72 from now on we will not explicitely write the superindex @xmath73 . in order to further simplify the problem we will redefine the diagonal nsi parameters so that @xmath74 , as one can easily see that subtracting a matrix proportional to the identity leaves the physics involved in the neutrino oscillation unaffected . neutrino propagation depends on the supernova matter and chemical profile through the effective potential . this profile exhibits an important time dependence during the explosion . [ fig : snprofiles ] shows the density @xmath75 and the electron fraction @xmath76 profiles for the sn progenitor as well as at different times post - bounce . progenitor density profiles can be roughly parametrized by a power - law function @xmath77 where @xmath78 g/@xmath79 , @xmath80 cm , and @xmath81 . the electron fraction profile varies depending on the matter composition of the different layers . for instance , typical values of @xmath43 between 0.42 and 0.45 in the inner regions are found in stellar evolution simulations @xcite . in the intermediate regions , where the msw @xmath82 and @xmath83-resonances take place @xmath84 . this value can further increase in the most outer layers of the sn envelope due to the presence of hydrogen . after the sn core bounce the matter profile is affected in several ways . first note that a front shock wave starts to propagate outwards and eventually ejects the sn envelope . the evolution of the shock wave will strongly modify the density profile and therefore the neutrino propagation @xcite . following ref . @xcite we shall assume that the structure of the shock wave is more complicated and an additional `` reverse wave '' appears due to the collision of the neutrino - driven wind and the slowly moving material behind the forward shock , as seen in the upper panel of fig . [ fig : snprofiles ] . on the other hand , the electron fraction is also affected by the time evolution as the sn explosion proceeds once the collapse starts the core density grows so that the neutrinos become eventually effectively trapped within the so called `` neutrinosphere '' . at this point the trapped electron fraction has decreased until values of the order of 0.33 @xcite . when the inner core reaches the nuclear density it can not contract any further and bounces . as a consequence a shock wave forms in the inner core and starts propagating outwards . when the newly formed supernova shock reaches densities low enough for the initially trapped neutrinos to begin streaming faster than the shock propagates @xcite , a breakout pulse of @xmath1 is launched . in the shock - heated matter , which is still rich of electrons and completely disintegrated into free neutrons and protons , a large number of @xmath1 are rapidly produced by electron captures on protons . they follow the shock on its way out until they are released in a very luminous flash , the breakout burst , at about the moment when the shock penetrates the neutrinosphere and the neutrinos can escape essentially unhindered . as a consequence , the lepton number in the layer around the neutrinosphere decreases strongly and the matter neutronizes @xcite . the value of @xmath43 steadily decreases in these layers until values of the order of @xmath85 . outside the neutrinosphere there is a steep rise until @xmath86 . this is a robust feature of the neutrino - driven baryonic wind . neutrino heating drives the wind mass loss and causes @xmath43 to rise within a few @xmath87 km from low to high values , between 0.45 and 0.55 @xcite , see bottom panel of fig . [ fig : snprofiles ] . inspired in the numerical results of ref . @xcite we have parametrized the behavior of the electron fraction near the neutrinosphere phenomenologically as , @xmath88~ , \label{eq : ye}\ ] ] where @xmath89 and @xmath90 . the parameters @xmath91 and @xmath92 describe where the rise takes place and how steep it is , respectively . as can be seen in fig . [ fig : snprofiles ] both decrease with time . ( yellow ) and the @xmath83 ( cyan ) resonance take place are also indicated , as well as the nsi - induced @xmath93 ( gray ) resonance for the parameters @xmath94 and @xmath95 , scaledwidth=50.0% ] in order to study the neutrino propagation through the sn envelope we will split the problem into two different regions : the inner envelope , defined by the condition @xmath96 with @xmath97 , and the outer one , where @xmath98 . from the upper panel of fig . [ fig : snprofiles ] one can see how the boundary roughly varies between @xmath99 cm and @xmath100 cm , depending on the time considered . this way one can fully characterize all resonances that can take place in the propagation of supernova neutrinos , both the outer resonant conversions related to neutrino masses and indicated as the upper bands in fig . [ fig : snprofiles ] , and the inner resonances that follow from the presence of non - standard neutrino interactions , indicated by the band at the bottom of the same figure . here we pay special attention to the use of realistic matter and chemical supernova profiles and three - neutrino flavors thus generalising previous studies . let us first write the hamiltonian in the inner layers , where @xmath101 . in this case the hamiltonian can be written as @xmath102 when the value of the @xmath103 is of the same order as the electron fraction @xmath43 internal resonances can arise @xcite . taking into account the current constraints on the @xmath31 s discussed in sec . [ sec : preliminaries ] one sees that small values of @xmath43 are required @xcite . as a result , these can only take place in the most deleptonised inner layers , close to the neutrinosphere , where the kinetic terms of the hamiltonian are negligible . given the large number of free parameters @xmath103 involved we consider one particular case where @xmath104 and @xmath105 are small enough to neglect a possible initial mixing between @xmath1 and @xmath106 or @xmath107 . barring fine tuning , this basically amounts to @xmath108 . according to the discussion of sec . [ sec : preliminaries ] @xmath109 automatically satisfies the condition , whereas one expects that the window @xmath110 will eventually be probed in future experiments . since the initial fluxes of @xmath106 and @xmath107 are expected to be basically identical , it is convenient to redefine the weak basis by performing a rotation in the @xmath111 sector : @xmath112 where @xmath113 and @xmath114 stand for @xmath115 and @xmath116 , respectively . the angle @xmath117 can be written as @xmath118 the hamiltonian becomes in the new basis @xmath119 where @xmath120 with our initial assumptions on @xmath54 one notices that the new basis @xmath121 basically diagonalizes the hamiltonian , and therefore coincides roughly with the matter eigenstate basis . a novel resonance can arise if the condition @xmath122 is satisfied , we call this @xmath93-resonance , @xmath93 standing for `` internal '' would give rise to another internal resonance which can be studied using the same method . for brevity , we will not pursue this in this paper . ] . the corresponding resonance condition can be written as @xmath123 where @xmath124 is defined as @xmath125 . as function of @xmath126 and @xmath127 according to eq . ( [ eq : ye_resi ] ) for different values of @xmath43 . the region in yellow represents the region of parameters that gives rise to @xmath93-resonance before the collapse . the arrows indicate how this region widens with time.,scaledwidth=45.0% ] in fig . [ fig : yei - ei ] we represent the range of @xmath126 and @xmath127 leading to the @xmath93-resonance for an electron fraction profile between different @xmath128 s and @xmath129 . it is important to notice that the value of @xmath128 depends on time . right before the collapse the minimum value of the electron fraction is around @xmath130 . hence the window of nsi parameters that would lead to a resonance would be relatively narrow , as indicated by the shaded ( yellow ) band in fig . [ fig : yei - ei ] . as time goes on @xmath128 decreases to values of the order of a few % , and as a result the region of parameters giving rise to the @xmath93-resonance significantly widens . for example , in the range @xmath131 possibly accessible to future experiments one sees that the @xmath93-resonance can take place for values of @xmath127 of the order of @xmath85 . this indicates that the potential sensitivity on nsi parameters that can be achieved in supernova studies is better than that of the current limits . as seen in fig . [ fig : snprofiles ] in order to fulfill the @xmath93-resonance condition for such small values of the nsi parameters the values of @xmath43 must indeed lie , as already stated , in the inner layers . several comments are in order : first , in contrast to the standard @xmath82 and @xmath83-resonances , related to the kinetic term , the density itself does not explicitly enter into the resonance condition , provided that the density is high enough to neglect the kinetic terms . analogously the energy plays no role in the resonance condition , which is determined only by the electron fraction @xmath43 . moreover , in contrast to the standard resonances , the @xmath93-resonance occurs for both neutrinos and antineutrinos simultaneously @xcite . finally , as indicated in fig . [ fig : scheme ] the @xmath1 s ( @xmath0 ) are not created as the heaviest ( lightest ) state but as the intermediate state , therefore the flavor composition of the neutrinos arriving at the @xmath82-resonance is exactly the opposite of the case without nsi . as we show in sec . [ sec : observables ] , this fact can lead to important observational consequences . in order to calculate the hopping probability between matter eigenstates at the @xmath93-resonance we use the landau - zener approximation for two flavors @xmath132 where @xmath133 stands for the adiabaticity parameter , which can be generally written as @xmath134 where @xmath135 . if one applies this formula to the @xmath136 box of eq . ( [ eq : hi1 ] ) assuming that @xmath137 and @xmath138^{1/2}$ ] one gets @xmath139 where the parametrization of the @xmath43 profile has been defined as in eq . ( [ eq : ye ] ) with @xmath140 . the density @xmath141 represents the density in units of @xmath142 g/@xmath79 , @xmath143 stands for @xmath92 in units of @xmath144 cm , and @xmath145 is a function whose value is of the order @xmath146 in the range of parameters we are interested in . taking all these factors into account it follows that the internal resonance will be adiabatic provided that @xmath147 , well below the current limits , in full numerical agreement with , e. g. , ref . @xcite . in fig . [ fig : adiab_i ] we show the resonance condition as well as the adiabaticity in terms of @xmath37 and @xmath38 assuming the other @xmath148 . in order to illustrate the dependence on time we consider profiles inspired in the numerical profiles of fig . [ fig : snprofiles ] at @xmath149 s ( upper panel ) and 15.7 s ( bottom panel ) . for definiteness we take @xmath128 as the electron fraction at which the density has value of @xmath150g/@xmath79 . for comparison with fig . [ fig : yei - ei ] we have assumed @xmath151 in the case of 15.7 s. we observe how the border of adiabaticity depends on @xmath37 through the value of the density at @xmath152 which in turn depends on time . before moving to the discussion of the outer resonances a comment is in order , namely , how does the formalism change for other non - standard interaction models . first note that the whole treatment presented above also applies to the case of nsi on up - type quarks , except that the position of the internal resonance shifts with respect to the down - quark case . indeed , in this case the nsi potential @xmath153 would induce a similar internal resonance for the condition @xmath154 . in contrast , for the case of nsi with electrons , the nsi potential is proportional to the electron fraction , and therefore no internal resonance would appear . in the outer layers of the sn envelope neutrinos can undergo important flavor transitions at those points where the matter induced potential equals the kinetic terms . in absence of nsi this condition can be expressed as @xmath155 . neutrino oscillation experiments indicate two mass scales , @xmath156 and @xmath157 @xcite , hence two different resonance layers arise , the so - called @xmath82-resonance and the @xmath83-resonance , respectively . the presence of nsi with values of @xmath158 modifies the properties of the @xmath82 and @xmath83 transitions @xcite . in particular one finds that the effects of the nsi can be described as in the standard case by embedding the @xmath31 s into effective mixing angles @xcite . an analogous `` confusion '' between @xmath159 and the corresponding nsi parameter @xmath38 has been pointed out in the context of long - baseline neutrino oscillations in refs . @xcite . in this section we perform a more general and complementary study for slightly higher values of the nsi parameters : @xmath160 , still allowed by current limits , and for which the @xmath93-resonance could occur . the phenomenological assumption of hierarchical squared mass differences , @xmath161 , allows , for not too large @xmath31 s , a factorization of the 3@xmath5 dynamics into two 2@xmath5 subsystems roughly decoupled for the @xmath82 and @xmath83 transitions @xcite . to isolate the dynamics of the @xmath82 transition , one usually rotates the neutrino flavor basis by @xmath162 , and extracts the submatrix with indices ( 1,3 ) @xcite . whereas this method works perfectly for small values of @xmath103 it can be dangerous for values above @xmath163 . in order to analyze how much our case deviates from the simplest approximation we have performed a rotation with the angle @xmath164 instead of just @xmath165 . by requiring that the new rotation diagonalizes the submatrix ( 2,3 ) at the @xmath82-resonance layer one obtains the following expression for the correction angle @xmath11 @xmath166/\nonumber \\ & & \left[(\delta_{\rm atm}+ \frac{1}{2}\delta_{\odot } ) c^2_{13}+\frac{1}{4}\delta_{\odot } c2_{12}(-3+c2_{13 } ) \right . \nonumber \\ & & \left.+ v_{\tau\tau}^{nsi } c2_{23 } + 2 v_{\mu\tau}^{nsi } s2_{23}\right]~,\end{aligned}\ ] ] where @xmath167 and @xmath168 . in our notation @xmath169 and @xmath170 represent @xmath171 and @xmath172 , respectively . the parameters @xmath173 and @xmath174 are analogously defined . in the absence of nsi @xmath11 is just a small correction to @xmath165 -resonance @xmath175 after neglecting the kinetic terms . ] , @xmath176 in order to calculate @xmath11 we need to know the @xmath82-resonance point . to calculate it one can proceed as in the case without nsi , namely , make the @xmath177 rotation and analyze the submatrix @xmath178 . the new hamiltonian @xmath179 has now the form @xmath180 + \delta_{\rm atm}s^2_{13 } \nonumber \\ & & + \delta_{\odot}(c^2_{13}s^2_{12}+s^2_{13})~ , \nonumber\\ h_{\tau\tau } '' & = & v_0\rho ( 2-y_e)\varepsilon_{\tau\tau } '' + \delta_{\rm atm}c^2_{13}c^2_{\alpha } \nonumber \\ & & + \delta_{\odot}\left[c^2_{13}c^2_{\alpha } + ( s_{\alpha}c_{12}+c_{\alpha}s_{12}s_{13})^2 \right]~ , \nonumber \\ h_{e\tau } '' & = & v_0\rho ( 2-y_e)\varepsilon_{e\tau } '' + \frac{1}{2}\delta_{\rm atm}s2_{13}c_{\alpha } \nonumber \\ & & + \frac{1}{2}\delta_{\odot}(-c_{13}s_{\alpha}s2_{12 } + c^2_{12}c_{\alpha}s2_{13})~. \end{aligned}\ ] ] we have defined @xmath181 , and @xmath182 , where @xmath183 , and @xmath184 . the resonance condition for the @xmath82 transition , @xmath185 can be then written as @xmath186 = \delta_{\rm atm}(c^2_{13 } c^2_{\alpha } -s^2_{13 } ) & \nonumber \\ & + \delta_{\odot}[c^2_{12}(c^2_{13}-c^2_{\alpha}s^2_{13})- s^2_{\alpha}s^2_{12}+\frac{1}{2}s2_{\alpha}s2_{12 } s_{13 } ] & . \label{eq : resh}\end{aligned}\ ] ] it can be easily checked how in the limit of @xmath187 one recovers the standard resonance condition , @xmath188 in the region where the @xmath82-resonance occurs @xmath189 . taking into account eqs . ( [ eq : alfa ] ) and ( [ eq : resh ] ) one can already estimate how the value of @xmath11 changes with the nsi parameters . in fig . [ fig : alpha ] we show the dependence of @xmath11 on the @xmath37 after fixing the value of the other nsi parameters . one can see how for @xmath190 the approximation of neglecting @xmath11 significantly worsens . assuming @xmath191 and a fixed value of @xmath192 one can easily see that @xmath37 basically affects the numerator in eq . ( [ eq : alfa ] ) . therefore one expects a rise of @xmath11 as the value of @xmath37 increases , as seen in fig . [ fig : alpha ] . the dependence of @xmath11 on @xmath192 is correlated to the relative sign of the mass hierarchy and @xmath192 . for instance , for normal mass hierarchy and positive values of @xmath192 the dependence is inverse , namely , higher values of @xmath192 lead to a suppression of @xmath11 . apart from this general behavior , @xmath11 also depends on the diagonal term @xmath126 as seen in fig . [ fig : alpha ] . this effect occurs by shifting the resonance point through the resonance condition in eq . ( [ eq : resh ] ) . one can now calculate the jump probability between matter eigenstates in analogy to the @xmath93-resonance by means of the landau - zener approximation , see eqs . ( [ eq : plz - i1 ] ) , ( [ eq : gamma - i1 - 1 ] ) , and [ eq : gamma - i1 - 2 ] , @xmath193 where @xmath194 represents the adiabaticity parameter at the @xmath82-resonance , which can be written as @xmath195 where the expressions for @xmath179 are given in eqs ( [ eq : h_h ] ) . let us first consider the case @xmath196 . in this case @xmath197 and one can rewrite the adiabaticity parameter as @xmath198 where @xmath199 in agreement with ref . for slightly larger @xmath31 s there can be significant differences . in fig . [ fig : p - h ] we show @xmath200 in the @xmath38-@xmath37 plane for antineutrinos with energy @xmath87 mev in the case of inverse mass hierarchy , using eq . ( [ eq : plz - h ] ) with ( upper panel ) and without ( bottom panel ) the @xmath11 correction . the values of @xmath53 and @xmath38 have been chosen so that the jump probability lies in the transition regime between adiabatic and strongly non adiabatic . in the limit of small @xmath37 , @xmath11 becomes negligible and therefore both results coincide . from eq . ( [ eq : gammah_lisi ] ) one sees how as the value of @xmath38 increases @xmath201 gets larger and therefore the transition becomes more and more adiabatic . for negative values of @xmath38 there can be a cancellation between @xmath38 and @xmath53 , and as a result the transition becomes non - adiabatic . an additional consequence of eq . ( [ eq : deg ] ) is that a degeneracy between @xmath38 and @xmath53 arises . this is seen in fig . [ fig : confusion ] , which gives the contours of @xmath202 in terms of @xmath38 and @xmath53 for @xmath203 . one sees clearly that the same landau - zener hopping probability is obtained for different combinations of @xmath38 and @xmath53 . this leads to an intrinsic `` confusion '' between the mixing angle and the corresponding nsi parameter , which can not be disentangled only in the context of sn neutrinos , as noted in ref . @xcite . we now turn to the case of @xmath204 . as @xmath205 increases the role of @xmath11 becomes relevant . whereas in the bottom panel @xmath200 remains basically independent of @xmath37 , one can see how in the upper panel @xmath200 becomes strongly sensitive to @xmath37 for @xmath204 . one sees that for positive values of @xmath37 it tends to adiabaticity whereas for negative values to non - adiabaticity . this follows from the dependence of @xmath206 on @xmath11 , essentially through the term @xmath207 , see eq . ( [ eq : h_h ] ) . for @xmath204 one sees that @xmath208 starts being important , and as a result this term eventually becomes of the same order as the others in @xmath206 . at this point the sign of @xmath37 , and so the sign of @xmath208 , is crucial since it may contribute to the enhancement or reduction of @xmath206 . this directly translates into a trend towards adiabaticity or non - adiabaticity , seen in fig . [ fig : p - h ] . thus , for the range of @xmath37 relevant for the nsi - induced internal resonance the adiabaticity of the outer @xmath82 resonance can be affected in a non - trivial way . -resonance in terms of @xmath38 and @xmath53 for @xmath203 . antineutrinos with energy @xmath87 mev and inverted mass hierarchy has been assumed.,scaledwidth=48.0% ] turning to the case of the @xmath83 transition a similar expression can be obtained by rotating the original hamiltonian by @xmath209 @xcite . however , in contrast to the case of the @xmath82-resonance , where the mixing angle @xmath53 is still unknown , in the case of the @xmath83 transition the angle @xmath52 has been shown by solar and reactor neutrino experiments to be large @xcite . as a result , for the mass scale @xmath210 this transition will always be adiabatic irrespective of the values of @xmath103 , and will affect only neutrinos . as mentioned in the introduction one of the major motivations to study nsi using the neutrinos emitted in a sn is the enhancement of the nsi effects on the neutrino propagation through the sn envelope due to the specific extreme matter conditions that characterize it . in this section we analyze how these effects translate into observable effects in the case of a future galactic sn . schematically , the neutrino emission by a sn can be divided into four stages : infall phase , neutronization burst , accretion , and kelvin - helmholtz cooling phase . during the infall phase and neutronization burst only @xmath1 s are emitted , while the bulk of neutrino emission is released in all flavors in the last two phases . whereas the neutrino emission characteristics of the two initial stages are basically independent of the features of the progenitor , such as the core mass or equation of state ( eos ) , the details of the neutrino spectra and luminosity during the accretion and cooling phases may significantly change for different progenitor models . as a result , a straightforward extraction of oscillation parameters from the bulk of the sn neutrino signal seems hopeless . only features in the detected neutrino spectra which are independent of unknown sn parameters should be used in such an analysis @xcite . the question then arises as to how can one obtain information about the nsi parameters . taking into account that the main effect of nsi is to generate new internal neutrino flavor transitions , one possibility is to invoke theoretical arguments that involve different aspects of the sn internal dynamics . in ref . @xcite it was argued that such an internal flavor conversion during the first second after the core bounce might play a positive role in the so - called sn shock reheating problem . it is observed in numerical simulations @xcite that as the shock wave propagates it loses energy until it gets stalled at a few hundred km . it is currently believed that after neutrinos escape the sn core they can to some extent deposit energy right behind and help the shock wave continue outwards . on the other hand it is also believed that due to the composition in matter of the protoneutronstar ( pns ) the mean energies of the different neutrino spectra obey @xmath211 . this means that a resonant conversion between @xmath212 and @xmath213 between the neutrinosphere and the position of the stalled shock wave would make the @xmath212 spectra harder , and therefore the energy deposition would be larger , giving rise to a shock wave regeneration effect . another argument used in the literature was the possibility that the @xmath47process nucleosynthesis , responsible for synthesizing about half of the heavy elements with mass number @xmath214 in nature , could occur in the region above the neutrinosphere in sne @xcite . a necessary condition is @xmath215 in the nucleosynthesis region . the value of the electron fraction depends on the neutrino absorption rates , which are determined in turn by the @xmath216 luminosities and energy distribution . these can be altered by flavor conversion in the inner layers due to the presence of nsi . therefore by requiring the electron fraction be below 0.5 one can get information about the values of the nsi parameters . while it is commonly accepted that neutrinos will play a crucial role in both the shock wave re - heating as well as the @xmath47process nucleosynthesis , there are still other astrophysical factors that can affect both . while the issue remains under debate we prefer to stick to arguments directly related to physical observables in a large water cherenkov detector . there are several possibilities . * the modulations in the @xmath0 spectra due to the passage of shock waves through the supernova @xcite * the modulation in the @xmath0 spectra due to the time dependence of the electron fraction , induced by the @xmath93-resonance * the modulations in the @xmath0 spectra due to the earth matter @xcite * detectability of the neutronization @xmath1 burst @xcite three of these observables , 1 , 3 and 4 have already been considered in the literature in the context of neutrino oscillations . here we discuss the potential of the above promising observables in providing information about the nsi parameters . it is important to pay attention to the possible ocurrence of the internal @xmath93-resonance and to its effect in the external @xmath82 and @xmath83-resonances . the first can induce a genuinely new observable effect , item 2 above . here we concentrate on neutral current - type non - standard interactions , hence there will be not effect in the main reaction in water cherenkov and scintillator detectors , namely the inverse beta decay , @xmath217 will contribute to the @xmath218 cross section . ] . for definiteness we take nsi with @xmath73 ( down ) quarks , in which case the nsi effects will be confined to the neutrino evolution inside the sn and the earth , through the vector component of the interaction . from all possible combinations of nsi parameters we will concentrate on those for which the internal @xmath93 transition does take place , namely @xmath219 , see fig . [ fig : yei - ei ] . concerning the fc nsi parameters we will consider @xmath220 between @xmath221 and @xmath163 , range in which the @xmath93-resonance is adiabatic , see fig . [ fig : adiab_i ] . in the following discussion we will focus on the extreme cases defined in table [ table : nuschemes ] . one of the motivations for considering these cases is the fact that the resonances involved become either adiabatic or strongly non adiabatic , and hence the survival probabilities in the absence of earth effects or shock wave passage , become energy independent . this assumption simplifies the task of relating the observables with the neutrino schemes . .definition of the neutrino schemes considered in terms of the hierarchy , the value of @xmath53 , and the presence of nsi , as described in the text . the values of the survival probabilities for @xmath1 ( @xmath222 ) and @xmath0 ( @xmath223 ) for each case are also indicated . [ cols="^,^,^,^,^,^",options="header " , ] the authors wish to thank h - th . janka , o. miranda , s. pastor , th . schwetz , and m. trtola for fruitful discussions . work supported by the spanish grant fpa2005 - 01269 and european network of theoretical astroparticle physics ilias / n6 under contract number rii3-ct-2004 - 506222 . a. e. has been supported by a fpu grant from the spanish government . r. t. has been supported by the juan de la cierva program from the spanish government and by an erg from the european commission . 1.31.2 kamland collaboration , k. eguchi _ et al . _ , rev . lett . * 90 * , 021802 ( 2003 ) , [ hep - ex/0212021 ] . s. pakvasa and j. w. f. valle , hep - ph/0301061 , proc . of the indian national academy of sciences on neutrinos , vol . 70a , no.1 , p.189 - 222 ( 2004 ) , eds . d. indumathi , m.v.n . murthy and g. rajasekaran . v. barger , d. marfatia and k. whisnant , hep - ph/0308123 . kamland collaboration , t. araki _ et al . lett . * 94 * , 081801 ( 2004 ) . m. maltoni , t. schwetz , m. a. tortola and j. w. f. valle , new j. phys . * 6 * , 122 ( 2004 ) , appendix c in hep - ph/0405172 ( v5 ) provides updated neutrino oscillation results taking into account new ssm , new sno salt data , latest k2k and minos data ; previous works by other groups are referenced therein . j. schechter and j. w. f. valle , phys . rev . * d22 * , 2227 ( 1980 ) . j. w. f. valle , j. phys . . ser . * 53 * , 473 ( 2006 ) , [ hep - ph/0608101 ] , review based on lectures at the corfu summer institute on elementary particle physics in september 2005 . j. schechter and j. w. f. valle , phys . rev . * d24 * , 1883 ( 1981 ) , err . d25 , 283 ( 1982 ) . c .- s . lim and w. j. marciano , phys . rev . * d37 * , 1368 ( 1988 ) . e. k. akhmedov , phys b213 * , 64 ( 1988 ) . l. wolfenstein , phys . rev . * d17 * , 2369 ( 1978 ) . , ( editions frontires , gif - sur - yvette , 1986 , p.355 . ) , 86 massive neutrinos in astrophysics and particle physics , proceedings of the sixth moriond workshop , ed . by fackler , o. and tran thanh van , j. j. w. f. valle , phys . lett . * b199 * , 432 ( 1987 ) . r. n. mohapatra and j. w. f. valle , phys . rev . * d34 * , 1642 ( 1986 ) . j. bernabeu _ et al . b187 * , 303 ( 1987 ) . g. c. branco , m. n. rebelo and j. w. f. valle , phys . lett . * b225 * , 385 ( 1989 ) . n. rius and j. w. f. valle , phys . b246 * , 249 ( 1990 ) . f. deppisch and j. w. f. valle , phys . rev . * d72 * , 036001 ( 2005 ) , [ hep - ph/0406040 ] . a. zee , phys . lett . * b93 * , 389 ( 1980 ) . k. s. babu , phys b203 * , 132 ( 1988 ) . l. j. hall , v. a. kostelecky and s. raby , nucl . phys . * b267 * , 415 ( 1986 ) . m. malinsky , j. c. romao and j. w. f. valle , phys . lett . * 95 * , 161801 ( 2005 ) , [ hep - ph/0506296 ] . a. b. mcdonald , astro - ph/0406253 . k. scholberg , astro - ph/0701081 . lsnd , l. b. auerbach _ et al . _ , phys . rev . * d63 * , 112001 ( 2001 ) , [ hep - ex/0101039 ] . munu , z. daraktchieva _ et al . _ , b564 * , 190 ( 2003 ) , [ hep - ex/0304011 ] . charm , j. dorenbosch _ et al . _ , phys . lett . * b180 * , 303 ( 1986 ) . charm - ii , p. vilain _ et al . _ , phys . lett . * b335 * , 246 ( 1994 ) . nutev , g. p. zeller _ et al . _ , phys . * 88 * , 091802 ( 2002 ) , [ hep - ex/0110059 ] . v. d. barger , r. j. n. phillips and k. whisnant , phys . d44 * , 1629 ( 1991 ) . s. davidson , c. pena - garay , n. rius and a. santamaria , jhep * 03 * , 011 ( 2003 ) , [ hep - ph/0302093 ] . j. barranco , o. g. miranda , c. a. moura and j. w. f. valle , phys . rev . * d73 * , 113001 ( 2006 ) , [ hep - ph/0512195 ] . z. berezhiani and a. rossi , phys . b535 * , 207 ( 2002 ) , [ hep - ph/0111137 ] . a. friedland , c. lunardini and c. pena - garay , phys . b594 * , 347 ( 2004 ) , [ hep - ph/0402266 ] . m. m. guzzo , p. c. de holanda and o. l. g. peres , phys . lett . * b591 * , 1 ( 2004 ) , [ hep - ph/0403134 ] . o. g. miranda , m. a. tortola and j. w. f. valle , jhep * 10 * , 008 ( 2006 ) , [ hep - ph/0406280 ] . n. fornengo _ et al . _ , phys . rev . * d65 * , 013010 ( 2002 ) , [ hep - ph/0108043 ] . a. friedland , c. lunardini and m. maltoni , phys . * d70 * , 111301 ( 2004 ) , [ hep - ph/0408264 ] . a. friedland and c. lunardini , phys . rev . * d72 * , 053009 ( 2005 ) , [ hep - ph/0506143 ] . g. mangano _ et al . _ , nucl . phys . * b756 * , 100 ( 2006 ) , [ hep - ph/0607267 ] . s. k. katsanevas , talk at workshop on neutrino oscillation physics ( now 2006 ) , otranto , lecce , italy , 9 - 16 sep 2006 . s. amanik , g. m. fuller and b. grinstein , astropart . phys . * 24 * , 160 ( 2005 ) , [ hep - ph/0407130 ] . p. s. amanik and g. m. fuller , astro - ph/0606607 . s. p. mikheev and a. y. smirnov , sov . * 42 * , 913 ( 1985 ) . s. p. mikheev and a. y. smirnov , nuovo cim . * c9 * , 17 ( 1986 ) . h. nunokawa , y. z. qian , a. rossi and j. w. f. valle , phys . rev . * d54 * , 4356 ( 1996 ) , [ hep - ph/9605301 ] . h. nunokawa , a. rossi and j. w. f. valle , nucl . phys . * b482 * , 481 ( 1996 ) , [ hep - ph/9606445 ] . s. mansour and t .- k . kuo , phys . rev . * d58 * , 013012 ( 1998 ) , [ hep - ph/9711424 ] . s. bergmann and a. kagan , nucl . * b538 * , 368 ( 1999 ) , [ hep - ph/9803305 ] . g. l. fogli , e. lisi , a. mirizzi and d. montanino , phys . rev . * d66 * , 013009 ( 2002 ) , [ hep - ph/0202269 ] . t .- k . kuo and j. t. pantaleone , phys . rev . * d37 * , 298 ( 1988 ) . s. bergmann , nucl . phys . * b515 * , 363 ( 1998 ) , [ hep - ph/9707398 ] . huber , t. schwetz and j. w. f. valle , phys . 88 * , 101804 ( 2002 ) , [ hep - ph/0111224 ] . p. huber , t. schwetz and j. w. f. valle , phys . rev . * d66 * , 013006 ( 2002 ) , [ hep - ph/0202048 ] . particle data group , w. m. yao _ et al . _ , j. phys . * g33 * , 1 ( 2006 ) . f. j. botella , c. s. lim and w. j. marciano , phys . rev . * d35 * , 896 ( 1987 ) . s. e. woosley , a. heger and t. a. weaver , reviews of modern physics * 74 * , 1015 ( 2002 ) . r. c. schirato , g. m. fuller , . lanl ) , ucsd and lanl ) , astro - ph/0205390 . g. l. fogli , e. lisi , d. montanino and a. mirizzi , phys . rev . * d68 * , 033005 ( 2003 ) , [ hep - ph/0304056 ] . r. tomas _ et al . _ , jcap * 0409 * , 015 ( 2004 ) , [ astro - ph/0407132 ] . c. y. cardall , astro - ph/0701831 . h. a. bethe , j. h. applegate and g. e. brown , astrophys . j. * 241 * , 343 ( 1980 ) . a. burrows and t. j. mazurek . j. * 259 * , 330 ( 1982 ) . h. th . janka , private communication . t .- k . kuo and j. t. pantaleone , rev . phys . * 61 * , 937 ( 1989 ) . m. kachelriess and r. tomas , hep - ph/0412100 . m. liebendoerfer _ et al . _ , phys . rev . * d63 * , 103004 ( 2001 ) , [ astro - ph/0006418 ] . m. rampp and h. t. janka , astron . astrophys . * 396 * , 361 ( 2002 ) , [ astro - ph/0203101 ] . t. a. thompson , a. burrows and p. a. pinto , astrophys . j. * 592 * , 434 ( 2003 ) , [ astro - ph/0211194 ] . k. sumiyoshi _ et al . _ , astrophys . j. * 629 * , 922 ( 2005 ) , [ astro - ph/0506620 ] . qian , prog . . phys . * 50 * , 153 ( 2003 ) , [ astro - ph/0301422 ] . j. pruet , s. e. woosley , r. buras , h .- janka and r. d. hoffman , astrophys . j. * 623 * , 325 ( 2005 ) , [ astro - ph/0409446 ] . c. lunardini and a. y. smirnov , nucl . b616 * , 307 ( 2001 ) , [ hep - ph/0106149 ] . a. s. dighe , m. t. keil and g. g. raffelt , jcap * 0306 * , 005 ( 2003 ) , [ hep - ph/0303210 ] . a. s. dighe , m. t. keil and g. g. raffelt , jcap * 0306 * , 006 ( 2003 ) , [ hep - ph/0304150 ] . a. s. dighe , m. kachelriess , g. g. raffelt and r. tomas , jcap * 0401 * , 004 ( 2004 ) , [ hep - ph/0311172 ] . k. takahashi , k. sato , a. burrows and t. a. thompson , phys . 113009 ( 2003 ) , [ hep - ph/0306056 ] . m. kachelriess _ _ , phys . rev . * d71 * , 063003 ( 2005 ) , [ astro - ph/0412082 ] . m. t. keil , phd thesis tu mnchen 2003 [ astro - ph/0308228 ] . m. t. keil , g. g. raffelt and h. t. janka , astrophys . j. * 590 * ( 2003 ) 971 [ astro - ph/0208035 ] . r. tomas , d. semikoz , g. g. raffelt , m. kachelriess and a. s. dighe , phys . rev . * d68 * , 093013 ( 2003 ) , [ hep - ph/0307050 ] . h. nunokawa , v. b. semikoz , a. y. smirnov and j. w. f. valle , nucl . * b501 * , 17 ( 1997 ) , [ hep - ph/9701420 ] . h. duan , g. m. fuller , j. carlson and y .- z . qian , phys . rev . * d74 * , 105014 ( 2006 ) , [ astro - ph/0606616 ] . h. duan , g. m. fuller , j. carlson and y .- z . qian , phys . lett . * 97 * , 241101 ( 2006 ) , [ astro - ph/0608050 ] . s. hannestad , g. g. raffelt , g. sigl and y. y. y. wong , phys . rev . * d74 * , 105010 ( 2006 ) , [ astro - ph/0608695 ] . g. g. raffelt and g. g. r. sigl , hep - ph/0701182 . a. b. balantekin , j. m. fetter and f. n. loreti , phys . rev . * d54 * , 3941 ( 1996 ) , [ astro - ph/9604061 ] . h. nunokawa , a. rossi , v. b. semikoz and j. w. f. valle , nucl . b472 * , 495 ( 1996 ) , [ hep - ph/9602307 ] . g. l. fogli , e. lisi , a. mirizzi and d. montanino , jcap * 0606 * , 012 ( 2006 ) , [ hep - ph/0603033 ] . a. friedland and a. gruzinov , astro - ph/0607244 .
solar coronal mass ejections ( cmes ) produce adverse space weather effects at earth . planets in the close habitable zone of magnetically active m dwarfs may experience more extreme space weather than at earth , including frequent cme impacts leading to atmospheric erosion and leaving the surface exposed to extreme flare activity . similar erosion may occur for hot jupiters with close orbits around solar - like stars . we have developed a model , forecasting a cme s altered trajectory ( forecat ) , which predicts a cme s deflection . we adapt forecat to simulate cme deflections for the mid - type m dwarf v374 peg and hot jupiters with solar - type hosts . v374 peg s strong magnetic fields can trap cmes at the m dwarfs s astrospheric current sheet , the location of the minimum in the background magnetic field . solar - type cmes behave similarly , but have much smaller deflections and do not get trapped at the astrospheric current sheet . the probability of planetary impact decreases with increasing inclination of the planetary orbit with respect to the astrospheric current sheet - 0.5 to 5 cme impacts per day for m dwarf exoplanets , 0.05 to 0.5 cme impacts per day for solar - type hot jupiters . we determine the minimum planetary magnetic field necessary to shield a planet s atmosphere from the cme impacts . m dwarf exoplanets require values between tens and hundreds of gauss . hot jupiters around a solar - type star , however , require a more reasonable g. these values exceed the magnitude required to shield a planet from the stellar wind , suggesting cmes may be the key driver of atmospheric losses .
1605.02683
recent observations by the mars atmosphere and volatile evolution ( maven ) spacecraft have shown that coronal mass ejections ( cmes ) can have a significant impact on the martian atmosphere , and may have influence mars long - term atmospheric evolution @xcite . increases in the external ram and magnetic pressure , due to the passage of the cme , compress the martian magnetosphere , which in turn affects the ionosphere and neutral atmosphere . comparison of maven observations and numerical modeling show that the impact of a cme on 08 march 2015 caused over an order of magnitude increase in the ion escape rate on the dayside , as compared to quiescent times . evidence from our own solar system suggests that space weather can affect the habitability of a planet through atmospheric losses @xcite . for other stellar systems , we expect space weather to be important . additionally , extrasolar space weather may differ significantly from our own system as exoplanets can orbit at very close distances , and the stellar activity and magnetic field strength can be greatly enhanced . in this work we combine knowledge of solar space weather with inferred properties of other systems to determine the frequency and severity of cme impacts for two specific cases - a hot jupiter orbiting a solar - type star , and an exoplanet in the habitable zone of a m dwarf . `` hot jupiters '' were originally one of the most frequently discovered types of exoplanet . hot jupiters tend to orbit solar - like stars ( f- , g- , and k - type ) at very close distances , typically of order 10 stellar radii ( @xmath1 , 0.046 au ) and as small as 3 @xmath1 @xcite . the close orbits lead to a systematic bias in the frequency of observed hot jupiters - these planets are the easiest to observe using radial - velocity or transits . for example , the transit of the exoplanet hd189733b results in a photometric depth of 3% in the light curve @xcite . while hot jupiters may account for approximately 20% of discovered exoplanets , @xcite suggest that only 1% of f , g , and k type stars host an hot jupiter . hot jupiters are not habitable exoplanets due to their close orbits and large size . however , hot jupiters present an opportunity to study planetary systems unlike anything else in our own solar system . the study of hot jupiters has yielded new insights on the evolution of planetary systems as hot jupiters are thought to form at farther radial distances and migrate in toward the star ( e.g. @xcite and @xcite ) . hot jupiters are expected to have very different planetary weather than seen in the solar system . the large asymmetry between the dayside and nightside temperatures can lead to extreme atmospheric winds as fast as 3000 m s@xmath2 @xcite . the small orbital distances lead to high levels of insolation , which can inflate the radii of hot jupiters and lead to lower planetary densities @xcite . the high levels of xuv radiation can cause the inflated atmospheres to escape at extreme rates . for hd189733b , @xcite determine a minimum atmospheric escape rate of 10@xmath3 g s@xmath2 , with observations consistent with values up to several orders of magnitude higher . more recent measurements by @xcite are consistent with a rate of 10@xmath3 g s@xmath2 . the hot , inflated atmosphere would be extremely vulnerable to erosion from cme impacts unless shielded by a planetary magnetic field @xcite . for comparison , @xcite estimate that the 08 march 2015 cme impact increased the martian atmospheric escape to 10@xmath4 g s@xmath2 . the quiescent atmospheric loss of a hot jupiter greatly exceeds values seen in the solar system , and could be significantly enhanced by cme impact . small , low mass stars , such as m dwarfs , vastly outnumber their more massive counterparts within the galaxy . m dwarfs great number , combined with their long main sequence lifetime , and small mass and radius , which results in a low luminosity , have made them popular targets in the search for habitable exoplanets . however , many of the aspects that make m dwarf exoplanets easier to find may also be a detriment to their habitability . the cool , low luminosity nature of m dwarfs leads to close habitable zones , as defined using the traditional requirement of planetary temperatures conducive to the existence of liquid water . m dwarf habitable zones range between 0.03 astronomical units ( au ) and 0.4 au , with the distance being the smallest for late - type m dwarfs @xcite . the largest of these orbital distances is roughly the same as mercury s orbit . for the early- and mid - type habitable zone exoplanets , we have no analogue orbit at such small distances within our own solar system . stellar activity tends to increase with the size of the stellar convection envelope @xcite and stellar rotation rates @xcite , although the activity saturates for sufficiently high rotational velocity @xcite . for mid- to late - type m dwarfs ( m4 to m8.5 ) the activity saturates at higher rotational velocities than for early - type m dwarfs , and above m9 the activity levels decrease significantly @xcite . accordingly , most m dwarf stars will have significantly enhanced stellar activity as compared to the sun . the effects of this enhanced activity should be more pronounced relative to solar exoplanets due to the close proximity of a `` habitable '' m dwarf exoplanet . m dwarfs have extremely long main sequence lifetimes and can remain active for periods of order gyrs @xcite potentially impinging on exoplanet habitability for long time scales . one manner in which space weather can adversely affect a potentially habit exoplanet is via the stellar x - ray and euv flux ( xuv ) . this radiation can heat the upper atmosphere or even ionize it , leading to atmospheric ion pick up loss @xcite . if the uv radiation can penetrate to the surface then it can damage any potential dna present @xcite . while the quiescent xuv flux for a m dwarf habitable zone planet is likely an order of magnitude less than that at earth , the flux will increase to 10 - 100 times that at earth during m dwarf flares @xcite . @xcite suggest that retaining a moderate atmosphere is critical for habitability and that this can be facilitated by the presence of a strong planetary magnetic field . however , @xcite suggest that planets orbiting m dwarfs at distances less than 0.2 au may be tidally locked and the slow rotation will affect the planetary dynamo , leading to little to no planetary magnetic field . @xcite show that in the case of a significant planetary magnetic field , the atmosphere can still be eroded when cme impacts compress the planetary magnetosphere . as the number of cme impacts increases , an exoplanet becomes less likely to retain an atmosphere to shield the surface from harmful xuv radiation . this can adversely affect the habitability of an m dwarf exoplanet in the habitable zone , or potentially erode the hot jupiter itself . these effects will be minimized for planets with an orbital plane corresponding to locations where cmes are less likely to impact . recent observations of m dwarfs indicate that multi - planetary systems are common , and that the planets tend to share an orbital plane @xcite . for a planet orbiting a star which formed from the same gaseous disk we expect low obliquity ( angle between the planetary orbit and the stellar rotation ) , however , many counterexamples with high obliquity have been observed @xcite . @xcite find that the orbits of single planets tend to be more oblique than multi - planet systems . in this work we assume low obliquity , which is representative of many planetary systems , but certainly not all . after assuming some planetary orbit , to be able to predict the likelihood of cme impact , we need to understand the path a cme takes as it leaves a star . observations show that solar cmes `` deflect , '' or deviate from a purely radial trajectory . these deflections correspond to changes in latitude , longitude , or both @xcite . the deflections tend to move cmes away from coronal holes ( chs ) toward the heliospheric current sheet ( hcs , @xcite ) . the hcs corresponds to where the solar magnetic field reverses radial polarity , and tends to be located near the solar equator during solar minimum conditions . magnetic forces deflecting cmes toward the minimum magnetic energy , the hcs on global scales , can explain the direction of observed cme deflections @xcite . in this paper , we adapt a solar cme deflection model , forecasting a cme s altered trajectory ( forecat ; @xcite ) , to predict the deflections of cmes for an m dwarf and hot jupiter system . considering the full range of plausible cme masses and velocities for both cases , we determine whether cme deflections can increase the likelihood of exoplanetary impacts , which may be detrimental to m dwarf exoplanet habitability or increase the atmospheric erosion of hot jupiters . @xcite developed forecat , a model for solar cme deflections and rotations based on the background solar magnetic forces ( magnetic tension and magnetic pressure gradients ) . forecat neglects some time - varying effects such as reconnection . in contrast to more sophisticated magnetohydrodynamic ( mhd ) models , forecat includes the minimum physics necessary to accurately reproduce observed solar cmes @xcite making large parameter space studies feasible on a standard gpu - enabled desktop computer . forecat cmes are initiated by specifying the initial position of a cme on the stellar surface ( latitude , longitude , and tilt , measured clockwise with respect to the equatorial plane ) , three shape parameters to describe an elliptical toroidal shape @xcite , as well as the cme mass , @xmath5 , and final propagation velocity , @xmath6 . forecat includes drag in the nonradial direction to slow the deflection motion . the standard hydrodynamic drag equation can describe mhd drag @xcite and works best with a drag coefficient set equal to a constant times the hyperbolic tangent of the plasma @xmath7 ( the ratio of thermal to magnetic pressure ) . the drag coefficient constant is taken to be one throughout this work . @xcite and @xcite show that forecat reproduces the observed trajectory of solar cmes and is able to constrain some of the cmes initial parameters and the lower corona background . forecat depends on empirical and analytic models to describe a cme s radial propagation and expansion . the cme follows a three - phase propagation model , similar to that presented in @xcite- a slow rise followed by rapid acceleration and finally constant radial propagation . forecat uses fixed radial distances to transition between the three phases . the cme slowly rises at 80 km s@xmath2 until 1.5 @xmath8 , followed by linear acceleration until it reaches @xmath6 at 3 @xmath8 . @xcite explore the effect of these chosen values , which are reasonable average values for solar cmes , and find that variations in these parameters can have small effects on the magnitude of the deflection . the global trends in cme deflection , however , do not change , and we expect that these chosen values should not significantly effect this work . these effects will be explored in a future work . we expect that the higher surface magnetic field strength of a m dwarf @xcite will provide more energy to accelerate cmes radially . this could cause an increase in the initial velocity if the cmes have comparable masses to solar cmes , as discussed in section [ dmparams ] . while individual deflections are sensitive to the chosen parameters of the three - phase propagation model @xcite , we find that the global trends of m dwarf cme deflections do not depend significantly on their specific values . all forecat cmes expand self - similarly in this work , maintaining a fixed angular width . observations of solar cmes suggest that their angular width , as viewed in a coronagraph , can rapidly increase close to the sun but remains constant beyond approximately 5 @xmath8 @xcite . however , self - similar expansion has also been observed much closer to the sun @xcite . the potential field source surface ( pfss ) model describes the solar background magnetic field using a sum of legendre polynomials weighted by harmonic coefficients determined from a photospheric magnetogram . this model assumes that the magnetic field can be described as current - free and becomes entirely radial above the `` source surface . '' the source surface is typically taken to be 2.5 @xmath8 for pfss models of the sun as this produces the most accurate representation of the global structure of the magnetic field @xcite . the pfss model has recently been used to describe the magnetic field of a m dwarf @xcite in stellar wind simulations @xcite . the appropriate source surface distance for an m dwarf is unknown , but @xcite use values of 5 or 4 @xmath1 , respectively . above the source surface , the solar rotation causes the magnetic field to follow an archimedian spiral @xcite . the solar wind density is determined by a modified version of the ( @xcite , hereafter g06 ) density model . the g06 density model combines empirical relations describing the polar or current sheet density versus radial distance . the radial distance from the current sheet determines the relative contribution of the two density profiles . finally , we assume the solar wind is purely radial so that the velocity can be determined using the density and a mass flux that does not vary with distance . similar to the g06 model , we allow the mass flux to vary with angular distance from the current sheet . we discuss the adaptation of this model to a m dwarf stellar wind in section [ stellarwind ] . v374 peg is an m4 star with radius @xmath9 0.34 @xmath8 , mass @xmath10 0.28 @xmath11 , and a 0.44 day rotation period . the habitable zone , as defined using the incident stellar flux @xcite , should exist around 0.1 au ( 20 @xmath8 , 60@xmath1 ) . while smaller mass m dwarfs will have even closer habitable zones that may lead to more extreme space weather , we use v374 peg in this work as it has a well - studied surface magnetic map , reconstructed through the use of zeeman doppler imaging @xcite . this map of the magnetic field has been used to drive several mhd simulations of the stellar wind @xcite . we use the v374 peg magnetogram and a source surface radius of 5 @xmath12 , the same as used in @xcite and @xcite , to determine the magnetic field with the pfss model . to study the effects on hot jupiters orbiting solar - type stars , we simulate a solar - type star with a magnetic background corresponding to carrington rotation ( cr ) 2029 ( 2005 april - may ) , used previously with forecat in @xcite . in the results of this paper , when referring to trends for `` m dwarfs '' versus `` hot jupiters '' we specifically mean mid - type m dwarfs and hot jupiters around solar - type stars . we expect the results to change for different spectral types . for the background stellar wind , we use the results of the mhd solutions presented in @xcite . @xcite consider two types of solar wind heating : a non - isothermal - driven wind where the wind is heated by a spatially - varying polytropic index @xcite , and an alfvn - driven model where the wind is accelerated and heated by alfvn waves that are damped by surface alfvn waves damp and/or turbulence @xcite . for v374 peg we use the alfvn and non - isothermal driven models @xcite , which we refer to as v374 alf and v374 ter in this work . for cr 2029 we consider one alfvn case and two non - isothermal cases , referred to as cr2029 alf and cr2029 ter1 and cr2029 ter2 in this work . cr2029 ter1 uses the same unscaled magnetic field and coronal boundary density and temperature as in cr2029 alf . for cr2029 ter2 , the background magnetic field is scaled up by a factor of 4 in the mhd simulation , a common practice with the non - isothermal driven heating . additionally , cr2029 ter2 has a larger coronal base density and hotter coronal temperature than cr2029 alf . for each case we determine the coefficients for the density model of @xcite ( hereafter g06 ) by fitting to the mhd models . the g06 density model determines the solar wind density using the combination of a current sheet and a polar radial density profile , @xmath13 \mathrm{,}\ ] ] where the @xmath14 represents either the current sheet ( cs ) or the polar ( p ) values . the relative contribution of each polynomial is weighted by @xmath15 , the distance from the current sheet , @xmath16 \mathrm{e}^{-\lambda^2 / w^2 } \mathrm{,}\ ] ] where @xmath17 is a measure of the angular width of the current sheet . we fit the width from the mhd solution using a second order polynomial below 4.5 @xmath8 or 4.5 @xmath1 @xmath18 where the radial distance @xmath19 has units of @xmath8 or @xmath1 for the solar and m dwarf cases , respectively . figure [ fig : dens ] shows the number density versus radial distance for the five density models used in this work , as well as the g06 model . for all cases the current sheet density ( solid lines ) exceeds the polar density ( dashed lines ) . v374 alf and v374 ter ( cyan and red lines , respectively ) behave similarly close to the star , but v374 alf exceeds v374 ter at farther distances . both m dwarf models exceed the solar models by several orders of magnitude at all distances . both the current sheet and polar components of cr2029 alf and cr2029 ter2 ( blue and magneta , respectively ) behave similarly to the g06 current sheet profile ( solid black line ) . cr2029 ter is orders of magnitude smaller than the other solar models , particularly at large distances . we determine the stellar wind speed using the density and the assumption of a constant mass flux along a radial path . the mass flux values are the same as those given in @xcite . as done for previous solar forecat results , we use a higher mass flux in the poles than the current sheet and determine the value at any given location using a weighting based on angular distance from the current sheet . ter1 produces mass fluxes significantly less than observed solar values . we find that forecat is not terribly sensitive to the chosen values , their effects only become significant at distances at which the deflection has become negligible . coronagraph images have been instrumental to our current understanding of solar cmes . a cme s plane - of - the - sky velocity can be measured as the time - derivative of its radial position and the increase in brightness relative to the solar background provides an estimate of the cme mass @xcite . directly imaging extrasolar cme s remains implausible for the foreseeable future so the properties of stellar cmes must be inferred by other means . the signatures of stellar cmes have been searched for via type ii radio bursts @xcite , x - ray dimmings @xcite , uv absorption @xcite , and more recently , an enhancement in the blue wing of a spectral line due to the doppler shift of a propagating cme @xcite . successful observations of blue - wing enhancements have yielded cme velocities between a few hundred to many thousands of km s@xmath2 . these observations suggest stellar cmes may have masses and velocities greatly exceeding the average solar values of a few times 10@xmath20 g and approximately 500 km s@xmath2 @xcite . @xcite infer an 8x10@xmath21 g cme erupting at 5800 km s@xmath2 from the young m3.5 dwarf ad leo and @xcite estimate a 10@xmath22 to 10@xmath23 g cme from a t - tauri star . @xcite recently infer a cme with mass greater than 10@xmath24 g from v374 peg . the plausible range of m dwarf cme parameters can be approximated based on observations of m dwarf flares combined with scaling relations between cmes and flares determined from solar measurements . @xcite determine an empirical relationship between solar flare energy and cme mass and show that it can be extrapolated to other stellar types with larger flare energies . @xcite analyze the flare rates of late - type stars ( spectral type f - m ) and determine that , on average , a few 10@xmath25 erg flares occur per day . this flare energy is equivalent to that of the largest observed solar flares @xcite . using the flare energy - cme mass relation from @xcite , we expect a m dwarf to release several 10@xmath21 g cmes per day . recent observations of v374 peg showed flare rates between 0.15 and 0.42 flares per hour , with maximum total energies as large as 10@xmath26 ergs @xcite . to determine the plausible range of m dwarf cme speeds , we first consider the range of solar cme speeds . it is generally assumed that some fraction of an active region s free magnetic energy is converted into a cme s kinetic energy ( or the quiet sun magnetic energy in the case of filament eruptions ) . @xcite and @xcite find a strong correlation between cme speed and the active region magnetic energy . using the data of @xcite and making the approximation of a constant mass for all cmes , we find between 0.5% and 5% of the available magnetic energy tends to be converted into kinetic energy depending on the chosen cme mass . @xcite consider the available energy in a large active region ( diameter 5 arcmin , total volume @xmath27=10@xmath28 @xmath29 ) with average photospheric magnetic field strength 200 g. this yields an available energy of 1.6x10@xmath26 ergs . for a more average size ar ( 1000 mm@xmath30 @xmath31 0.8 arcmin diameter , @xcite ) we find an available energy of 2.5x10@xmath32 ergs . we determine a typical cme mass by assuming a half - torus of major and minor radius 0.2 and 0.05 @xmath8 filled with the coronal base density ( 8.35x10@xmath33 g @xmath34 , @xcite ) which yields a mass , @xmath5 , of 1.4x10@xmath35 g which is slightly larger , but similar to the average observed cme mass of 1.3x10@xmath35 g @xcite . assuming some fraction @xmath36 of the total magnetic energy becomes kinetic energy , we can determine the cme speed , @xmath37 . @xmath38 for @xmath36=0.05 and using the large active region volume , we find a maximum speed of approximately 3400 km s@xmath2 , and for the average active region volume , we find a speed of approximately 425 km s@xmath2 . these speeds are in good agreement with the maximum and average observed solar cme velocities . we now apply this estimation , shown to be reasonable for solar cmes , to mid - type m dwarf cmes . we assume the active region and cme retain the same size relative to their host star . the m dwarf volumes correspond to @xmath39 , or 3.9% , of their solar values due to the smaller radius of v374 peg ( @xmath1=0.34@xmath8 ) . our m dwarf density model has a coronal base density of 1.67x10@xmath40 g @xmath34 , which yields a cme mass of 1.66x10@xmath24 g , about an order of magnitude larger the average solar cme mass . observations of m dwarf surface magnetic field strength tend to lack sufficient resolution to resolve individual active regions . the average active region b of 200 g used in the previous calculation is approximately 20 times the average solar quiet sun photospheric value , so we scale the observed m dwarf magnetic field of 1 kg to 20 kg in a m dwarf active region . using equation [ eq : vmax ] with these values yields an average speed of 4,013 km s@xmath2 for a 1.66x10@xmath24 g cme , and a maximum speed of 31,600 km s@xmath2 . to cover a reasonable range of the plausible m dwarf parameter space , we simulate cmes between 10@xmath20 g to 10@xmath23 g. the lower limit represents the least massive solar cmes relevant for space weather and the upper limit corresponds to the maximum stellar values inferred from observations . we consider m dwarf speeds from 300 km s @xmath2 up to 10,000 km s@xmath2 . for the more massive cmes we restrict the speed to the upper limit , @xmath41 , determined using eq . [ eq : vmax ] and the large m dwarf active region volume ( 3.9x10@xmath42 @xmath29 ) with an average b of 20 kg . @xmath43 for the solar - type simulations we use the observed solar cme ranges of masses between 10@xmath20 g to 10@xmath24 g and speeds between 300 km s @xmath2 and 1,500 km s@xmath2 . using forecat , we simulate the deflections of cmes out to 0.1 au ( @xmath4460 @xmath12 ) . all cmes begin at a longitude of 150@xmath45 , and we consider high , mid , and low latitude cmes , which respectively start at 70@xmath45 , 40@xmath45 , and the equator . the cmes are oriented parallel to the equator and we do not consider any effects of rotation in this work . for each initial position we initiate cmes with masses between 10@xmath20 g to 10@xmath23 g and speeds between 300 and 10,000 km s@xmath2 , or the maximum speed determined by the mass and equation [ eq : vmax ] . before looking at results for the full mass and speed parameter space , we consider a few individual cases to better understand the differences between solar and m dwarf cme deflection . we emphasize that in this work we are looking at a single mid - type m dwarf , and that the behavior may differ greatly for other types , which we discuss in section [ disc ] . figure [ fig : vels ] shows the trajectory of mid - latitude cmes with masses of 10@xmath20 g and speeds of 300 km s@xmath2 ( cyan ) , 1,000 km s@xmath2 ( white ) , and 5,000 km s@xmath2 ( purple ) . note that the cyan curve for the lowest mass case barely extends past the white curve . in each panel the color contours show the radial magnetic field and the line contours show the magnetic field strength at the source surface . the black contour line gives the approximate location of the current sheet , the minimum in the magnetic energy at farther distances . we refer to the stellar analogue of the heliospheric current sheet as the astrospheric current sheet . based on solar cme deflections , we expect stellar cmes to deflect to the astrospheric current sheet . stellar rotation causes the background to change with respect to a cmes position , even if the cme is not deflecting . cmes wtih different final propagation speeds will experience different amounts of stellar rotation so rather than shifting the background , we add the translation due to rotation into the cmes trajectory in figure [ fig : vels ] . a cme that does not deflect would appear as a line of constant latitude in figure [ fig : vels ] . cme ( cyan ) , a 1,000 km s@xmath2 cme ( white ) , and a 5,000 km s@xmath2 cme ( purple ) . a shift in longitude due to stellar rotation is incorporated into the cme trajectory.,width=480 ] the cmes initially behave the same , deflecting eastward and toward lower latitudes , as they have the same speed during the slow rise phase . in figure [ fig : vels ] the trajectories overlap until the acceleration phase begins . all three cmes move toward the astrospheric current sheet and upon reaching it they become `` trapped '' as they can not penetrate the potential barrier on the opposite side . this is evident in figure [ fig : vels ] as the combined rotation and deflection motion clearly traces the path of the astrospheric current sheet . for all cases , as stellar rotation changes the cme s stellar longitude changes , the latitude of the astrospheric current sheet changes , causing the cme to deflect latitudinally . the stellar magnetic field is strong enough to force low mass cmes to continue sliding along the acs for the duration of the propagation out to 0.1 au . faster cmes experience less change in longitude due to the decrease in propagation time . this behavior is more extreme than what we have previously seen for solar cme deflection @xcite . while solar cmes deflect toward the heliospheric current sheet , the solar magnetic forces are not strong enough to produce this trapping behavior , even for the less massive solar cmes . we also consider the effects of different cme masses and initial starting locations . figure [ fig : masspos ] shows results in the same format as figure [ fig : vels ] but for cmes with a speed of 1000 km s@xmath2 . analogous to figure [ fig : vels ] , the effects of rotation have been incorporated into the cme trajectory . the top , middle , and bottom panels correspond to the high , mid , and low initial latitudes , respectively , and the white trajectories correspond to a 10@xmath20 g cme and the purple trajectories to a 10@xmath22 g cme . . the effects of stellar rotation have been incorporated into the cme trajectory . the top , middle , and bottom panels correspond to high , mid , and low initial cme latitudes . the white lines correspond to 10@xmath20 g cmes and the purple lines to 10@xmath22 g cmes.,height=672 ] the white line in the middle panel of figure [ fig : masspos ] corresponds to the same case as the white line in figure [ fig : vels ] . the top and bottom panels show that the low mass cmes behave similarly , regardless of their initial location . although the cmes deflect to the astrospherical current sheet , the actual path taken will vary . upon reaching the astrospherical current sheet the cmes remain trapped , experiencing very little longitudinal motion in the inertial frame . the effect of stellar rotation on the position of the astrospherical current sheet then forces the cme to change latitude . as seen in @xcite the deflection decreases with cme mass , however , significant deflections ( @xmath4620@xmath45 ) can still occur for 10@xmath22 g cmes . for these high mass cmes , the deflection tends to cease by 5 @xmath1 . in the cme trajectories in figure [ fig : masspos ] this corresponds to a horizontal line toward the east . when the astrospheric current sheet is inclined with respect to the stellar equator , the motion due to rotation can move the cme away from the magnetic minimum . unlike for the low mass cmes , the magnetic forces are not strong enough to cause the high mass cmes to slide along the astrospheric current sheet . and 0.1 au ( 60 @xmath12 ) , respectively . each circle represents a cme and the size and color indicate the cme mass and velocity . the color and line contours represent the background magnetic field as in figure [ fig : vels ] . the effects of stellar rotation have been incorporated into the cme positions.,height=576 ] figure [ fig : mvv ] shows the results of 115 simulations for each initial latitude spanning a range of cme masses and final propagation speeds . the color and line contours represent the magnetic background as in figures [ fig : vels ] and [ fig : masspos ] . each circle represents a single cme , the size of the circle indicates the cme mass with more massive cmes being larger . the color of each circle represents the cme s final speed . the left panels show the position of each cme at 2 @xmath12 and the right panels at the habitable zone distance , 0.1 au . as in figures [ fig : vels ] and [ fig : masspos ] we have incorporated a change in longitude due to stellar rotation to better show the position of the cmes with respect to the appropriate portion of the astrospheric current sheet . the most massive cmes , which have the smallest total deflections , show a negligible deflection between 2 @xmath12 and 0.1 au , their motion in this range is predominantly due to the stellar rotation . accordingly , we only include cme masses as large as 10@xmath21 g in the right panels of figure [ fig : mvv ] . the top , middle , and bottom panels show results for the low , mid , and high latitude cases , respectively . the left panels of figure [ fig : mvv ] show a cluster of cmes around the initial location corresponding to the most massive cmes , which deflect the least . for solar cmes , the deflection is typically determined below 2 @xmath8 . for the less massive cmes , we find that while much deflection occurs below 2 @xmath12 , a significant amount occurs at farther distances . the lower mass cmes are more tightly clustered around the astrospherical current sheet at 0.1 au than 2 @xmath12 for all initial latitudes . for each initial latitude , we see that the cmes with the same final speed tend to reach the same final longitude at 0.1 au so their final longitude is determined by the propagation time . forecat results show that the strong magnetic fields of v374 peg can cause significant deflections , which cause the majority of cmes to move closer to the astrospheric current sheet . to understand the effect on exoplanet impact , we need to know the average distance of a m dwarf cme from the astrospheric current sheet , which will depend on the cme mass . for each initial latitude we determine the average distance from the astrospheric current sheet ( including the effects of stellar rotation ) as a function of cme mass . figure [ fig : csdist ] shows this quantity for the high ( green ) , mid ( red ) , and low ( blue ) initial latitudes . the error bars correspond to one half of the standard deviation . small masses have small error bars because all the cmes deflect close to the equator and show little scatter . the error bars decrease for large masses because little to no deflection occurs for these cmes . below approximately 5x10@xmath21 g the distance from the astrospheric current sheet decreases with mass for all cases . above this mass the trend is less clear and the distance depends more strongly on the initial distance ( marked with a dashed line for each case ) . we fit a quadratic polynomial to all three sets of data below 10@xmath21 g to get a relation between cme mass , @xmath47 , in @xmath48 , and distance from the astrospheric current sheet , @xmath49 , in @xmath45 , at 0.1 au . @xmath50 the solid black line in figure [ fig : csdist ] shows this fit . above 10@xmath21 g we assume the cme deflection is negligible and cme remains at its initial distance from the astrospheric current sheet . the averages in figure [ fig : csdist ] tend to be close to the initial distances for each mass . as figure [ fig : masspos ] shows , these massive cmes do initially deflect toward the astrospheric current sheet but after the deflection ceases the stellar rotation can cause an increase in the distance from the astrospheric current sheet . we repeat the analysis of the previous section for hot jupiters orbiting solar - type stars at a distance of 10 @xmath8 ( 0.05 au ) , a distance representative of typical hot jupiter orbits . we simulate cmes with masses between 10@xmath51 g and 10@xmath24 g and speeds between 300 and 1500 km s@xmath2 , the same range seen for solar cmes @xcite . since we use a high resolution solar magnetogram we can resolve active regions and place the cmes at their polarity inversion lines . we consider one cme erupting from an active region ( initial latitude and longitude of -15.4@xmath45 and 17@xmath45 with a tilt of -72@xmath45 , hereafter active region cmes ) and one cme erupting from the quiet sun ( initial latitude and longitude of 37.2@xmath45 and 121.9@xmath45 with a tilt of -31.9@xmath45 . hereafter quiet sun cmes ) . for both initial positions we consider all three background densities : alf , ter1 , and ter2 . figure [ fig : hjdens ] shows simulations for a 300 km s@xmath2 , 10@xmath20 g cmes erupting from both initial locations for all three backgrounds ( alf in red , ter1 in white , and ter2 in blue ) . . , width=480 ] for both initial locations we see that the cmes deflect toward the astrospheric current sheet . in this case we see larger deflections from the quiet sun cmes than the active region cmes . the active region cmes initially deflect to the west until their longitudinal motion is halted and the cmes begin a small latitudinal deflection . the initial westward deflection results from imbalances in the local magnetic gradients in the active region . this motion continues until the cmes approach a region of enhanced magnetic field strength around 50@xmath45 longitude at 1.45@xmath8 . this enhancement slows their longitudinal motion . the ensuing latitudinal motion is a result of the global magnetic gradients determined by the location of the astrospheric current sheet . much weaker local magnetic gradients deflect the quiet sun cmes so their trajectory more closely resembles the direction of the global magnetic gradients . for each initial position the cr2029 alf and cr 2029 ter2 density models produce similar results , but the cr 2029 ter1 model yields a significant difference . as the mass loss rate of the background stellar wind decreases the total amount of deflection increases due to the decrease in the drag , which results from the decrease in the stellar wind density . note that the @xcite density model , used previously for solar forecat simulations , produces results nearly identical to the alf model . hereafter we only consider the cr2029 alf and cr 2029 ter1 backgrounds and assume the cr2029 ter2 results do not differ from the cr2029 alf results . figure [ fig : hjmvv ] shows the deflected positions at 10 @xmath8 for cmes sampling a range of solar cme mass and velocities for the solar - type star , analogous to figure [ fig : mvv ] . we include results for both initial positions using the cr2029 alf and cr 2029 ter1 density models . for the active region cmes ( top panels ) we find that the smallest masses behave similar to the cases in figure [ fig : hjdens ] - they deflect westward until the region of enhanced magnetic field slows their longitudinal motion and the global gradients create a latitudinal motion . the more massive cmes deflect slower causing them to reach the enhanced magnetic field at farther distances where the enhancement has weakened and is unable to halt the longitudinal motion . as we have seen for the m dwarf ( and solar case in @xcite ) the deflection brings the cmes closer to the astrospheric current sheet and the amount increases with decreasing cme mass and speed . this effect is more visible in the quiet sun cmes ( bottom panels ) where the local gradients are relatively weak as compared to near the active region . the behavior of these cme is the same as the solar cme we have previously studied @xcite . since the majority of the deflection occurs below 10 @xmath8 @xcite , we do not expect significant difference between the deflected cme positions at hot jupiter and solar habitable zone distances . . the left and right panels correspond respectively to results using the cr2029 alf and cr 2029 ter1 background density . the top and bottom panels show results for two different initial locations . , width=624 ] to quantify the amount of deflection we determine the average distance from the astrospheric current sheet after the deflection as a function of cme mass , similar to figure [ fig : csdist ] . the results for the cr2029 alf density model are shown in the top panel of figure [ fig : hjcsdist ] . the blue line corresponds to the active region cmes and the red line corresponds to the quiet sun cmes . the dashed lines show the initial distance of the cmes . as expected the distance from the current sheet decreases as the cme mass decreases , however we do not see a unique relation between mass and distance as seen for the m dwarf . the m dwarf has much stronger forces that push the lowest mass cmes entirely to the astrospheric current sheet , the weaker solar forces can only cause motion toward the astrospheric current sheet . shows the distance from the astrospheric current sheet , binned by mass for the two different initial locations . the bottom panel shows deflection toward the astrospheric current sheet , equivalent to difference between the final and initial distances . the black line corresponds to the polynomial - best - fit to both cases.,height=576 ] instead , for the solar case , it is more instructive to look at the deflection toward the astrospheric current sheet as a function of cme mass . we define the deflection toward the astrospheric current sheet as the difference between the final distance , @xmath52 , and initial distance from the astrospheric current sheet , @xmath53 . the bottom panel of figure [ fig : hjcsdist ] shows the deflection toward the astrospheric current sheet for both initial positions . this quantity is better fit by a single function than the distance - the black line shows the polynomial best fit for masses below 5x10@xmath35 g. @xmath54 we do not find a significant difference in the best fit polynomial between the cr2029 alf and cr2029 ter1 results . from the forecat cme deflections , we expect all cmes to deflected toward the astrospherical current sheet , but the magnitude of the deflection varies . the less massive cmes to deflect to and reach the astrospheric current sheet for the m dwarf . more massive m dwarf cmes and all stellar - type cmes deflect toward the astrospheric current sheet , but do not necessarily reach it . accordingly cme impacts should occur less frequently for planetary orbits inclined with respect to the astrospheric current sheet . here we combine the results from forecat with scaling laws from the literature to estimate the frequency of cme impacts as a function of a planet s orbital inclination , @xmath55 , for both habitable zone m dwarf exoplanets and hot jupiters orbiting a solar - type star . in this work we refer to inclination from the plane of the astrospheric current sheet , rather than with respect to the stellar equator , or relative to the plane of the sky . alternatively , this is equivalent to assuming the astrospheric current sheet lies in the equatorial plane . @xcite use geometrical arguments to estimate the frequency of cme impacts for a planet with an equatorial orbit , assuming cmes of angular width @xmath56 are isotropically released between latitudes @xmath57 . @xcite include an additional term , @xmath58 , representing the planet s angular width but find it has a negligible effect so we do not include it . the probability of impact , @xmath59 , is calculated as @xmath60 which is the product of a longitudinal and a latitudinal probability of the cme impacting the planet . for orbital inclinations less than or equal to @xmath61 , the entirety of the orbit is contained in @xmath57 , and there is no modification to the probability . for inclinations greater than @xmath61 , we multiply eq . [ eq : kho07 ] by the fraction of the orbit between @xmath57 . this results in a probability of impact as a function of orbital inclination and cme latitude range . @xmath62 we expect cmes to erupt over a wide range of latitudes but the deflections will effectively collimate the cmes into a smaller range of latitudes about the current sheet . figure [ fig : latprob](a ) shows equation [ fig : latprob ] for different values of @xmath55 and @xmath61 . @xmath56 is set to 60@xmath45 as in @xcite . ) for different values of the cme latitude range , @xmath61 . panels ( b ) and ( c ) show the change in the probability when the effects of deflections are included for a habitable zone m dwarf planet and a solar hot jupiter.,height=768 ] fig . [ fig : latprob](a ) shows that when @xmath63 the probability is uniform as the orbit is fully contained within @xmath57 . if @xmath64 the probability decreases as @xmath55 increases . as @xmath61 decreases , @xmath65 increases and @xmath66 decreases . we use equations [ eq : deltacs ] and [ eq : deltacshj ] to incorporate the effects of deflection and determine the actual probability of impact . every cme mass has a different probability of planetary impact and the net probability is the sum of the individual probabilities weighted by the probability , @xmath67 , of each mass . we use discrete mass bins and approximate the post - deflection cme range @xmath61 as equivalent to the post - deflection distance from the astrospheric current sheet @xmath68 . for the m dwarf , a specific value of @xmath52 is determined from equation [ eq : deltacs ] . @xmath69 for the hot jupiter we subtract the amount given by equation [ eq : deltacshj ] from the initial cme distance from the astrospheric current sheet . a minimum distance of 0.5@xmath45 , the smallest value from the m dwarf values , is assumed for the hot jupiter results . equation [ eq : ptot ] requires the probability of a cme having a specific mass . using the observed m dwarf flare rates , the relationship between flare energy and cme mass , and the distribution of cme masses , and the probability of impact versus inclination , we can estimate the distribution of m dwarf cme masses . for the solar - type cmes we assume the solar distribution from @xcite . as discussed in section [ dmparams ] , @xcite estimate approximately five 10@xmath25 erg flares per day which corresponds to five 10@xmath21 g cmes per day using the flare - cme mass relation from @xcite . @xcite determine the probability @xmath70 of a cme having mass , @xmath5 . @xmath71 using lasco observations of 7668 solar cmes , @xcite determine an average mass , @xmath72 , of 1.55x10@xmath35 g , and a standard deviation , @xmath73 , of 1.114 . in section [ dmparams ] , we estimate an average m dwarf cme mass of 1.66x10@xmath24 g which we set as our new @xmath74 , causing a shift in eq . [ eq : pm ] toward higher masses . using this new distribution , we determine a probability of 0.097 for a cme with mass 10@xmath21 g. if there are five 10@xmath21 g cmes then we expect a total number of cmes per day , @xmath75 , of 51 . using similar scaling relations and observations of v374 peg flare rates , @xcite estimate 15 - 60 cmes per day with mass greater than 10@xmath24 g. these rates are about an order of magnitude higher than solar values during solar maximum , but this does not seem completely unreasonable given the increased activity of m dwarfs . panel ( b ) of figure [ fig : latprob ] shows the values of equation [ eq : ptot ] versus the initial cme latitude ranges and orbital inclination for the m dwarf . compared to the probability without deflection ( figure [ fig : latprob](a ) ) , the deflection causes the probability to increase for small inclinations and decrease for large inclinations . for large initial cme latitude ranges , the probability roughly doubles for low inclinations , and decreases by a factor of 5 for high inclinations . while subtle variations exist different initial cme latitude ranges , all have a probability of impact near 10% for low inclinations and 1% for high inclinations . this corresponds to approximately 5 cme impacts per day for low inclinations and one impact every two days for high inclinations . for comparison , roughly 10% of solar cmes are halo cmes , half of which will propagate toward earth opposed to away from earth @xcite . an average of 5 cmes occur per day during solar maximum , which corresponds to an impact frequency of 0.25 cmes per day . planets in an equatorial orbit of an habitable mid - type m dwarf may be impacted by cmes 20 times more frequently than the average solar maximum rates at earth . figure [ fig : latprob](c ) shows the probabilities for hot jupiters orbiting a solar - type star when the effects of cme deflections are considered . comparison with panel ( a ) shows that deflections have less of an effect for the hot jupiters than the mid - type m dwarf planets . the probability increases slightly for low orbital inclinations and decreases slightly for high orbital inclinations . assuming solar - like cme rates , we expect between 0.05 and 0.5 cme impacts per day . cmes are certainly not isotropically released from the sun , but they tend to occur below 50@xmath45 ( polar crown filaments being a notable exception ) . for an equatorial orbit and an initial latitude range of 50@xmath45 , and assuming no changes between 10 @xmath8 and 1 au ( which we expect to be true ) , we would determine an expected cme impact rate of 0.35 cmes per day during solar maximum . this agrees well with the number estimated from the fraction of halo cmes . we have not accounted for the short orbital period of hot jupiters , which can be as short as 0.4 days @xcite . the above analysis gives a probability of impact that does not depend on the specific longitude of planet - all longitudes have equal probability . a rapidly orbiting planet , however , will cover many different longitudes in the time it takes the cme to traverse the planet s orbital distance , increasing the chance of cme impact . so far we have shown that cme deflection can increase the number of cme impacts for both mid - type m dwarf exoplanets , and to a lesser extent , hot jupiters . in this section we determine what effect these impacts may have on a planetary magnetosphere . @xcite use a thermal balance model to determine the atmospheric profile of an exoplanet , which they then combine with a model for atmospheric pick - up ion loss driven by the external plasma conditions when a cme impacts the exoplanet . they find significant atmospheric losses for magnetospheres that extend less than one planetary radii above the planet s surface . here , using a simple analytic model , we determine the minimum planetary magnetic field , @xmath76 , required to maintain a magnetosphere of radius @xmath77 greater than 2 @xmath78 , and we compare these values with the expected magnetic field strengths of rocky and gaseous planets . as in many previous works ( @xcite , @xcite , @xcite , and references within ) , we balance pressure between the magnetosphere and the stellar wind or cme . we assume the magnetospheric pressure is dominated by magnetic pressure and the external stellar pressure comes from both magnetic and ram pressure . this yields @xmath79 ^ 2}{8 \pi } = \frac{[b_{sw / cme}]^2}{8 \pi } + \frac{1}{2 } \rho_{sw / cme } \ ; v^2_{sw / cme}\ ] ] where @xmath80 is the magnetospheric field strength at a radial distance @xmath77 away from the planet s center , and @xmath81 , @xmath82 , and @xmath83 are the magnetic field strength , density , and speed of the stellar wind or cme upon reaching the planet . note that the magnetic field strengths in equation [ eq : pbal ] corresponds only to the component of the magnetic field vectors tangential to the magnetopause . for simplicity we consider the equator and assume the magnetopause normal parallels the radial direction . as in @xcite , we model the planetary magnetosphere as a dipole : @xmath84 where @xmath85 is the magnetic field strength on the surface at the equator . at the equator @xmath86 has no radial component . for the stellar wind or cme magnetic field strength we consider only the tangential components of the magnetic field ( toroidal and poloidal ) , which we describe below . we first consider the m dwarf case . our v374 alf stellar wind model yields a density of 1.54x10@xmath87 g @xmath34 and a speed of 200 km s@xmath2 at the habitable zone distance of 0.1 au . the v374 ter model yields slightly different values with a density of 3.83x10@xmath88 g @xmath34 and a speed of 500 km s@xmath2 at the same distance . the v374 alf and v374 ter models produce ram pressures of 3.0x10@xmath89 dyn @xmath90 and 4.9x10@xmath89 dyn @xmath90 , respectively . above the source surface , where only open magnetic field exists , the stellar magnetic field , @xmath91 , as a function of radial distance , @xmath19 , can be determined from the parker interplanetary magnetic field @xcite @xmath92 where @xmath93 and @xmath94 are the radial and toroidal directions . @xmath95 is the colatitude , @xmath96 is the magnetic field at the source surface distance , @xmath97 , and @xmath98 is the stellar rotation rate . we use the magnitude of the @xmath94 component of @xmath99 in equation [ eq : pbal ] . previous pressure balance studies , such as @xcite , have only considered the radial component of the interplanetary magnetic field , which , in its unperturbed state , is typically not tangential to the magnetopause . it can , however , be argued that the draping of the interplanetary magnetic field about the magnetosphere will convert the radial component into a tangential component . previous studies have also typically neglected the toroidal component of the interplanetary magnetic field , which can become important in the case of fast rotators . additionally , any presence of a bow shock will affect the interplanetary magnetic field strength seen at the magnetopause . here we consider only the unperturbed tangential magnetic field and determine the minimum magnetic field stregth needed to prevent atmospheric losses due to the stellar wind . because @xmath91 depends on the stellar wind speed , we obtain two different values for the two different background models : 1.9 g for v374 alf , and 0.74 g for v374 ter , which correspond to magnetic pressures of 0.144 dyn @xmath90 and 0.0218 dyn @xmath90 . we determine @xmath76 by setting @xmath100 = 2 @xmath78 . we find minimum magnetic field intensities of 15 g and 6.6 g for v374 alf and v374 ter , both roughly an order of magnitude larger than the earth s magnetic field of 1 g @xcite and twice that of jupiter @xcite . instead of stellar wind impact we now consider the effects of the ram pressure and magnetic pressure of the cme itself . cme - driven shocks , which we do not consider , should compress the upstream stellar wind , enhancing the external pressure upon the magnetosphere , and increasing the minimum planetary magnetic field . we estimate the cmes magnetic field as it is not explicitly included in forecat . we adopt initial cme magnetic field strengths of either 1 kg or 20 kg , corresponding to the stellar quiet region or active region value . ignoring any reconnection , magnetic flux is conserved , so the magnetic field strength will decrease proportional to the increase in the cross - sectional area perpendicular to the magnetic field direction . it can be shown for a self - similarly expanding toroidal cme and both poloidal or toroidal magnetic field that the cross - sectional area increases as @xmath101 . for cmes with 1 kg or 20 kg initial magnetic fields , we expect a magnetic pressure of 0.0088 dyn @xmath90 or 3.5 dyn @xmath90 at 20 @xmath8 . assuming self - similar expansion , our toroidal cmes will have a volume proportional to @xmath102 . given the initial cme mass , which we approximate as constant , and volume , which we know from the cme shape , we can determine the density at any distance . the ram pressure will vary greatly over the range of cme masses and velocities we consider , between 4.6x10@xmath103 dyn @xmath90 and 5100 dyn @xmath90 . for comparison , the average solar wind pressure at 1 au is of order a few 10@xmath104 dyn @xmath90 , and observations of earth - impacting cmes reach dynamic pressures as high as 2x10@xmath103 dyn @xmath90 ( e. g. @xcite and @xcite ) . the 08 march 15 cme had a dynamic pressure of 1.5x10@xmath103 dyn @xmath90 , and the 2003 halloween cme had a dynamic pressure of 7x10@xmath104 dyn @xmath90 at mars and magnetic pressures of 1.6x10@xmath105 dyn @xmath90 and 4.3x10@xmath105 dyn @xmath90 @xcite . this implies that mid - type m dwarf exoplanets will need magnetic fields much stronger than found in our own solar system just to withstand the combined magnetic and ram pressure of the cmes . figure [ fig : planetb ] shows @xmath76 for different cme masses and velocities . the top row corresponds to mid - type m dwarf cases , and the left and right panels correspond to cmes with initial magnetic field strengths of 1 kg and 20 kg . note the logarithmic scale color bar . the faster , more massive cmes require stronger planetary magnetic fields . for these cmes the ram pressure greatly exceeds the magnetic pressure . below cme masses of @xmath4410@xmath35 g and @xmath4410@xmath21 g the magnetic pressure exceeds the ram pressure for initial magnetic field strengths of 1 kg and 20 kg . for cmes with an initial magnetic field strength of 1 kg , we find that a planetary magnetic field of a few tens of gauss can withstand a significant fraction of these cmes . @xcite use the scaling law of @xcite to estimate the dipolar moment of exoplanets . for fast rotators , @xcite predict dipole moments up to 80 times larger than earths , and determine dipole moment up to 18 times earth s for specific observed exoplanets . it seems unlikely , however , for a rocky body to generate a magnetic field of order hundreds to thousands of gauss , so the fastest , most massive cmes are capable of compressing the planetary magnetosphere below 2 @xmath78 . these cmes , however , are the ones that are the least likely to impact a planet . deflection forces can not significantly influence their motion so there is no enhancement in the impact probability . figure [ fig : planetb](b ) shows that when the initial magnetic field strength is increased to 20 kg only planets with magnetic field strengths larger than 100 g will be shielded from the majority of cmes . this includes the slowest , least massive cmes , which do have an enhanced probability of impacting the planet . we conclude that knowledge of the magnetic field of m dwarf cmes is critical for understanding whether a planet s atmosphere could possibly withstand continual impacts for any extended period of time . we repeat the analysis in the previous section for a hot jupiter orbiting a solar - type star at 10 @xmath8 . as compared to the m dwarf exoplanet , the hot jupiter experiences weaker ram and magnetic pressure from the solar - like wind . for the cr2029 alf background , we find a magnetic pressure of 1.53x10@xmath106 dyn @xmath90 , and a ram pressure of 1.87x10@xmath103 , which corresponds to a minimum planetary magnetic field of 0.055 g. the cr2029 ter2 background does not differ significantly from the cr2029 alf background . for the less dense cr2029 ter background we determine a magnetic pressure of 1.03x10@xmath106 dyn @xmath90 , and a ram pressure of 1.22x10@xmath103 dyn @xmath90 , which corresponds to a magnetic field strength of 0.044 g. for all backgrounds we find a minimum planetary magnetic field strength significantly less than found for the m dwarf exoplanet , and less than that of the gaseous planets within our own solar system . we compare the minimum planetary magnetic field strength needed to withstand the stellar wind with the magnitude needed to withstand the ram and magnetic pressure experienced during a cme impact . again we determine values for the range of cme masses and speeds considered in the analysis of cme deflections . this range corresponds to ram pressures between 5.88x10@xmath103 dyn @xmath90 and 0.14 dyn @xmath90 . we consider initial cme magnetic field strengths of 10 g and 200 g , which corresponds to the magnitude of the solar quiet sun and active regions in the photosphere . this yields magnetic pressures of 0.001 dyn @xmath90 and 0.45 dyn @xmath90 . figure [ fig : planetb](c ) and ( d ) show contours of the minimum planetary magnetic field for the exoplanet cases . for the lower magnetic field cme ( figure [ fig : planetb](c ) ) , we find that the ram pressure dominates for most of the cmes , leading to significant variation across parameter space . the least massive , slowest cmes require a planetary magnetic field of 1.32 g , and this increases to 15.5 for the most massive , fastest cmes . for the stronger magnetic field cme , the magnetic pressure dominates leading to a larger minimum planetary magnetic field , and less variation with cme parameters . in this case the minimum planetary magnetic field varies between 26.9 g and 30.9 g. while these magnetic field strengths may slightly exceed those within our own solar system , they may not be unreasonable for hot jupiters . @xcite extends a scaling model derived in @xcite , which relates the energy flux available to drive a dynamo with the resulting magnetic field strength . this model reproduces the magnetic field strength of the earth and jupiter , as well as some low - mass stars , and predicts a magnetic field strength 10 times that of jupiter for a planet 10 times as massive as jupiter . this suggests that jupiter - like planets could have a magnetic field of order 50 g. recent work by @xcite infers a 28 g planetary magnetic field based on the standoff distance of the planetary bow shock . this suggests that hot jupiters may have magnetospheres that can shield them from the effects of cme impacts . while cme deflection will cause a small increase in the frequency of cme impact for a hot jupiter orbiting a solar - type star , and the hot jupiter may need a slightly stronger magnetic field , the system is not terribly different from our own solar system . in comparison , cmes from a mid - type m dwarf present an extreme version of solar cmes . for a mid - type m dwarf habitable zone planet to be truly habitable requires planetary magnetic field strengths orders of magnitude greater than found in our own solar system . in this work , however , we have only considered one m dwarf of spectral type m4 , and at only one point in its main - sequence lifetime . activity tends to increase from early- to late - type m dwarfs @xcite , until above type m9 when activity decreases @xcite . late - type m dwarfs tend to have the longest activity lifetimes @xcite . @xcite show that young , rapidly - rotation , early - type m dwarfs can have kg magnetic fields , and these magnetic field strengths are common among mid- and late - type m dwarfs over a wider range of ages . we expect cmes to accumulate around the astrospheric current sheet for young early - type m dwarfs and most mid- and late - type m dwarfs . depending on a star s activity lifetime , there may be some time after which the stellar magnetic field and/or activity weakens , constant cme bombardment ceases , and an previously inhabitable exoplanet can start developing an atmosphere . the activity of m dwarfs tends to correlate with stellar rotation for early - type m dwarfs ( e.g. @xcite ; @xcite ; and @xcite ) . @xcite develop a model for the spin - down time of a low - mass star due to angular momentum loss via the stellar wind . we suggest that constant , massive cmes , could also have an effect on the angular momentum loss of a star , which in turn would affect the activity levels . the analysis in section 7 suggests that as many as 50 cmes per day may occur for an active m dwarf , which agrees with a recent estimation for v374 peg by @xcite . while this order - of - magnitude estimate is extreme by solar standards , we still expect that less frequent cmes could affect the angular momentum loss . better understanding will only occur with improved observations of inferred cmes outside our own solar system . estimating cme s contribution also requires knowledge of the alfvn radius . we can make an estimate from our simple static backgrounds , but we expect dynamic effects may strongly affect this value . if the cme legs remain attached to the surface of a star , they will begin to wrap around it as the star rotates as much as several times a day . barring any reconnection , this could significantly enhance the magnetic field strength close to the star as it crease a large toroidal magnetic field . this would increase the alfvn radius , allowing for more efficient magnetic breaking of the stellar rotation . similarly one might wonder the effect of the large numbers of cmes on the stellar mass loss rate . using the m dwarf cme mass distribution used in this work , and assuming 50 cmes a day yields a mass loss rate of 1.5x10@xmath40 m@xmath107 yr@xmath2 . m dwarfs main sequence lifetimes exceed the solar main - sequence lifetime of 10 billion years , but we find that an m dwarfs mass would change by 0.0015 m@xmath107 yr@xmath2 over a solar lifetime . while this is not a insignificant percentage of the total m dwarf mass , we do not expect mass loss from cmes to cause an m dwarf to slowly change spectral type . however , if the actual mass - loss rate happens to be much higher than our calculated value , it could start to become a concern . finally , throughout this work we have focused on how cme impacts may adversely affect an exoplanet . in contrast , frequent cme impacts benefit astronomers searching for signatures of aurora on exoplanets . radio detections of extrasolar could identify previously undiscovered exoplanets , as well as information about exoplanetary magnetospheres . recently @xcite found signatures of aurora on a planet orbiting an m8.5 dwarf . continued observations with existing , and new facilities such as the owens valley long wavelength array ( ov - lwa ) , should find more signatures of planetary magnetospheres . we suggest the best candidates for extrasolar auroral detections are those where cme impacts occur the most frequently . exoplanets orbiting m dwarfs should experience the most frequent cme impacts , and we expect these cmes to have a stronger magnetospheric impact than solar cmes . we have shown that hot jupiters require a significant , yet not implausible , planetary magnetic field , which , combined with their larger size , would lead to aurora brighter than those in our own solar system . we have adapted forecat , a solar cme deflection model , to simulate cme deflections for the m4 dwarf v374 peg . m dwarf cme parameters are highly uncertain so we consider cme masses between 10@xmath20 g and 10@xmath23 g and final propagation speeds between 300 km s@xmath2 and 10,000 km s@xmath2 . the deflection depends strongly on the cme mass , and , to a much weaker extent , the cme speed . the least massive cmes quickly deflect to the astrospheric current sheet and remain trapped there . these extreme deflections exceed those seen in our own solar system . we also apply forecat to the deflections of cmes for solar - type stars at hot jupiter distances . we find the same dependence on mass and speed as seen for the m dwarf but the magnitude of the deflections is significantly smaller . while the cmes continue to deflect toward the astrospheric current sheet , only the slowest , least massive cmes actually reach it by 10 @xmath8 , analogous to what we see in our own solar system . we determine the average distance between the cme s deflected position and the astrospheric current sheet as a function of cme mass . for the mid - type m dwarf cmes we find that the distance does not depend on the initial position for cme masses below 10@xmath21 g. for the solar - like cmes the distance does depend on the initial position so we determine the amount of deflection toward the astrospheric current sheet as a function of cme mass . we combine these relations between the deflection and the cme mass with geometrical arguments and an estimation of the m dwarf cme mass distribution to determine the probability and frequency of cme impacts . for both habitable zone mid - type m dwarf exoplanets and hot jupiters the probability of impact decreases if the exoplanet s orbit is inclined with respect to the astrospheric current sheet . the sensitivity to the inclination is much greater for the mid - type m dwarf exoplanets due to the extreme deflections to the astrospheric current sheet . for low inclinations we find a probability of 10% whereas the probability decreases to 1% for high inclinations . from our estimation of 50 cmes per day , we expect habitable mid - type m dwarf exoplanets to be impacted 0.5 to 5 times per day , 2 to 20 times the average at earth during solar maximum . the frequency of cme impacts may have significant implications for exoplanet habitability if the impacts compress the planetary magnetosphere leading to atmospheric erosion . for the hot jupiters , the impact probability has a similar range but more inclinations have moderate values , near a few percent , as opposed to a clear division between the two extremes as seen for the mid - type m dwarf . we quantify the effect of these cmes on an exoplanetary magnetosphere by determining the minimum planetary magnetic field need to maintain a magnetosphere twice the size of the planetary body . this size corresponds to the minimum necessary to retain an atmosphere according to @xcite . we find the mid - type m dwarfs exoplanets require planetary magnetic fields between tens to hundreds of gauss , whereas hot jupiters only require magnitudes between a few and 30 g. the magnetic field needs to shield a planet from cme impacts greatly exceeds that required for shielding from the stellar wind . we expect that rocky exoplanets can not generate sufficient magnetic field to shield their atmosphere from mid - type m dwarf cmes , but hot jupiters around solar - type stars could likely remain shielded . we expect that the minimum magnetic field strength will change with m dwarf spectral type as the amount of stellar activity and stellar magnetic field strength change , and that early - type m dwarfs would be more likely to retain an atmosphere than mid or late - type m dwarfs .
for standard leptogenesis from the decay of singlet right - handed neutrinos , we derive source terms for the lepton asymmetry that are present in a finite density background but absent in the vacuum . these arise from cuts through the vertex correction to the decay asymmetry , where in the loop either the higgs boson and the right - handed neutrino or the left - handed lepton and the right - handed neutrino are simultaneously on shell . we evaluate the source terms numerically and use them to calculate the lepton asymmetry for illustrative points in parameter space , where we consider only two right - handed neutrinos for simplicity . compared to calculations where only the standard cut through the propagators of left - handed lepton and higgs boson is included , sizable corrections arise when the masses of the right - handed neutrinos are of the same order , but the new sources are found to be most relevant when the decaying right - handed neutrino is heavier than the one in the loop . in that situation , they can yield the dominant contribution to the lepton asymmetry . ttk-10 - 50 + * leptogenesis : the other cuts * bjrn garbrecht + 0.3 cm _ institut fr theoretische teilchenphysik und kosmologie , + rwth aachen university , + d52056 aachen , germany _ +
1011.3122
leptogenesis @xcite is often studied in parametric regimes where the masses of the right - handed neutrinos are either hierarchical or degenerate . the hierarchical limit is particularly useful for gaining valuable analytical insights into the connections between leptogenesis and the observed neutrino oscillations @xcite . to be specific , we discuss here the simple case of two right - handed neutrinos @xmath0 with masses @xmath1 and yukawa couplings @xmath2 to the higgs and lepton doublets , _ i.e. _ a model as specified in ref . the key simplification in the hierarchical limit , @xmath3 , is that the evolution of the lepton asymmetry during leptogenesis as a function of @xmath4 depends up to a proportionality factor only on the ratio @xmath5 , which characterises the washout strength . here , @xmath6 is the temperature and @xmath7 is the planck mass . the remaining proportionality factor characterising the amount of @xmath8 violation is @xmath9m_1/m_2 $ ] , up to corrections of order @xmath10 , which one neglects in the hierarchical limit . on the other hand , mass - degenerate right - handed neutrinos lead to resonant leptogenesis @xcite . this corresponds to a phenomenologically attractive scenario , since the decay asymmetry is enhanced , which allows for lower temperatures at which leptogenesis takes place . thus , the production of unwanted relics , most notoriously of gravitinos within supersymmetric models @xcite , along with the lepton asymmetry asymmetry can be avoided . a lower energy scale might also give rise to new experimentally accessible signals connected with leptogenesis , see _ e.g. _ @xcite . however , the origin of the masses of the right - handed neutrinos is yet unknown and their masses may well be neither hierarchical nor degenerate . when the hierarchical limit @xmath3 no longer applies , it is well known that the decay asymmetry of @xmath11 is not simply proportional to @xmath12 with negligible corrections , as can be verified by inspection of the vertex and wave - function contributions to the decay asymmetry of @xmath11 @xcite . in the finite - temperature background , there are additional corrections due to new cuts . while in the vacuum background , the @xmath8-violating contribution from the vertex function arises exclusively from the cut through @xmath13 , where the internal lepton and higgs boson are on - shell , at finite temperature also the two other possible cuts through @xmath14 or @xmath15 in the vertex diagram contribute , _ cf . _ figure [ fig : doublediamond ] ( b ) . this is because in the finite - temperature background , the cut - particles do not need to correspond to stimulated ( suppressed ) emission processes for bosons ( fermions ) , but they can also correspond to absorption processes of particles from the plasma . the presence of these cut contributions has been mentioned in ref . @xcite . however , by now , neither analytical expressions for these terms have been derived nor have these been evaluated numerically in order to compute effective decay asymmetries or the resulting lepton asymmetry . these are the main goals of the present work . the new cut contributions are a finite density effect , and a powerful method of describing out - of - equilibrium field theory is given by the schwinger - keldysh closed - time - path ( ctp ) formalism @xcite . this approach has been applied to leptogenesis and has resulted in some recent activity which we build upon within the present work @xcite . main advantages of the ctp approach to leptogenesis over the conventional description by semi - classical boltzmann equations may be seen in the absence of the need of an explicit subtraction procedure for real intermediate states ( ris ) @xcite and in the systematic treatment of finite - density corrections . within the ctp approach , the vertex diagram in figure [ fig : doublediamond ] ( b ) appears as a subdiagram in the self - energy figure [ fig : doublediamond ] ( a ) , which is a contribution to the lepton self - energy . this self - energy in turn enters the collision term of the kadanoff - baym equation for the lepton , that can be reduced to a kinetic equation which describes the gain and the loss and therefore the time evolution of the lepton number density . in order to simplify the collision term to a manageable form , it is useful to substitute equilibrium propagators for @xmath16 and @xmath17 and to employ kubo - martin - schwinger ( kms ) relations . the present work relies strongly on ref . @xcite , where these simplification strategies are explained and justified in detail and where also many definitions and quantities that we use here are introduced . an additional simplifying assumption that we introduce here , but that will not hold true in general , is that also @xmath18 is maintained in equilibrium ( _ e.g. _ through interactions with an additional lepton flavour ) , such that @xmath11 is the only out - of - equilibrium particle . we leave a study of the situation where more than one of the right - handed neutrinos deviates from equilibrium to future work . provided the initial distribution function for @xmath11 is thermal , as we assume in the present work , the main contributions to the lepton asymmetry occur at times when the value of @xmath4 is in the range of around one up to a few , for an in - detail discussion , see ref . this is because at these temperatures , @xmath11 becomes non - relativistic and therefore deviate from equilibrium . as a consequence , we anticipate the new cuts to be important under the following conditions : * since the effect is due to the finite densities , @xmath19 should not be much larger than @xmath20 , because otherwise the new contributions are maxwell - suppressed , just as the equilibrium distribution functions for energies of order @xmath19 much larger than @xmath6 . * the new effects will be most pronounced in case leptogenesis occurs at comparably low values for @xmath21 , when it is around one . this is the case in the transitional regime between strong and weak washout . in such a situation , the finite density effects that include the contributions from the new cuts , take their largest relevance . * a loophole to these arguments are situations where @xmath22 . ( we denote within this work by @xmath11 the neutrino that deviates from equilibrium , whereas @xmath18 is assumed to be very close to equilibrium at the time relevant for leptogenesis . this definition differs from what is commonly used in the literature , where @xmath11 corresponds to the lightest right - handed neutrino , _ cf . _ refs @xcite . in this case , the finite density effects from the @xmath18 are also important when leptogenesis occurs at larger values of @xmath21 , since the distribution of @xmath18 can still be unsuppressed . the plan of this paper is as follows : in section [ sec : vertex ] , we give the expressions for the @xmath8-violating source terms that bias the lepton number and that are valid for finite ratios of @xmath12 . for the standard cut through @xmath23 , we quote the result from ref . @xcite , while for the new cuts through @xmath24 and @xmath25 , respectively , we derive new expressions . in section [ sec : examples ] , we present the first numerical results for the cut through @xmath23 at finite @xmath12 and finite density , as well as for the new cuts through @xmath24 and @xmath25 . we first define and evaluate expressions for the effective @xmath8 violation for several ratios of @xmath12 . then , we solve the kinetic equations for the lepton asymmetry for the same values of @xmath12 and specific illustrative choices of @xmath26 , @xmath27 . the results for the lepton asymmetry are compared with the effective @xmath8 violation as well as with what is stated above on the parametric regions where we anticipate the new cut contributions to be most relevant . within flavoured leptogenesis , a new cut in the wave - function of the lepton @xmath16 contributes to the asymmetry . a rough estimate of this effect is presented in section [ sec : flavour ] , and it is found to be negligibly small . we summarise and conclude in section [ sec : conclusions ] . within the present work , we extend the results of ref . @xcite . there , the kinetic evolution equation for the lepton asymmetry is expressed as @xmath28 where @xmath29 is the conformal time in the friedmann background , @xmath30 the washout term , @xmath31 the source term and @xmath32 ( @xmath33 ) the comoving ( anti-)lepton number density . in the radiation - dominated universe , the scale factor is given by @xmath34 , where @xmath35 is an arbitrary constant . the washout term @xmath30 is discussed in detail in ref . it encompasses the tree - level decay and inverse decay processes of the right - handed neutrino . the source term accounts for the @xmath8-violating loop effects . it decomposes as @xmath36\,,\end{aligned}\ ] ] where @xmath37 is the wave - function contribution to the collision term , as it is given in ref . @xcite . in this work , we are primarily concerned with the vertex contribution @xmath38 and the new corrections that it acquires when compared with ref . this term can be expressed in the usual kadanoff - baym form @xmath39\,,\end{aligned}\ ] ] for @xmath40 . the wightman self - energy @xmath41 has the diagrammatic representation of figure [ fig : doublediamond ] ( a ) . it is given by @xcite @xmath42^tc^\dagger { \rm i}s^t_{nj}(-q ) { \rm i}\delta_\phi^<(-p - k ) { \rm i}\delta_\phi^t(-q - k ) \\\notag -&{\rm i}s_{ni}^{\bar t}(-p ) c\left[{\rm i}s_\ell^<(p+k+q)\right]^tc^\dagger { \rm i}s^t_{nj}(-q ) { \rm i}\delta_\phi^<(-p - k ) { \rm i}\delta_\phi^<(-q - k ) \\\notag -&{\rm i}s_{ni}^>(-p ) c\left[{\rm i}s_\ell^>(p+k+q)\right]^tc^\dagger { \rm i}s^>_{nj}(-q ) { \rm i}\delta_\phi^{\bar t}(-p - k ) { \rm i}\delta_\phi^t(-q - k ) \\\notag + & { \rm i}s_{ni}^{\bar t}(-p ) c\left[{\rm i}s_\ell^{\bar t}(p+k+q)\right]^tc^\dagger { \rm i}s^>_{nj}(-q ) { \rm i}\delta_\phi^{\bar t}(-p - k ) { \rm i}\delta_\phi^<(-q - k ) \big\}\,,\end{aligned}\ ] ] where we sum over the indices @xmath43 . the self - energy @xmath44 follows when applying the replacements @xmath45 and @xmath46 . for the propagators @xmath47 , @xmath48 and @xmath49 , we use the zero - width approximations as written down in ref . @xcite . to be specific , let us now consider the term with @xmath50 and @xmath51 . the @xmath8-violating contributions from the decays and inverse decays of @xmath11 arise when two out of the three propagators @xmath52 , @xmath53 and @xmath54 are on shell . these are the cut propagators . as a consequence of this , the third propagator is off shell . since the @xmath55 and @xmath56 propagators are purely on shell , it follows that only terms where the off - shell propagator is of the time - ordered @xmath6 or anti - time - ordered @xmath57 type contribute . the collision term ( [ collision : kb ] ) can therefore be split into the portions @xmath58 where the superscripts indicate through which two of the loop propagators in the vertex correction the cut goes . likewise , we decompose the vertex contribution to the source term as @xmath59 within the present work , we make the simplifying assumption that @xmath18 is in thermal equilibrium at all times , whereas @xmath11 is in equilibrium initially but then deviates from equilibrium when it becomes non - relativistic . we note that in general , both @xmath11 and @xmath18 will deviate from equilibrium at the same time . in fact , situations are conceivable where the initial abundance of @xmath18 is zero and @xmath27 is so small that @xmath18 does not equilibrate before becoming non - relativistic . therefore , the equilibrium deviation of @xmath18 can be large when compared to @xmath11 , and inverse decays of @xmath18 may largely enhance the lepton asymmetry . for now , leave this interesting possibility for future work and note that @xmath18 may be maintained in equilibrium by a stronger coupling to a different lepton flavour within which no asymmetry is produced . the term @xmath60 arises from those terms within @xmath61 , eq . ( [ sigma : v ] ) , where the propagators @xmath62 occur . it can be further simplified when approximating the distribution functions of @xmath16 and @xmath17 by fermi - dirac and , respectively , bose - einstein equilibrium distributions @xmath63 and @xmath64 . accounting for the fact that @xmath16 deviates from equilibrium only leads to higher order corrections in the gradient expansion , as it is explained in ref . @xcite . the gradient expansion corresponds to an expansion in powers of @xmath65 or equivalently @xmath66 , where @xmath67 denotes the hubble rate . the contribution of the cut through @xmath13 to the source term can be factorised into @xcite @xmath68\,\end{aligned}\ ] ] and @xmath69\end{aligned}\ ] ] with the result @xmath70\int \frac{d^3 k}{(2\pi)^3 2 \sqrt{{\mathbf k}^2+m_1 ^ 2}}\ , \delta f_{n1}(\mathbf k ) \,v^{\ell \phi}(k;m_1,m_2 ) \,.\end{aligned}\ ] ] the term @xmath71 denotes the deviation of @xmath72 from the equilibrium fermi - dirac distribution , @xmath73 . in ref . @xcite , this result has been evaluated in the hierarchical limit @xmath74 , which has the virtue that , as it is explained in the introduction , the evolution of the lepton number density is independent of @xmath19 up to an overall proportionality . furthermore , when @xmath74 , the collision term ( [ coll : vertex : final ] ) can be reduced analytically to a one - dimensional integral , which can easily be evaluated numerically . for the present work , we numerically evaluate a multi - dimensional integral for @xmath75 that remains when exploiting the @xmath76-functions . we note that also within the related articles @xcite , all source terms have been evaluated in the hierarchical limit , such that here , we present the first quantitative results for effects of finite @xmath12 in a finite - density background . now the cut goes through the internal @xmath18 and @xmath17 lines of figure [ fig : doublediamond ] ( a ) , and the internal @xmath16 is off - shell . the according contributions to the collision term are @xmath77^tc^\dagger { \rm i}s_{nj}^t(-q ) { \rm i}\delta_\phi^<(-p - k ) { \rm i}\delta_\phi^{t}(-q - k ) \\\notag - & { \rm i}s_\ell^<(k ) { \rm i}s_{ni}^{\bar t}(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^>(-q ) { \rm i}\delta_\phi^{\bar t}(-p - k ) { \rm i}\delta_\phi^{<}(-q - k ) \\\notag + & { \rm i}s_\ell^>(k ) { \rm i}s_{ni}^<(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^{\bar t}(-q ) { \rm i}\delta_\phi^>(-p - k ) { \rm i}\delta_\phi^{\bar t}(-q - k ) \\\notag - & { \rm i}s_\ell^>(k ) { \rm i}s_{ni}^{t}(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^<(-q ) { \rm i}\delta_\phi^{t}(-p - k ) { \rm i}\delta_\phi^{>}(-q - k ) \big\}\,.\end{aligned}\ ] ] as explained above , we assume here that besides @xmath16 and @xmath17 , @xmath18 is in equilibrium , such that its number density is given by the fermi - dirac distribution function @xmath78 . this allows us to apply kms relations and two further replacements , which become identities under the integrals ( _ cf . @xcite , where this is explained in more detail ) : @xmath79\ , , \\\notag { \rm i}s_{nj}^{t,\bar t}(-q ) { \rm i}\delta_\phi^{t,\bar t}(-q - k ) & \to \frac 12 \left [ { \rm i}s_{nj}^{<}(-q ) { \rm i}\delta_\phi^{>}(-q - k ) + { \rm i}s_{nj}^{>}(-q ) { \rm i}\delta_\phi^{<}(-q - k ) \right ] \,.\end{aligned}\ ] ] the result of these simplifications is @xmath80\int \frac{dk^0}{2\pi } \frac{d^4 p}{(2\pi)^4}\frac{d^4q}{(2\pi)^4 } \\\notag \frac 12{\rm tr}\big\ { & \big [ { \rm i}s_\ell^<(k ) { \rm i}s_{n1}^{>}(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}\delta_\phi^{<}(-p - k ) \\\notag - & { \rm i}s_\ell^>(k ) { \rm i}s_{n1}^{<}(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}\delta_\phi^{>}(-p - k ) \big ] \\\notag \times & \big [ { \rm i}s_{n2}^<(-q ) { \rm i}\delta_\phi^{>}(-q - k ) - { \rm i}s_{n2}^>(-q ) { \rm i}\delta_\phi^{<}(-q - k ) \big ] \big\}\,.\end{aligned}\ ] ] the final contribution to the source term , @xmath81 then follows when substituting the explicit forms of the propagators ( as they can be found in ref . @xcite ) as @xmath82\int\frac{d^3 p^\prime}{(2\pi)^32|\mathbf p^\prime| } \frac{d^3 p}{(2\pi)^3 2\sqrt{\mathbf p^2+m_1 ^ 2 } } \frac{d^3 p^{\prime\prime}}{(2\pi)^3|\mathbf p^{\prime\prime}| } \frac{d^3 k^{\prime\prime}}{(2\pi)^3 2|\mathbf k^{\prime\prime}| } \frac{d^3 k^\prime}{(2\pi)^3 2 \sqrt{{\mathbf k^\prime}^2+m_2 ^ 2 } } \\ \label{s : ell } \times & ( 2\pi)^4\delta^4(p - p^\prime - p^{\prime\prime } ) ( 2\pi)^4\delta^4(k^\prime - k^{\prime\prime}-p^\prime ) { p^\prime}^\mu\frac{(p+k^{\prime\prime})_\mu}{(p+k^{\prime\prime})^2}m_1m_2 \\\notag \times & \delta f_{n1}({\mathbf p } ) \left [ 1-f^{\rm eq}_\ell(\mathbf p^\prime)+f^{\rm eq}_\phi(\mathbf p^{\prime\prime } ) \right ] \times \left [ -f_\phi^{\rm eq}(\mathbf k^{\prime\prime})-f^{\rm eq}_{n2}(\mathbf k^\prime ) \right]\,.\end{aligned}\ ] ] to our knowledge , this is the first report of a result for a source term for leptogenesis that is present at finite density but absent in the limit of a vacuum background . we remark that the last factor results from the expression @xmath83f^{\rm eq}_{n2}-f_\phi^{\rm eq}[1-f^{\rm eq}_{n2}]$ ] . this could also be guessed starting from a hypothetical loop factor when the decay @xmath84 was kinematically allowed , @xmath85=[1+f_\phi^{\rm eq}][1-f^{\rm eq}_{n2}]+f_\phi^{\rm eq}f^{\rm eq}_{n2}$ ] , and applying the replacements @xmath86 $ ] . this argument may be considered as a consistency check for our derived result ( [ s : ell ] ) . it is important to notice that for @xmath87 , the source term ( [ s : ell ] ) is strongly maxwell suppressed . while in such a situation , @xmath88 is always suppressed because of the large mass of @xmath18 , the energy - momentum conserving @xmath76-functions always imply that then at least one of the momenta @xmath89 or @xmath90 is much larger than @xmath6 , such that @xmath91 or @xmath92 are suppressed as well . we finally consider the cuts through the propagators @xmath18 and @xmath16 within the loop and take @xmath17 to be off - shell . the contribution to the collision term is @xmath93^tc^\dagger { \rm i}s_{nj}^>(-q ) { \rm i}\delta_\phi^t(-p - k ) { \rm i}\delta_\phi^{t}(-q - k ) \\\notag - & { \rm i}s_\ell^<(k ) { \rm i}s_{ni}^{\bar t}(-p ) c\left[s_\ell^{\bar t}(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^>(-q ) { \rm i}\delta_\phi^{t}(-p - k ) { \rm i}\delta_\phi^{<}(-q - k ) \\\notag + & { \rm i}s_\ell^>(k ) { \rm i}s_{ni}^<(-p ) c\left[s_\ell^<(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^{<}(-q ) { \rm i}\delta_\phi^t(-p - k ) { \rm i}\delta_\phi^{\bar t}(-q - k ) \\\notag - & { \rm i}s_\ell^>(k ) { \rm i}s_{ni}^{t}(-p ) c\left[s_\ell^t(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^<(-q ) { \rm i}\delta_\phi^{t}(-p - k ) { \rm i}\delta_\phi^{>}(-q - k ) \big\}\,.\end{aligned}\ ] ] again , we substitute for @xmath94 , @xmath47 and @xmath95 the equilibrium propagators and employ kms relations . under the integrals , we make the replacements @xmath96^tc^\dagger \\\notag \to & \frac12 \left [ { \rm i}s_{ni}^{>}(-p ) { \rm i}c\left[s_\ell^{>}(p+k+q)\right]^tc^\dagger + { \rm i}s_{ni}^{<}(-p ) { \rm i}c\left[s_\ell^{<}(p+k+q)\right]^tc^\dagger \right]\,.\end{aligned}\ ] ] furthermore , it is useful to notice that when substituting the equilibrium propagator for @xmath16 , @xmath97^tc^\dagger { \rm i}s_{nj}^>(-q ) \\\notag - & { \rm i}s_\ell^>(k ) { \rm i}s_{ni}^<(-p ) c\left[s_\ell^<(p+k+q)\right]^tc^\dagger { \rm i}s_{nj}^<(-q)\end{aligned}\ ] ] is odd under the exchange @xmath98 , while @xmath99 $ ] ( which is the off - shell contribution ) is even . making use of these additional remarks , the collision term simplifies to @xmath100 \int \frac{dk^0}{2\pi } \frac{d^4 p}{(2\pi)^4}\frac{d^4q}{(2\pi)^4 } \\\notag { \rm tr}\big\ { & \left [ { \rm i } s_{n2}^<(-q ) { \rm i } s_\ell^ > ( k ) -{\rm i } s_{n2}^>(-q ) { \rm i } s_\ell^ < ( k ) \right ] { \rm i}\delta_\phi^t(-p - k ) \\\notag \times & \big [ { \rm i}s_{n1}^<(-p ) c\left[s_\ell^<(p+k+q)\right]^tc^\dagger { \rm i } \delta_\phi^<(-q - k ) \\\notag - & { \rm i}s_{n1}^>(-p ) c\left[s_\ell^>(p+k+q)\right]^tc^\dagger { \rm i } \delta_\phi^>(-q - k ) \big ] \big\}\,.\end{aligned}\ ] ] the source term is @xmath101 which becomes , when substituting the finite - density propagators , @xmath102 \int\frac{d^3 k^{\prime\prime}}{(2\pi)^3 2|\mathbf k^{\prime\prime}| } \frac{d^3 p}{(2\pi)^3 2\sqrt{\mathbf p^2+m_1 ^ 2 } } \frac{d^3 p^{\prime\prime}}{(2\pi)^3 2|\mathbf p^{\prime\prime}| } \frac{d^3 k^\prime}{(2\pi)^3 2\sqrt{{\mathbf k^\prime}^2+m_2 ^ 2 } } \frac{d^3 p^{\prime}}{(2\pi)^3 2|\mathbf p^{\prime}| } \\ \label{s : phi } \times & ( 2\pi)^4\delta^4(p - k^{\prime\prime}-p^{\prime\prime } ) ( 2\pi)^4\delta^4(k^\prime - k^{\prime\prime}-p^{\prime } ) \frac{p^{\prime\prime}_\mu { p^\prime}^\mu}{(p^{\prime\prime}+k^\prime)^2}m_1 m_2 \\\notag \times & \delta f_{n1}({\mathbf p } ) \left [ 1-f^{\rm eq}_\ell(\mathbf p^{\prime\prime } ) + f^{\rm eq}_\phi(\mathbf k^{\prime\prime } ) ] \right ] \times \left[f^{\rm eq}_\ell ( \mathbf p^\prime ) -f^{\rm eq}_{n2}(\mathbf k^\prime)\right]\,.\end{aligned}\ ] ] again , we note that the factor @xmath103=-[1-f^{\rm eq}_\ell]f^{\rm eq}_{n2}+f^{\rm eq}_\ell[1-f^{\rm eq}_{n2}]$ ] could also be guessed from a would - be factor @xmath104[1-f^{\rm eq}_{n2}]-f^{\rm eq}_\ell f^{\rm eq}_{n2}=1-f^{\rm eq}_\ell f^{\rm eq}_{n2}$ ] , which would arise if @xmath105 was kinematically allowed , and the replacements @xmath86 $ ] . we also note that @xmath106 is maxwell suppressed for @xmath87 , in analogy with what is discussed @xmath107 . we now present results from the numerical evaluation of the source terms ( [ coll : vertex : final],[s : ell],[s : phi ] ) . in section [ subsection : effective ] , we define an effective measure for the @xmath8 violation from the various cut contributions through a benchmark out - of - equilibrium distribution for @xmath11 . then , we evaluate this effective @xmath8 violation as a function of @xmath4 for various values of @xmath12 . we find that in case @xmath108 the new corrections are only significant for @xmath109 or smaller . if @xmath110 , the corrections from the new cuts can be relevant for larger values of @xmath21 . we proceed in section [ subsection : asymmetry ] with the calculation of the lepton asymmetry by solving the kinetic boltzmann - type equations ( [ kin : eq ] ) with the new source terms . our choice of the washout strength is motivated by the wish to exhibit models where the contributions from the new cuts are sizable , that is where relevant contributions to the final asymmetry arise at values for @xmath21 around one . this is the case in the transitional regime from weak to strong washout . we find that indeed , the new cut contributions can have a sizable impact on the final asymmetry . however , in case sizable contributions to the asymmetry originate from @xmath111 in our simulations , the quantitative results need to be interpreted with care . this is because in these regions , thermal corrections to the masses and widths of @xmath17 and @xmath16 will become relevant . in order to obtain a quantitative comparison of the amount of @xmath8 violation from the various source terms , it is useful to define a benchmark form for the distribution @xmath71 . this is necessary , since the precise form of @xmath71 depends on time , washout strength and initial conditions . we follow ref . @xcite by taking for @xmath72 a fermi - dirac distribution with a pseudo - chemical potential @xmath112 . the deviation from equilibrium is then obtained by expanding to linear order in @xmath113 , @xmath114 when the only interactions of the neutrino @xmath11 are mediated by @xmath26 , the actual distribution function is not exactly described by the pseudo - chemical potential , and eq . ( [ delta : fn1:appr ] ) should indeed only be regarded as a useful benchmark for the purpose of comparing the various contributions to the source term . note however that in case there are fast elastic scatterings between the neutrinos @xmath11 , eq . ( [ delta : fn1:appr ] ) should be a very accurate description for the actual distribution . we substitute eq . ( [ delta : fn1:appr ] ) into eqs . ( [ coll : vertex : final],[s : ell],[s : phi ] ) and into @xmath115 and @xmath116 as given in ref . @xcite , where @xmath116 is the source term including both , vertex and wave function terms , evaluated in the hierarchical limit . out of these , we take the ratios @xmath117 and @xmath118 as functions of @xmath4 and with @xmath119 defined in eq . ( [ s : v ] ) . the ratio ( [ ratio : v ] ) allows for a comparison of the source term at finite density and the standard cut through @xmath16 and @xmath17 only with the source term in the hierarchical limit . through the ratio ( [ ratio : full ] ) , we compare the source with all cuts at finite density with the hierarchical limit . to quantify the effect of the new cuts , both ratios ( [ ratio : v ] ) and ( [ ratio : full ] ) are compared with one another . before we present the numerical results for eqs . ( [ ratio : v ] ) and ( [ ratio : full ] ) , a few remarks on their relevance and range of validity are in order . first , note that the relevant contributions to leptogenesis are generated when @xmath109 or larger . ( this is to be understood in the sense that while @xmath21 is _ much _ smaller than one , no sizable contributions to the final asymmetry are generated ) . this is a consequence of the fact that @xmath11 must become non - relativistic and maxwell suppressed before it equilibrates , because otherwise the lepton asymmetry would be completley washed out . therefore , the effect of the new contributions is more relevant if it extends to larger values of @xmath21 . second , we remark that that for small @xmath21 , finite temperature effects should become relevant , which we do not take into account within the present work . the most important correction is the contribution of the top quarks to the debye mass - square of the higgs boson , @xmath120 , where @xmath121 is the top - quark yukawa coupling . besides , also the @xmath122 gauge couplings of the higgs bosons and the leptons are of relevance . a full evaluation of these finite - temperature effects in the context of leptogenesis has not been performed yet , but it would be of great importance for obtaining quantitatively accurate results in models where asymmetries generated at values of @xmath21 that are somewhat smaller than one are relevant . moreover , the size of the top - quark yukawa coupling at the energy scale of leptogenesis depends due to running on the mass of the higgs boson , which is yet unknown . for the present discussion , we therefore keep in mind that for the contributions that are generated at @xmath123 , we have to anticipate an inaccuracy of order one . c ( a ) : @xmath124 + c ( b ) : @xmath125 + + c ( c ) : @xmath126 + c ( d ) : @xmath127 + in figure [ fig : epseff ] , the ratios ( [ ratio : v ] ) and ( [ ratio : full ] ) are plotted for several values of @xmath12 . the main features for each of the particular values can be summarised as follows : * for @xmath124 , the hierarchical limit is of course not a good approximation . the total asymmetry is dominated by the resonantly enhanced contribution @xmath115 . as a consequence of this , @xmath128 is much larger than one . since the distribution of @xmath18 is only very weakly thermally suppressed when compared to @xmath11 , @xmath129 and @xmath130 agree only for comparably large values of @xmath21 . * for @xmath125 , we note that the hierarchical limit becomes a better approximation to @xmath129 , while yet deviating by about 20% . compared to case ( a ) , @xmath129 and @xmath130 start to agree for smaller values of @xmath21 , as the distribution of @xmath18 now suffers from stronger thermal suppression . we note that the very substantial deviation of @xmath129 from @xmath130 for very small values of @xmath21 needs to be interpreted with care in the light of the thermal corrections mentioned above . * for @xmath126 , the results obtained from the hierarchical limit @xmath131 and the source @xmath129 with the standard cuts only are now in good agreement , as anticipated . this serves also as a consistency check for our numerical evaluation of @xmath130 in the form of a multi - dimensional integral as compared to @xmath116 through the one - dimensional integral given in ref . the new cut contributions become suppressed for even smaller values of @xmath21 along with the stronger maxwell suppression of @xmath18 . the comment regarding scenario ( b ) , that the deviations that originate for values @xmath132 need to be interpreted with care due to thermal corrections , is even more relevant for the present case . * in this scenario , @xmath127 . since @xmath18 is now lighter than @xmath11 , the coupling @xmath27 must be chosen sufficiently small , such that @xmath18 does not equilibrate and wash out the lepton asymmetry before it becomes non - relativistic and maxwell suppressed . in this situation , for the whole relevant range of @xmath21 , neither the hierarchical limit nor the sources @xmath133 are an accurate approximation to the full result @xmath130 . from the comparison of the effective amount of @xmath8 violation in presence of the new cuts with the case when the new cuts are neglected , we see that when @xmath108 , the deviations are only sizable for @xmath21 is of order a few or smaller . for larger values of @xmath21 , the distribution of the @xmath18 is strongly maxwell suppressed , and the new cut contributions become irrelevant . in order to exhibit the effect of the new cuts , we therefore choose the parameters @xmath20 and @xmath26 such that leptogenesis takes place in the transitional regime from weak to strong washout . in this situation , sizable contributions to the final lepton asymmetry are generated for @xmath109 , such that the effects of the new cuts becomes relevant . we choose for @xmath0 thermal initial conditions . in the early universe , these may be established through interactions via heavy gauge bosons that freeze out at times before leptogenesis takes place . we assume that @xmath18 is maintained in equilibrium due to a yukawa coupling with an additional lepton flavour , within which no asymmetry is produced . the coupling @xmath27 is chosen smaller than @xmath26 , such that washout effects from inverse decays of @xmath18 are negligible . ( explicitly , for the scenarios with @xmath108 , the largest error from washout through @xmath27 occurs for @xmath124 . since the thermal suppression of @xmath18 compared to @xmath11 is very small in this case and we have chosen @xmath134 , we expect that the washout is underestimated by 20% . the accuracy improves for larger ratios of @xmath108 due to the thermal suppression of @xmath18 and its irrelevance for washout ) . while this appears as a somewhat special setup , we note that even when @xmath27 is large , the vector of couplings of @xmath18 to the various left - handed lepton flavours defines a particular linear combination @xmath135 of leptons that are washed out through inverse decays of @xmath18 . this linear combination can in general be linearly independent of the linear combination @xmath16 within which the lepton asymmetry through decays and inverse decays of @xmath11 is produced . the contribution to the asymmetry in @xmath16 that is orthogonal to @xmath135 is then unaffected by the washout through @xmath18 @xcite . ( note the different assignment of the heavy neutrinos to the indices @xmath136 in that work ) . therefore , the qualitative features of our particular setup should be relevant for parametrically more generic models of leptogenesis . however , it would still be interesting to study the effect of different initial conditions and the possibility of @xmath18 being out - of - equilibrium within future work . as it is described in detail in ref . @xcite , we obtain the numerical results as follows : first , we solve the evolution equations for @xmath72 as a function of @xmath21 . these , we feed into the washout term @xmath30 and the source term @xmath31 for the leptons , in order to solve for the lepton asymmetry in eq . ( [ kin : eq ] ) . the expansion of the universe is taken into account when inserting the scale factor according to @xmath137 . c ( a ) : @xmath124 , @xmath138 + c ( b ) : @xmath125 , @xmath138 + + c ( c ) : @xmath126 , @xmath138 + c ( d ) : @xmath139 , @xmath140 + the results for the lepton - number to entropy ratio @xmath141 as defined in ref . @xcite are presented in figure [ figure : asymmetries ] . note that it is instructive to compare the particular panels with those in figure [ fig : epseff ] . again , we summarise some features for each of the particular values of @xmath142 : * for @xmath124 , the hierarchical limit clearly underestimates the full result . this is anticipated , because the wave - function contribution to the full result is resonantly enhanced for @xmath143 . the new cuts give rise to relative corrections at the 10% level . note however that the absolute correction is larger when compared to scenarios ( b ) and ( c ) . the relative correction is marginalised due to the resonant enhancement of @xmath115 . * for @xmath125 , there are sizable deviations of the full result from the result with the cut through @xmath13 only , that arise in the region @xmath109 . * for @xmath126 , the new contributions are of importance for smaller values of @xmath21 when compared to scenarios ( a ) and ( b ) . in the light of the thermal corrections that we anticipate to be important for small values of @xmath21 , the quantitative result needs to be interpreted with care . the good agreement between the result from the standard cut through @xmath13 only and the hierarchical limit serves again as a consistency check for our numerical evaluations . * for @xmath127 , we choose a smaller value for @xmath27 , motivated by the requirement that the @xmath18 must not wash out the lepton asymmetry . the full result and the result with the cut through @xmath144 only receive different contributions even for values of @xmath21 of the order of a few , _ cf . _ figure [ fig : epseff ] ( d ) . we expect therefore only a small contamination from theoretical uncertainties due to thermal corrections . note that in this scenario most of the lepton asymmetry of the universe originates from the new cuts . when we insert a loop of @xmath17 and @xmath18 as a wave - function correction into the propagator @xmath47 in the vacuum , no branch cut term due to on - shell @xmath17 and @xmath18 arises for kinematic reasons . again , this holds no longer true in a finite temperature background . therefore , @xmath8-violating source terms can arise from the diagram in figure [ fig : flavourwf ] . since there is no lepton number violation , this does not yield a contribution to the asymmetry in models of unflavoured leptogenesis . however , this diagram may appear as a source in flavoured scenarios . in this section , we give a rough estimate of this contribution , leading to the conclusion that it is generically negligible . within the ctp - formalism , the form of the wightman type self - energy , that is given diagrammatically in figure [ fig : flavourwf ] , reads @xmath145 compared to the unflavoured scenario , we have promoted the yukawa couplings of the right - handed neutrinos @xmath146 to a matrix @xmath147 , where the first index refers to the right - handed neutrino and the second index to the left - handed lepton flavour . for the definitions of the model lagrangian and the lepton propagator @xmath148 , we refer to ref . @xcite . next , we insert @xmath149 into the collision term ( [ collision : kb ] ) . this again simplifies when substituting equilibrium propagators for @xmath18 , @xmath16 and @xmath17 and exploiting kms relations . we furthermore choose to work in the flavour basis where the matrix of standard model lepton yukawa couplings @xmath150 is diagonal . this is advantageous since off - diagonal components of @xmath151 are damped away quickly in this basis , provided the interactions mediated by @xmath150 are fast compared to the hubble rate @xcite , _ cf . _ @xcite for a description within the ctp approach and for a numerical study of this effect . note that this also implies that the off - diagonal components of the equilibrium propagator for @xmath16 are vanishing . we eventually obtain for the collision term @xmath152 \int\frac{dk^0}{2\pi}\frac{d^4 p}{(2\pi)^4}\frac{d^4 q}{(2\pi)^4 } \\\notag \frac 12{\rm tr } \big\ { & \left ( { \rm i}s_{\ell aa}^<(k ) { \rm i } s_{ni}^>(p ) { \rm i } \delta_\phi^<(p - k ) - { \rm i}s_{\ell aa}^>(k ) { \rm i } s_{ni}^<(p ) { \rm i } \delta_\phi^>(p - k ) \right ) { \rm i}{\rm re}\left[\frac{1}{k^2+{\rm i\varepsilon}}\right ] \\\notag \times & \left ( { \rm i } s_{nj}^<(q ) { \rm i } \delta_\phi^>(q - k ) - { \rm i } s_{nj}^>(q ) { \rm i } \delta_\phi^<(q - k ) \right ) \big\}\,,\end{aligned}\ ] ] where @xmath153 is infinitesimal . in the zero - width limit , this term contains a divergence from the factor @xmath154 that originates from the propagator @xmath155 , since @xmath156 is proportional to @xmath157 . at finite temperature , this is regulated through the replacements @xmath158 and @xmath159\,.\end{aligned}\ ] ] here , the thermal masses of the leptons are @xmath160 , where @xmath161 is the @xmath122 gauge coupling . the functions @xmath162 and @xmath163 are of order @xmath164 for @xmath165 of order @xmath6 . they account for thermal mass corrections from flavour - sensitive and flavour blind interactions , and are discussed in more detail in ref . likewise @xmath166 is the finite - temperature width of the leptons , where @xmath167 is of order @xmath6 when @xmath165 of order @xmath6 . when leptogenesis occurs at temperatures of roughly below @xmath168 to @xmath169 but above @xmath170 to @xmath171 , the @xmath172-lepton yukawa coupling @xmath173 is in equilibrium , while the electron @xmath174 and muon @xmath175 couplings are yet out - of - equilibrium . in this situation , flavoured leptogenesis distinguishes effectively between two flavours , where flavour 1 is identified with @xmath172 and the coupling @xmath176 and flavour 2 with a linear combination of @xmath174 and @xmath175 and with negligible yukawa coupling @xmath177 . the integral ( [ c : flav : wf ] ) could possibly again be evaluated numerically . however , we can estimate that from the factors ( [ subst : st ] ) and ( [ subst : delta ] ) , the @xmath178-integration yields a factor @xmath179 for @xmath180 . the latter estimate follows from above assumption that only @xmath173 appears as a relevant standard model yukawa coupling . above suppression factor is to be compared to @xmath181 for a leptogenesis scenario with @xmath182 but no pronounced resonant enhancement . since @xmath183 , we conclude that this contribution to flavoured leptogenesis is suppressed due to the large width of the leptons @xmath16 at finite temperature . in other words , the contribution to the @xmath8-asymmetry is rendered ineffective because the separation between the resonances of the lepton quasi - particles is well within their overlap due to the finite widths . in this paper , we have presented in the form of eqs . ( [ s : ell ] ) and ( [ s : phi ] ) the first results for source terms that contribute to the lepton asymmetry in a finite - density background , but that are absent in the vacuum . in order for these to be relevant , we have seen that @xmath19 should not be much larger than @xmath6 at the time of leptogenesis . the main features of the numerical evaluations are easily understood . first , @xmath19 should not be much larger than @xmath20 for the new effects to be sizable at larger values of @xmath21 , _ cf . _ figure [ fig : epseff ] . we reemphasise that the quantitative results for the scenario in figure [ figure : asymmetries ] ( c ) should be considered with great care since a substantial amount of the deviations incurred by the new effects is generated at very small values of @xmath21 , where we expect thermal corrections to become relevant . second , the effect is most pronounced in scenarios where a sizable amount of the asymmetry is produced for comparably small values of @xmath21 , as it is the case for the transitional regime between weak and strong washout . as a loophole , we find that the largest effects arise within a somewhat unconventional scenario with @xmath110 , since in this situation the new cuts are important throughout the time of the out - of - equilibrium decays of @xmath11 . therefore , relevant contributions from the new cuts also result for larger values of @xmath21 . however , in regard of ref . @xcite , where it is shown that decay asymmetries from the heavier singlet neutrinos generically survive subsequent washout , such a scenario may not only appear as a mere loophole . provided the reheat temperature is large enough to produce the heavier singlet neutrino in the early universe , the present work implies that a large contribution to the baryon asymmetry of the universe generically results from the new cuts . we mention that while this work has been in preparation , ref . @xcite appeared , where non - standard cuts are discussed as well . since finite density effects are neglected , it is found there that the new cuts are only allowed within certain scattering processes . more precisely , in order to satisfy the kinematic thresholds for the presence of cut contributions in the vacuum , diagrams with additional radiation of standard model particles are considered . note that if the resulting amplitudes were substituted as source terms into the kinetic equations for the lepton asymmetry , they would be subject to the same maxwell suppression that occurs also in the present case for @xmath87 . however , compared to the new source terms that we have derived in the present work , the terms discussed in ref . @xcite are subject to further suppression because of the insertion of additional coupling constants . * the results ( [ s : ell ] ) and ( [ s : phi ] ) should be generalised to include effects from deviations of the distribution @xmath184 from equilibrium , which we expect to be non - zero in generic scenarios of leptogenesis . * systematic investigations of the parameter space for scenarios where the new cuts are relevant would be desirable . we expect the new cut contributions to be of crucial importance for phenomenological studies where it is assumed that @xmath110 , such as refs . @xcite , where @xmath18 may or may not be in equilibrium . ( note the different definitions of @xmath0 in these works ) . note that in case @xmath185 , thermal corrections may become important again , since the thermal masses of @xmath17 and @xmath16 at the time of the decay of @xmath11 may exceed @xmath19 . we expect the new cuts still to be important in such a situation , but the interpretations of the particular cut particles in terms of absorption processes may change to emissions and vice versa . * in the present work , we have restricted ourselves to compute the lepton asymmetries for thermal initial conditions for @xmath0 . this is in part motivated by the fact that for a vanishing initial density of @xmath11 , the final asymmetry is a remainder of an incomplete cancellation of a contribution that is created initially at small @xmath21 , when @xmath11 is underabundant , and an opposite one through later decays when @xmath21 is larger and @xmath11 overabundant . the fact that the cancellation is incomplete is because the opposite asymmetries are affected differently by the washout , since they occur at different temperatures . in order to obtain a quantitatively reliable result for the remaining asymmetry , a rather accurate prediction of the asymmetry that is created through inverse decays at small @xmath21 is necessary . due to the uncertainties because of thermal corrections , that we have emphasised in this paper , such an accurate prediction is presently not available . the situation somewhat improves for the thermal initial conditions that we consider within the present work , because @xmath11 is always overabundant and the produced asymmetry is therefore of the same sign for all values of @xmath21 . in regard of these points , we briefly comment on possible technical improvements that may prove very useful in order to increase the accuracy of the predictions for leptogenesis from the new cuts as well as from the standard cuts . it would be particularly interesting , if the following issues were addressed : * as it should be clear from the discussions within this paper , the uncertainties due to thermal corrections for small @xmath21 are problematic for the predictivity of the new cut contributions as well as more generically for leptogenesis in the weak washout regime , in particular for vanishing initial conditions for @xmath11 . in order to resolve this issue , a calculation of the rates @xmath186 for temperatures that are of order @xmath20 or larger is necessary . both , the tree - level rates as well as the @xmath8-violating loop effects need to be calculated . for that purpose , in particular the thermal masses and finite widths of @xmath16 and @xmath17 should be taken into account . first work into this direction has been reported in ref . @xcite . * in ref . @xcite , sizable effects from the finite width of @xmath11 in the @xmath8-violating source term have been reported . it needs to be verified , whether the initial conditions chosen in that work are applicable to the situation in the early universe . furthermore , the finite widths of @xmath16 and @xmath17 , which are much larger than for @xmath11 and have been neglected so far , need to be taken into account in a future calculation . considering these comments and a number of related papers , the present work may be regarded as a contribution to present efforts to improve the theoretical description of leptogensis , to increase the accuracy of quantitative predictions and to extend their range of applicability . the author would like to thank martin beneke and pedro schwaller for useful comments on the manuscript . this work is supported by the gottfried wilhelm leibniz programme of the deutsche forschungsgemeinschaft . s. davidson and a. ibarra , `` a lower bound on the right - handed neutrino mass from leptogenesis , '' phys . b * 535 * ( 2002 ) 25 [ arxiv : hep - ph/0202239 ] . w. buchmller , p. di bari and m. plmacher , `` cosmic microwave background , matter - antimatter asymmetry and neutrino masses , '' nucl . b * 643 * ( 2002 ) 367 [ erratum - ibid . b * 793 * ( 2008 ) 362 ] [ arxiv : hep - ph/0205349 ] . w. buchmller , p. di bari and m. plmacher , `` the neutrino mass window for baryogenesis , '' nucl . b * 665 * ( 2003 ) 445 [ arxiv : hep - ph/0302092 ] . m. beneke , b. garbrecht , m. herranen and p. schwaller , `` finite number density corrections to leptogenesis , '' nucl . b * 838 * ( 2010 ) 1 [ arxiv:1002.1326 [ hep - ph ] ] . m. flanz , e. a. paschos and u. sarkar , `` baryogenesis from a lepton asymmetric universe , '' phys . b * 345 * ( 1995 ) 248 [ erratum - ibid . b * 382 * ( 1996 ) 447 ] [ arxiv : hep - ph/9411366 ] . m. flanz , e. a. paschos , u. sarkar and j. weiss , `` baryogenesis through mixing of heavy majorana neutrinos , '' phys . b * 389 * ( 1996 ) 693 [ arxiv : hep - ph/9607310 ] . a. pilaftsis , `` cp violation and baryogenesis due to heavy majorana neutrinos , '' phys . d * 56 * ( 1997 ) 5431 [ arxiv : hep - ph/9707235 ] . l. covi , e. roulet and f. vissani , `` cp violating decays in leptogenesis scenarios , '' phys . b * 384 * ( 1996 ) 169 [ arxiv : hep - ph/9605319 ] . a. pilaftsis and t. e. j. underwood , `` resonant leptogenesis , '' nucl . b * 692 * ( 2004 ) 303 [ arxiv : hep - ph/0309342 ] . m. y. khlopov and a. d. linde , `` is it easy to save the gravitino ? , '' phys . b * 138 * ( 1984 ) 265 . j. r. ellis , j. e. kim and d. v. nanopoulos , `` cosmological gravitino regeneration and decay , '' phys . b * 145 * , 181 ( 1984 ) . j. r. ellis , d. v. nanopoulos and s. sarkar , `` the cosmology of decaying gravitinos , '' nucl . b * 259 * , 175 ( 1985 ) . a. pilaftsis and t. e. j. underwood , `` electroweak - scale resonant leptogenesis , '' phys . d * 72 * ( 2005 ) 113001 [ arxiv : hep - ph/0506107 ] . s. blanchet , z. chacko , s. s. granor and r. n. mohapatra , `` probing resonant leptogenesis at the lhc , '' arxiv:0904.2174 [ hep - ph ] . s. blanchet , t. hambye and f. x. josse - michaux , `` reconciling leptogenesis with observable @xmath187 rates , '' jhep * 1004 * ( 2010 ) 023 [ arxiv:0912.3153 [ hep - ph ] ] . g. f. giudice , a. notari , m. raidal , a. riotto and a. strumia , `` towards a complete theory of thermal leptogenesis in the sm and mssm , '' nucl . b * 685 * ( 2004 ) 89 [ arxiv : hep - ph/0310123 ] . j. s. schwinger , `` brownian motion of a quantum oscillator , '' j. math . * 2 * ( 1961 ) 407 . l. v. keldysh , `` diagram technique for nonequilibrium processes , '' zh . * 47 * ( 1964 ) 1515 [ sov . phys . jetp * 20 * ( 1965 ) 1018 ] . w. buchmller and s. fredenhagen , `` quantum mechanics of baryogenesis , '' phys . b * 483 * ( 2000 ) 217 [ arxiv : hep - ph/0004145 ] . a. de simone and a. riotto , `` quantum boltzmann equations and leptogenesis , '' jcap * 0708 * ( 2007 ) 002 [ arxiv : hep - ph/0703175 ] . m. garny , a. hohenegger , a. kartavtsev and m. lindner , `` systematic approach to leptogenesis in nonequilibrium qft : vertex contribution to the cp - violating parameter , '' phys . d * 80 * ( 2009 ) 125027 [ arxiv:0909.1559 [ hep - ph ] ] . m. garny , a. hohenegger , a. kartavtsev and m. lindner , `` systematic approach to leptogenesis in nonequilibrium qft : self - energy contribution to the cp - violating parameter , '' phys . d * 81 * , 085027 ( 2010 ) [ arxiv:0911.4122 [ hep - ph ] ] . m. garny , a. hohenegger and a. kartavtsev , `` medium corrections to the cp - violating parameter in leptogenesis , '' phys . d * 81 * ( 2010 ) 085028 [ arxiv:1002.0331 [ hep - ph ] ] . m. garny , a. hohenegger and a. kartavtsev , `` quantum corrections to leptogenesis from the gradient expansion , '' arxiv:1005.5385 [ hep - ph ] . e. w. kolb and s. wolfram , `` baryon number generation in the early universe , '' nucl . b * 172 * ( 1980 ) 224 [ erratum - ibid . b * 195 * ( 1982 ) 542 ] . w. buchmller , p. di bari and m. plmacher , `` leptogenesis for pedestrians , '' annals phys . * 315 * ( 2005 ) 305 [ arxiv : hep - ph/0401240 ] . p. di bari , `` seesaw geometry and leptogenesis , '' nucl . b * 727 * ( 2005 ) 318 [ arxiv : hep - ph/0502082 ] . g. engelhard , y. grossman , e. nardi and y. nir , `` importance of the heavier singlet neutrino in leptogenesis , '' phys . * 99 * ( 2007 ) 081802 [ arxiv : hep - ph/0612187 ] . e. bertuzzo , p. di bari and l. marzola , `` the problem of the initial conditions in flavoured leptogenesis and the tauon @xmath18-dominated scenario , '' arxiv:1007.1641 [ hep - ph ] . t. endoh , t. morozumi and z. h. xiong , `` primordial lepton family asymmetries in seesaw model , '' prog . * 111 * ( 2004 ) 123 [ arxiv : hep - ph/0308276 ] . a. abada , s. davidson , f. x. josse - michaux , m. losada and a. riotto , `` flavour issues in leptogenesis , '' jcap * 0604 * , 004 ( 2006 ) [ arxiv : hep - ph/0601083 ] . e. nardi , y. nir , e. roulet and j. racker , `` the importance of flavor in leptogenesis , '' jhep * 0601 * , 164 ( 2006 ) [ arxiv : hep - ph/0601084 ] . c. s. fong , m. c. gonzalez - garcia and j. racker , `` cp violation from scatterings with gauge bosons in leptogenesis , '' arxiv:1010.2209 [ hep - ph ] . c. p. kiessig , m. plmacher and m. h. thoma , `` decay of a yukawa fermion at finite temperature and applications to leptogenesis , '' phys . d * 82 * ( 2010 ) 036007 [ arxiv:1003.3016 [ hep - ph ] ] .
a theory of signatures for odd - dimensional links in rational homology spheres is studied via their generalized seifert surfaces . the jump functions of signatures are shown invariant under appropriately generalized concordance and a special care is given to accommodate 1-dimensional links with mutual linking . furthermore our concordance theory of links in rational homology spheres remains highly nontrivial after factoring out the contribution from links in integral homology spheres .
math0108197
signature invariants of knots and links have been studied in many articles , and turned out to be effective in various problems in knot theory , especially in studying concordance . signatures were first defined as numerical invariants from seifert matrices of classical knots in the three space by trotter @xcite . murasugi formulated a definition of signatures for links @xcite . tristram introduced link signature functions @xcite . kauffman and taylor found a geometric approach to signatures of links using branched covers @xcite . milnor defined a signature function from a duality in the infinite cyclic cover of a knot complement @xcite . it is known that all the approaches are equivalent @xcite . signature invariants are known to be concordance invariants of codimension two knots and links in odd dimensional spheres . in the remarkable works of levine @xcite , signature invariants are used to classify concordance classes of knots . levine showed that the knot concordance groups are isomorphic to the cobordism groups of seifert matrices in higher odd dimensions , and signatures are complete invariants for the matrix cobordism groups modulo torsion elements . in this paper , we develop signature invariants of arbitrary links of codimension two in odd dimensional rational homology spheres , motivated from the work of cochran and orr @xcite . our signatures are invariant under the following equivalence of links in rational homology spheres . an @xmath0-_link _ @xmath1 in a rational homology @xmath2-sphere @xmath3 is a submanifold diffeomorphic to an ordered union of disjoint standard @xmath0-spheres . such a link is sometimes denoted by a pair @xmath4 . two @xmath0-links @xmath5 and @xmath6 are _ rational homology concordant _ ( or _ @xmath7-concordant _ ) if there is an @xmath8-manifold @xmath9 whose boundary is the disjoint union of @xmath10 and @xmath11 such that the triad @xmath12 has the rational homology of @xmath13 , s^{n+2}\times 0 , s^{n+2}\times 1)$ ] , and there is a proper submanifold @xmath14 in @xmath9 diffeomorphic to @xmath15 $ ] such that the boundary of the @xmath16-th component of @xmath14 is the union of the @xmath16-th components of @xmath17 and @xmath18 . concordant links are clearly rational homology concordant as well , and so the signatures studied in this paper are invariant under ordinary concordance as well . as done in @xcite for links in @xmath19 , we extract invariants of links from the signature function @xmath20 and its jump function @xmath21 where @xmath22 is a rational seifert matrix , @xmath23 and @xmath24 are reals , and @xmath25 . we follow the convention that the signature of an anti - hermitian matrix @xmath26 is the signature of the hermitian matrix @xmath27 . section [ sec : matrix - sign - poly ] is devoted to an algebraic study of @xmath28 and @xmath29 . especially , we prove a _ reparametrization formula _ for @xmath28 and @xmath29 of a seifert matrix of a union of parallel copies of a seifert surface ( see lemma [ lem : alg - sign - formula ] , theorem [ thm : sign - formula ] ) . a seifert surface of a link @xmath1 is defined to be an oriented submanifold bounded by @xmath1 in the ambient space . @xmath1 has a seifert surface if and only if there exists a homomorphism of the first integral homology of the exterior of @xmath1 onto the infinite cyclic group sending each meridian of @xmath1 to a fixed generator . in general , such a homomorphism does not exist and links bound no seifert surfaces , unless the ambient space is an integral homology sphere . in the case of rational homology spheres , we show that a certain union of parallel copies of a link @xmath1 bounds a seifert surface even though @xmath1 itself does not . more precisely , for an @xmath30-component link @xmath1 and @xmath31 where @xmath32 are relative prime , we call an oriented submanifold @xmath33 a _ generalized seifert surface of type @xmath34 for @xmath1 _ if for some nonzero integer @xmath35 , @xmath33 is bounded by the union of @xmath36 parallel copies ( or negatively oriented copies if @xmath37 ) of the @xmath16-th component of @xmath1 for all @xmath16 . if @xmath1 is a 1-link , the parallel copies are taken with respect to some framing @xmath38 on @xmath1 and the surface @xmath33 and the framing @xmath38 must induce the same framing on the parallel copies . we call @xmath35 the _ complexity of @xmath33_. a higher dimensional link admits a generalized seifert surface of any type @xmath34 . but a 1-link admits a generalized seifert surface of the types satisfying a certain condition and a framing is uniquely determined by a given type . furthermore @xmath7-concordant links admit generalized seifert surfaces of the same type . ( see theorem [ thm : seifert - surface - existence - highdim ] and theorem [ thm : seifert - surface - existence - dim1 ] . ) we define a seifert matrix to be a matrix associated to the rational - valued seifert pairing on the middle dimensional rational homology of a generalized seifert surface , and derive invariants of a link from the signature jump function of a seifert matrix as follows . let @xmath39 for @xmath40 . the _ signature jump function of type @xmath34 _ for a @xmath41-link @xmath1 is defined by @xmath42 where @xmath22 is a seifert matrix of a generalized seifert surface @xmath33 of type @xmath34 for @xmath1 and @xmath35 is the complexity of @xmath33 . our main result is that the signature jump function @xmath43 is a well - defined invariant under @xmath7-concordance which is independent of the choice of generalized seifert surfaces . [ thm : link - sign - inv ] if @xmath44 and @xmath18 are @xmath7-concordant , @xmath45 for all @xmath24 . for @xmath46 , we denote @xmath43 by @xmath47 . if @xmath1 is a link in an integral homology sphere , @xmath1 bounds a seifert surface @xmath33 and @xmath48 can be computed from a seifert matrix of @xmath33 . hence our signature invariant is a generalization of the well - known signature invariant of links in the spheres . for 1-links that are null - homologous modulo @xmath49 in a rational homology 3-sphere , gilmer defined the @xmath49-signature that depends on the choice of a specific element in the first homology of the ambient space @xcite . we remark that it is equal to the evaluation of the signature function associated to a generalized seifert matrix of type @xmath50 at the @xmath49-th roots of unity , and so it can be derived from our signature jump function . a link is called a _ slice link _ if its components bound properly embedded disjoint disks in a rational homology ball bounded by the ambient space . for a link @xmath1 and an @xmath0-tuple @xmath51 with @xmath52 , let @xmath53 be the union of @xmath0 parallel copies of @xmath1 where the @xmath16-th copy is oriented according to the sign of @xmath54 , and let @xmath55 be the sum of @xmath54 . complexity @xmath56 of type @xmath34 _ for @xmath1 is defined to be the minimum of the absolute values of complexities of generalized seifert surfaces of type @xmath34 . for @xmath57 , let @xmath58 be an @xmath0-dimensional link with @xmath59-components and @xmath60 be an @xmath2-ball in @xmath61 such that for some @xmath62 , @xmath60 intersects @xmath63 components of @xmath64 at a trivial @xmath63-component disk link in @xmath60 . by pasting @xmath65 and @xmath66 together along an orientation reversing homeomorphism on boundaries , a @xmath67-component link is obtained and is called a _ connected sum of @xmath18 and @xmath68_. a connected sum depends on the choice of @xmath60 if @xmath69 . some useful properties of our signature invariant are listed in the following theorem . [ thm : link - sign - prop ] 1 . if @xmath1 is slice , @xmath70 . 2 . if @xmath71 , @xmath72 . 3 . @xmath73 is a periodic function of period @xmath74 . let @xmath1 be a connected sum of @xmath41-links @xmath18 and @xmath68 . if either @xmath75 or @xmath18 , @xmath68 are knots , then @xmath76 . in section [ sec : sign - of - link ] , we prove the above results by combining the techniques of signatures of links in @xmath19 with the reparametrization formula in section [ sec : matrix - sign - poly ] that plays a key role in resolving the difficulty that links do not bound seifert surfaces . we give examples of links in rational homology spheres whose signature jump functions have arbitrary rational multiples of @xmath77 as periods , and show that the theory of @xmath7-concordance of links in rational homology spheres is not reduced to the concordance theory of links in spheres . in fact we construct a family of infinitely many @xmath7-concordance classes of knots that are independent modulo knots in the spheres in the following sense . let @xmath78 and @xmath79 be the set of concordance classes of @xmath0-knots in @xmath80 and the set of @xmath7-concordance classes of @xmath0-knots in rational homology spheres , respectively . they are abelian groups under connected sum . we have a natural homomorphism @xmath81 since concordant links in a sphere are @xmath7-concordant . [ thm : independent - knots ] there exist infinitely many @xmath41-knots in rational homology spheres that are linearly independent in the cokernel of @xmath82 . we remark that knots constructed by cochran and orr using the covering link construction @xcite are nontrivial in the cokernel since they are not concordant to knots of complexity one . our result shows that the cokernel is large enough to contain a subgroup isomorphic to @xmath83 . on the other hand , we illustrate that our theory is also useful for links in spheres by offering alternative proofs of a result of litherland @xcite about signatures of satellite knots and a result of kawauchi @xcite about concordance of split links . in a separated paper @xcite , we will compute signature jump functions of covering links and study concordance of boundary links and homology boundary links . in particular , results of cochran and orr @xcite , gilmer and livingston @xcite , and levine @xcite will be generalized . in this section , we study algebraic properties of the signature function @xmath29 and its jump function @xmath28 that will play a key role in our link signature theory . it is known that if @xmath22 is a complex square matrix such that @xmath84 is nonsingular , the function @xmath85 is constant except at finitely many jumps where @xmath86 is a zero of @xmath87 , and so the jump function @xmath28 is well - defined . in general , @xmath28 is well - defined for any complex square matrix @xmath22 . we accompany a proof since we did not find one in the literature . let @xmath88 denote the usual involution @xmath89 on the laurent polynomial ring @xmath90 $ ] . [ lem : sign - local - const ] let @xmath91 be a square matrix over @xmath90 $ ] that satisfies @xmath92 . then for any @xmath93 , the signature of @xmath94 is locally constant on a deleted neighborhood of @xmath95 , i.e. , it is constant on each of @xmath96 and @xmath97 for some @xmath98 . first we show that the rank of @xmath94 is constant on a deleted neighborhood of @xmath95 . let @xmath99 be a square matrix obtained by deleting some rows and columns from @xmath91 and denote its determinant by @xmath100 . then either @xmath101 or there is a deleted neighborhood @xmath102 of @xmath95 where @xmath103 since @xmath100 is a laurent polynomial over @xmath104 . since there are finitely many @xmath99 , the intersection of all @xmath102 is a deleted neighborhood of @xmath95 where @xmath105 is constant . now it suffices to show that @xmath106 is locally constant on any open interval @xmath107 where @xmath105 is constant . near a fixed @xmath108 in @xmath107 , @xmath94 is congruent to a diagonal matrix whose diagonals @xmath109 are real - valued continuous functions on @xmath23 . we can choose a neighborhood @xmath110 of @xmath108 in @xmath107 such that if @xmath111 , then @xmath112 and its sign is not changed in @xmath110 . since @xmath105 is constant , @xmath113 on @xmath110 if @xmath114 . therefore @xmath106 is constant on @xmath110 . from the lemma , it is immediate that for an arbitrary @xmath22 , the function @xmath85 is locally constant except at finitely many jumps on the unit circle and so @xmath28 is well - defined . as mentioned in the introduction , signature invariants of links are defined from the signature functions of seifert matrices of generalized seifert surfaces . note that a union of parallel copies of a generalized seifert surface is again a generalized seifert surface . since both of them can be used to define signature invariants of links , we are naturally led to investigate a relation of the signature functions of them . explicitly , for an @xmath115-tuple @xmath116 with @xmath117 , consider the union of @xmath115 parallel copies of a seifert surface @xmath33 of a @xmath41-link , where the @xmath0-th copy is oriented according to the sign of @xmath118 . if @xmath22 is a seifert matrix of @xmath33 , the seifert matrix of the union of parallel copies is given as follows . @xmath119 where @xmath120 the following result shows that the signature jump functions of @xmath121 and @xmath22 are essentially equivalent up to a change of parameters . to simplify statements , we define @xmath122 to be zero . [ thm : sign - formula ] if @xmath71 or @xmath123 is nonsingular , then @xmath124 . the rest of this section is devoted to the proof of theorem [ thm : sign - formula ] . for later use , we introduce some notations . for a nonzero integer @xmath115 , let @xmath125 where @xmath126 is an @xmath127-tuple whose entries are @xmath128 . we denote @xmath129 , @xmath130 by @xmath131 , @xmath132 , respectively , and use similar notations for @xmath133 and @xmath134 . since seifert pairing of links in rational homology spheres is rational - valued ( see section [ sec : link - in - rhs ] ) , we assume that @xmath22 is a square matrix over @xmath7 . for a positive integer @xmath135 , let @xmath136 be the step function of period @xmath77 given by @xmath137 for negative @xmath135 , let @xmath138 and let @xmath139 . [ lem : alg - sign - formula ] suppose that @xmath140 for @xmath141 . if @xmath142 or @xmath123 is nonsingular , we have 1 . @xmath143 . 2 . @xmath144 . let @xmath145 . first we note that if @xmath146 , @xmath147 we will prove some matrix congruence relations . consider the matrix @xmath148 where @xmath149 or @xmath150 . subtracting the second row from the first row and performing the corresponding column operation , a new matrix is obtained . subsequently we perform the following row operations and corresponding column operations : if @xmath151 and @xmath149 , we add the first row times @xmath152 to the second row . if @xmath153 and @xmath149 , we assume @xmath123 is nonsingular and add the top row times @xmath154 to the bottom row . then we obtain @xmath155 respectively . if @xmath156 , replace @xmath157 by @xmath158 and @xmath63 by @xmath159 , respectively . then we obtain @xmath160 for @xmath161 and @xmath162 , respectively . now we prove the lemma . note that @xmath163 and @xmath164 , where @xmath165 . we may assume that @xmath166 and furthermore by rearranging @xmath54 we may assume @xmath167 and @xmath168 where @xmath169 . for , if @xmath170 and the lemma is proved for the case of @xmath166 , @xmath171 and the conclusion follows . suppose @xmath172 . by applying repeatedly the first and third congruence relations in the above to the matrix @xmath173 where @xmath174 we obtain a matrix that is a block sum of @xmath175 and @xmath176 if @xmath177 , the contribution of ( [ equ : reducedmatrix ] ) to the signature is zero since @xmath178 . ( recall that @xmath22 is a matrix over @xmath7 . ) this proves the conclusion for @xmath172 and @xmath177 . if @xmath179 , let @xmath180 . then we have @xmath181 and therefore @xmath182 by an elementary calculation , it is easily shown that @xmath183 is zero at @xmath184 , @xmath185 , and @xmath186 on @xmath187 and @xmath188 on @xmath189 . hence by induction on @xmath55 the sum in the above equation is equal to @xmath190 . this proves the conclusion for @xmath172 and @xmath179 . suppose @xmath191 . in this case , the proof also proceeds in a similar way . we transform the matrix ( [ equ : cablematrix ] ) to a block sum of matrices in ( [ equ : reducedmatrix ] ) and @xmath192 by applying all of the four congruence relations shown before . the signature of ( [ equ : hyperbolicmatrix ] ) is zero since @xmath193 is nonsingular . hence we have @xmath194 . on the other hand , @xmath195 for @xmath177 , and @xmath196 for @xmath179 . this completes the proof . now we are ready to prove the main theorem of this section . if @xmath191 and @xmath123 is nonsingular , @xmath197 is the zero function by lemma [ lem : alg - sign - formula ] , and so @xmath198 for all @xmath24 . suppose @xmath172 . by lemma [ lem : sign - local - const ] , we can choose @xmath199 such that @xmath200 is constant on each of @xmath201 , @xmath202 $ ] , @xmath203 and @xmath204 $ ] . we may assume that @xmath205 and @xmath206 are irrational . for @xmath177 , by lemma [ lem : alg - sign - formula ] @xmath207 for @xmath179 , if @xmath208 then the conclusion is trivial since @xmath209 for any @xmath22 . if not , we have @xmath210 by lemma [ lem : alg - sign - formula ] . if @xmath211 for any @xmath63 , we may assume that @xmath212 is constant near @xmath24 , and the theorem is proved from the above equation . if @xmath213 for some @xmath63 that is not a multiple of @xmath55 , then @xmath214 since @xmath215 for all sufficiently small @xmath216 . if @xmath170 , we have @xmath217 , as in the proof of lemma [ lem : alg - sign - formula ] . we remark that a similar argument shows an analogous result for alexander polynomials defined by @xmath218 : if @xmath123 is nonsingular , then @xmath219 up to units of @xmath220 $ ] . this result is mentioned and used in @xcite . we finish this section with a remark on the matrix cobordism groups . consider the set of rational matrices @xmath22 such that @xmath123 is nonsingular . define the block sum operation and cobordisms of such matrices as in @xcite . then arguments of @xcite show that the cobordism of such matrices is an equivalence relation and that the set @xmath221 of cobordism classes becomes an abelian group under the block sum . then @xmath222 induces a group homomorphism on @xmath223 and both @xmath28 and @xmath222 are well - defined on @xmath223 . thus lemma [ lem : alg - sign - formula ] also makes sense when @xmath22 is regarded as its cobordism class . in this section , we study links and their seifert surfaces in rational homology spheres . we start with a review of rational - valued linking numbers . let @xmath224 , @xmath225 be disjoint @xmath226 , @xmath227-cycles respectively in a @xmath2-rational homology sphere @xmath3 where @xmath228 . since @xmath229 , @xmath230 are torsion , there are @xmath231 , @xmath232-chains @xmath233 , @xmath234 in @xmath3 such that @xmath235 for some nonzero integers @xmath135 , @xmath236 . the _ linking number _ of @xmath224 and @xmath225 is defined to be @xmath237 where @xmath238 denotes the intersection number in @xmath3 . the linking number @xmath239 is determined by @xmath224 and @xmath225 and is independent of the choices of @xmath135 , @xmath236 , @xmath233 and @xmath234 . when the ambient space is obvious , we will denote the linking number by @xmath240 . we remark that @xmath241 is well - defined only for disjoint cycles , and its value modulo @xmath242 is well - defined for homology classes . the following properties of the linking number are easily checked . first , if @xmath243 for some @xmath244 and @xmath231-chain @xmath157 disjoint to @xmath225 , then @xmath245 assumes the same value on @xmath246 . @xmath247 has the same property . second , suppose that @xmath3 is a boundary component of an @xmath8-manifold @xmath248 such that @xmath249 . if @xmath224 , @xmath225 are as above and @xmath233 , @xmath234 are relative @xmath231 , @xmath232-cycles of @xmath250 such that @xmath251 for some @xmath252 , @xmath253 . we give a formula for linking numbers in a rational homology 3-sphere described by a kirby diagram . for an @xmath30-component framed link , we define the linking matrix @xmath22 to be a square matrix of dimension @xmath30 whose @xmath254-entry is the linking number of the @xmath16-th component and the @xmath255-th longitude with respect to the framing . [ thm : lk - kirby - diagram - dim1 ] let @xmath256 be a framed link in @xmath257 such that the result of surgery on @xmath257 along @xmath1 is a rational homology sphere @xmath3 . let @xmath258 be the linking matrix of @xmath1 . for disjoint @xmath259-cycles @xmath236 in @xmath260 , we have @xmath261 where @xmath262 , @xmath263 are column vectors with @xmath264 , @xmath265 , respectively . let @xmath266 be a meridian curve on a tubular neighborhood of @xmath267 . let @xmath233 and @xmath268 be 2-chains in @xmath257 that are transverse to @xmath1 and bounded by @xmath236 and a 0-linking longitude of @xmath267 , respectively . let @xmath269 and @xmath270 be chains obtained by removing from @xmath233 and @xmath268 the intersection with an open tubular neighborhood of @xmath1 , respectively . let @xmath271 be the core of the 2-handle attached along @xmath267 . let @xmath272 be a chain in @xmath3 , where @xmath63 , @xmath273 are integers to be specified later . then @xmath274 is homologous to @xmath275 . thus if @xmath276 satisfies @xmath277 , then @xmath274 is homologous to @xmath278 . since @xmath3 is a @xmath7-homology sphere , @xmath22 is invertible over @xmath7 . so we can take as @xmath63 a nonzero common multiple of denominators of entries of @xmath279 , and @xmath280 . now we have @xmath281 now we turn our attention to links . the main goal of this section is to study the existence of a generalized seifert surface of a given type @xmath34 for links . let @xmath1 be an @xmath0-link in a rational homology sphere , and suppose @xmath282 . @xmath1 can be viewed as a framed manifold since all framings are equivalent . generally , for a framed submanifold @xmath283 of codimension 2 in a manifold @xmath284 , we identify a tubular neighborhood of @xmath283 with @xmath285 with respect to the framing , and in particular identify @xmath286 with a submanifold in the boundary of the exterior @xmath287 . we say @xmath288 is _ primitive _ if there is a homomorphism of @xmath289 whose composition with @xmath290 is the same as the projection @xmath291 . when @xmath284 is obvious , we say @xmath283 is primitive . by a standard use of obstruction theory , transversality and thom - pontryagin construction , @xmath1 is primitive if and only if @xmath1 bounds a seifert surface . our main observation is that certain unions of parallel copies of @xmath1 is primitive and hence bound seifert surfaces . we use the following notations to denote parallel copies of a framed submanifold @xmath283 in @xmath284 . for an @xmath115-tuple @xmath116 where @xmath292 , let @xmath55 be the sum of @xmath54 as before , and @xmath293 be the union of @xmath115 parallel copies of @xmath283 in @xmath284 with respect to the framing where the @xmath16-th copy is oriented according to the sign of @xmath54 . for a nonzero integer @xmath115 , let @xmath294 where @xmath295 is the @xmath115-tuple @xmath296 . we assert that @xmath53 is primitive if and only if @xmath55 is a multiple of @xmath35 , where @xmath35 is a positive integer given as follows . let @xmath297 be the homomorphism sending each meridian to 1 , which is uniquely determined by the alexander duality . let @xmath26 be the quotient group of @xmath298 by its torsion subgroup . since @xmath299 , we can view @xmath26 as a subgroup of @xmath300 . choose a basis @xmath301 of @xmath26 , and let @xmath35 be the positive least common multiple of denominators of the reduced fractional expression of @xmath302 . we call @xmath35 be the _ complexity _ of @xmath1 . @xmath35 is independent to the choice of the basis . note that @xmath303 and hence a homomorphism @xmath23 from @xmath26 into @xmath242 is induced by @xmath304 . suppose that @xmath53 is primitive for @xmath116 . assuming that @xmath53 is the product of @xmath1 and @xmath115 points in the tubular neighborhood @xmath305 , we can view @xmath306 as a subspace of @xmath307 . there is a homomorphism of @xmath308 into @xmath242 sending each meridian of @xmath53 to 1 , and it induces a homomorphism @xmath309 of @xmath310 to @xmath242 sending each meridian of @xmath1 to @xmath55 . the homomorphisms @xmath311 and @xmath312 on @xmath313 are equal since they assume the same value on meridians . hence @xmath314 and so @xmath55 is a multiple of the denominator of @xmath302 . this shows that @xmath55 is a multiple of @xmath35 . conversely , suppose @xmath315 for some integer @xmath63 . then the homomorphism @xmath316 sends each meridian of @xmath1 to @xmath55 , and it extends to a homomorphism of @xmath308 to @xmath242 sending each meridian of @xmath53 to 1 . this proves the assertion . for a type @xmath317 for @xmath1 , let @xmath318 be the union of @xmath32 parallel copies ( negatively oriented if @xmath319 ) of the @xmath16-th component of @xmath1 over all @xmath16 . we call the complexity of @xmath318 the _ complexity of type @xmath34 for @xmath1 _ and denote it by @xmath56 . a generalized seifert surface of type @xmath34 with complexity @xmath115 for @xmath1 is a seifert surface of @xmath320 , which exists if and only if @xmath321 is primitive . this holds exactly when @xmath115 is a multiple of @xmath56 . in particular , [ thm : seifert - surface - existence - highdim ] for @xmath282 , an @xmath0-link @xmath1 admits a generalized seifert surface of any type @xmath34 . [ thm : compute - complexity ] fix a basis of @xmath310 , and let @xmath322 be a matrix whose @xmath16-th column represents the @xmath16-th meridian of @xmath1 with respect to the basis . then @xmath56 is equal to the least common multiple of denominators ( of a reduced fractional expression ) of entries of the row vector @xmath323 . ( we view @xmath34 as a column vector . ) first of all , @xmath322 is invertible since meridians are independent . consider the homomorphism of @xmath324 to @xmath7 sending meridians to 1 . viewing @xmath306 as a subspace of @xmath325 , a homomorphism of @xmath300 to @xmath7 sending the @xmath16-th meridian of @xmath1 to @xmath32 is induced . suppose that the @xmath16-th basis of @xmath26 is sent to @xmath326 under the induced homomorphism . by the definition of the complexity , the least common multiple of the denominators of @xmath326 is equal to @xmath56 . the row vector @xmath262 satisfies @xmath327 , and so @xmath328 as desired . now we consider 1-links in a rational homology sphere . our result for 1-links is expressed in terms of a linking matrix of @xmath1 ( for any choice of a framing ) as follows . we say a 1-link _ @xmath1 admits type @xmath34 _ if there is a generalized seifert surface of type @xmath34 for @xmath1 . [ thm : seifert - surface - existence - dim1 ] a 1-link @xmath1 admits type @xmath34 if and only if for a linking matrix @xmath258 with respect to a ( or any ) framing , @xmath329 is an integer for all @xmath16 . first we will investigate when a generalized seifert surface inducing a given fixed framing exists , and then show theorem [ thm : seifert - surface - existence - dim1 ] . let @xmath1 be a 1-link , @xmath38 be a framing on @xmath1 and @xmath258 be the linking matrix with respect to @xmath38 . [ lem : seifert - surface - existence - framed ] there exists a generalized seifert surface of type @xmath317 for @xmath1 that induces the same framing on its boundary as the framing induced by @xmath38 if and only if @xmath330 for all @xmath16 . let @xmath318 be the link obtained by taking @xmath32 parallel copies of the @xmath16-th components of @xmath1 with respect to the given framing @xmath38 . @xmath1 admits a desired generalized seifert surface of complexity @xmath115 if and only if @xmath321 is primitive . since @xmath324 is generated by meridians of @xmath318 , for any multiple @xmath115 of the complexity of @xmath318 we obtain a homomorphism @xmath23 of @xmath331 to @xmath242 sending each meridian of @xmath320 to 1 , by the arguments for higher dimensional links . in order to show that @xmath332 is primitive , we need to check whether the longitudes of @xmath332 is sent to zero by @xmath23 , or equivalently , the longitudes of @xmath1 is sent to zero by the unique homomorphism from @xmath333 sending meridians @xmath334 of @xmath1 to @xmath335 . since the @xmath16-th longitude of @xmath1 is equal to @xmath336 in @xmath300 , the conclusion follows . let @xmath22 be a linking matrix of a 1-link @xmath1 with respect to a fixed framing @xmath38 . then the @xmath16-th longitude with respect to an arbitrary framing @xmath337 is homologous to the sum of the @xmath16-th longitude with respect to @xmath38 and a multiple of the @xmath16-th meridian in the complement of @xmath1 . hence the linking matrix with respect to @xmath337 is of the form @xmath338 where @xmath339 is a diagonal matrix of integers . conversely , @xmath339 determines a framing since framings are in 1 - 1 correspondence with elements of @xmath340 . therefore , by lemma [ lem : seifert - surface - existence - framed ] , there is a generalized seifert surface of type @xmath34 if and only if @xmath341 for some diagonal matrix @xmath339 of integers , or equivalently , @xmath342 is an integer for all @xmath16 . we remark that generalized seifert surfaces of the same type induce the same framing , since in the above proof the numbers @xmath343 are determined by @xmath317 . from now on , for a 1-link @xmath1 and a type @xmath34 admitted by @xmath1 , we denote by @xmath318 the union of @xmath32 parallel copies of the @xmath16-th component of @xmath1 with respect the framing determined by @xmath34 . if a linking matrix of a link @xmath1 is zero , then any type is admitted . \2 . if @xmath1 is a link in a homology sphere , each entry of a linking matrix is an integer . hence type @xmath50 is admitted . let @xmath344 be an @xmath30-component unlink in @xmath257 and @xmath1 be the union of meridian curves of components of @xmath344 . a rational homology sphere @xmath3 is obtained by @xmath0-surgery on each component of @xmath344 for @xmath345 , and @xmath1 can be viewed as a link in @xmath3 . by theorem [ thm : lk - kirby - diagram - dim1 ] , the diagonal matrix with diagonals @xmath346 is a linking matrix , and by theorem [ thm : seifert - surface - existence - dim1 ] , @xmath1 does not admit any type . \4 . for any given type @xmath347 , there is a link that admits only type @xmath348 . for , suppose that a link @xmath1 has linking matrix @xmath349 . if a type @xmath350 is admitted by @xmath1 , @xmath351 and @xmath352 are integers by theorem [ thm : seifert - surface - existence - dim1 ] . this implies that @xmath353 . by the following lemma that classifies linking matrices , such a link @xmath1 exists . for any symmetric square matrix @xmath22 with rational entries , there is a framed 1-link in a rational homology sphere whose linking matrix is @xmath22 . we claim that for any symmetric nonsingular @xmath30 by @xmath30 matrix @xmath354 and @xmath30 by @xmath0 matrix @xmath355 with integral entries , there is an @xmath0-component framed 1-link in a rational homology sphere with linking matrix @xmath356 . consider the @xmath0-component standard unlink @xmath1 in @xmath257 with zero - linking framing . we can construct an @xmath30-component framed link @xmath357 in @xmath257 with linking matrix @xmath358 such that @xmath1 and @xmath357 are disjoint and the linking number of the @xmath16-th component @xmath357 and the @xmath255-th component of @xmath1 is @xmath359 . the result of surgery along @xmath357 is a rational homology 3-sphere @xmath3 since @xmath358 is nonsingular . @xmath1 can be viewed as a link in @xmath3 , and the linking matrix of @xmath1 in @xmath3 is @xmath356 , by theorem [ thm : lk - kirby - diagram - dim1 ] . any symmetric matrix @xmath22 over @xmath7 can be written as @xmath360 where @xmath339 is a nonsingular diagonal matrix and each @xmath361 represents a zero matrix of suitable size . choose a nonzero common multiple @xmath0 of denominators of entries of @xmath26 and @xmath362 and let @xmath363 , @xmath364 . then @xmath322 and @xmath354 are matrices over @xmath242 , and we have @xmath365 . by the claim , @xmath22 is realized as a linking matrix of a 1-link in a rational homology sphere . we finish our discussion on 1-links by showing that the set of admitted types of a link is invariant under link concordance . for two @xmath7-concordant 1-links , a framing on a concordance between the links induces framings on the links , and in fact this is a 1 - 1 correspondence between the framings of the two links . if 1-links @xmath44 and @xmath18 are @xmath7-concordant , then @xmath44 admits type @xmath34 if and only if so does @xmath18 . furthermore , @xmath34 determines corresponding framings on @xmath44 and @xmath18 . since linking numbers are @xmath7-concordance invariants , linking matrices of @xmath44 and @xmath18 with respect to corresponding framings are the same . hence by theorem [ thm : seifert - surface - existence - dim1 ] , the conclusion is proved . we remark that our complexity is motivated from a similar notion in @xcite . the complexity in @xcite is similar to @xmath56 for @xmath366 . in general the former is a multiple of the latter . from now on , we will consider only odd dimensional links . let @xmath367 be a @xmath41-link and @xmath33 be a generalized seifert surface of type @xmath34 . define the _ seifert pairing _ @xmath368 by @xmath369 , [ y])=\lk(x^+ , y)$ ] for two @xmath227-cycles @xmath224 , @xmath225 in @xmath33 where @xmath370 denotes the cycle obtained by pushing @xmath224 slightly along the positive normal direction of @xmath33 . it is easily verified that @xmath371 is well - defined . choosing a basis of @xmath372 , a matrix @xmath22 over @xmath7 is associated to @xmath371 . @xmath22 is called a _ seifert matrix _ of @xmath33 . we remark that for @xmath373 , another convention is found in some literature where seifert pairings and seifert matrices are defined on the cokernel of @xmath374 . these different conventions give different results in the case of links ( e.g. consider a full - twisted annulus with hopf link boundary in @xmath257 ) . we remark that if @xmath22 is a seifert matrix of @xmath33 , the matrix @xmath121 is a seifert matrix of @xmath375 . for @xmath75 , we can view a seifert matrix @xmath22 as a representative of an element @xmath376 $ ] of the matrix cobordism group @xmath221 , since @xmath123 represents the intersection pairing on @xmath372 that is non - degenerated . the following is an analogous result of the well - known result of @xcite . recall that a @xmath7-concordance between two links is called primitive if there is a homomorphism of the first homology of its exterior into @xmath242 that sends each meridian and each longitude to @xmath259 and @xmath361 , respectively . [ lem : prim - conc - matrix ] let @xmath377 and @xmath378 be seifert matrices of seifert surfaces for two @xmath41-links @xmath379 and @xmath380 that are @xmath7-concordant via a primitive @xmath7-concordance . if @xmath75 , then @xmath381 = [ a_1]$ ] in @xmath221 . in this case , everything is primitive and an argument in @xcite can be used to prove the conclusion . details are as follows . let @xmath382 be a primitive @xmath7-concordance between @xmath379 and @xmath380 , and @xmath383 be the exterior of @xmath14 in @xmath9 . we may assume @xmath384 and @xmath385 . for @xmath386 , let @xmath387 be a seifert surface of @xmath64 where @xmath388 is defined . let @xmath344 be a parallel copy of @xmath14 in @xmath389 bounded by @xmath390 . let @xmath391 , a closed oriented submanifold of @xmath392 of codimension 1 . define a map @xmath393 by a thom - pontryagin construction for @xmath394 . since @xmath14 is primitive , there is a homomorphism @xmath395 sending each meridian to 1 . let @xmath396 denote the map induced by the inclusion . then @xmath397 and @xmath398 assume the same value on each meridian and so @xmath399 since @xmath242 is torsion free . therefore @xmath309 is extended to a map @xmath400 . by a transversality argument , we construct a @xmath401-submanifold @xmath99 of @xmath402 such that @xmath403 . since @xmath75 , @xmath404 . let @xmath371 be the bilinear pairing on @xmath405 represented by the block sum of @xmath377 and @xmath378 . @xmath406 is a half - dimensional subspace of @xmath372 by a duality argument , and @xmath371 vanishes on @xmath407 by the previous remarks on the linking number as in @xcite . therefore @xmath377 and @xmath378 are cobordant matrices . we note that the above proof does not work in the case of @xmath373 , since @xmath405 is not isomorphic to @xmath408 . [ thm : conc - matrix ] let @xmath378 , @xmath409 be seifert matrices of two @xmath7-concordant @xmath41-links with @xmath75 . then @xmath410 = i^q_r[a_1]$ ] in @xmath223 for some integer @xmath98 . let @xmath14 be a @xmath7-concordance between two links whose seifert matrices are @xmath378 and @xmath409 . by the same argument as that for links , there is a nonzero integer @xmath115 ( the `` complexity '' of @xmath14 ) such that @xmath411 is a primitive . @xmath411 is a @xmath7-concordance of two links with seifert matrices @xmath412 and @xmath413 . by lemma [ lem : prim - conc - matrix ] , @xmath414 = [ i^q_ra_1]$ ] . in this section we prove the invariance of signatures under link concordance . in the case of @xmath75 , seifert matrices can be viewed as an element of @xmath223 and the invariance of signatures may be shown directly using theorem [ thm : conc - matrix ] . because this does not work for @xmath373 , we will interpret signatures via finite branched covers following ideas of @xcite to show the invariance of signatures in any odd dimensions . suppose that @xmath49 is a positive integer , @xmath288 is a primitive pair , and @xmath415 . then a homomorphism @xmath416 sending each meridian to 1 exists uniquely , and the composition of @xmath23 and the canonical projection @xmath417 determines a @xmath49-fold cover @xmath418 of @xmath284 branched along @xmath283 with a transformation @xmath419 on @xmath418 corresponding to @xmath420 such that @xmath421 . when @xmath284 is even dimensional , define @xmath422 to be the signature of the restriction of the intersection form of @xmath418 on the @xmath423-eigenspace of the induced homomorphism @xmath424 on the middle dimensional homology of @xmath418 with complex coefficients . put @xmath428 $ ] and identify @xmath3 with @xmath429 . the map @xmath430 given by primitiveness is extended to a map @xmath431 in an obvious way , and the union of it and a constant map @xmath432 is extended to a map @xmath433 . then @xmath283 is obtained by transversality at an interior point of @xmath434 . note that the above argument is similar to that of the well - known special case @xmath435 where @xmath436 can be taken as @xmath284 . hence it seems to be natural to take rational homology balls as @xmath284 in our general case , however it is not known even when @xmath3 is null - cobordant . this is why we allow boundary components of @xmath284 other than @xmath3 . we consider @xmath440 . the idea is that @xmath288 bounds a primitive pair @xmath441 so that the signature of the branched cover of @xmath284 vanishes . then by the additivity of signatures , the signatures of branched covers of @xmath442 and @xmath443 are the same and the proof is completed . this is similar to the well - known argument for the case of @xmath435 and @xmath444 . in that case , we can take @xmath445 and construct @xmath446 by the argument of the proof of the previous lemma . for our general case , we take @xmath447 $ ] and proceed in the same way . even though @xmath354 is not closed , all extra boundary components have vanishing middle dimensional rational homology and hence have no contributions to signatures . in ( * ? ? ? * chapter xii ) , kauffman shows this lemma when @xmath1 is a link in @xmath257 and @xmath449 is defined from @xmath451 where @xmath283 is a properly embedded surface in @xmath452 bounded by @xmath1 . in fact the same argument can be applied to show that if @xmath453 and @xmath283 is obtained by pushing into @xmath284 the interior of a seifert surface where @xmath22 is defined , then @xmath454 . we omit the details . the following result as well as lemma [ lem : prim - conc - matrix ] is based on the idea that most of the techniques of signature invariants for links in spheres also work for links in rational homology spheres provided everything is primitive . in fact , for @xmath75 , this result is a consequence of lemma [ lem : prim - conc - matrix ] . we use a geometric argument to prove this result in any odd dimension . [ lem : prim - conc - sign ] if @xmath377 and @xmath378 are seifert matrices defined on seifert surfaces of primitive @xmath41-links @xmath44 and @xmath18 that are @xmath7-concordant via a primitive @xmath7-concordance , @xmath455 for all @xmath24 . let @xmath456 be a primitive @xmath7-concordance between two links @xmath44 and @xmath18 . choose a primitive pair @xmath437 where @xmath457 can be defined . let @xmath458 . then @xmath283 is also primitive . we will compute @xmath459 from @xmath288 . note that for any sufficiently large prime @xmath226 , the pair @xmath456 can be considered as a concordance between links in @xmath460-homology spheres and in particular @xmath461 . for such a prime @xmath226 , let @xmath462 and @xmath463 be the @xmath226-fold branched covers of @xmath456 and @xmath379 , respectively . we need the following consequence of milnor s exact sequence for infinite cyclic coverings @xcite . suppose that @xmath464 be a @xmath226-fold cyclic cover of a @xmath465-complex @xmath466 induced by a map @xmath467 that factors through the projection @xmath468 . then if @xmath469 is zero , @xmath470 is also zero . in our case , @xmath471 is zero and so @xmath472 is also zero , for any large prime @xmath226 . from a mayer - vietoris sequence for @xmath473 , it follows that @xmath474 is zero . hence @xmath475 contributes nothing to the signature of the branched covering of @xmath288 , and @xmath476 by additivity of signatures . therefore @xmath477 and by lemma [ lem : sign - branch - cover ] , @xmath478 if @xmath479 for some large prime @xmath226 and some integer @xmath63 . since the set of such @xmath24 is dense in the real line , @xmath480 for all @xmath24 . now we are ready to show the invariance of the signature jump function under @xmath7-concordance . the key to handle the non - primitive case is to combine the geometric result for the primitive case with the reparametrization formula in section [ sec : matrix - sign - poly ] . we may similarly proceed as in the proof of theorem [ thm : conc - matrix ] . ( indeed , for the case of @xmath75 , the result is a consequence of theorem [ thm : conc - matrix ] . ) let @xmath44 and @xmath18 be @xmath7-concordant links admitting the given type @xmath34 . let @xmath377 and @xmath378 be seifert matrices of seifert surfaces of @xmath481 and @xmath482 , respectively . let @xmath483 be the union of @xmath32 parallel copies of the @xmath16-th component of a @xmath7-concordance between @xmath44 and @xmath18 . if @xmath373 , we need to take parallel copies with respect to the framing determined by @xmath34 . as before , we can choose a positive integer @xmath115 such that @xmath484 is primitive . then @xmath485 and @xmath486 are @xmath7-concordant via the primitive @xmath7-concordance @xmath487 . note that the links @xmath485 and @xmath488 are viewed as the boundaries of unions of parallel copies of seifert surfaces of the links @xmath481 and @xmath482 . in the case of @xmath373 , the framing used to take @xmath489 and the framing induced by the seifert surface of @xmath489 must be compatible in order for @xmath485 and @xmath490 to be @xmath7-concordant . this is where our framing condition in the definition of generalized seifert surfaces for 1-links is necessary . we remark that our framing condition for generalized seifert surfaces of 1-links seems to be the minimal requirement for signatures to be well - defined @xmath7-concordance invariants . for example , as an attempt to remove the framing condition , one may fix corresponding framings on @xmath494 for @xmath495 and consider @xmath496 taken with respect to the framings , where @xmath115 is a nonzero multiple of the complexity of concordance . it can be shown that they admit seifert surfaces , and are @xmath7-concordant via a primitive concordance . the signature jump functions of seifert matrices of @xmath497 are equal by lemma [ lem : prim - conc - sign ] . but the above argument can not be applied to prove that the reparametrizations @xmath498 of the signature jump functions are invariants of the original links @xmath494 and in fact they are not in general . we also remark that it is meaningful to consider types for higher dimensional links as well as 1-links . in fact the signature jump function of type @xmath34 for a link @xmath1 is no more than the signature jump function of type @xmath50 for the link @xmath318 obtained by taking parallel copies , however , more information on @xmath1 is obtained from types other than @xmath50 even for higher dimensional links in the sphere . for example , suppose @xmath499 is a knot that is not torsion in the ( algebraic ) knot concordance group ( so that @xmath500 is nontrivial @xcite ) and @xmath1 be a split union of @xmath499 and @xmath501 . then @xmath502 is trivial , but if @xmath503 , @xmath504 is nontrivial for some @xmath24 . choose a positive multiple @xmath115 of @xmath508 , and let @xmath22 be a seifert matrix for @xmath321 . then since @xmath121 is a seifert matrix of @xmath509 , @xmath510 by theorem [ thm : sign - formula ] . \4 . for @xmath57 , let @xmath60 be a @xmath401-ball in the ambient space @xmath61 of @xmath64 such that the connected sum @xmath4 is obtained by gluing @xmath515 along their boundaries . we view @xmath516 as a subspace of @xmath3 . let @xmath35 be a positive common multiple of @xmath517 and @xmath518 . then there is a generalized seifert surface @xmath387 of type @xmath34 with complexity @xmath35 for @xmath64 . we may assume that each component of @xmath519 is a @xmath520-disk whose boundary contains exactly one component of @xmath521 so that the intersection of @xmath522 and @xmath523 consists of disjoint disks , and their union is a generalized seifert surface @xmath33 of type @xmath34 with complexity @xmath35 for @xmath1 in @xmath3 . for @xmath373 , @xmath18 and @xmath68 are 1-knots by hypothesis . we need additional arguments since @xmath527 is not decomposed as above . for a manifold @xmath528 , let denote the cokernel of @xmath529 by @xmath530 . we assert that for any 1-knot and a generalized seifert surface @xmath528 of complexity @xmath35 , the seifert pairing on @xmath531 induces a well - defined `` seifert pairing '' on @xmath532 that induces the signature jump function of the knot . for , if @xmath224 is a boundary component of @xmath528 , a parallel copy of @xmath528 is a 2-chain whose boundary is homologous to @xmath533 in the complement of @xmath534 and so the linking number of @xmath224 and any 1-cycle on @xmath534 is zero . this proves the assertion . assume that @xmath387 is connected for @xmath57 . let @xmath535 be boundary components of @xmath536 . @xmath537 are ordered so that parallel cycles in @xmath538 are obtained by pushing @xmath326 and @xmath539 slightly along the negative and positive normal directions of @xmath33 , respectively . let @xmath540 be curves on @xmath541 such that the intersection number of @xmath542 and @xmath543 on @xmath33 is @xmath544 , where @xmath545 is the kronecker symbol ( see figure [ fig : csum - generator ] for a schematic picture ) . then @xmath546 is the direct sum of @xmath547 , @xmath548 and a free abelian group generated by @xmath549 and @xmath540 . we will show that the generators @xmath326 and @xmath542 have no contribution to the signature . similarly to the proof of the previous assertion , it is shown that the linking number of @xmath326 and any 1-cycle on the interior of @xmath550 is zero . furthermore , the value of the seifert pairing at @xmath326 and @xmath551 can be computed by counting intersections of the 1-cycle obtained by pushing @xmath542 slightly and the 2-chain obtained by attaching @xmath552 annuli bounded by @xmath553 ( @xmath554 ) to @xmath522 . by this argument we can verify that the seifert matrix on @xmath546 with respect to generators of @xmath555 and @xmath540 , @xmath549 is given by @xmath556\ ] ] where @xmath388 is a seifert matrix defined on @xmath555 and empty blocks represent zero matrices . the bottom - center block and middle - right block are nonsingular . in fact , subtracting the @xmath16-th row from the @xmath557-th row for each @xmath558 , the bottom - center block becomes an upper - triangular nonsingular matrix , and similarly for the right - middle block . hence the top - center block and middle - left block can be eliminated by row and column operations . this shows that @xmath22 is congruent to the block sum of @xmath378 , @xmath409 and a metabolic matrix . therefore @xmath376=[a_1]+[a_2]$ ] in the matrix cobordism group @xmath559 , and we have @xmath560 as desired . by theorem [ thm : link - sign - prop ] , the signature jump function of a link @xmath1 in an integral homology sphere always has the period @xmath77 since @xmath318 is primitive . however the period of the signature jump function of a link with nontrivial complexity in a rational homology sphere is not @xmath77 in general . in the following example , for any nonzero rational number @xmath115 we construct a knot whose signature jump function has the period @xmath561 . let @xmath499 be a knot in @xmath19 and @xmath22 be a seifert matrix defined on a seifert surface @xmath33 of @xmath499 . for @xmath562 , a knot @xmath563 with seifert matrix @xmath564 is obtained by fusing the link @xmath565 along @xmath566 bands whose interiors are disjoint to @xmath567 . ( we may take as @xmath563 a @xmath568-cable knot with companion @xmath499 for @xmath373 . ) let @xmath371 be an embedded @xmath41-sphere with knot type @xmath563 and @xmath107 be an unknotted @xmath41-sphere that bounds a ball disjoint to @xmath371 in @xmath19 . let @xmath344 be a simple closed curve in @xmath569 whose linking numbers with @xmath371 and @xmath107 are @xmath0 and @xmath570 , respectively , as shown in figure [ fig : period - knot ] . for @xmath571 , the result of surgery along @xmath371 and @xmath344 ( with respect to zero - framing if @xmath373 ) is a rational homology sphere @xmath3 , and @xmath107 can be considered as a knot in @xmath3 . @xmath572 is an infinite cyclic group generated by a meridian @xmath233 of @xmath371 . the meridian @xmath573 of @xmath107 is given by @xmath574 in @xmath572 . hence @xmath575 by theorem [ thm : compute - complexity ] . in fact , there is a submanifold in @xmath19 bounded by the union of a parallel of @xmath371 and @xmath576 as in figure [ fig : period - surface ] , and a generalized seifert surface @xmath528 with complexity @xmath0 for @xmath107 is obtained by gluing a @xmath520-disk in @xmath3 to the submanifold along the parallel of @xmath371 . for @xmath577 , @xmath564 is a seifert matrix of @xmath528 . for @xmath373 , the block sum of @xmath564 and a zero matrix of dimension @xmath578 is a seifert matrix of @xmath528 . in any case , we have @xmath579 . as a special case of this example , there are knots in rational homology spheres whose signature jump functions do not have the period @xmath77 . such knots are not @xmath7-concordant to knots in ( homology ) spheres , by theorem [ thm : link - sign - prop ] . this illustrates that the @xmath7-concordance theory of links in rational homology spheres is not reduced to that of links in spheres . choose infinitely many distinct primes @xmath580 greater than 7 . for odd @xmath227 , let @xmath581 and for even @xmath227 , let @xmath582 it is easily checked that @xmath512 is nonzero if and only if @xmath583 for some integer @xmath63 , where @xmath584 for even @xmath227 or @xmath585 for odd @xmath227 . since @xmath22 is a seifert matrix of a knot in @xmath19 by @xcite , a @xmath41-knot @xmath267 in a rational homology sphere such that @xmath586 is constructed by the above example . ( for @xmath587 , we need to use the block sum of 8 copies of @xmath22 to satisfy the extra condition @xcite forced by a result of rochlin . since the signature jump function multiplies by 8 , this is irrelevant to our purpose . ) in order to show that @xmath267 s are independent , we consider a connected sum @xmath499 of @xmath588 over all @xmath16 , where all but finitely many @xmath589 are zero and @xmath590 for some @xmath255 . we will show that @xmath499 is not @xmath7-concordant to a knot in a integral homology sphere . this will complete the proof . by the additivity of signature jump function , @xmath591 . we claim that @xmath592 but @xmath593 . if a summand @xmath594 of @xmath595 is nonzero , then @xmath596 for some @xmath63 , or equivalently @xmath597 . this implies @xmath598 and @xmath599 . this shows the first claim . on the other hand , if a summand @xmath600 of @xmath601 is nonzero , then we have @xmath602 for some @xmath63 , and @xmath603 or @xmath604 . this is a contradiction , and the second claim is proved . from the claims , @xmath500 is not of period @xmath77 . by theorem [ thm : link - sign - prop ] , @xmath499 is not @xmath7-concordant to knots of period @xmath77 , in particular , to knots in integral homology spheres . another natural question is whether the homomorphism @xmath605 is injective . more generally , one may ask whether the natural map from the set of concordance classes of links in spheres into that of links in rational homology spheres is injective . we do not address these questions in this paper . we finish this paper with two applications illustrating that our results are also useful for links in spheres . one is a simple proof of the following result of litherland @xcite : let @xmath499 be a satellite knot with companion @xmath107 , that is , @xmath606 where @xmath107 is a 1-knot in a standard solid torus @xmath354 in @xmath257 and @xmath309 is a homeomorphism from @xmath354 onto a tubular neighborhood of a knot @xmath563 sending a 0-linking longitude on @xmath607 to a 0-linking longitude of @xmath563 . let @xmath115 be the winding number of @xmath107 in @xmath354 . then @xmath608 . for , if @xmath22 and @xmath322 are seifert matrices of @xmath107 and @xmath563 , then the block sum @xmath609 is a seifert matrix of @xmath499 for some tuple @xmath295 such that @xmath610 , and litherland s formula is proved by theorem [ thm : sign - formula ] . we remark that a formula for alexander polynomials of satellite knots @xcite can also be proved using an analogous result of theorem [ thm : sign - formula ] mentioned in section [ sec : matrix - sign - poly ] . another is a new proof of ( a weaker version of ) a result of kawauchi @xcite : if a @xmath41-knot @xmath499 is not ( algebraically if @xmath373 ) torsion , then @xmath611 is a boundary link that is never concordant to any split union for any @xmath98 . for , if @xmath612 is concordant to a split link , @xmath613 is concordant to a split union @xmath614 , and then by theorem [ thm : link - sign - inv ] and [ thm : link - sign - prop ] , @xmath615 . this implies that @xmath616 for all @xmath24 and @xmath499 is ( algebraically if @xmath373 ) torsion by a result of levine @xcite . r. litherland , _ signatures of iterated torus knots _ , topology of low - dimensional manifolds ( proc . second sussex conf . , chelwood gate , 1977 ) , pp . 7184 , lecture notes in math . , 722 , springer , berlin , 1979 .
the quantum fluctuation of the relative location of two-dimensional de sitter branes ( i.e. , of spacetime dimensions ) embedded in the-dimensional anti - de sitter bulk , which we shall call the quantum radion , is investigated at the linear perturbation level . the quantization of the radion is done by deriving the effective action of the radion . assuming the positive tension brane is our universe , the effect of the quantum radion is evaluated by using the effective einstein equations on the brane in which the radion contributes to the effective energy momentum tensor at the linear order of the radion amplitude . specifically , the rms effective energy density arising from the quantum radion is compared with the background energy density . it is found out that this ratio remains small for reasonable values of the parameters of the model even without introducing a stabilizing mechanism for radion , although the radion itself has a negative mass squared and is unstable . the reason behind this phenomenon is also discussed . * quantum radion on de sitter branes * uchida gen and misao sasaki +
gr-qc0201031
based on the idea of a brane - world suggested from string theory@xcite , randall and sundrum proposed an interesting scenario that we may live on either of two boundary 3-branes with positive and negative tensions in the 5-dimensional anti - de sitter space ( ads ) @xcite . one of the attractive features of the randall - sundrum ( rs ) scenario is that the gravity on the brane is confined within a short distance from the brane even for an infinitely large extra - dimension@xcite . this applies to the positive tension brane and it is because the ads bulk on both sides of the positive tension brane shrinks exponentially as one goes away from the brane . since the rs scenario gives an exciting , new picture of our universe , it is clearly important to study various aspects of this scenario to test it or to give constraints on its parameters . one good example is an analysis done by garriga and tanaka@xcite , in which they have shown that the radion in the original rs scenario acts as a brans - dicke scalar on the branes at the linear perturbation order and the effective gravity is that of a brans - dicke theory with positive and negative brans - dicke parameters on the positive and negative tension branes , respectively , where the values of the brans - dicke parameters are determined by the distance between the two branes . in a previous paper , we have shown that essentially the same situation arises in the case of two de sitter ( ds ) branes embedded in the ads bulk . however , we have also shown that the radion effectively has a negative mass squared with its absolute value proportional to the curvature of the ds brane , hence is unstable if it can fluctuate by itself without the matter energy momentum tensor . the phenomenon we shall study in this paper is the quantum fluctuation of this mode , called the quantum radion . to make clear what we mean by the quantum radion , let us describe the mode in more detail . our brane universe may be displaced from the 0th order trajectory of a homogeneous and isotropic brane . by appropriately fixing the coordinate gauge , the displacement perpendicular to the brane can be described by a scalar function on the brane . assuming there are two branes that are fixed points of the @xmath5 symmetry , it can be shown that only the relative displacement of the two branes is physical , which we call the radion . there are two distinctively different kinds of displacement of a brane : the `` bend '' and `` fluctuation '' . the bend is a type of displacement due to inhomogeneities of the matter energy - momentum tensor on the brane . the trace of the energy - momentum acts as an additional tension of the brane , and the brane must `` bend '' accordingly . the relative bend is described by the mode of radion that couples with the source on the branes , and it can be written as a functional of the energy - momentum tensor@xcite . in the context of the quasi - localized gravity discussed by gregory , rubakov & sibiryakov@xcite , the role of this type of radion has been extensively studied@xcite . on the other hand , the `` fluctuation '' is a type of displacement that is purely geometrical , which obeys a free wave equation without source . the relative displacement of this kind is the mode of radion which we shall discuss in this paper . this mode of radion was studied first by charmousis , gregory and rubakov for the rs branes whose effective radion mass is zero by solving the field equations for the rs branes @xcite and by chacko and fox@xcite for the ds and ads branes whose effective radion mass squared are negative and positive , respectively . the fact that the radion mass squared is negative ( or zero ) suggests the ( marginal ) instability of the two brane system . in fact , in the case of the original rs flat two brane model@xcite , the negativity of the brans - dicke parameter on the negative tension brane@xcite can be regarded as a result of this marginal instability . to recover the stability , goldberger & wise introduced a bulk scalar field@xcite that couples to the branes in such a way that the distance between the two branes is stabilized . however , it should be also noted that the effective brans - dicke parameter on the positive tension brane is positive and it can be large enough to be consistent with experiments for the separation of the branes larger than the ads curvature radius at least at the linear perturbation order@xcite . hence a stabilization mechanism may be unnecessary if we live on the positive tension brane . we therefore do not introduce a stabilization mechanism . we shall work on the system that consists of an @xmath2-dimensional ads bulk spacetime bounded by two branes of constant curvature that are fixed points of the @xmath5-symmetry . the zero - curvature branes correspond to the flat rs branes@xcite , while the positive - constant curvature branes correspond to the ds branes@xcite . our main concern is of course the ds brane case , but we treat the flat brane case simultaneously to make clear the similarities and differences between the two cases . the ds brane case is of particular interest because it gives a good model of braneworld inflation . in the standard 4-dimensional inflation , the quantum vacuum fluctuations play a very important role . it is therefore natural to ask if the quantum radion fluctuations play an important role , if not disastrous , in the braneworld inflation . it should be mentioned that historically a very similar situation was analyzed by garriga and vilenkin @xcite in which they considered the fluctuations of a thin domain wall in @xmath6 spacetime dimensions . although they assumed the minkowski background , many of the results obtained there apply equally to the present case . in particular , they showed that the wall fluctuation mode is represented by a scalar field living on the @xmath7-dimensional de sitter space which describes the internal metric on the domain wall , and the scalar field has the negative mass squared @xmath8 . the paper is organized as follows . in sec . [ sec : bg ] , we describe the background spacetime and our notation . in sec . [ sec : pert ] , we solve the perturbation equation in the bulk that describes the radion mode . we find there is a gauge degree of freedom that should be carefully treated in the ds brane case in contrast to the flat brane case where no such subtlety arises . in sec . [ sec : qradion ] , assuming the dependence on the extra dimensional coordinate that solves the perturbation equation in the bulk , we derive the effective action for the radion and quantize it . in sec . [ sec : omega ] , based on the result obtained in sec . [ sec : qradion ] , we evaluate the effective energy density of the radion on the brane which is present at the linear order in the radion amplitude , and estimate its effect by calculating the rms value . we find the effect remains small for reasonable values of the model parameters even though the radion itself is unstable . in sec . [ sec : summary ] , we summarize our results and discuss the implications . first , we summarize the basic equations of the system and explain our notation . the system we consider is a @xmath5 symmetric @xmath2-dimensional bulk spacetime with two @xmath0-dimensional branes as the fixed points of the symmetry . the bulk metric @xmath9 obeys the @xmath2-dimensional einstein equation : @xmath10 where @xmath11 is the @xmath2-dimensional gravitational constant , @xmath12 is the @xmath2-dimensional cosmological constant which we assume to be negative , and @xmath13 is localized on the branes . we use the latin indices for @xmath2-dimensional tensor fields in the bulk and the greek indices for @xmath1-dimensional tensor fields on the brane . denoting the @xmath1-dimensional metric on the brane by @xmath14 , we decompose the localized energy momentum tensor @xmath13 into the tension part @xmath15 and the matter part @xmath16 . then eq . ( [ bulkein ] ) reduces to @xmath17 on the brane@xcite , where @xmath18 is a tensor field quadratic in @xmath16 , and @xmath19 is the projected @xmath2-dimensional weyl tensor defined by @xmath20 with @xmath21 being the unit vector normal to the brane . the constants @xmath22 and @xmath23 are related to the basic constants of the systems as @xmath24 it may be noted that although the decomposition of @xmath13 is not unique , there is no arbitrariness in the metric on the brane . we consider branes with @xmath25 . looking at eq . ( [ branein ] ) , we see that the @xmath19 term can be considered as the effective energy - momentum tensor on the brane induced by the bulk gravitational field , @xmath26 the background we consider is the rs brane system and the ds brane system . to be more precise , the bulk is the @xmath2-dimensional anti - de sitter spacetime ( ads@xmath27 ) whose metric may be written as @xmath28\ , ; \qquad b(z)=\frac{\ell\sqrt{\cal k}}{\sinh \sqrt{\cal k}z}\ , , \quad \ell=\left[-\frac{n(n-1)}{2\lambda_{n+1}}\right]^{1/2}\end{aligned}\ ] ] where @xmath29 is the metric of a lorentzian @xmath30dimensional constant curvature space with curvature @xmath31 or @xmath32 , and @xmath33 is the curvature radius of ads@xmath27 . the function @xmath34 is called the warp factor . the two branes are placed at the coordinates @xmath35 and @xmath36 ( @xmath37 ) with their tensions and cosmological constants given by @xmath38 which satisfy the effective einstein equations with @xmath39 and @xmath40 . the choice @xmath31 corresponds to the rs brane system while @xmath32 corresponds to the ds brane system . note that the @xmath1-dimensional gravitational constant @xmath23 on the positive tension ds brane is related to that on the positive tension rs brane , say @xmath41 , as @xmath42 . although we perform most of our calculation in the coordinate system with @xmath43 , sometimes it is more convenient to use the coordinate system with the proper distance coordinate @xmath44 defined by @xmath45 it is also useful to introduce the rate of the change in the warp factor @xmath34 in the @xmath44 coordinate , which we shall denote by @xmath46 , @xmath47 in this section , we consider the gravitational perturbation on the background described in the previous section . our analysis is a generalization and a reformulation of @xcite which discussed the rs brane system @xmath48 . the 4-dimensional ds and ads brane system was discussed in @xcite . we denote the gravitational perturbation by @xmath49 to the background metric @xmath50 , i.e. , @xmath51 where @xmath9 is the full metric . we choose the rs gauge with respect to the positive or negative tension branes and denote the gravitational perturbation in this gauge by @xmath52}$ ] and the coordinates by @xmath53},x^\mu_{[\pm]}\}$ ] . the rs gauge condition is @xmath54}=h_{5\mu}^{[\pm]}= h^{[\pm]\mu}_{\ \ \ \ \mu}=h^{[\pm]\,;\nu}_{\mu\nu}=0\ , . \label{rsgaugedef}\end{aligned}\ ] ] in the rs gauge , the einstein equations ( [ bulkein ] ) simplify to @xmath55 \frac{h_{\mu\nu}^{[\pm]}}{b^2}=0\ , . \label{eomhmnrs}\end{aligned}\ ] ] since @xmath56}n^a=0 $ ] , we may use eq . ( a23 ) of @xcite to obtain the boundary condition on @xmath56}$ ] . expressing the perturbed locations of the branes in terms of @xmath57}(x)$ ] as @xmath58}=r_{+}+\varphi^{[+]}(x ) \quad\mbox{and}\quad r_{[-]}=r_{-}+\varphi^{[-]}(x)\,,\end{aligned}\ ] ] where @xmath59 denote the positions of the background branes , we have @xmath60}}{b^2}\right]= \frac{2}{b}{{\cal l}_{\mu\nu}\varphi^{[\pm ] } } \qquad \mbox{at}\quad r_{[\pm]}=r_{\pm}\mp0\ , , \label{junccond}\end{aligned}\ ] ] where @xmath61 note that @xmath62 is set to zero as we are interested in the gravitational perturbation that does not couple with the matter on the brane . it should be noted here that eq . ( a23 ) of @xcite is defined from the side with the smaller value of @xmath44 , while in this paper the boundary condition at the negative tension brane is given from the larger value of @xmath44 ; @xmath63}=r_{-}+0 $ ] , hence the signature is reversed . in passing , we also note that eq . ( [ junccond ] ) may be derived by first introducing the gaussian normal coordinates with respect to the branes and transforming them to the rs gauge , as was done in @xcite . with the junction condition ( [ junccond ] ) at hand , we consider a particular solution of the form , @xmath64}_{\mu\nu } = b^{2}u(z){{\cal l}_{\mu\nu}\varphi^{[\pm]}}(x^\mu)\ , . \label{formhmn}\end{aligned}\ ] ] the traceless condition on the metric perturbation in the rs gauge demands @xmath57}$ ] to satisfy @xmath65\varphi^{[\pm]}=0\,,\end{aligned}\ ] ] which implies @xmath66{{\cal l}_{\mu\nu}\varphi^{[\pm]}}=0\ , . \label{varphidef}\end{aligned}\ ] ] then eq . ( [ eomhmnrs ] ) reduces to @xmath67 u(z)=0\,.\end{aligned}\ ] ] the general solution to this equation is obtained as follows . first , it is easy to see that @xmath68 is a solution . note here the identity @xmath69 . then the other independent solution , @xmath70 , may be found using the wronskian of the above equation , which is @xmath71 thus , the general solution of the form ( [ formhmn ] ) is @xmath72}_{\mu\nu}= b^2 \ell^{-1}\left(c^{[\pm ] } u^{(1)}(z ) + d^{[\pm ] } u^{(2)}_n(z)\right){{\cal l}_{\mu\nu}\varphi^{[\pm]}}\ , , \label{h+-mn}\end{aligned}\ ] ] where @xmath73}$ ] and @xmath74}$ ] are constants . note that there is an ambiguity in @xmath70 which depends on the choice of the integration constant . the change of it induces the change of the coefficient @xmath73}$ ] . this implies the solution @xmath75 represents a gauge degree of freedom . in fact , if we evaluate @xmath76}_{\mu\nu}$ ] , we obtain @xmath77}_{\mu\nu}=-\frac1{n-1}[{\partial}_z^2-{\cal k } ] \left(\frac{h^{[\pm]}_{\mu\nu}}{b^2}\right ) = -(b/\ell)^{2-n}\ell^{-1}d^{[\pm]}{{\cal l}_{\mu\nu}\varphi^{[\pm]}}\ , , \label{e+-mn}\end{aligned}\ ] ] which is independent of @xmath73}$ ] . note also that the @xmath43-dependence of @xmath19 is clearly consistent with the boundary condition at the branes with no source on it , @xmath78@xcite . since @xmath19 is a gauge - invariant tensor field , we see that the displacement of the branes should be related as @xmath79}\varphi^{[+]}=d^{[-]}\varphi^{[-]}\ , . \label{dvp = dvp}\end{aligned}\ ] ] with the above relation at hand , we can use a residual gauge degree of freedom in the rs gauge to bring the two different sets of the coordinates @xmath80}\}$ ] to the same coordinates , say @xmath81 . this is done by an infinitesimal coordinate transformation , @xmath82}-\alpha d^{[\pm]}\right ] \varphi^{[\pm]}\,,\qquad \hat{\xi}_{\mu}=\frac{1}{2\ell } \left[c^{[\pm]}-\alpha d^{[\pm]}\right]\varphi^{[\pm]}_{|\mu}\,,\end{aligned}\ ] ] where @xmath83 is the yet remaining gauge parameter and the function @xmath84 is defined as @xmath85^{-1 } = \frac{2}{\ell^2}\frac{\cosh\sqrt{\cal k}z-1}{\cal k } = \left\ { \begin{array}{cl } \displaystyle b^{-2 } & ( { \cal k}=0)\\ \displaystyle \frac2{\ell}(j-\ell^{-1 } ) & ( { \cal k}=1)\ , . \end{array } \right.\end{aligned}\ ] ] the metric perturbation in the unified coordinates @xmath81 is @xmath86}{{\cal l}_{\mu\nu}\varphi^{[\pm]}}\,,\end{aligned}\ ] ] and the branes are now placed at @xmath87}-\alpha d^{[\pm]}\right]\right ) \varphi^{[\pm]}\,.\end{aligned}\ ] ] we note that the positions of the two branes can not fluctuate independently in this unified gauge because of eq . ( [ dvp = dvp ] ) . this shows the nature of the radion that it describes the relative displacement of the branes . with the above form of the metric perturbation and the locations of the branes , we find that the boundary condition on @xmath49 gives @xmath88}=-\frac{2\ell } { \alpha{\cal k}\ell+b_\pm{\partial}_z u^{(2)}(z_\pm ) } \left(1+\frac{\cal k}{2}\left[c^{[\pm]}-\alpha d^{[\pm]}\right]\right)\ , , \label{d+-}\end{aligned}\ ] ] which implies @xmath89 } + { b_{\pm}\partial_z u^{(2)}(z_{\pm})\over2\ell}d^{[\pm]}=0\ , . \label{cdrel}\end{aligned}\ ] ] an apparent complication in the case of @xmath32 is that this seems to constrain the values of @xmath73}$ ] which is just unphysical , gauge parameters . this can be resolved by considering the integration constant term in @xmath90 defined in eq . ( [ psi(2)def ] ) . for a change of @xmath73}$ ] as @xmath91}\to c^{[\pm]}+\beta\,,\end{aligned}\ ] ] we consider the simultaneous change of the integration constant term in @xmath90 as @xmath92}\beta)\,u^{(1)}\,,\end{aligned}\ ] ] which results in @xmath93}\beta { { \cal k}\ell\over b}\ , . \label{dzpchange}\end{aligned}\ ] ] this combined change of @xmath73}$ ] and the integration constant term in @xmath70 leaves eq . ( [ cdrel ] ) as well as @xmath94}_{\mu\nu}$ ] unchanged . one can also see from eq . ( [ dzpchange ] ) the reason for the simplification in the case of @xmath31 , i.e. , the change of @xmath90 does not affect @xmath95 if @xmath31 . introducing @xmath96 defined as @xmath97}\varphi^{[+]}\over2\ell } = -{d^{[-]}\varphi^{[-]}\over2\ell}\,,\end{aligned}\ ] ] the results are summarized as @xmath98 where @xmath99 and @xmath100 satisfies @xmath101{\tilde}\varphi(x^\mu)=0\ , . \label{eomvarphi}\end{aligned}\ ] ] as it should be clear by now , the parameter @xmath83 describes the residual freedom of the rs gauge in the unified coordinates . corresponding to the coordinate transformation , @xmath102 and @xmath103 with @xmath104 satisfying @xmath105 . given the initial data for @xmath106 , we obtain the full time evolution of the fluctuations of the brane by solving eq . ( [ eomvarphi ] ) . as the initial data , it is natural to assume that @xmath107 is classically null , and only the quantum vacuum fluctuations are present . to consider the quantum fluctuations of the brane , we need the action for @xmath106 in the first place . we obtain it by substituting the classical solution ( [ result ] ) into the original action of the system , but without constraining @xmath108 to satisfy eq . ( [ eomvarphi ] ) . this is because thus obtained action is already maximized except for the degree of freedom for the brane to fluctuate and hence is adequate for the quantization of such a degree of freedom@xcite . our system consists of identical bulk spacetimes @xmath109 and @xmath110 with branes @xmath111 and @xmath112 that are the boundary hypersurfaces of @xmath109 and @xmath113 with tensions of @xmath114 and @xmath115 , respectively ; @xmath116 and @xmath117 . the action of such a system is given by @xmath118\ , , \quad i_{k,1}:=\frac{1}{\kappa^2}\int_{{\partial}{\cal m}_1 } \hspace{-3.5ex}\sqrt{-q}\,d^n\!x\ , k\ , , \quad i_{\sigma\pm } : = -\int_{\sigma_\pm}\hspace{-2.5ex}\sqrt{-q}\ , d^n\!x\ , \sigma_{\pm}\ , , \label{basicaction}\end{aligned}\ ] ] where @xmath119 , @xmath120 and @xmath21 is a unit vector normal to the branes , pointing from @xmath109 to @xmath113 . the substitution of @xmath121 and the locations of the branes given by eq . ( [ result ] ) into the action ( [ basicaction ] ) , we find @xmath122 & = & \frac{(n-1)(n-2)\ell^{n-1}}{\kappa^2}n_{(n-2){\cal k } } \int d^n\!x \sqrt{-\gamma}\ , { \tilde}\varphi\bigl[\box_n+n{\cal k}\bigr]{\tilde}\varphi\ , , \label{pre - action}\end{aligned}\ ] ] where @xmath123.\end{aligned}\ ] ] the normalization factor @xmath124 is found to be identical to the one introduced in @xcite for the normalization of the radion mode of @xmath19 . details of the derivation and its comparison with those by chiba@xcite and by chacko and fox@xcite are given in appendix [ calaction ] . since @xmath125 , we may normalize @xmath106 as @xmath126 such that the action reduces to that of a scalar field @xmath127 with the mass squared @xmath128 on the positive tension brane : @xmath129 & = & \int d^n\ ! x \sqrt{-q_+}\ , \bigg [ -\frac12q_+^{\mu\nu}{\partial}_\mu\psi{\partial}_\nu\psi-\frac12m^2\psi^2 \bigg]\,,\label{actionpsi}\end{aligned}\ ] ] where @xmath130 is the intrinsic metric of the positive tension brane . here and below , we focus on the positive tension brane , but the parallel arguments apply to the negative tension brane except for the difference in the normalization factor . in terms of @xmath131 , we see from eq . ( [ result ] ) that @xmath19 is given by @xmath132 where @xmath133 here it may be appropriate to make a comment on the form of eq . ( [ emnphi ] ) in which the operator @xmath134 is replaced by its traceless part . in the previous section , we considered only the classical solution for the radion fluctuations , hence we were allowed to use the equation of motion to express @xmath19 in terms of the radion field , and the equation of motion for the radion was derived from the traceless condition of the rs gauge . when we consider the quantum radion , however , @xmath19 must be traceless at the off shell level as well . this means that the traceless condition of the rs gauge should not be used to express @xmath19 in terms of the radion field . instead , one has to go back to the original definition of the weyl tensor in terms of the metric perturbation whose trace may not be null . this is the reason for the appearance of the traceless part of @xmath135 in eq . ( [ emnphi ] ) . now that we have obtained the action for the fluctuation of the brane , which is of the form of the scalar field action , we may canonically quantize it in the ordinary way . adopting the flat coordinate chart on the brane , which is specified as @xmath136 \qquad \left ( a(\eta):=\frac{b_+}{(1-{\cal k})-{\cal k}\eta}\ , , \quad h_+:=\frac{{\partial}_\eta a}{a^2}={\cal k}b_+^{-1}\right)\,,\end{aligned}\ ] ] we obtain @xmath137,\end{aligned}\ ] ] where @xmath138 and @xmath139 satisfy the commutation relations , @xmath140=\delta(\bm{k}-\bm{k}')\ , , \quad [ \hat{a}_{\bm{k}},\hat{a}_{\bm{k } ' } ] = [ \hat{a}_{\bm{k}}^\dagger,\hat{a}_{\bm{k}'}^\dagger]=0\,,\end{aligned}\ ] ] and @xmath141 satisfies the equation of motion , @xmath142\psi_{k}(\eta)=0\ , , \label{psikeom}\end{aligned}\ ] ] with the normalization condition , @xmath143 as a natural choice for the vacuum we choose the bunch - davis vacuum , for which we have @xmath144 where @xmath145 is the hankel function of the first kind . for @xmath32 , this would correspond to choosing the de sitter invariant vacuum if the mass squared were positive . note that the rms amplitude of the vacuum fluctuations per logarithmic interval of @xmath146 exhibits an infrared divergence in the limit @xmath147 , @xmath148 we now focus on the case @xmath32 , which is of cosmological interest , and consider the effect of the quantum radion on the brane . as noted , @xmath149 shows an instability which grows as @xmath150 . however , what one can detect on the brane is @xmath19 given by eq . ( [ emnphi ] ) which acts as the effective energy - momentum tensor , eq . ( [ efftau ] ) . to see its behavior , let us spell out the components of @xmath19 explicitly . the components for each @xmath146 mode are @xmath151 \psi_ky\ , , \nonumber\\ e_{0j } & & = n_e(-k)\left[{\partial\over\partial\eta}+{1\over\eta}\right ] \psi_ky_{j}\ , , \nonumber\\ e_{ij } & & = n_ek^2\psi_ky_{ij } + n_e\delta_{ij}\left[{1\over\eta}\left({\partial\over\partial\eta } + { 1\over\eta}\right)-{k^2\over n-1}\right]\psi_ky \label{emnform}\end{aligned}\ ] ] where eq . ( [ psikeom ] ) has been used , and @xmath152 , @xmath153 , and @xmath154 stand for @xmath155 inserting the explicit form of @xmath156 given in eq . ( [ bdvac ] ) , we find @xmath157 where @xmath158 and @xmath159 denote the trace and traceless part of @xmath160 , respectively , and @xmath161 from the above , and noting @xmath162 for @xmath163 , we see that @xmath159 , which represents the effective anisotropic stress , dominates over the other components for @xmath164 , i.e. , on super horizon scales , though its physical amplitude ( @xmath165 ) decays as @xmath166 . as for the effective energy density induced by the radion , it decays rapidly as @xmath167 . however , it may be said that the decay rate is not fast enough in the following sense . since @xmath19 is traceless and conserved at the linear perturbation order , one can regard it as the energy momentum tensor of a radiation - like fluid . if it should behave exactly as radiation , the energy density would decay as @xmath168 instead of @xmath167 . of course the decay rate @xmath167 does not mean @xmath19 behaves as a dust matter . the fact that the decay rate is slower than the standard radiation fluid ( for @xmath169 ) is caused by a strong anisotropic stress present in the case of radion . thus , in this sense , the instability of radion indeed shows up on the brane , but the effect turns out to be rather mild because of the nature of de sitter space in which the background energy density remains constant in time . we defer the detailed discussions on this point to appendix [ decayemn ] . to summarize , we conclude that the instability of the quantum radion induces a large anisotropic stress on the brane , but because of the rapid expansion of the brane it is not strong enough to give rise to a gravitational instability on the brane . in the previous section , we have seen that the quantum radion fluctuations do not cause a gravitational instability on the brane . however , this does not necessarily mean the effect is always small . in particular , if the rms value of the effective energy density due to the radion field is comparable to the background vacuum energy density of the de sitter brane , the evolution of the brane will be significantly affected . if we regard our background to represent a braneworld undergoing cosmological inflation , this would imply a significant modification of the braneworld inflationary scenario . in this section , assuming the positive tension brane describes our universe , we consider the cosmological constraints arising from the effect of the radion quantum fluctuations . to quantify the effect , we consider the power spectrum of the effective energy density , @xmath170 focusing on the positive tension brane , we define the power spectrum @xmath171 of @xmath172 normalized by the background energy density @xmath173 as @xmath174 where @xmath175 is the surface area of the @xmath0-dimensional unit sphere ; @xmath176 . we find @xmath177 let us introduce the density parameter @xmath178 for the rms effective energy density of the radion field as @xmath179 which is obtained by integrating @xmath180 over @xmath146 and taking the square root of it . when @xmath178 is comparable to , or larger than the order of unity , the fluctuation of the brane is non - negligible . if we integrate the above over @xmath146 , we encounter the divergence from large @xmath146 . however , this ultraviolet divergence is the same as the usual one in the minkowski background . hence , to discard the divergence , we cut off the integration at the hubble horizon scale ; @xmath181 . the result is @xmath182 here and below we fix the dimensions to @xmath183 . note that @xmath184 is the ratio of the ads@xmath185 curvature radius to the ds@xmath186 curvature radius , and @xmath187 . if we introduce the @xmath188-dimensional planck scale in the single flat brane limit , @xmath189 , we have the well - known relation , @xmath190 . below we assume @xmath191 is the present - day planck length in our universe . to discuss the cosmological constraints , it is convenient to introduce the @xmath192-dimensional gravitational length scale @xmath193 . in terms of @xmath194 , eq . ( [ omegae ] ) is re - expressed as @xmath195 we see that the quantum radion effect will be negligible if @xmath196 given the ratios @xmath197 and @xmath198 , this constrains the value of @xmath199 to be larger than a critical value , @xmath200 since @xmath194 is the fundamental scale of the 5-dimensional theory , it is natural to assume @xmath201 ; otherwise the whole system will be in the quantum gravitational regime . note that @xmath202 . with this assumption , let us consider the following two limiting cases . @xmath203 : + in this case , eq ( [ critical ] ) is satisfied for practically all values of @xmath204 except for the range very close to @xmath205 , that is , the constraint becomes @xmath206 @xmath207 : + in this case , we have @xmath208 using the equality @xmath209 , the right hand side is rewritten as @xmath210 . for any reasonable scenario of inflation , we must have @xmath211 . hence the constraint on the value of @xmath199 is very mild in this case as well . thus , with the assumptions that @xmath211 and @xmath212 , the constraints obtained in the above two cases can be concisely expressed as @xmath213 to summarize , as far as the quantum radion fluctuations are concerned , the negative tension brane can be very close to the positive tension brane , even within a distance smaller than the ads@xmath185 curvature radius . in this paper , taking up the randall - sundrum type two - brane scenario in an @xmath2-dimensional spacetime , we have investigated the quantum fluctuations of the relative displacement of the branes , which we called the quantum radion . we have considered the cases of the flat two - brane and de sitter two - brane systems simultaneously . adopting the so - called randall - sundrum gauge , we have first solved the linear gravitational perturbation equations that describe the radion mode . then assuming the perturbation of the form with a fixed @xmath43-dependence that solves the gravitational equation in the @xmath43-direction , where @xmath43 is the extra - dimensional coordinate orthogonal to the branes , we have derived the effective action for the radion . with the effective action at hand , we have quantized the radion assuming the radion state is the bunch - davis vacuum . we have analyzed the effect of the quantum radion on the brane using the effective einstein equations derived by shiromizu , maeda and sasaki@xcite in which the radion adds an effective energy - momentum tensor at the linear order in the field amplitude . although the radion has the negative mass squared @xmath214 on the de sitter brane where @xmath215 is the hubble parameter , we have found that the corresponding instability does not show up on the brane . we have noted , however , that the anisotropic stress induced by the radion is unusually large , though it still decays in time as @xmath216 for a fixed @xmath146 , where @xmath217 is the cosmic scale factor and @xmath146 is the comoving wavenumber , and the rms value of the effective energy density for a fixed @xmath146 decays as @xmath167 irrespective of the spacetime dimensions @xmath1 . focusing on the positive tension de sitter brane with the hubble rate @xmath205 , which models the braneworld inflation , we have estimated the rms total energy density of the quantum radion by integrating over @xmath146 up to @xmath218 . we have introduced the density parameter @xmath178 which describes the relative magnitude of the radion energy density to the background energy density , and discussed the condition @xmath219 on the model parameters . for @xmath211 and @xmath220 , which are reasonable to assume for the background to be not in the quantum regime , we have found that practically any choice of the location of the negative tension brane is allowed . this implies the quantum radion does not seriously affect the braneworld inflation scenario , at least at the linear perturbation order . the most intriguing question remained now is if the analysis here at the order of the linear perturbation level is sufficient . naively , one may regard the effective action of the radion we have obtained as a piece to be added to the total effective action for an effective 4-dimensional theory that includes the gravity . then the variation of the radion effective action with respect to the 4-metric will give the energy - momentum tensor of a scalar field with a negative mass squared , which would grow and diverge in contrast to the decaying effective energy - momentum tensor we analysed in this paper . however , such an energy - momentum tensor is quadratic in the radion , which is beyond the accuracy of our analysis at the linear - order perturbation level . we do need to investigate the higher order perturbation to truly see the effect of the quantum radion on the brane this issue is left for future study . this work is supported in part by the monbukagakusho grant - in - aid for scientific research no . 12640269 and by the yamada science foundation . ug is supported by the jsps research fellowships for young scientists , no . we would like to thank t. chiba , j. garriga , k. koyama , t. shiromizu and t. tanaka for discussions . in this appendix , we give an outline of the derivation of the action given in eq . ( [ pre - action ] ) and compare it with the effective actions of radion derived in @xcite . the action is obtained by substituting the results in eq . ( [ result ] ) into the action of the system given in eq . ( [ basicaction ] ) . note that @xmath29 denotes the metric of the @xmath1-dimensional space - time with a constant curvature @xmath221 and @xmath222 the covariant derivative with respect to @xmath29 . we do not demand @xmath223 to satisfy the field equation ( [ eomvarphi ] ) , so that the degree of freedom associated with the brane fluctuation remains . we first substitute the results given in eq . ( [ result ] ) into the bulk actions @xmath224 and @xmath225 of eq . ( [ basicaction ] ) . carrying out the calculation with the help of the commutation rule , @xmath226\omega_\rho = 2{\cal k}\gamma_{\mu[\rho}\gamma_{\nu]\sigma}\omega^\sigma\ , , \label{commutation}\end{aligned}\ ] ] results in vanishing of all the second - order terms except for the terms that can be cast into the surface term on @xmath227 . thus , to the second order in @xmath100 , the action ( [ basicaction ] ) is found to have support only on @xmath227 : @xmath228\ , ; \qquad \hat{k}_{\mu\nu}:=\frac12b^{2}{\partial}_r\!\left(b^{-2}h_{\mu\nu}\right ) . \label{actionsubs}\end{aligned}\ ] ] note that @xmath107 here is defined as the displacement of the corresponding brane measured from the side of smaller values of @xmath44 , that is , @xmath229 here , it is worth noting that the above action @xmath230 is different from the one obtained in @xcite ( see eq . ( a.19 ) ) . this is because the above action @xmath230 is obtained by substituting a particular form of the metric to the action of the system , while the action of @xcite is constructed without any restriction on the metric . the particular form of the metric in eq . ( [ result ] ) solves the equation of motion in @xmath43-direction , and thus already extremize the action in the @xmath43-direction . therefore , @xmath230 is an action to be maximized in the @xmath231-directions only , which is to give the equation of motion for @xmath121 as a function of @xmath232 . meanwhile , the action obtained in @xcite is an action to be maximized in both the @xmath233direction and @xmath231-directions independently to give the equations of motion for both @xmath121 and @xmath107@xcite . let us define @xmath234 and @xmath235 as @xmath236 then @xmath237 is express as @xmath238\,.{\nonumber}\end{aligned}\ ] ] we find that the variation of @xmath234 with respect to @xmath121 gives the boundary condition of @xmath121 on the brane , while that of @xmath239 with respect to @xmath107 gives the equation of motion for @xmath107 . with the substitution of eq . ( [ result ] ) to @xmath237 , @xmath234 vanishes and only @xmath235 remains . using the commutation rule ( [ commutation ] ) , @xmath235 is simplified to give @xmath240{\tilde}\varphi\,,{\nonumber}\end{aligned}\ ] ] which is eq . ( [ pre - action ] ) . the effective action for the radion has been derived previously by chiba@xcite and by chacko and fox@xcite in the longitudinal gauge . however , in our opinion , there seems to exist some subtleties in their derivations that may be worth clarified , apart from the difference in the choice of gauge . we therefore compare their derivations with ours . first , we need to transform the metric perturbation in eq . ( [ result ] ) , which is given in the rs gauge , to that in the longitudinal gauge in which the locations of the branes are unperturbed , @xmath241 and @xmath242 . with the coordinate transformation of the form , @xmath243 we obtain the metric perturbation in the desired gauge : @xmath244 let us first consider chiba s effective action of radion given in eq . ( 14 ) of @xcite . the metric ansatz chiba uses to derive the action is @xmath245 in which @xmath246 is a general 4-dimensional metric that depends solely on @xmath231 . here , the coordinate @xmath43 in his paper is replaced with the coordinate @xmath44 to avoid confusion . linearizing the metric ( [ cgrmetric ] ) with respect to @xmath104 , it reduces to the metric perturbation of the form ( [ longgauge ] ) with the following correspondence between his notation and ours : @xmath247 where we have fixed @xmath183 . ( it may be noted that eq . ( [ cgrmetric ] ) can be extended to the general @xmath1 dimensions simply by replacing @xmath248 with @xmath249 in the case of @xmath31 . ) with the above correspondence , we find his result , when linearized , does not agree with ours . let us examine where the difference comes from by following his derivation . substituting the decomposition of the ricci scalar of the metric ( [ cgrmetric ] ) , @xmath250 & = & -20k^2+e^{2kh}r[\bar{g } ] + 6k^2e^{4kr+2kh}h_{,r}^{-1}(-1 + 2kfe^{2kr}){\nabla\!}^\rho f { \nabla\!}_\rho f + ( \mbox{total divergence term in $ x^\mu$})\ , , \label{rdecomp}\end{aligned}\ ] ] into his starting action of the system , eq . ( 1 ) of @xcite , and integrating it with respect to @xmath44 , we obtain @xmath251{\nonumber}\\ & & \hspace{30ex } -3km^3(\alpha-1)({\nabla\!}_{(+)}f)^2 -3km^3(\alpha^{-2}e^{-4(\alpha-1)kf}-1 ) \bigg\}{\nonumber}\\ & & \hspace{14ex } -\sigma_{(+)}\int d^4\!x\sqrt{-{g}_{(+ ) } } -\sigma_{(-)}\int d^4\!x\sqrt{-{g}_{(-)}}\ , , \label{cact}\end{aligned}\ ] ] where the following correspondence is understood : @xmath252 comparing the action ( [ cact ] ) with eq . ( 5 ) in @xcite , it seems that the first term @xmath253 is missing there , which comes from the integration of the ricci scalar of the background rs bulk spacetime , @xmath254 in eq . ( [ rdecomp ] ) . furthermore , if the correspondence ( [ chibacorresp2 ] ) we deduced is correct , the values of the brane tensions @xmath255 as stated in eq . ( 3 ) of @xcite are wrong by the factor 2 ; i.e. , the correct values are @xmath256 . besides , in his starting action where the integral with respect to the extra dimension is given by twice the integral over @xmath44 in the range @xmath257 , that is , from the positive tension brane to the negative tension brane , the geometrical boundary terms of the bulk spacetime , @xmath258 and @xmath259 as in our action of the system in eq . ( [ basicaction ] ) , should be added . with these corrections we find @xmath260 -3km^3(\alpha-1)({\nabla\!}_{(+)}f)^2 \bigg\}\,,{\nonumber}\end{aligned}\ ] ] where we have used the fact that @xmath261 . since our effective action of the radion is obtained by fixing @xmath262 , the action to be compared is @xmath263 using eqs . ( [ chibacorresp1 ] ) and ( [ chibacorresp2 ] ) and the correspondence @xmath264 which follows from them , we find that the above action agrees with our result . let us next consider the derivation by chacko and fox and their result . the correspondence between their notation and ours are @xmath265 from which we deduce @xmath266 using the above correspondence , we find that their result given in eq . ( 34 ) of @xcite exactly agrees with ours . there is , however , a slight subtlety which may be worth pointing out . it is the following . chacko and fox state that they work in the compact extra dimension with @xmath5 symmetry . therefore there is no need to introduce the geometrical boundary terms in the action . in fact , the form of the linearized action given by eq . ( 33 ) of @xcite is clearly free from such terms . however , if we perform the integral over @xmath44 of the linearized action , there appear contributions from the branes due to the curvature singularities . the interesting fact is that the contributions cancel out with those from the tension terms in the longitudinal gauge . perhaps it is related with the fact that the coordinated locations of the branes are unperturbed in the longitudinal gauge . this implies that one may arrive at the correct answer by simply ignoring the singular contributions from the branes and simultaneously the tension terms , and then integrating the linearized action only over the bulk between branes . in the present paper , we took a different approach . that is , we divided the covering space of the bulk into the two patches bounded by the branes . this division requires the additional geometrical boundary terms for each patch to the action of the system . if we apply our method to their starting action , eq . ( 1 ) of @xcite , it is rewritten as @xmath267 where @xmath268 is the extrinsic curvature of the @xmath44-constant hypersurface defined with the normal vector pointing toward the increasing direction of @xmath44 . apart from the difference in notation as expressed in the above , this form of the action is the one we used as the starting point of our calculation . ( [ basicaction ] ) . substituting the perturbation given in eq . ( [ longgauge ] ) into the above action , we find that the terms arising from the first integral in eq . ( [ ( 1)rw ] ) that can be cast into the total divergence form cancel out with the perturbation of the last four terms of eq . ( [ ( 1)rw ] ) , i.e. , the tension terms and the geometrical boundary terms . this cancellation gives the same form of action as the one in their eq . ( 33 ) by restricting the integral only over the bulk , and consequently leads to their final result , eq . in this appendix , we derive the effective brans - dicke parameter on the @xmath1-dimensional positive tension brane with curvature @xmath31 , @xmath269 . the action for the brans - dicke theory in the @xmath1-dimensional space - time is @xmath270 + \int \sqrt{-q}\,d^n\!x{\cal l}_{m}\end{aligned}\ ] ] where @xmath14 is the metric of the n - dimensional space - time , @xmath271 is the brans - dicke parameter , and @xmath272 is the lagradian density of matter fields . let @xmath273 $ ] . then , the equations of motion linearized in @xmath274 are @xmath275 w(x^\mu)\ , , \label{einsteinbd}\\ & & \hspace{2ex } f_0 \box_q w(x^\mu)=\frac{1}{n-1+(n-2)\omega_{\rm bd } } \left(8\pi g_n\tau-2\lambda f_0\right)\ , , \label{eombd}\end{aligned}\ ] ] where @xmath16 is the energy - momentum tensor that arises from @xmath272 . using eq . ( [ eombd ] ) , eq . ( [ einsteinbd ] ) can be put into the following form : @xmath276 w(x^\mu ) \label{bd - einstein}\end{aligned}\ ] ] where @xmath277 . to obtain the effective einstein equations on the positive tension brane for the energy density small compared with the tension of the brane , we substitute eq . ( 3.8 ) of @xcite into eq . ( [ branein ] ) of this paper . the result is @xmath278 \phi\ , . \label{effectiveeinstein}\end{aligned}\ ] ] comparing eq . ( [ bd - einstein ] ) with eq . ( [ effectiveeinstein ] ) , we find that the effective gravity on the brane takes the brans - dicke form with the following identifications : @xmath279 eliminating @xmath280 from eq . ( [ omega ] ) , we obtain @xmath281 while eliminating @xmath271 from eq . ( [ omega ] ) , we obtain @xmath282 \label{gn}\end{aligned}\ ] ] for @xmath32 and @xmath183 , we have @xmath283 and @xmath284\ , . \label{bdparagnk=1}\end{aligned}\ ] ] for @xmath31 and @xmath183 , we have @xmath285 and eqs . ( [ bdpara ] ) and ( [ gn ] ) recover the result of garriga and tanaka@xcite . in this appendix , we express the effect of the radion on the brane in the language of cosmological perturbation theory . we follow the notation of @xcite . to begin with , we list a few important facts to be kept in mind : the points ( a ) and ( b ) are somewhat related . as long as the background bulk spacetime is anti - de sitter , the point ( b ) is always true . independent of this fact , from the brane point of view , the matter perturbation variables are automatically gauge - invariant on the pure de sitter space . for example , if we consider the density perturbation , the gauge transformation law for @xmath289 is given by @xmath290 where @xmath291 is the gauge function describing the shift in the time coordinate ; @xmath292 . thus the density perturbation on exactly de sitter space where @xmath293 is always gauge - invariant , and the same is true for all the matter perturbation variables . it should be noted , however , this special property of the de sitter background can lead to an erroneous result if one tries to consider the perturbation in a gauge specified by conditions involving the matter variables . we shall come back to this issue later . for the moment , what we can claim is that the gauge - invariance is guaranteed provided one considers the perturbations in a class of gauges whose conditions involve only the geometrical variables such as the metric . let us call this class of gauges the geometrical gauges . first we introduce the matter perturbation variables : @xmath294 where @xmath295 , @xmath296 and @xmath297 are the spatial harmonics satisfying @xmath298y(x^i)\,,\end{aligned}\ ] ] and @xmath299 is just the effective radion energy momentum tensor , @xmath300 note that the usual velocity perturbation , @xmath301 , can not be defined because of @xmath302 on the background . also note that the traceless nature of @xmath288 implies @xmath303 . as for the metric perturbation , we introduce the variables as @xmath304dx^i dx^j\,,\end{aligned}\ ] ] where we have chosen the spatially flat chart of the de sitter space . as noted above , the choice of gauge is irrelevant as long as we choose a geometrical gauge . the perturbation equations are then derived by simply writing down the energy momentum conservation law @xmath305 on the unperturbed background . therefore , for definiteness , let us choose the newton ( or longitudinal ) gauge in which the shear of the constant time hypersurfaces vanishes , i.e. , @xmath306 , and denote the matter variables in this gauge as @xmath307 adopting the scale factor @xmath217 as the time variable , we find @xmath308 these equations can be combined to give a second order equation for @xmath309 , @xmath310 we see from the above equation that if there were no anisotropic stress @xmath311 , then @xmath312 or @xmath313 after horizon - crossing . however , in the present case , we have @xmath314 . this slow decay rate of the anisotropic stress acts as a source to the energy density @xmath315 . with @xmath316 , the right hand side of eq . ( [ evoldelta2 ] ) behaves as @xmath317 , which implies @xmath318 ( for @xmath319 ) . the behavior of @xmath320 is found as @xmath321 . to be a bit more precise , the leading order behaviors of @xmath309 , @xmath320 and @xmath311 are given by @xmath322 where @xmath323 is a constant . these results are of course fully consistent with those obtained in the main text , eq . ( [ emnform ] ) . thus one may say that this unusual behavior of @xmath309 is caused by the large ( though decaying ) anisotropic stress . it is instructive to re - express the above result in the so - called comoving gauge in which the @xmath324constant hypersurfaces are chosen in such a way that @xmath325 note that this condition involves the matter variable @xmath326 . a peculiarity of the comoving gauge is that the density perturbation in this gauge is not equal to @xmath309 , although the matter perturbation variables should be gauge - invariant on the pure de sitter space - time as noted above . the density perturbation on the comoving hypersurface , which we denote by @xmath327 , is related to @xmath309 as @xcite @xmath328 one immediately sees from eq . ( [ leading ] ) that the @xmath320 term dominates and the leading order behavior of @xmath327 is given by @xmath329 which decays much slower than @xmath309 . the cause of this seemingly inconsistent result is the comoving gauge condition ( [ comoving ] ) which forces @xmath326 to be zero . as @xmath326 is a gauge - invariant quantity , this condition would never be fulfilled and thus the comoving slice does not exist . in fact , it is straightforward to show that all the metric perturbation variables are ill - defined in the comoving gauge . therefore , although the variable @xmath327 is well - defined as in eq . ( [ defdelta ] ) , this @xmath327 does not represent a density perturbation in any gauge . let us analyze the behaviors of the metric perturbation variables in the newton gauge . although not essential , for simplicity , we take @xmath330 which implies @xmath331 . we denote the metric variables in the newton gauge by @xmath332 then the @xmath333-components and the traceless part of the @xmath334-components of the einstein equations give@xcite @xmath335 inserting the leading behaviors of @xmath327 and @xmath311 given above to these equations , we find @xmath336 one may find it a bit surprising that these variables grow as @xmath217 , seemingly indicating an instability . the resolution is that the newton gauge is not really a good gauge . one should examine if there is a different choice of gauge in which the metric perturbations behave regularly , and there exists indeed such a gauge . by the shift of the time slice @xmath292 , @xmath337 and @xmath338 transform as @xmath339 then it is easy to see that the leading terms of @xmath340 and @xmath341 are simultaneously eliminated by @xmath342constant . thus the above apparent instability is just a reflection of the bad choice of gauge . 22 i. antoniadis , phys . lett . * b246 * , 377 ( 1990 ) l. randall and r. sundrum , , 4690 ( 1999 ) . l. randall and r. sundrum , , 3370 ( 1999 ) . j. garriga and t.tanaka , , 2778 ( 2000 ) . u. gen and m. sasaki , prog . * 105 * , 591 ( 2001 ) . r. gregory , v. a. rubakov and s. m. sibiryakov , phys . lett . * 84 * , 5928 ( 2000 ) [ arxiv : hep - th/0002072 ] . c. csaki , j. erlich and t. j. hollowood , phys . lett . * 84 * , 5932 ( 2000 ) [ arxiv : hep - th/0002161 ] . g. r. dvali , g. gabadadze and m. porrati , phys . b * 484 * , 112 ( 2000 ) [ arxiv : hep - th/0002190 ] ; g. r. dvali , g. gabadadze and m. porrati , dimensions , phys . b * 484 * , 129 ( 2000 ) [ arxiv : hep - th/0003054 ] . c. csaki , j. erlich and t. j. hollowood , with quasi - localized gravity , phys . b * 481 * , 107 ( 2000 ) [ arxiv : hep - th/0003020 ] . l. pilo , r. rattazzi and a. zaffaroni , jhep * 0007 * , 056 ( 2000 ) [ arxiv : hep - th/0004028 ] . i. i. kogan , s. mouslopoulos , a. papazoglou and l. pilo , arxiv : hep - th/0105255 . c. charmousis , r. gregory and v. a. rubakov , phys . d * 62 * , 067505 ( 2000 ) [ arxiv : hep - th/9912160 ] . z. chacko and p. j. fox , phys . rev . d * 64 * , 024015 ( 2001 ) [ arxiv : hep - th/0102023 ] . w. d. goldberger and m. b. wise , phys . lett . * 83 * , 4922 ( 1999 ) [ arxiv : hep - ph/9907447 ] . j. garriga and m. sasaki , phys . d * 62 * , 043523 ( 2000 ) [ arxiv : hep - th/9912118 ] . j. garriga and a. vilenkin , phys . d * 44 * , 1007 ( 1991 ) . j. garriga and a. vilenkin , phys . d * 45 * , 3469 ( 1992 ) . t. shiromizu , k. maeda and m. sasaki , , 024012 ( 2000 ) . private communication with r.m.wald t. chiba , phys . d * 62 * , 021502 ( 2000 ) [ arxiv : gr - qc/0001029 ] . h. kodama and m. sasaki , _ cosmological perturbation theory , _ prog . * 78 * ( 1984 ) 1 . scalar field and a self - tuning problem ,
we present the results of six suzaku observations of the recurrent black hole transient 4u during its decline from outburst from february 8 to march 23 in 2006 . all observations show the typical high / soft state spectral shape in the 250 kev band , roughly described by an optically thick disk spectrum in the soft energy band plus a weak power - law tail that becomes dominant only above kev . the disk temperature decreases from 1.4 kev to 1.2 kev as the flux decreases by a factor 2 , consistent with a constant radius as expected for disk - dominated spectra . all the observations reveal significant absorption lines from highly ionized ( h - like and he - like ) iron k at 7.0 kev and 6.7 kev . the brightest datasets also show significant but weaker absorption structures between 7.8 kev and 8.2 kev , which we identify as a blend of iron k and nickel k absorption lines . the energies of these absorption lines suggest a blue shift with an outflow velocity of . the h like iron k equivalent width remains approximately constant at ev over all the observations , while that of the he like k line increases from 7 ev to 20 ev . thus the ionization state of the material decreases , as expected from the decline in flux . we fit the profiles with voigt functions ( curve of growth ) to derive absorbing columns , then use these together with detailed photo - ionization calculations to derive physical parameters of the absorbing material . the data then constrain the velocity dispersion of the absorber to 200 , and the size of the plasma as assuming a source distance of 10 kpc .
astro-ph0610496
in recent years , a growing number of x - ray binaries have been found to exhibit absorption lines from highly ionized elements ( e.g. , @xcite and references therein ; @xcite ) . these systems range from microquasars such as gro j1655@xmath840 @xcite and grs 1915 + 105 @xcite to low - mass x - ray binaries such as gx 13 + 1 @xcite , x 1658@xmath8298 @xcite and x 1254@xmath8690 @xcite . these systems are viewed at high inclination angles , and the absorption structure is visible throughout the orbital period ( e.g. @xcite ; 2002 ) . the absorption features are therefore thought to originate in material that is associated with and extends above the outer accretion disk . this is illuminated by the x - rays produced from the innermost regions of the accretion flow ( both the disk and hard x - ray coronal emission ) . the reprocessed emission and scattered flux from the extended wind can be seen directly in the accretion disk corona sources , where the intrinsic x - rays are obscured ( e.g. kallman et al . 2003 ) , but for the majority of highly inclined sources , the wind material is seen in absorption against the much brighter intrinsic central x - ray source . multiple absorption lines seen there then give an excellent probe of the physical conditions in the wind ( e.g. ueda et al . 2004 ; miller et al . 2006 ) , where the spectra indicate the presence of significant amounts of highly ionized outflowing material . such accretion disk winds can be produced by thermal driving ( e.g. , begelman et al . 1983 ) , and/or by magnetic forces @xcite as inferred by @xcite from chandra hetgs data of gro j@xmath9 . 4u 1630@xmath8472 is a black hole candidate ( e.g. , @xcite ) , known for x - ray outbursts that repeat with an interval of roughly 600690 days ( jones et al . 1976 ; parmar , angelini & white 1995 ) . the source has been observed with every major mission from exosat ( @xcite ; parmar et al . 1986 ) , ginga ( parmar et al . 1997 ) , batse cgro ( bloser et al . 1996 ) , to beppo - sax ( oosterbroek et al . 1998 ) , rxte ( tomsick et al . 1998 ; kuulkers et al . 1998 ; hjellming et al . 1999 ; cui et al . 2000 ; dieters et al . 2000 ; trudolyubov et al . 2001 ; kalemci et al . 2004 ; @xcite ) , asca ( @xcite ) , and recently chandra ( j. miller , private communication ) . these x - ray observations showed that the source is absorbed heavily with @xmath10@xmath11 ( tomsick et al . 1998 ; kuulkers et al . 1997 ; cui et al . 2000 ; trudolyubov et al . 2001 ; @xcite ) . while the ir counterpart was found during a 1998 outburst by augusteijn et al . ( 2001 ) , no optical counterpart is known for 4u 1630@xmath8472 , most likely due to its high reddening and location in a crowded star field ( parmar , stella & white 1986 ) . thus , the distance , inclination angle , and mass of the compact object are unknown for 4u 1630@xmath8472 . this source is classified as a black hole not from dynamic mass measurements , but because of the similarity of its x - ray spectral and timing properties to those of systems with measured black hole masses . this source was observed six times with suzaku during its most recent outburst as part of a program to study discrete spectral structures as a function of the changing accretion conditions . the suzaku monitoring observation led to successful detection of absorption lines on all occasions . though one of the absorption lines may have been detected by chandra ( j. miller , private communication ) , the suzaku xis data reveal the line with a much higher significance , thanks to its larger effective area at 7 kev than that of chandra . in this paper , we focus mainly on the absorption line structures , and how these discrete features evolve over the two months span of our observations . considering the large absorption column density , we assume the source distance , @xmath12 , at @xmath13 kpc in this paper . in addition , by comparison to the other black hole binaries reported to show absorption lines , an inclination angle , @xmath14 , of @xmath15 is assumed . errors quoted in this paper represent 90% confidence limit for a single parameter unless otherwise specified . an outburst of 4u @xmath0 was reported by the rxte all sky monitor ( asm ; @xcite ) on 2005 december . from 2006 february 8 through march 23 , we observed the source six times with a series of 2123 ks exposures as part of the initial performance verification of suzaku . suzaku is the 5th x - ray astronomy satellite of japan launched on july 10 , 2005 , with an m - v rocket from the uchinoura space center ( mitsuda et al . it carries 4 sets of x - ray telescopes ( @xcite ) each with a focal - plane x - ray ccd camera ( xis , x - ray imaging spectrometer ; @xcite ) operating in the energy range of 0.212 kev , together with a non - imaging hard x - ray detector ( hxd , takahashi et al . 2006 ) , which covers the 10600 kev energy band with si pin photo - diodes and gso scintillation counters . three of the xis ( xis0 , 2 , 3 ) have front - illuminated ( fi ) ccds , while xis1 utilizes a back - illuminated ( bi ) ccd , achieving an improved soft x - ray response but poorer hard x - ray sensitivity . considering the large column density to 4u @xmath0 , we concentrate on the harder x - ray band in this paper , and only analyze the data from the fi ccds . figure [ fig : asm ] shows a lightcurve of 4u @xmath0 obtained by the rxte asm . the suzaku observations are indicated with downward arrows . because the optical axes of the xis and the hxd are slightly offset , the source was put at the center of the hxd field of view during the observations , to maximize the hxd count rate , and to reduce the xis count rate as pileup is an issue for such a bright source ( see below ) . the hxd was operated in the nominal mode throughout the observations . because the source was very bright , the xis was operated with the 1/4 window option , in which a smaller field of view of @xmath16 is read out every 2 s. a burst option was added in the first four observations . this decreases the effective exposure by a factor of two . the editing modes of the fi ccds ( xis023 ) were set to @xmath17 and @xmath18 for high and medium data rates , respectively . table [ tab : obs ] summarizes the details of the observations . clllcrcc epoch & & & & & + & & & & total@xmath19 & high@xmath19 & 210 kev & 2040 kev + 1 & feb 8 8:14 & feb 9 3:20 & 1/4 window , 1-s burst & 22.2&22.2&@xmath20&@xmath21 + 2 & feb 15 23:08 & feb 16 13:08 & 1/4 window , 1-s burst & 21.5&9.6&@xmath22&@xmath23 + 3 & feb 28 23:00 & mar 1 16:43 & 1/4 window , 1-s burst & 21.3&21.3&@xmath24&@xmath25 + 4 & mar 8 01:39 & mar 8 17:31 & 1/4 window , 1-s burst & 21.2&20.7&@xmath26&@xmath27 + 5 & mar 15 5:19 & mar 15 17:51 & 1/4 window & 23.2&5.2&@xmath28&@xmath29 + 6 & mar 23 9:57 & mar 23 22:21 & 1/4 window & 21.7&15.4&@xmath30&@xmath31 + ( 80mm,60mm)figure1.ps for both the xis and the hxd , we use version 0.7 screened data provided by the suzaku team . the screening of the version 0.7 xis data are based on the following standard criteria : a ) only grade0 , 2 , 3 , 4 , 6 events are accumulated , b ) the time interval after passage through the south atlantic anomaly is larger than 436 seconds , c ) the object is at least @xmath32 and @xmath33 above the rim of the earth during night and day , respectively . we limit our analysis to the @xmath17 mode data ( high data rate ) , because the @xmath18 mode data ( medium data rate ) are not well calibrated at this time . during the 2nd and 5th observations , more than half of the exposure was acquired in @xmath18 mode , and thus these data have poorer statistics . the exposure times for the total and high data rate are summarized in table [ tab : obs ] . the xis events were extracted from a circular region with a radius of @xmath34 centered on the image peak . this extraction circle is larger than the window size . the effective extraction region is therefore the intersection of the window and this circle . the background was entirely negligible throughout the six observations ( less than 1% of the signal at 9 kev ) , so was not subtracted from the data . using xis2 as a reference , we find that the 210 kev count rate declined over the six observations , with @xmath35 , @xmath36 , @xmath37 , @xmath38 , @xmath39 , and @xmath40 respectively , roughly corresponding to 4.6 , 4.0 , 3.1 , 2.8 , 2.5 , and @xmath41 in the same energy band . within each observation , we find no significant intensity variation intrinsic to the source . the source was so bright during these observations that the xis suffered from photon pileup at the image center . a rough threshold above which the photon pileup becomes significant is @xmath42 cts exposure@xmath43 for a point source . the count rates of 4u @xmath0 during the observations were around this level . furthermore , because 4u @xmath0 has a very soft energy spectrum , the effect of photon pileup can be significant in the hard band spectra . therefore , we excluded a circular region of @xmath44 at the image center from the event extraction region to minimize these effects . this did not remove the photon pileup completely , and an artificial structure is seen as a hard excess above @xmath45 kev at the @xmath46510% level . because photon pileup tends to produce x - ray events which have twice ( or a little more ) as high an energy as the most abundant events ( in this case , about 4 kev ) , we modeled the artificial structure as a gaussian centered at 10 kev with a fixed width of @xmath47 kev . as seen in the next section , this phenomenological model works quite well to compensate for the effect of pileup . because the event extraction region has a complex shape ( an annulus truncated with the window size ) , we calculated the effective area for each xis sensor by using xissimarfgen ( version 2006 - 04 - 24 ) , which calculates the arf file through monte carlo simulations . the three fi ccds have almost the same performance , so we have added their data to increase the statistics in the spectra . hereafter we refer to the summed data as xis023 . we calculated the corresponding instrumental responses by summing the latest redistribution matrices of ae_xi[0,2,3]_20060213.rmf using the ftool addrmf and the arf files using the ftool addarf . though the main objective of this paper is the absorption structure in the xis energy band , it is also useful to understand the continuum emission since this provides a knowledge of spectral states and photoionising flux . we thus analyzed the hxd data using only the pin data , since the source is dim in the gso band . both high and medium data rate are used in the pin data analysis . the version 0.7 pin data are screened by the following selection criteria : a ) the object is at least @xmath32 above the earth rim , b ) the time interval after passage through the south atlantic anomaly is longer than 500 seconds , c ) the cutoff rigidity is greater than @xmath48 . the pin background events were provided by the hxd team for each observation based on two different models , a and b ( kokubun et al . here we constructed the pin background spectra based on model b filtered by the same good time intervals . the latest version of the response matrix , ae_hxd_pinhxnom_20060419.rmf , was used in our spectral analysis . even if we allow 5% systematic uncertainty in the background level , the source is clearly detected up to 40 kev and 35 kev in pin data from the first four and last two observations , respectively . dead time fractions were calculated from the pseudo event rates ( @xcite ) as 5.3 , 4.3 , 4.6 , 5.1 , 4.6 , and 4.6% , in the first to last observations , respectively . after the dead time correction , the background subtracted pin count rates are estimated to be 0.16 , 0.13 , 0.093 , 0.065 , 0.055 , and 0.043 @xmath49 in the range of 2040 kev , corresponding to 6.9 , 6.7 , 5.0 , 3.5 , 2.6 , and @xmath50 . because the xis calibration is still in progress , we include a brief description of the calibration uncertainties . the xis energy scale is relatively well calibrated with an uncertainty of less than 5 ev , or 0.2% when the standard mode is used ( @xcite ) . however , when a window option is applied , the gain can change slightly , because the number of frame - store and parallel transfers are different than in the standard mode . this difference is not taken into account in the version 0.7 processing software . since a window option was used for all the observations of 4u @xmath0 , the xis instrument team estimates that this produces a systematic gain difference of @xmath110 ev at 6 kev at the time of 4u @xmath0 observations . when the window option is used , the onboard @xmath51fe calibration source is not directly visible but some events can be scattered into the active window in xis2 , which has the strongest calibration sources . the scattered @xmath51fe events are accumulated from the regions free from the source photons ( @xmath52 of the window area for each side ) . we obtain an average energy of @xmath53 ev . this is consistent with the @xmath110 ev offset estimated above . we also examined the data for a possible gain shift due to the large surface brightness ( local event density ) of the image , by comparing the absorption line energies in the observed spectra for different radii of extraction annuli . we find no significant differences . thus a conservative estimate is that the xis gain at the iron band is correct to within 20 ev during the 4u @xmath0 observations . the energy resolution of the xis is gradually degrading due to the radiation damage on the ccd . however , this effect is not yet included in the currently available response matrices . at the time of the 4u @xmath0 observations , a narrow line at 6 kev would appear broadened with @xmath54 ev , where @xmath55 is an equivalent gaussian width ( standard deviation ) . thus , when we evaluate a line width , we take this effect into account . it is important to note that this does not affect the measurement of the equivalent width . the effective areas of the xis and pin are calibrated using observations of the crab nebula . the accuracy is expected to be 510% . this is sufficient to estimate various emission parameters of 4u @xmath0 . however , because of the very good statistics of the data , the systematic errors in calibration could degrade the model fitting significantly . a major contribution to the systematic error is the calibration uncertainty at gold m - ii edge in the instrumental response . to account for this residual , we added a gaussian at @xmath56 kev with a fixed width of @xmath57 kev . figure [ fig : rawspec_xis - pin]a shows the 250 kev spectra of 4u @xmath0 obtained with the xis023 and the pin . we do not use data below 2 kev , because of large interstellar absorption . an instrument - independent way to view the spectra is to construct the ratio of the source spectra to the corresponding data from the crab nebula , which has an approximately power - law spectrum with @xmath58 , and interstellar absorption of @xmath59 ( @xcite ) . figure [ fig : rawspec_xis - pin]b shows the crab ratios calculated using the crab data obtained with suzaku on 2006 april 5 at the hxd nominal position and the xis operated with the 1/4 window option . this plot reveals several noticeable features in the broad band source spectra . the source spectra are characterized by a dominant soft thermal component in the xis band , which is heavily absorbed , while the harder x - ray pin data clearly show a weak power - law tail . the dominant soft component is generally observed from bright black hole binaries in the high / soft state @xcite , and is believed to be emission from the optically thick standard accretion disk ( @xcite ) . in the crab ratio ( figure [ fig : rawspec_xis - pin]b ) , the weak power - law tail is roughly characterized as @xmath60 in the first two observations . when the 2050 kev pin spectra of these two observations are fitted with the single power - law model under the current response matrix , the values of @xmath61 are obtained as @xmath62 and @xmath63 , respectively . the extrapolated 20100 kev fluxes are estimated as @xmath64 and @xmath65 , respectively , which are only 3 % of the absorbed disk flux in the range of 210 kev . though the steeper power - law tails may be suggested on the latter four observations from the crab ratio plot ( figure [ fig : rawspec_xis - pin]b ) , they are difficult to quantify because of their faintness under the current background uncertainty . further detailed analysis of the hxd data will be presented in a separate paper . the dominance of the disk and the weak power - law tail are both characteristic of the high / soft state . in addition , the crab ratios also reveal the presence of complex absorption structures in the iron k band . in this paper , we concentrate on these absorption line structures and their relation to the accretion disk parameters . hereafter , we perform detailed spectral analyses on the xis data . ( 80mm,60mm)figure2a.ps ( 80mm,60mm)figure2b.ps figure [ fig : spec_xis2 - 10 ] shows the 210 kev xis raw spectra . to reproduce the continuum spectra , we use the multi - color disk model ( diskbb in xspec , @xcite ) modified by interstellar absorption ( wabs model ; @xcite ) , following the canonical modeling for the high / soft state black hole binaries . here , the diskbb model approximates emission from the standard accretion disk , and is known to successfully reproduce the dominant soft spectra of the high / soft state black holes . as shown in figure [ fig : rawspec_xis - pin]b , the contribution of the power - law tail is negligible in the xis energy band , so is not included in the spectral fitting . the residual effects of pileup are modeled by a broad gaussian at 10 kev ( see 2.2 ) , and the current uncertainties in the instrument response at the gold m - ii edge by another gaussian at 3.2 kev ( see 2.3 ) . we start with the first observation , which showed the brightest spectrum . the continuum is well reproduced by the diskbb model with a disk temperature of @xmath66 kev . this is amongst the highest disk temperatures observed with rxte from this source in previous high / soft states ( e.g. , @xcite ) . a disk bolometric luminosity of @xmath67 and an apparent disk inner radius of @xmath68 km are obtained , where @xmath69 is defined as @xmath70 . the absorbing column density , @xmath71 , is estimated to be @xmath72 . this is close to that seen with previous observations by asca and rxte ( e.g. , @xcite ) . the fit with this simple model is however far from acceptable ( @xmath73 ) because of the complex structure in the iron k band . residuals between the data and the best - fit continuum model are shown in the middle panel of figure [ fig : spec_xis2 - 10]a . the most obvious features are two narrow dips at 7.0 kev and 6.8 kev . we therefore add two negative gaussians to the model . the fit is dramatically improved with @xmath74 . the line center energies are estimated to be @xmath75 kev and @xmath76 kev , consistent with the h - like and he - like iron k@xmath2 lines at 6.966 kev and 6.697 kev and k@xmath77 by referring to the national institute of standards and technology ( nist ; see http://aeldata.phy.nist.gov/archive/data.html ) and @xcite for h - like and he - like k@xmath2 , respectively . ] ; the lines are possibly blue - shifted with @xmath78 . the equivalent widths are 30 ev and 7 ev , for the h - like and he - like lines , respectively . the lines are narrow with an upper limit to @xmath55 of 40 ev and 50 ev . these values are consistent with what we expect for narrow lines given the current response matrix , which does not include the degradation of the energy resolution ( 2.3 ) . we thus fix the value of @xmath55 at 10 ev to represent intrinsically narrow lines in this subsection . in addition to these obvious line structures , the residuals reveal weaker absorption structures at 7.8 kev and 8.2 kev . these energies are very similar to the rest frame energies of he - like and h - like iron k@xmath3 lines , namely 7.88 kev and 8.25 kev . however , they are also close in energy to the he - like and h - like nickel k@xmath2 lines at 7.80 kev and 8.09 kev , respectively , which can have comparable equivalent width to that of the iron k@xmath3 lines ( kotani et al 2000 ) . the data can marginally constrain two weak lines at these energies , but not four . hence we include two additional negative gaussians , again with the width @xmath55 fixed at 10 ev , and allow the energies to vary freely . this decreases @xmath79 to @xmath80 , indicating that the two negative gaussians are statistically significant at the 99.98% confidence level using an @xmath81-test ( where @xmath82 ) . the best - fit parameters are summarized in table [ tab : gaussfit ] . residuals between the data and the best - fit model are shown in the bottom panel in figure [ fig : spec_xis2 - 10]a . the inferred energies are almost consistent with h - like and he - like iron k@xmath3 , though they can be contaminated by nickel k@xmath2 . we performed the same spectral analysis on the other five observations . the xis spectra , the best - fit models , and the residuals are shown in figure [ fig : spec_xis2 - 10]b f . the best - fit continuum and absorption parameters are summarized in table [ tab : gaussfit ] , and the time histories of the continuum parameters are given in figure [ fig : disk ] . through the six observations over two months , the value of @xmath83 remained approximately constant at @xmath84 km , while the disk temperature gradually decreased to 1.18 kev , corresponding to a factor 1.75 decrease in the disk bolometric luminosity to @xmath85 . the constancy of @xmath83 is another characteristic feature of the high / soft state of black hole binaries ( e.g. , @xcite and references therein ) , and confirms that the object was indeed in the high / soft state . the correlated decrease in the derived value of @xmath71 by @xmath13% could be an artifact of the over - simplified diskbb model ; it assumes that the disk radiates locally as a blackbody , but distortions from this are suggested both observationally ( e.g. , @xcite ; @xcite ) and theoretically ( @xcite ; @xcite ) from such luminous binaries . irrespective of the continuum modeling , the two strong iron k@xmath2 lines are clearly detected throughout the observations , whereas the weaker lines at higher energies are significantly detected in the first four datasets ( @xmath81-test significance values are given in table [ tab : gaussfit ] ) . figure [ fig:6 - 9 ] shows an enlargement of the 69 kev spectra , and residuals between the data and the continuum model . the time histories of the individual absorption - line equivalent widths and their energies are plotted in figure [ fig : gauss ] . while the equivalent width of the h - like line was almost constant throughout the observations , that of the he - like line increased by a factor of two between the first and second observations , and then perhaps increased slightly more through the subsequent observations . these imply that the ionization state of the absorbing gas decreased over time , as expected from the declining source luminosity . the time averaged center energies of the two strong iron k@xmath2 lines are estimated to be @xmath86 kev and @xmath87 kev , suggesting a blue shift velocity of @xmath88 and @xmath89 , respectively . if we take only the first observation , the blue shifts become more significant with @xmath90 and @xmath91 , for the h - like and he - like irons , respectively . here , the errors are statistical only . given the current uncertainty in the gain calibration of @xmath92 ev or @xmath93 ( see 2.3 ) , the blue shift can not be measured with higher significance . however , a blue shift is physically expected , since the material responsible for the absorption lines is certainly a wind from the disk ( 1 ) , and these values are consistent with the observed blue shift from other accreting systems like gro j@xmath9 @xcite and gx @xmath94 @xcite . in the next subsection , the absorption line parameters and their time evolution are studied in detail . ( 70mm,60mm)figure3a.ps ( 70mm,60mm)figure3b.ps ( 70mm,60mm)figure3c.ps ( 70mm,60mm)figure3d.ps ( 70mm,60mm)figure3e.ps ( 70mm,60mm)figure3f.ps ( 80mm,40mm)figure4.ps ( 80mm,40mm)figure5.ps lllllll epoch & continuum & gaussian 1 & gaussian 2 & gaussian 3 & gaussian 4 & @xmath95 + & @xmath71 [ @xmath96 & @xmath97 [ kev ] & @xmath97 [ kev ] & @xmath97 [ kev]&@xmath97 [ kev ] & + & @xmath98 [ kev]&ew [ ev ] & ew [ ev ] & ew [ ev]&ew [ ev ] & + & @xmath83 [ @xmath99 km]&@xmath100)$ ] @xmath101 & @xmath100)$ ] @xmath101 & & + 1 & @xmath102 & @xmath76 & @xmath75 & @xmath103 & @xmath104 & 686.3/566 + & @xmath105 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & + & @xmath110 & @xmath111&@xmath112&@xmath113&99.98% & + 2 & @xmath114 & @xmath115 & @xmath116 & @xmath117 & @xmath118 & 628.6/566 + & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & + & @xmath124 & @xmath125&@xmath126&@xmath127&@xmath128% & + 3 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & 538.8/566 + & @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & + & @xmath139 & @xmath111&@xmath140&@xmath141&99.998% & + 4 & @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & 636.7/566 + & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & + & @xmath152 & @xmath153&@xmath154&@xmath155&@xmath128% & + 5 & @xmath156 & @xmath157 & @xmath158 & ( 7.87.9)@xmath159&(8.28.3)@xmath159&599.9/566 + & @xmath160 & @xmath161 & @xmath162 & @xmath163&@xmath164 & + & @xmath165 & @xmath166&@xmath167&@xmath168&15% & + 6 & @xmath169 & @xmath170 & @xmath171 & @xmath172&(8.28.3)@xmath159 & 616.0/566 + & @xmath173 & @xmath174 & @xmath175 & @xmath176&@xmath177 & + & @xmath178 & @xmath179&@xmath180&@xmath181&62% & + ( 80mm,40mm)figure6.ps the simple gaussian analysis above shows significant he - like and h - like iron k@xmath2 absorption lines in the data , and suggests the presence of the weaker absorption structures at 7.88.2 kev . since the parameters of the weaker high - energy lines are difficult to constrain , below we focus on the stronger two lines , and study them using 6.07.5 kev spectra . the limited energy range means that we do not need to include the ( instrumental ) broad gaussian at 10 kev or the narrow gaussian at @xmath182 kev . we model the continuum by diskbb as before , with free parameters @xmath183 and @xmath83 , but fix @xmath71 to @xmath184 , which is the average value obtained from the 210 kev fits . this does not affect the derived line center energies or equivalent widths , since the differences of @xmath71 from the fixed value , at most @xmath185 , produce a rather small change to the iron edge at 7.1 kev . to conduct a detailed analysis of these absorption lines , we use a full voigt profile as implemented in the kabs model . the kabs model is constructed by @xcite ( see their 3.2.1 ) as a local model for xspec , and has three fitting parameters for each specific ion ; the ion column density , @xmath186 , the velocity dispersion , @xmath187 , and the bulk outflow velocity characterized by a blue shift , @xmath188 . the line shape is produced through a convolution of a lorentzian profile , describing natural and pressure broadening with a gaussian describing doppler motions ( specified by the velocity dispersion including thermal and kinetic motion ) . except for forbidden lines , each individual line of the k@xmath2 multiplet ( k@xmath189 and k@xmath77 ) is included separately , with the appropriate energy , oscillator strength , and einstein coefficient . this properly models the fine structure in the line even though these details of the line profile can not be resolved by the ccd . the values of chi - squared are the same as for the previous negative gaussian fits , but this model gives the ion column @xmath186 for an assumed velocity dispersion , @xmath187 , as opposed to simply an equivalent width . we require the two lines to have the same @xmath187 and the same @xmath188 , and neglect any line emission which could fill in the absorption . although we can not measure @xmath187 from line profiles , its upper and lower bounds may be obtained . the upper limit is set at @xmath190 by the upper limits on the line width , @xmath19140 ev ( 3.2 ) . the lower limit to @xmath187 is given by the thermal velocity . at such high ionization , the self - consistent temperature of the photoionized material must be of order the compton temperature , @xmath192 . considering the balance between compton heating by the dominant disk illumination , and compton , bremsstrahlung and ion coolings , @xmath192 is roughly estimated as @xmath193 ( i.e. , 0.5 kev in this case ) . the flux in the power - law tail has only a very small effect on this as it is so much weaker than the disk ( 20100 kev power - law flux is only @xmath194% of the disk bolometric flux ) , and does not contribute significantly on the compton heating ( e.g. , @xcite ) . this temperature implies thermal ( random kinetic ) velocties as @xmath195 , and thus @xmath196 can be obtained as a secure lower limit . table [ tab : kabs ] gives the derived @xmath186 for these two limiting velocities , together with velocities of @xmath197 , 200 , and @xmath198 . as expected from the curve - of - growth analysis , the observed equivalent width can be produced by a lower value of @xmath186 if a higher value of @xmath187 is assumed . in table [ tab : kabs ] , we can put further constraints on @xmath187 by using the observed equivalent widths of weaker lines at 7.8 kev and 8.2 kev ( gaussian 3 and 4 in table[tab : gaussfit ] ) . the table includes the model predictions for the equivalent width of the iron k@xmath3 absorption lines . there , models with 1 " in the last column ( note " ) are inconsistent , because the predicted iron k@xmath3 line equivalent widths exceeds the actually observed values , leaving no room for the nickel k@xmath2 lines to contribute . by scanning @xmath187 with finer steps , we found that @xmath187 of @xmath199@xmath200 significantly over predicts the k@xmath3 equivalent width , and is hence ruled out . the top three panels of figure [ fig : kabs ] show the time histories of the best - fit kabs parameters , assuming @xmath201 which is based on the marginally resolved line width in the chandra hetgs data of gx @xmath94 @xcite . the column density of the h - like iron , @xmath202 , is almost constant at @xmath203 while that of the he - like iron , @xmath204 , increases significantly from the first to the second observation . this is consistent with the behavior of the equivalent width estimated using the negative gaussian fit ( table [ tab : gaussfit ] ) . the weighted average blue shift is about @xmath205 , corresponding to an outflow velocity of @xmath206 . again , the first observation shows @xmath207 or @xmath208 ( see table [ tab : kabs ] ) . they are consistent with the results obtained with the negative gaussian fits in 3.2 . ccccccccccc epoch & & & @xmath186 ratio@xmath19&&note@xmath159 + & @xmath204 & @xmath202 & @xmath209 & @xmath210 & fe xxv & fe xxvi&he / h&@xmath211 & @xmath212 & + & @xmath213 & @xmath213 & @xmath214&@xmath215 & ev & ev&@xmath216&@xmath217 & @xmath218 & + + 1 & @xmath219 & @xmath220 & @xmath221 & 116.1 & 4.1 & 12.5 & @xmath222 & @xmath223 & @xmath224&2 + 2 & @xmath225 & @xmath226 & @xmath227 & 97.9 & 9.3 & 11.3 & @xmath228 & @xmath229 & @xmath230&2 + 3 & @xmath231 & @xmath232 & @xmath233 & 98.8 & 7.9 & 12.6 & @xmath234 & @xmath235 & @xmath236&2 + 4 & @xmath237 & @xmath238 & @xmath239 & 102.4 & 10.1 & 11.8 & @xmath240 & @xmath241 & @xmath242&2 + 5 & @xmath243 & @xmath244 & @xmath245 & 85.7 & 8.1 & 10.3 & @xmath246 & @xmath247 & @xmath248&2 + 6 & @xmath249 & @xmath250 & @xmath251 & 108.2 & 10.3 & 12.0 & @xmath252 & @xmath253 & @xmath254&1,2 + + 1 & @xmath255 & @xmath256 & @xmath257 & 115.9 & 3.1 & 17.1 & @xmath258 & @xmath259 & @xmath260 & 1 , 2 + 2 & @xmath261 & @xmath262 & @xmath263 & 98.0 & 11.3 & 17.6 & @xmath264 & @xmath265 & @xmath266&2 + 3 & @xmath267 & @xmath268 & @xmath269 & 99.2 & 11.1 & 17.1 & @xmath270 & @xmath271 & @xmath272&2 + 4 & @xmath273 & @xmath274 & @xmath275 & 103.5 & 12.9 & 16.8 & @xmath276 & @xmath277&@xmath278 & 2 + 5 & @xmath279 & @xmath280 & @xmath281 & 87.5 & 11.9 & 16.4 & @xmath282 & @xmath283&@xmath284 & 1 , 2 + 6 & @xmath285 & @xmath286 & @xmath287 & 107.2 & 13.9 & 17.0 & @xmath288 & @xmath289 & @xmath290&1 , 2 + + 1 & @xmath291 & @xmath292 & @xmath293 & 115.6 & 2.2 & 15.6 & @xmath294 & @xmath295 & @xmath296 & 1 + 2 & @xmath297 & @xmath298 & @xmath299 & 99.8 & 8.1 & 17.8 & @xmath300 & @xmath229 & @xmath301 & + 3 & @xmath302 & @xmath303 & @xmath269 & 98.9 & 8.0 & 15.2 & @xmath304 & @xmath305 & @xmath306 & + 4 & @xmath307 & @xmath308 & @xmath309 & 103.8 & 9.9 & 14.3 & @xmath310 & @xmath311&@xmath312 & + 5 & @xmath313 & @xmath314 & @xmath315 & 87.2 & 9.1 & 12.3 & @xmath316 & @xmath317 & @xmath318&1 + 6 & @xmath319 & @xmath320 & @xmath321 & 107.4 & 11.6 & 15.0 & @xmath322 & @xmath294 & @xmath323 & 1 + + 1 & @xmath324 & @xmath325 & @xmath326 & 115.6&1.9&8.5 & @xmath327 & @xmath328 & @xmath329 & + 2 & @xmath330 & @xmath331 & @xmath332 & 99.8&4.7 & 9.7 & @xmath333 & @xmath334 & @xmath335 & + 3 & @xmath336 & @xmath337 & @xmath338 & 98.7&4.9&8.3 & @xmath339 & @xmath340 & @xmath341 & + 4 & @xmath342 & @xmath343 & @xmath344 & 103.7&5.4 & 8.0 & @xmath345 & @xmath346 & @xmath347 & + 5 & @xmath348 & @xmath349 & @xmath350 & 86.3&5.4 & 7.7 & @xmath351 & @xmath352 & @xmath353 & + 6 & @xmath354 & @xmath355 & @xmath356 & 108.3&6.5 & 8.5 & @xmath357 & @xmath358&@xmath347 & + + 1 & @xmath359 & @xmath360 & @xmath361 & 123.7 & 1.9 & 7.2 & @xmath362 & @xmath363 & @xmath364 & + 2 & @xmath365 & @xmath366 & @xmath367 & 95.0 & 4.1 & 8.5 & @xmath368 & @xmath369 & @xmath347 & + 3 & @xmath370 & @xmath371 & @xmath372 & 101.1 & 4.5 & 7.1 & @xmath373 & @xmath374 & @xmath375 & + 4 & @xmath376 & @xmath377 & @xmath378 & 100.2 & 4.9 & 7.5 & @xmath379 & @xmath380 & @xmath381 & + 5 & @xmath382 & @xmath383 & @xmath384 & 82.6 & 4.4 & 6.9 & @xmath385 & @xmath386 & @xmath387 & + 6 & @xmath388 & @xmath389 & @xmath390 & 110.4 & 5.1 & 6.9 & @xmath391 & @xmath392 & @xmath393 & + ( 80mm,40mm)figure7.ps we observed very significant absorption lines from he like and h like iron k@xmath2 in the disk dominated x - ray spectrum of 4u @xmath0 during its decline from outburst . the inferred blue shift of the absorption lines supports the presence of an outflow . this source thus joins the growing number of galactic binaries with such absorption features , suggesting that winds are a generic feature of bright accretion disks observed at high inclination angles . through the analyses in 3 , we found that iron is always predominantly h like , requiring that the ionization parameter , @xmath394 ( where @xmath395 and @xmath396 are the number density and the distance of the absorbing material from the illuminating source ) , be very high . in this section , we use the xstar photoionization code ( version 2.1kn5 , @xcite ) to calculate the ionization balance of the line - producing material , to convert the ion column densities into physical parameters including total column density , number density , and location , for each observation . ( 80mm,60mm)figure8a.ps ( 80mm,60mm)figure8b.ps we derive the physical parameters of the absorber using the first observation as a specific example to illustrate the method . we calculate the relative ion populations under illumination by a diskbb spectrum , and use this to predict a theoretical ratio of @xmath204 to @xmath202 . the observed ratio from the data then enables us to estimate the @xmath211-parameter , and so convert the observed ion column densities of h - like and he - like irons into a total column density , @xmath212 . we use the illuminating spectrum derived from the suzaku data , i.e. , a diskbb spectrum of @xmath397 kev , to calculate the self - consistent ion populations of iron using xstar . the solid lines in figure [ fig : fepop]a shows the calculated relative fractions of each ionization stage of iron , @xmath398 , against @xmath211 , while figure [ fig : fepop]b shows @xmath399 . the ratio can be well approximated by a power law , as @xmath400 . similarly , we obtain the fraction of h - like iron as @xmath401 at high values of @xmath211 ( roughly @xmath402 ) . since a value of @xmath212 is obtained as @xmath403 , where @xmath404 is the solar iron abundance , it can be estimated as @xmath405 . by using these relations , we derive @xmath211 and @xmath212 for the range of @xmath187 given in table [ tab : kabs ] . the derived values of @xmath211 , @xmath212 and @xmath406 are also summarized in the same table . in general , all these quantities are anti - correlated with @xmath187 , and at low velocity dispersions of @xmath407 and @xmath408 , @xmath212 exceeds @xmath409 i.e. , is optically thick to electron scattering . this case is indicated as 2 " in the last column ( note " ) of the table . thus , the case of @xmath410 is ruled out by both the k@xmath3 absorption structures and the condition of optically thin scattering . we can thus conclude that there must be higher velocities present than expected simply from the temperature of the material , pointing to the importance of either a structured velocity field as expected in an accelerating outflow , and/or turbulence . we repeat the above analysis for all the datasets . since the value of @xmath98 changes as the source decline , we calculate xstar models @xmath183 spanning the observed range of 1.2 , 1.3 , and 1.4 kev , and then use this to interpolate the ion populations for each observed temperature . the dashed lines in figure [ fig : fepop ] show the results with @xmath411 kev , in comparison with the solid lines which assumes @xmath397 kev . we approximated the behavior of the ratio of he - like to h - like ion columns and @xmath412 by following two formulae , @xmath413 @xmath414 where @xmath412 peaks at @xmath415 . using these formulae , we converted the observed values of @xmath406 and @xmath416 into @xmath212 and @xmath211 , as given in table [ tab : kabs ] . for all the data sets , @xmath196 ( @xmath417 kev ) and @xmath418 are again ruled out because @xmath212 becomes optically thick , as well as generally over - predicting the iron k@xmath3 absorption ( 3.3 ) . the inferred time histories of the values of @xmath211 and @xmath212 are shown in figure [ fig : kabs ] assuming @xmath201 , together with upper and lower bounds corresponding to @xmath419 and @xmath420 . assuming that the velocity dispersion stays constant at @xmath201 then there is a marginal decrease in @xmath212 as the luminosity decreases , from @xmath421 in the first observation to @xmath422 in the final dataset , as well as a marginal decrease in @xmath211 from @xmath423 to @xmath424 . we can roughly estimate @xmath395 and @xmath396 of the absorbing plasma , by using the definitions of @xmath425 and @xmath394 where @xmath426 is the launching radius of the absorber . in this section , we use @xmath201 to estimate @xmath211 and @xmath212 , and choose to replace luminosity with flux , so as to make the dependence on the poorly constrained source distance explicit . we parameterize this in terms of @xmath427 , denoting distance in units of 10 kpc , and note there is no uncertainty in @xmath211 parameter associated with the inclination , since the absorber is along the same line of sight as we are . the description of @xmath212 can be simplified as @xmath428 with a mean number density @xmath429 and a characteristic thickness of the absorber @xmath430 . using @xmath431 , the average distance of the absorber , @xmath432 , is thus estimated as @xmath433 and @xmath434 , for the first and the last observations , respectively . the densities are also estimated as @xmath435 and @xmath436 . assuming @xmath437 by following the previous research ( e.g. , @xcite ; 2004 ) , the values of @xmath432 and @xmath429 are estimated to be on the order of @xmath438 and @xmath439 . they depend on the plasma thickness @xmath440 , and the longer distance and higher density are realized on the the condition of smaller @xmath440 . however , since this structure is continuous , it seems unlikely to have too small thickness ( e.g. , @xmath441 ) . thus the range of the distance can be securely estimated as @xmath442 . this value is well consistent with those estimated from the other binaries which showed absorption line structures . because of lack of any evidence for low ionization material , it is reasonable and convenient to assume a constant ionization state throughout the matter . we thus assume that @xmath395 drops as @xmath443 to keep @xmath211 constant at different distances . this is also physically realistic for a geometrically thick absorber , since mass conservation requires this dependency for a constant outflow velocity with constant opening angle . in this case , @xmath212 and @xmath211 are described as @xmath444 and @xmath445 , respectively , where @xmath446 is the launching density at @xmath426 . the launching distance is thus estimated as @xmath447@xmath448 cm throughout the observations , with the associated density of @xmath449@xmath450 @xmath451 . these values are in good agreement with the previous estimation of the mean distance @xmath432 as expected as averaging the density profile gives an effective width of @xmath452 . there are no significant changes over the six observations , but the data are consistent with a small increase in @xmath426 and/or a small decrease in @xmath446 as the luminosity decreases . there are some uncertainties on these numbers due to the statistical uncertainties in @xmath212 and @xmath211 , but the larger source of uncertainty is systematic from the assumed velocity dispersion of @xmath201 . the lower limit on @xmath187 gives an upper limit on @xmath212 and @xmath211 , and thus results in smaller launching radius with larger density . considering our range in @xmath187 of 200@xmath420 , the allowed range of @xmath426 is estimated to be ( 0.24)@xmath453 cm and ( 15)@xmath453 cm for the first and the last observations , respectively . the estimated launch radii are much smaller than the distance traveled by the absorbing material over a single 20 ks observation from the measured blue shift of @xmath454 , @xmath455 . thus the persistence of the wind over the two - month monitoring period , where the source was always in the high / soft state , shows that the material is not associated with a one off event such as a state transition . neither can it be connected to the jet since the radio emission is strongly suppressed in the high / soft state ( fender et al . 1999 ) . instead it must be produced by the disk in a continuous or quasi - continuous manner . here we use the estimated physical parameters of the absorber to constrain the driving mechanism for this wind . one of the candidate mechanisms is radiation pressure on electrons , which becomes dynamically important as @xmath456 approaches @xmath457 , reducing the effective gravity by a factor @xmath458 . this can be made much more efficient if the cross - section for interaction between the matter and radiation is enhanced by line opacity . there are multiple line transitions in the uv region of the spectrum , so uv emitting disks can drive a powerful wind at luminosities far below eddington . such line driven disk winds are seen in cv s ( @xcite ) and are probably also responsible for the broad absorption line ( bal ) outflows seen in agn ( @xcite ) . however , the disk temperature for black hole binaries is in the x - ray regime , so line driving is probably unimportant here ( proga & kallman 2002 ) . we do indeed see lines , but the momentum absorbed in these iron transitions is very small , completely insufficient to drive the wind . another type of outflow from a disk is a thermally driven wind ( @xcite ) . here again the central illumination is important , but the process is less direct . the illumination heats the upper layers of the disk to a temperature of order the compton temperature , @xmath459 . this will expand due to the pressure gradient , and at large enough radii the thermal energy driving the expansion is larger than the binding energy , leading to a wind from the outer disk . simple estimates of the launching radius of this wind give @xmath460 cm ( @xcite ) . this is a much larger distance than inferred for our wind . however , this is an overestimate , and a more careful analysis shows that thermal winds are launched at a radius a factor 510 smaller than this ( @xcite ; @xcite ) . secondly , the _ combination _ of thermal driving and radiation pressure from moderately sub - eddington illumination can be much more effective at producing a wind . the hydrodynamic calculation of proga & kallman ( 2002 ) shows that thermal driving with @xmath461 gives a powerful wind from well within the expected thermal driving radius . their simulation is not exactly matched to the situation here , as they model a neutron star rather than a black hole . nonetheless , their wind is launched from a distance of around @xmath462 , has a column between @xmath463 @xmath464 and ionization of @xmath465 for inclinations larger than @xmath15 . this is very close to our inferred parameters for the wind seen here in 4u @xmath0 . the simulation even has an outflow velocity of @xmath454 as currently indicated by the data . the last type of outflow is a magnetically driven wind . these are much harder to quantitatively study as the magnetic field configuration is not known , yet they are almost certainly present at some level as the underlying angular momentum transport is known to be due to magnetic fields ( see e.g. balbus & hawley 2002 ) . winds ( and jets ) are clearly present in magnetohydrodynamical ( mhd ) simulations which include these magnetic stresses self consistently . these generically show that the mass loss is stochastic , with large fluctuations both spatially and temporally , but that the time averaged properties are well defined , so this magnetic wind is quasi - continuous ( e.g. , @xcite ) . these calculations are still in their infancy , especially for describing the properties of a geometrically thin disk as appropriate here , so can not yet be used for quantatiative predictions of the wind in the high / soft state . instead , an approximation to the properties of the self - consistent magnetic wind from an accretion disk can be made by imposing an external field geometry ( proga 2000 ; 2003 ) . the mass loss rates depend on this field configuration , but in general these allow steady , powerful winds to be launched from any radius . this lack of diagnostic power means that they can be reliably identified only when all other potential mechanisms for a wind are ruled out . while the physical parameters required for the absorbing material seen in 4u @xmath0 are somewhat more extreme than produced by the simple analytic estimates of a thermally driven wind , they seem to be potentially within the range of composite thermal / radiation pressure driven winds if @xmath466 . unfortunately there are no firm estimates for the distance and black hole mass for 4u @xmath0 , and thus the eddington ratio ( @xmath467@xmath468 ) is hard to constrain . neither are there any directly applicable simulations of these winds , and how their properties might change with @xmath469 decreasing by a factor 1.7 . hence there is no unambiguous requirement for magnetic driving from these data . nonetheless , this wind is clearly very similar in properties ( though with a slightly higher ionization state ) to that that seen in a similar high / soft state from another black hole , gro j1655 - 40 , where @xcite inferred a very small launching radius from the chandra grating data requiring a magnetic wind . given the continuity of properties seen between the wind in 4u @xmath0 and gro j@xmath9 then it seems likely that if the wind in gro j@xmath9 is magnetically driven , then this process is important also in the 4u @xmath0 absorber examined here . we are grateful to all the suzaku team members . we also thank tim kallman for making the xstar package publicly available . is supported by a special postdoctoral researchers program in riken . is supported by the japan - russia research cooperative program of jsps . thanks isas for hospitality during the period over which this work was carried out . the present work is supported in part by grant - in - aid no.1703011 from ministry of education , culture , sports , science and technology of japan . abe , y. , fukazawa , y. , kubota , a. , kasama , d. , & makishima , k. 2005 , pasj , 57 , 629 augusteijn , t. , kuulkers , e. , & van kerkwijk , m. h. 2001 , , 375 , 447 balbus , s. a. , & hawley , j. f. 2002 , , 573 , 749 begelman , m. c. , mckee , c. f. , & shields , g. a. 1983 , , 271 , 70 bloser , p. f. , barret , d. , grindlay , j. e. , zhang , s. n. , harmon , b. a. , fishman , g. j. , & paciesas , w. s. 1996 , , 120 , 191 boirin , l. , & parmar , a. n. 2003 , , 407 , 1079 boirin , l. , parmar , a. n. , barret , d. , paltani , s. , & grindlay , j. e. 2004 , , 418 , 1061 blandford , r. d. , & payne , d. g. 1982 , , 199 , 883 cui , q. , chen , w. , & zhang , s. n. 2000 , , 529 , 952 church , m. j. , reed , d. , dotani , t. , bauciska - church , m. , & smale , a. p. 2005 , , 359 , 1336 davis , s. w. , blaes , o. m. , hubeny , i. , & turner , n. j. 2005 , apj , 621 , 372 dieters , s. w. , et al . 2000 , apj , 538 , 3307 drake , g. w. 1988 , canadian j. phys . , 66 , 586 fender , r. , et al . 1999 , , 519 , l165 hawley j. f. , krolik j. h. , 2001 , apj , 548 , 348 hjellming , r. m. , et al . 1999 , , 514 , 383 jones , c. , forman , w. , tananbaum , h. , & turner , m. j. l. 1976 , , 210 , l9 kalemci , e. , tomsick , j. a. , rothschild , r. e. , pottschmidt , k. , & kaaret , p. 2004 , , 603 , 231 kallman t. r. , angelini l. , boroson b. , cottam j. , 2003 , apj , 583 , 861 kallman , t. , & bautista , m. 2001 , , 133 , 221 kuulkers , e. , van der klis , m. , & parmar , a. n. 1997 , , 474 , l47 kuulkers , e. , wijnands , r. , belloni , t. , mendez , m. , van der klis , m. , & van paradijs , j. 1998 , apj , 494,753 kokubun , m. , et al . 2006 , , submitted kotani , t. , ebisawa , k. , dotani , t. , inoue , h. , nagase , f. , tanaka , y. , & ueda , y. 2000 , , 539 , 413 erratum : , 2006 , apj , in press , astro - ph/0003237 koyama , k. et al . 2006 , , accepted kubota , a. , & makishima , k. , 2004 , apj , 601 , 428 lee , j. c. , reynolds , c. s. , remillard , r. , schulz , n. s. , blackman , e. g. , & fabian , a. c.2002 , , 567 , 1102 levine , a. m. , bradt , h. , cui , w. , jernigan , j. g. , morgan , e. h. , remillard , r. , shirey , r. e. , & smith , d. a. 1996 , , 469 , l33 makishima , k. , maejima , y. , mitsuda , k. , bradt , h.v . , remillard , r. a. , tuohy , i. r. , hoshi , & r. , nakagawa , m. 1986 , apj 308 , 635 mcclintock , j. e. , remillard , r. a. 2003 , in compact stellar x - ray sources , eds . lewin w. h. g. , & van der klis , m. , ( cambridge university press , cambridge ) , ( astro - ph/0306213 ) miller , j. m. , raymond j. , fabian , a. , steeghs , d. , homan , j. , reynolds , c. , van der klis , m. , & wijnands , r. 2006 , nature , 441 , 953 mitsuda , k. et al . 1984 , pasj , 36 , 741 mitsuda , k. et al . 2006 , , in press morrison , r. , & mccammon , d. 1983 , , 270 , 119 oosterbroek , t. , parmar , a. n. , kuulkers , e. , belloni , t. , van der klis , m. , frontera , f. , & santangelo , a. 1998 , , 340 , 431 parmar , a. n. , williams , o. r. , kuulkers , e. , angelini , l. , & white , n. e. 1997 , , 319 , 855 parmar , a. n. , stella , l. , & white , n. e. 1986 , apj , 304 , 664 parmar , a. n. , angelini , l. , & white , n. e. 1986 , , 304 , 664 pereyra , n. a. , kallman , t. r. , & blondin , j. m. 2000 , , 532 , 563 proga , d. 2000 , , 538 , 684 proga , d. 2003 , , 585 , 406 proga . d. , & kallman , t.r . 2002 , , 565 , 455 proga , d. , stone , j. m. , & kallman , t. r. 2000 , , 543 , 686 reynolds , c. s. , & wilms , j. 2000 , , 533 , 821 serlemitsos , p. , et al . 2006 , , in press schattenburg , m. l. , & canizares , c. r. 1986 , apj , 301 , 759 shakura , n. i. , & sunyaev , r.a . 1973 , a&a , 24 , 337 sidoli , l. , oosterbroek , t. , parmar , a. n. , & lumb , d. 2001 , , 379 , 540 sidoli , l. , parmar , a. n. , & oosterbroek , t. , lumb , d. 2002 , , 385 , 940 takahashi , t. , et al . 2006 , , in press tanaka , y. , & lewin , w. h. g. 1995 , in x - ray binaries , eds . w. h. g. lewin , j. van paradijs , and w. p. j. van den heuvel ( cambridge university press , cambridge ) , p126 tomsick , j. a. , lapshov , i. , & kaaret , p. 1998 , , 494 , 747 toor , a. , seward , f. d. 1974 , aj , 79 , 995 trudolyubov , s. p. , borozdin , k. n. , & priedhorsky , w. c. 2001 , mnras , 322 , 309 ueda , y. , asai , k. , yamaoka , k. , dotani , t. , & inoue , h. 2001 , , 556 , l87 ueda , y. , inoue , h. , tanaka , y. , ebisawa , k. , nagase , f. , kotani , & t. , gehrels , n. 1998 , apj , 492 , 782 ueda , y. , murakami , h. , yamaoka , k. , dotani , t. , & ebisawa , k. 2004 , apj , 609 , 325 watarai , k. , fukue , j. , takeuchi , m. , & mineshige , s. 2000 , pasj , 52 , 133 woods , d. t. , klein , r. i. , castor , j. i. , mckee , c. f. , & bell , j. b. 1996 , , 461 , 767
we first obtain the most general and compact parametrization of the unitary transformation diagonalizing any hermitian matrix , as a function of its elements and eigenvalues . we then study a special class of fermion mass matrices , defined by the requirement that all of the diagonalizing unitary matrices ( in the up , down , charged lepton and neutrino sectors ) contain at least one mixing angle much smaller than the other two . our new parametrization allows us to quickly extract information on the patterns and predictions emerging from this scheme . in particular we find that the phase difference between two elements of the two mass matrices ( of the sector in question ) controls the generic size of one of the observable fermion mixing angles : i.e. just fixing that particular phase difference will `` predict '' the generic value of one of the mixing angles , irrespective of the value of anything else .
0910.3132
in the absence of flavor symmetries , the yukawa couplings between the standard model ( sm ) fermions and the higgs field are in general complex arbitrary matrices which , after electroweak symmetry breaking ( ewsb ) , become the mass matrices of the quarks and charged leptons . in the case of neutrinos , the mass matrix will be in general complex symmetric . all these matrices contain more parameters than physical observables and an explicit computation of these observables ( fermion masses and mixings ) in terms of the original matrix elements can be quite cumbersome in general . indeed this would require us to solve a @xmath2 eigenvalue problem for each fermion matrix , and then compose the unitary transformations ( formed with the calculated eigenvectors ) of the up and down quark sectors and then also of the charged lepton and neutrino sectors . we observe however that it may be useful to address the question not as an eigenvalue problem , but as an eigenvector problem , treating the eigenvalues as input parameters and not as output . the first reason for this is that except for the neutrino sector , all the mass eigenvalues are quite well known . but the main point we make is that by keeping explicitly the mass eigenvalues as input parameters , the eigenvector solutions of each mass matrix become surprisingly simple and can be written as compact functions of both the mass matrix elements and the eigenvalues ( the fermion masses ) . such a parametrization of the mixing matrices , directly in terms of the original mass parameters and the fermion masses might prove to be useful in the studies aiming to explain the observed flavor structure of the sm by way of symmetries or textures or patterns @xcite at the level of the fermion mass matrices . it is true , though , that to proceed we need to work with hermitian matrices , but it is always possible to render the quark mass matrices hermitian in the standard model without loss of generality @xcite . the procedure to obtain hermitian matrices is quite standard , and it involves either working with the hermitian matrix @xmath3 , where @xmath4 is the original fermion mass matrix , or using its polar decomposition , i.e. solving @xmath5 , where @xmath6 is a unitary matrix converting the general complex matrix @xmath4 into a positive semi - definite hermitian matrix @xmath7 ( if @xmath4 is invertible , with distinct non - vanishing eigenvalues , then @xmath7 is positive definite and therefore @xmath6 is unique ) . we will first present our parametrization ( in fact 9 different types ) for the unitary matrix @xmath8 which diagonalizes a general hermitian matrix @xmath1 in the most compact way possible . then , to start taking advantage of it , we then propose a simple and mildly constraining ansatz for the flavor structure of the sm fermion sector . it assumes that the unitary transformations diagonalizing @xmath9 and @xmath10 can each be decomposed as only two rotations , instead of three . the idea is to assume that the third rotation angle is zero ( or much smaller than the other two ) and therefore one of the entries of each transformation matrix @xmath11 and @xmath12 will be zero or close to zero . we call this setup the two - angle ansatz and we will concentrate in only one of the many possible cases . our parametrization allows us to quickly obtain very simple dependences of the fermion mixing matrices @xmath13 and @xmath14 @xcite in terms of the masses and the original mass matrix elements . we can thus study easily the interesting properties of this type of ansatz as well as the consequences it has in the original mass matrices , in both the quark and the lepton sectors . a particularly interesting observation is that the specific value of some elements of @xmath9 and @xmath10 has no effect ( or very mild effect ) on the observed values of masses and mixings . as explained before , we are going to concentrate on hermitian matrices with the assumption that the fermion mass matrices are either hermitian or that one can construct a hermitian matrix out of them . all the results presented in this work are valid for general hermitian matrices , but for simplicity we will only consider the case of positive definite hermitian matrices . let @xmath1 be one such matrix : @xmath15 with eigenvalues @xmath16 and @xmath17 . a compact parametrization of the unitary matrix @xmath8 which diagonalizes it , is or @xmath18 ) is explicitly absent in each vector column ( see appendix a for details ) . in the parametrization shown here , the rotation matrix @xmath8 has the correct limit when the off - diagonal entries in the original mass matrices @xmath1 are set to zero , avoiding ( apparent ) divergences in this limit . ] @xmath19 after some algebra the normalization parameters are found to have the simple form @xmath20 , \\ n_2 ^ 2&=&(\lambda_3-\lambda_2)(\lambda_2-\lambda_1 ) \left[(a-\lambda_2)(\lambda_2-\gamma)+|g|^2\right ] , \\ n_3 ^ 2&=&(\lambda_3-\lambda_2)(\lambda_3-\lambda_1 ) \left[(\lambda_3-\gamma)(\lambda_3-\alpha)-|x|^2\right]\end{aligned}\ ] ] the surprisingly simple and compact form of this parametrization might make it suitable to treat flavor models keeping always an explicit dependence on all the matrix elements of the hermitian mass matrices . of course , if the three eigenvalues @xmath21 , @xmath22 and @xmath17 are fixed , there must be three constraint equations on the elements of the matrix @xmath1 . these equations are found from the three invariants @xmath23 , @xmath24 and @xmath25 : @xmath26 by choosing @xmath27 as an independent variable , it is possible to rewrite these constraint relations on the rest of variables as or @xmath28 as the independent variable . ] @xmath29 the interesting thing of this notation is that the constraint formulae on @xmath30 and @xmath28 actually become algebraic solutions for both @xmath30 and @xmath28 when the term @xmath31 vanishes identically . in particular , this is the case when one deals with mass matrices with texture zeroes in the off - diagonal elements . equipped with an exact and simple parametrization of the fermion mixing matrix in both up and down sectors ( or charged lepton and neutrino sectors ) , we look for economical patterns among the mixing matrices by following a bottom - up approach in the hope that it might be complementary to more top - down approaches such as imposing flavor symmetries or texture zeroes mass matrices ( see @xcite as well as the probably incomplete surveys of @xcite ) . one avenue is to find a similar ansatz for the flavor structure of the mixing matrices in both quark and leptonic sectors . such possibility exists in the sense that in both sectors , the mixing elements @xmath32 and @xmath33 are known to be small . this feature is very interesting and it is known to lead to a simple parametrization of the mixing matrix . note that in the limit @xmath34 we have two additional conditions , since @xmath35 is a complex number . this means that in this limit , the mixing matrix will have only two independent parameters instead of four . we can choose these four independent parameters to be @xmath36 , @xmath37 , @xmath38 and @xmath39 @xcite . in the limit of @xmath34 , we have the extra constraint @xmath40 . this means that the whole mixing matrix is described by @xmath36 and @xmath37 . note also that in this limit , there is no @xmath41 violation la dirac in both quark and leptonic sectors . in what follows , the subscript @xmath42 stands for the values of the mixing matrix elements in the limit @xmath34 . since @xmath35 is known to be very small , we believe that the zeroes values of the mixing matrix elements are not far from their measured values . for the quarks , we have : @xmath43 note that this zero - order @xmath44 can be decomposed as a product of two rotations , namely one is purely the cabbibo angle and the other one is purely made out of beauty namely @xmath45 : @xmath46 with @xmath47 in the leptonic sector we have : @xmath48 which can be decomposed also as a product of two rotations , namely one is purely solar and the other one is purely atmospheric : @xmath49 with , @xmath50 the diagonal phase matrix @xmath51 which contains the majorana phases is defined as : @xmath52 this structure for the physical fermion mixing matrices , decomposed mainly into just two rotations is quite suggestive and we will use this observation as the starting point of our analysis . in this context we ask ourselves how many economical possibilities one has to restore minimally and fully the mixing elements and @xmath41 violation . we start with the structure @xmath53 and then to be as general as possible while still keeping the original motivation we consider all mixing patterns emerging from the original structure , but with only one zero in each mixing matrix . now , we establish all the corrected mixing patterns for both up and down quark sectors : @xmath54 @xmath55 where @xmath56 stands for a non - zero value and @xmath57 for a corrected originally zero mixing matrix element . these mixing matrices with one texture zero in them can be decomposed themselves into only two rotations instead of three . we call this the two - angle ansatz , in which both the up and down quark sectors are diagonalized by unitary transformations containing just two angles ( i.e. having one vanishing element ) . there are obviously many possibilities for this ansatz but in particular there are 16 cases such that one can recover in a specific limit the case @xmath34 . in this work we are going to focus only on one specific example of this type of ansatz , although a full case by case study is underway . we feel that the main features of this ansatz do reveal themselves in the example studied here , and we prefer to continue elsewhere a more systematic exploration . + * notation * + before we proceed further , we will set up our notation for all the matrices in both up and down quark sectors as well as charged lepton and neutrino sectors . we will define the hermitian matrices @xmath9 , @xmath10 , @xmath58 and @xmath59 defined by : h_u= , h_d= .[quarksmass ] with eigenvalues @xmath60 and @xmath61 and h_l= , h_= . [ leptonsmass ] with eigenvalues @xmath62 and @xmath17 . in the text , we will also need to refer to the phases of the off - diagonal terms , denoted as & & = _ x,=_g , = _ b , + & & = _ y,=_h , = _ f , and similarly for the leptons . the matrices diagonalizing these mass matrices will be denoted respectively as @xmath11 , @xmath12 , @xmath63 and @xmath64 . we will now consider the mass matrices from eqs . ( [ quarksmass ] ) and ( [ leptonsmass ] ) with the extra constraints of @xmath66 and @xmath67 in the quark sector and @xmath68 and @xmath69 in the lepton sector . and @xmath70 @xcite and similar limits were considered in the lepton sector in @xcite ] for example in the down quark sector imposing @xmath67 corresponds to the requirement @xmath71 . after a short computation , we obtain simpler relations among the elements of the mass matrix : @xmath72 these relations will simplify significantly the general form of the down quark mixing matrix @xmath12 as well as the up quark mixing matrix @xmath11 ( see eq . ( [ wparametrization ] ) and appendix b for details ) . it is then straightforward to show that the quark mixing matrix takes the form v&=&w_u^w_d = p^_u v__ckm p_d where @xmath73 and @xmath74 are unphysical diagonal phase matrices given below and v__ckm=. [ vckm ] the complex rotation parameters @xmath75 , @xmath76 , @xmath77 and @xmath78 are given by c&= & ( i.e. |c| |v_tb| ) [ ceq ] + s&=&e^i _ b [ seq ] + c_&= & [ cgammaeq ] + c_&=&[crhoeq ] note that @xmath79 and that @xmath80 . the cp phase @xmath81 and the two unphysical phase matrices @xmath73 and @xmath74 are given respectively by @xmath82 , @xmath83 and @xmath84 , with @xmath85 . the previous form of the mixing matrix implies the following exact relations for the quark sector = , = . [ vubvcb ] and = , = , [ vubvtd ] which really correspond to two independent constraints ( i.e. the third and fourth relations can be obtained using the first two along with the unitary constraints ) . as can be seen only two original mass matrix elements @xmath86 ( from @xmath9 ) and @xmath87 ( from @xmath10 ) appear explicitly in these last four relations showing the first effect of the @xmath88 ansatz , i.e. linking each of the previous ratios to a different quark mass matrix , and in particular to the first diagonal elements of each mass matrix . once we fix these two elements to fit the experimental value of the ratios given in eqs . ( [ vubvcb ] ) and ( [ vubvtd ] ) , we will still have 6 free parameters ( including 4 phases ) to fit the rest of the data ( i.e we need to fit two more scalar observables from experiment , which for example can be taken to be the absolute value of @xmath89 and the value of the cp phase of the fermion mixing matrix ) . we can choose to use @xmath90 and the phases of @xmath91 and @xmath28 as the free parameters . it is also useful to define the phase combinations : _ 23 = -&=&_f-_b , + _ 12 = -&=&_y-_x + _ 13 = -&=&_h-_g . which have the constraint _ 23 + _ 12 = _ 13 . due to the @xmath88 ansatz imposition . of course , we have more than enough free parameters to fit the two remaining observables , but as we will shortly see , of the 4 free phases , only two phase differences can be relevant , and the 2 real parameters @xmath92 and @xmath93 turn out to be statistically irrelevant if the two phases are properly chosen . in other words , in most of the parameter space of the six free parameters needed to produce the two remaining physical observables , 4 directions are more or less irrelevant , with two phase differences being the two parameters required to obtain a good experimental fit . this situation is somewhat surprising because the amount of useful " free parameters is less than the total number of free parameters . let s see how this is played out . with random @xmath92 , @xmath93 and @xmath94 with respect to @xmath94 ( left panel ) , where @xmath92 and @xmath93 are two diagonal elements of the quark mass matrices @xmath9 and @xmath10 , and @xmath94 is the difference between the phases of the @xmath95 elements of these matrices . on the right panel , we present the distribution of the jarlskog invariant @xmath96 with random @xmath90 and @xmath97 , for @xmath98 . on the two panels , the black curves represent the @xmath99 and @xmath100 quantiles of the distribution for fixed @xmath94 ( left ) and @xmath97 ( right ) . in other words @xmath101 of the random points lie between the curves , with @xmath99 above it and @xmath99 below it . , title="fig:",width=272 ] with random @xmath92 , @xmath93 and @xmath94 with respect to @xmath94 ( left panel ) , where @xmath92 and @xmath93 are two diagonal elements of the quark mass matrices @xmath9 and @xmath10 , and @xmath94 is the difference between the phases of the @xmath95 elements of these matrices . on the right panel , we present the distribution of the jarlskog invariant @xmath96 with random @xmath90 and @xmath97 , for @xmath98 . on the two panels , the black curves represent the @xmath99 and @xmath100 quantiles of the distribution for fixed @xmath94 ( left ) and @xmath97 ( right ) . in other words @xmath101 of the random points lie between the curves , with @xmath99 above it and @xmath99 below it . , title="fig:",width=272 ] eq . ( [ vubvcb ] ) shows that @xmath102 and @xmath103 are required in order to obtain a good fit with experimental data @xcite . this has interesting implications for @xmath104 since it means that @xmath105 and @xmath106 after using the trace identity of @xmath9 and @xmath10 . we can therefore write v_tb(a , d,_23 ) it turns out that when the phase @xmath107 is small ( modulo @xmath108 ) , statistically we find that @xmath109 for any randomly chosen value of @xmath92 and @xmath93 . in fact , the generic value of @xmath110 is very much correlated with the value of @xmath107 , with little dependence on the values of the other two variables , at least for small enough @xmath107 . this is shown in figure [ vtbdelta ] , where we plot the distribution of @xmath110 with respect to @xmath107 for randomly chosen values of @xmath90 and @xmath94 . it is apparent that there is a clear correlation between the value of the phase @xmath107 and the value of @xmath89 . the two black curves correspond to the @xmath99 and @xmath100 quantiles of the distribution of @xmath89 for a given value of @xmath107 ( i.e. @xmath101 of the randomly generated points lie between the two curves , with @xmath99 above them and @xmath99 below ) . where @xmath111 , for three different values of @xmath94 . the dashed curve is a contour of the experimental fit for @xmath112 in the case @xmath98 . one sees quite clearly that as soon as @xmath94 is increased , the parameter space favorable for large values of @xmath104 shrinks dramatically . ] all this shows that the generic value of @xmath104 in this ansatz is actually governed by the specific value of the phase @xmath107 , with very mild dependence on the other 2 parameters @xmath92 and @xmath93 . it is not clear though , that demanding @xmath113 will be enough to fit the observed data in the quark sector since @xmath114 , i.e it is quite close to 1 . the random scan of figure [ vtbdelta ] does not show the distribution of @xmath104 for @xmath115 , but once we fix @xmath94 we just have two free parameters , @xmath92 and @xmath93 . in figure @xmath116 we show the complete allowed phase space for @xmath92 and @xmath93 , which are subject to the experimental constraints @xmath117 and @xmath118 . the different shaded regions are the points where @xmath119 , for three different values of the phase difference @xmath94 . when this last one is zero , the region is quite large , and it decreases very fast as the phase is increased . because the area of parameter space can be quite large we may still say that the value of @xmath89 consistent with experimental data could be considered `` generic '' as long as the phase difference @xmath94 is vanishingly small . finally , one can also compute quite easily the jarlskog invariant @xcite in this context . from eq . ( [ vckm ] ) , and using for example the definition @xmath120 , it is easy to see that it will have the form : j&= & c_c _ s_s _ |c||s|^2 [ j1313 ] where @xmath121 , @xmath75 and @xmath76 were given in eqs . ( [ ceq ] ) , ( [ seq ] ) , ( [ cgammaeq ] ) and ( [ crhoeq ] ) , and where @xmath122 with @xmath123 . before analyzing in more detail the dependence on the original mass matrix elements , it is interesting to relate the phase @xmath81 not just to the jarlskog invariant but also to the angles of the unitarity triangle @xmath124 , @xmath125 and @xmath126 . these angles are defined as : _ 1 & = & = , + _ 2 & = & = , + _ 3 & = & = and the relation between them and the jarlskog j is : j = rewrite our j as j = |v_ub||v_td| |v_tb| c_c _ = |v_td||v_cb| |v_tb| c_s _ = |v_ub||v_cb||v_tb| c_s _ [ j1313 ] for example , from the first identities of eq . ( [ jalpha ] ) and ( [ j1313 ] ) one sees that we must have & = & since @xmath127 , it follows from the above equation that @xmath128 . the experimental constraints on @xmath129 are such that @xmath130 @xcite , which basically means that the phase @xmath81 is constrainted to be @xmath131 . a more revealing way to see what this means for the original elements of the mass matrices , we can actually rewrite the jarlskog in a more explicit way as j&= & |v_ub||v_td| c_c _ + & & |v_ub||v_td| c_c _ [ jfunct ] where the approximation of the second line comes from assuming that @xmath132 and @xmath133 . we showed earlier that the imposition of @xmath134 gives a nice statistical reason for the large value of @xmath89 . with that extra condition , one actually has @xmath135 , which then means that @xmath136 due to experimental bounds . the experimental best fit value of the jarlskog invariant is @xmath137 @xcite ; in the right panel of figure [ vtbdelta ] we present a scan of values of the function of @xmath138 shown in eq . ( [ jfunct ] ) , for random values of @xmath92 , @xmath93 and @xmath97 ( assuming @xmath98 ) . it is apparent that the observed value of @xmath138 can be obtained quite generically when @xmath139 , in a quite insensitive way to the specific values of @xmath92 and @xmath93 . it is quite suggestive that some specific phase differences between the elements @xmath140 and @xmath141 , and between the elements @xmath142 and @xmath143 and between the elements @xmath144 and @xmath145 , given by : & & _ 23 = _ f-_b = 0 + & & _ 12 = _ y-_x = + & & _ 13 = _ h-_g = do lead to good generic values of both @xmath104 and @xmath138 . we are left with four parameters @xmath90 and two combinations of phases independent of @xmath97 and @xmath94 , all from the original mass matrices , which do not seem to play any important role in obtaining a good overall fit in the quark sector . before we finish this subsection on the quark sector , we would like to point out that since @xmath92 and @xmath93 can take almost any value ( inside their allowed range ) , one might actually get very close to a symmetric limit , namely a @xmath146 family symmetry , relating second and third families ( see @xcite for implementations mostly in the lepton sector ) . forgetting for a moment our ansatz but assuming the @xmath146 limit for both up and down quark mass matrices , one is then guaranteed to have vanishing elements @xmath147 and @xmath148 ( i.e. we recover our ansatz ) . moreover some of the elements of the mass matrix are subject to the constraints x = g & & y = h + = a & & = d + _ b = & & = _ f. from these equations , we must have @xmath149 and using eq . ( [ ceq ] ) it is easy to see that @xmath150 exactly , which means also that @xmath151 . since the fitted values of @xmath152 , @xmath45 , @xmath153 and @xmath154 are at most @xmath155 and much smaller than @xmath104 , then it seems plausible that a small deviation from this symmetric limit can easily restore the experimental values of these @xmath156 s . to obtain a correct experimental fit , one would also have to require the phases of @xmath30 and @xmath157 to be separated by @xmath158 ( as remarked earlier ) , and moreover the values of @xmath87 and @xmath86 will have to be chosen so as to obtain the correct ratios @xmath159 and @xmath160 ( see eq . ( [ vubvcb ] ) ) . although we imposed a flavor ansatz mainly for empirical reasons and simplicity , it is interesting that the patterns emerging from it do actually lead to a possible type of symmetry . of course we just considered one of the possible cases of our scheme , and others cases might lead to different patterns or reveal some other feature and for this a more systematic study is required , some of it being already underway . in the lepton sector , the mixing matrix takes the same form as for the quarks in the @xmath88 ansatz , i.e. v&= & w_l^w_= p_l^ v__pmns where @xmath161 is an unphysical diagonal phase matrix given below and v__pmns= p _ where @xmath162 is the majorana phase matrix given below and the complex rotation parameters @xmath163 and @xmath164 are given by c&= & ( i.e. |c| |v_3| ) [ vtau3 ] + s&=&e^i_b + c_&= & + c_ & = & note that @xmath165 and that @xmath166 . the cp phase @xmath167 and the majorana phase matrix @xmath162 , as well as the unphysical phase matrix @xmath161 are given respectively by @xmath168 , @xmath169 and @xmath170 , where @xmath171 . again , the previous form of the mixing matrix implies the relations : = , = , [ vtau1vtau2 ] and = , = . with again only two parameters from the original mass matrices separately controlling each ratio . but , the lepton case is different , obviously , because the lepton mixing matrix contains two large angles and also because the neutrino mass structure is unknown and might not be as hierarchical as in the quark sector ( see for example @xcite for the latest global fits coming from neutrino oscillation experiments ) . in the ( 13 - 13 ) ansatz we see that we must still enforce the charged lepton mass matrix to have a very small first diagonal element , i.e @xmath172 in order to obtain a small value for the ratio @xmath173 . this will imply that @xmath174 ( from the trace identity of @xmath58 ) which will simplify the functional form of @xmath175 of eq . ( [ vtau3 ] ) . before doing so , we need to look also into the required value of @xmath176 to obtain a correct fit for @xmath177 in the approximate tri - bimaximal scheme @xmath178 @xcite . this leads to @xmath179 , and we can now use this relation to obtain the simplified form of @xmath175 : v_3(a,d,_23,_1)&= & where @xmath180 is fixed by the measured atmospheric neutrino mass difference and @xmath17 for the neutrino eigenvalues , these could in fact correspond to the squared physical masses if we consider the neutrino mass matrix squared as our starting point . in this case , @xmath181 . ] . we now have 7 free parameters ( including 4 phases ) , since one of the neutrino masses is unknown , and we choose it to be @xmath182 . with random @xmath183 , @xmath184 , @xmath185 and @xmath186 ( defined as the mass of the lightest neutrino ) with respect to @xmath185 ( left panel ) . on the right panel , we present the distribution of the jarlskog invariant @xmath187 normalized to the value of @xmath188 , with random @xmath189 , @xmath190 and @xmath186 , for fixed @xmath191 . on the two panels , the black curves represent the @xmath99 and @xmath100 quantiles of the distribution for fixed @xmath185 ( left ) and @xmath190 ( right ) . in other words @xmath101 of the random points lie between the curves , with @xmath99 above it and @xmath99 below it.,title="fig:",width=272 ] with random @xmath183 , @xmath184 , @xmath185 and @xmath186 ( defined as the mass of the lightest neutrino ) with respect to @xmath185 ( left panel ) . on the right panel , we present the distribution of the jarlskog invariant @xmath187 normalized to the value of @xmath188 , with random @xmath189 , @xmath190 and @xmath186 , for fixed @xmath191 . on the two panels , the black curves represent the @xmath99 and @xmath100 quantiles of the distribution for fixed @xmath185 ( left ) and @xmath190 ( right ) . in other words @xmath101 of the random points lie between the curves , with @xmath99 above it and @xmath99 below it.,title="fig:",width=272 ] as seen in figure [ vtau3delta ] , when @xmath192 ev , the correlation between the phase @xmath193 and the value of @xmath194 is basically the same as it was in the quark sector , and this is basically due to the fact that this feature happens for masses which are hierarchical ( and in fact , for larger values of @xmath186 the correlation between the phase @xmath193 and the value of @xmath194 starts to wash out ) . it is quite suggestive , that if @xmath192 ev and @xmath195 , then we obtain a generic size of @xmath196 , as shown in figure [ vtau3delta ] . because @xmath197 is small , @xmath198 and we will also have @xmath199 . these two values of @xmath200 and @xmath201 are consistent with the tri - bimaximal scheme @xmath178 where @xmath202 and so we find that in the lepton case the preferred value of the phase combination @xmath203 is @xmath204 , which statistically predicts a large value for both the mixing angles @xmath205 and @xmath206 . it is also interesting to note that the charged fermion mass matrix must have the same structure as the up and down quark mass matrices , namely that the first diagonal element is of the order of the lightest eigenvalue . since the mixing element @xmath207 is not small , we also conclude that the neutrino mass structure must be different from the other three matrices . this could be due to the fact that the eigenvalues are not as hierarchical as for the charged fermions . as in the quark case , it is interesting to note that further imposing some flavor symmetry might simplify the relations obtained in this ansatz . in particular one could ask what is the effect of imposing the @xmath146 family symmetry ( see for example @xcite ) only in the neutrino sector . in our case this is a natural question to ask since this symmetry imposes automatically the element @xmath208 to be zero , which is the defining condition of our ansatz . but it also imposes the following constraints on the mass matrix elements ( h_)_12=(h_)_13 , & i.e. & y=h + ( h_)_22=(h_)_33 , & i.e. & =d + = , & i.e. & _ f=this will have interesting consequences on the lepton mixing matrix because in this case we will have v_3 = c&= & + o(m_e / m _ ) where we have assumed that @xmath209 . now one can see that when @xmath210 , independent of anything else , we will have @xmath211 , as well as @xmath212 . thus the @xmath213 symmetry , along with the requirement of the phase of @xmath214 to be @xmath158 , forces the appearance of large mixing angle in the @xmath14 matrix , and we remind the reader that we are not considering a diagonal charged lepton mass matrix , although all our matrices are taken in the @xmath88 ansatz . the form of the @xmath14 matrix under these assumptions is v__pmns= p_where as before , @xmath215 and @xmath216 and where @xmath167 is the dirac cp phase given by @xmath217 and the majorana phase matrix @xmath162 is given by p_= the phase @xmath218 is given by @xmath219 . finally , in the more general case where one does not consider @xmath146 symmetry in the neutrino sector , the leptonic jarlskog invariant @xmath138 is computed to be : j&= & & = & |v_e3||v_1|c_ c_ . for the preferred value of @xmath191 , such that statistically it is more favorable to obtain a @xmath175 mixing angle close to @xmath220 , we obtain j&= & |v_e3||v_1|c_ c_ . this suggests that now , the value @xmath221 will maximize the possible value of @xmath138 ( normalized to @xmath188 ) , and this is confirmed in the random scan shown in figure [ vtau3delta ] . in this work , we started by obtaining the simplest parametrization for the diagonalization of a @xmath222 hermitian matrix , as a function of both its matrix elements and its eigenvalues . since the masses of fermions in the sm are well known , the problem to attack is not an eigenvalue problem , but an eigenvector problem . in other words , we are able to obtain the eigenvectors of a hermitian matrix in an algebraically compact form because we treated the eigenvalues as known parameters , instead of unknown . with this parametrization in hand , we wanted then to show how it could simplify and make quite transparent the analysis and study of some flavor schemes . to this end we defined a new flavor scheme which imposes some ( arguably obscure ) constraints on the flavor structure in the fermion sector . the constraint imposed is the requirement of the vanishing of one of the mixing angles of the diagonalization matrix of each hermitian fermion mass matrix of the sm . we called this the `` two - angle '' ansatz ( i.e. out of three angles to diagonalize a hermitian matrix , we consider the family of hermitian matrices diagonalized by only two angles ) , and it is bottom - up motivated , i.e. it is inspired on the observed structure of both @xmath13 and @xmath14 matrices . nevertheless we also observe that in flavor models where one requires a full symmetry between two of the three families , like the @xmath223 symmetry models , one of the consequences is precisely the vanishing of one of the mixing angles . but we mainly decided to keep the bottom - up approach motivation and study the patterns emerging from our `` two - angle '' ansatz and focused on just one possible implementation , the @xmath88 case , in which the @xmath224 entries of all the diagonalizing matrices happen to vanish . using our parametrization , one can actually write both @xmath13 and @xmath14 in terms of the original matrix elements and eigenvalues . in particular we found a peculiar dependence of one of the mixing angles with the model parameters . we observed that by fixing the phase difference @xmath94 between two parameters of the up and down mass matrices ( or the neutrino and charged lepton mass matrices ) , the mixing angle @xmath104 ( or @xmath175 ) can be `` predicted '' in a statistical sense , i.e. if one makes a random scan allowing the remaining free parameters to take any possible value , including the eigenvalues , one obtains a narrow distribution for the value of @xmath89 ( or @xmath194 ) , and the central value is a monotonic function of the phase difference @xmath94 . this can be seen by rewriting the general formula for @xmath104 in the @xmath88 ansatz ( see eq . ( [ ceq ] ) ) as @xmath225 , @xmath226 and @xmath227 , and @xmath228 and @xmath229 are diagonal elements of the up and down mass matrices respectively . if the mass eigenvalues @xmath230 are unconstrained , the only constraints on the parameters required for a random scan are ^u_1/^u_3 < , < 1 & and & ^d_1/^d_3 < , < 1 since the diagonal elements of a hermitian matrix must be bounded by its largest and lowest eigenvalues . the result of a scan over all these parameters ( including the eigenvalues ) but for fixed @xmath231 is a highly peaked distribution centered at some value . this means that the generic value of @xmath89 is controlled almost exclusively by @xmath94 . of course one may think that after fixing the masses to the experimental values , as well as the other mixing angles , maybe we might loose this statistical prediction . this is not the case as was shown in figure [ vtbdelta ] , where the scan is performed now with only two free parameters , the rest having been fixed by other experimental observables . there is still a clear correlation between the value of the phase difference and the value of the angle @xmath104 . the imposition of this specific two - angle ansatz ( the ( @xmath232 ) case ) on the fermion mass matrices amounts to 2 constraints per mass matrix , and therefore 4 constraints in the quark sector . what we have noted is that with only one more constraint , i.e. the fixing of the phase difference @xmath94 , we are able to give a statistical prediction for the value of the angle @xmath89 ( or @xmath194 ) , and this , irrespective of any other parameter . in particular if @xmath98 we would expect @xmath233 and in the lepton sector , if we fix @xmath191 , then we would expect @xmath234 , both cases being close to the experimental fits . two hermitian matrices contain 18 free parameters altogether and so it is nontrivial that after imposing only 5 constraints on the whole set we obtain one prediction , irrespective of anything else . once noted this nontrivial property , we continued analyzing the rest of consequences of our scheme and showed for example how the cp violating phases in both quark and lepton sectors depend on the phases of the original mass matrix elements . another interesting outcome was the realization of how to treat in a similar way the quarks and charged lepton mass matrices , and use the special case of the neutrino matrix to explain in a transparent way the observed differences between the lepton and quark sectors . in particular we also analyzed the consequences of further imposing a @xmath146 symmetry in the neutrino sector . it is true that out of many possible implementations of the two - angle ansatz we chose to study only one case . we felt that this one case would show the main features of the scheme in a transparent way , and so we leave the systematic case by case study of the scenario for the future , although part of this work is already underway . m.t . would like to thank alex azatov , joseph schechter and lijun zhu for useful discussions . wishes to thank vestislav apostolov , franois bergeron and christophe reutenauer for many interesting discussions and helpful comments . this research was partially funded by nserc of canada . 9 s. weinberg , trans . new york sci . * 38 * , 185 ( 1977 ) ; f. wilczek and a. zee , phy . * b 70 * , 418 ( 1977 ) [ erratum - ibid . * 72b * , 504 ( 1978 ) ; h. fritzsch , phys . lett . * b 73 * 317 ( 1978 ) ; f. wilczek and a. zee , phy . rev . lett . * 42 * , 421 ( 1979 ) ; h. fritzsch , nucl . phys . * b 155 * 189 ( 1979 ) ; c. d. froggat , h. b. nielsen , nucl . phys . * b 147 * , 277 ( 1979 ) . y. nir , n. seiberg , phys . * b 309 * , 337 ( 1993 ) ; k. s. babu and r. n. mohapatra , phys . rev . lett . * 70 * , 2845 ( 1993 ) ; m. leurer , y. nir , n. seiberg , nucl . phys . * b 398 * , 319 ( 1993 ) and nucl . phys . * b 420 * , 468 ( 1994 ) ; d. b. kaplan , m. schmaltz , phys . rev . * d 49 * , 3741 ( 1994 ) ; l. e. ibanez , g. g. ross , phys . lett . * b 332 * , 100 ( 1994 ) ; p. bintruy , p. ramond , phys . lett . * b 350 * , 49 ( 1995 ) ; e. dudas , s. pokorski , c. a. savoy , phys . lett . * b 356 * , 45 ( 1995 ) . p. ramond , r.g . roberts and g.g . ross , nucl . b406 * , 19 ( 1993 ) . d. s. du , z. z. xing , phys . d 48 * , 2349 ( 1993 ) ; l. j. hall , a. rain , phys . * b 315 * 164 ( 1993 ) ; h. fritzsch , z. z. xing , prog . nucl . phys . * 45 * , 1 ( 2000 ) ; p.h . frampton and c. jarlskog , phys . b154 * , 421 ( 1985 ) . branco , l. lavoura and f. mota , phys . rev . * d39 * , 3443 ( 1989 ) ; g.c . branco , d. emmanuel - costa , r. gonzalez felipe , phys . * b 483 * , 87 ( 2000 ) ; g.c . branco , d. emmanuel - costa , r. gonzalez felipe , phys . * b 477 * , 147 ( 2000 ) . n. cabibbo , phys . * 10 * , 531 ( 1963 ) ; m. kobayashi , m. maskawa , prog . . phys . * 49 * , 652 ( 1973 ) ; b. pontecorvo , sov . jetp * 6 * , 429 ( 1957 ) ( zh . eksp . fiz * 33 * , 549 ( 1957 ) ) ; b. pontecorvo , sov . jetp * 7 * , 172 ( 1958 ) ( zh . eksp . fiz * 34 * , 247 ( 1957 ) ) ; z. maki , m. nakagawa , s. sakata , prog . * 28 * , 870 ( 1962 ) . s. n. gupta and j. m. johnson , phys . * d44 * , 2110 ( 1991 ) ; s. rajpoot , mod . lett . * a7 * , 309 ( 1992 ) ; h. fritzsch and d. holtmannspotter , phys . b * 338 * , 290 ( 1994 ) ; h. fritzsch and z. z. xing , phys . b * 353 * , 114 ( 1995 ) ; p. s. gill and m. gupta , j. phys . g * 21 * , 1 ( 1995 ) ; phys . d * 56 * , 3143 ( 1997 ) ; h. lehmann , c. newton and t. t. wu , phys . b * 384 * , 249 ( 1996 ) ; z. z. xing , j. phys . g * 23 * , 1563 ( 1997 ) ; k. kang and s. k. kang , phys . d * 56 * , 1511 ( 1997 ) ; t. kobayashi and z. z. xing , mod . a * 12 * , 561 ( 1997 ) ; int . j. mod . a * 13 * , 2201 ( 1998 ) ; j. l. chkareuli and c. d. froggatt , phys . b * 450 * , 158 ( 1999 ) ; j. l. chkareuli , c. d. froggatt and h. b. nielsen , nucl . b * 626 * , 307 ( 2002 ) ; a. mondragon and e. rodriguez - jauregui , phys . d * 59 * , 093009 ( 1999 ) ; h. nishiura , k. matsuda and t. fukuyama , phys . rev . d * 60 * , 013006 ( 1999 ) ; s. h. chiu , t. k. kuo and g. h. wu , phys . d * 62 * , 053014 ( 2000 ) ; h. fritzsch and z. z. xing , phys . d * 61 * , 073016 ( 2000 ) ; s. h. chiu , t. k. kuo and g. h. wu , phys . d * 62 * , 053014 ( 2000 ) ; b. r. desai and a. r. vaucher , phys . d * 63 * , 113001 ( 2001 ) ; h. fritzsch and z. z. xing , phys . b * 506 * , 109 ( 2001 ) ; r. rosenfeld and j. l. rosner , phys . b * 516 * , 408 ( 2001 ) ; r. g. roberts , a. romanino , g. g. ross and l. velasco - sevilla , nucl . b * 615 * , 358 ( 2001 ) ; e. ma and g. rajasekaran , phys . d * 64 * , 113012 ( 2001 ) ; k. s. babu , e. ma and j. w. f. valle , phys . b * 552 * , 207 ( 2003 ) ; j. w. mei , z. z. xing , phys . d 67 * , 077301 ( 2003 ) ; h. fritzsch , z. z. xing , phys . * b 555 * 63 ( 2003 ) ; z. z. xing and h. zhang , j. phys . g * 30 * , 129 ( 2004 ) ; h. d. kim , s. raby , l. schradin , phys . d 69 * , 092002 ( 2004 ) ; a. datta , l. everett and p. ramond , phys . b * 620 * , 42 ( 2005 ) ; g. altarelli and f. feruglio , nucl . b * 720 * , 64 ( 2005 ) , nucl . b * 741 * , 215 ( 2006 ) ; r. jora , s. nasri and j. schechter , int . j. mod . a * 21 * , 5875 ( 2006 ) ; m. hirsch , a. s. joshipura , s. kaneko and j. w. f. valle , phys . lett . * 99 * , 151802 ( 2007 ) ; g.c . branco , m. n. rebelo , j. i. silva - marcos , phys . rev . * d 76 * , 033008 ( 2007 ) ; j. c. romao , m. a. tortola , m. hirsch and j. w. f. valle , phys . d * 77 * , 055002 ( 2008 ) ; a. dighe and n. sahu , arxiv:0812.0695 [ hep - ph ] ; j. l. diaz - cruz , j. hernandez - sanchez , s. moretti , r. noriega - papaqui and a. rosado , arxiv:0902.4490 [ hep - ph ] ; d. emmanuel - costa and c. simoes , phys . rev . d * 79 * , 073006 ( 2009 ) ; g. altarelli , arxiv:0905.3265 [ hep - ph ] , arxiv:0905.2350 [ hep - ph ] ; r. jora , j. schechter and m. n. shahid , arxiv:0909.4414 [ hep - ph ] ; b. dutta , y. mimura and r. n. mohapatra , arxiv:0910.1043 [ hep - ph ] . p. f. harrison and w. g. scott , phys . b * 547 * , 219 ( 2002 ) ; t. kitabayashi and m. yasue , phys . d * 67 * , 015006 ( 2003 ) ; r. n. mohapatra , jhep * 0410 * , 027 ( 2004 ) ; i. aizawa , m. ishiguro , t. kitabayashi and m. yasue , phys . d * 70 * , 015011 ( 2004 ) ; a. ghosal , mod . phys . lett . * a 19 * , 2579 ( 2004 ) ; r. n. mohapatra and w. rodejohann , phys . d * 72 * , 053001 ( 2005 ) ; t. kitabayashi and m. yasue , phys . b * 621 * , 133 ( 2005 ) ; r. n. mohapatra , s. nasri and h. b. yu , phys . lett . b * 615 * , 231 ( 2005 ) ; r. n. mohapatra and s. nasri , phys . d * 71 * , 033001 ( 2005 ) ; s. nasri , int . j. mod . a * 20 * , 6258 ( 2005 ) ; i. aizawa , m. ishiguro , t. kitabayashi and m. yasue , j. korean phys . soc . * 46 * , 597 ( 2005 ) ; b. adhikary , phys . d * 74 * , 033002 ( 2006 ) ; w. grimus , arxiv : hep - ph/0610158 ; z. z. xing , h. zhang and s. zhou , phys . b * 641 * , 189 ( 2006 ) ; n. haba and w. rodejohann , phys . rev . d * 74 * , 017701 ( 2006 ) ; r. n. mohapatra , s. nasri and h. b. yu , phys . b * 636 * , 114 ( 2006 ) ; y. h. ahn , s. k. kang , c. s. kim and j. lee , phys . rev . d * 73 * , 093005 ( 2006 ) ; k. fuki and m. yasue , phys . d * 73 * , 055014 ( 2006 ) ; i. aizawa and m. yasue , phys . d * 73 * , 015002 ( 2006 ) ; i. de medeiros varzielas and g. g. ross , nucl . b * 733 * , 31 ( 2006 ) ; t. baba and m. yasue , phys . d * 75 * , 055001 ( 2007 ) ; t. baba , int . j. mod . e * 16 * , 1373 ( 2007 ) ; w. grimus , l. lavoura , jhep * 0809 * , 106 ( 2008 ) ; j. c. gomez - izquierdo and a. perez - lorenzana , phys . d * 77 * , 113015 ( 2008 ) ; a. s. joshipura and b. p. kodrani , phys . b * 670 * , 369 ( 2009 ) ; b. adhikary , a. ghosal and p. roy , arxiv:0908.2686 [ hep - ph ] . c. hamzaoui , _ phys . lett . _ * 61 * , 35 ( 1988 ) ; g.c . branco and l. lavoura , _ phys . lett . _ * b205 * , 123 ( 1988 ) ; c. jarlskog and r. stora , _ phys . lett . _ * b208 * , 268 ( 1988 ) ; c. hamzaoui , in rare decay symposium , vancouver , canada , november 1988 , d. bryman , j. ng , t. numao and j .- m . poutissou , edts . , p. 469 ; a. campa , c. hamzaoui and v. rahal , , in beyond the standard model symposium , ames , iowa , u.s.a , november 1988 , kerry whisnant and bing - lin young , edts . , world scientific , p. 470 ; a. campa , c. hamzaoui and v. rahal , phys . rev . * d39 * , 3435 ( 1989 ) ; g. blanger , c. hamzaoui and y. koide , phys . rev . * d45 * , 4186 ( 1992 ) . g.c . branco , m. n. rebelo , j. i. silva - marcos , phys . * b 597 * , 155 ( 2004 ) . s. antusch and s. f. king , phys . b * 631 * , 42 ( 2005 ) . c. amsler _ et al . _ [ particle data group ] , phys . b * 667 * , 1 ( 2008 ) . m. gronau , j. schechter , _ phys . lett . _ * 54 * , 385 ( 1985 ) , erratum - ibid * 54 * , 1209 ( 1985 ) ; c. jarlskog , _ phys . rev . lett . _ * 55 * , 1039 ( 1985 ) ; i. dunietz , o.w . greenberg , d .- wu , _ phys . lett . _ * 55 * , 2935 ( 1985 ) . g. l. fogli , e. lisi , a. marrone , a. palazzo and a. m. rotunno , phys . * 101 * , 141801 ( 2008 ) ; t. schwetz , m. tortola and j. w. f. valle , new j. phys . * 10 * , 113011 ( 2008 ) ; p. f. harrison , d. h. perkins , w. g. scott , phys . lett . * b 530 * , 167 ( 2002 ) ; z. z. xing , phys . lett . * b 533 * , 85 ( 2002 ) ; p. f. harrison and w. g. scott , phys . lett . * b 535 * , 163 ( 2002 ) ; w. grimus , a. s. joshipura , s. kaneko , l. lavoura , h. sawanaka and m. tanimoto , nucl . phys . b * 713 * , 151 ( 2005 ) chi - kwong li and roy mathias , journal of inequal . and appl . , * 3 * , 137 ( 1999 ) . we first note that the diagonal elements of a hermitian matrix are bounded by its smallest and largest eigenvalue respectively . this means that in order for a hermitian matrix to have zeroes in the diagonal entries , we must have at least one eigenvalue positive and one negative , the third one can take either sign , to accommodate the zeroes in the diagonal elements . for simplicity , we do not consider this case here and instead concentrate on positive definite hermitian matrices such that the eigenvalues are all positive . of course the results can be trivially extended for the case of a more general hermitian matrix , not necessarily definite positive . let s introduce our notation by considering the positive definite hermitian matrix @xmath1 @xmath235 the diagonal entries of @xmath1 must be real and positive and are bounded by its smallest and largest eigenvalue respectively . taking @xmath236 the bounds are 0<_1 _ 3 + 0<_1 _ 3 + 0<_1 a _ 3 the off - diagonal entries of @xmath1 are also bounded by its smallest and largest eigenvalues in the following way ; since a @xmath222 hermitian matrix has only three blocks of off - diagonal elements , namely @xmath237 , @xmath238 and @xmath239 up to conjugation @xcite : 0 |_3-_1| + 0 |_3-_1| + 0 |_3-_1| which means that : 0 |x| |_3-_1| + 0 |g| |_3-_1| + 0 |b| |_3-_1| the above results are valid for any @xmath240 hermitian matrix . these bounds might not be very revealing in the quark and charged lepton sectors , since the difference between the heaviest and the lightest eigenvalues is of the order of the largest eigenvalue , and so the constraint on the off - diagonal entries is quite mild . on the other hand , if the difference between the heaviest and lightest eigenvalue is very small then one sees that the off - diagonal entries must actually be smaller than half that difference , and so the hermitian matrix must be close to diagonal form . this case might be possible in the neutrino sector in the case of quasi - degenerate masses . we now write the three invariants @xmath23 , @xmath24 and @xmath25 : @xmath26 they can be rewritten as latexmath:[\ ] ] of course depending on the specific scenario studied , some other parametrization might be used . for example , one can use only the eigenvectors of eq . ( [ e1row ] ) , i.e. @xmath260 or use the ones from eq . ( [ e2row ] ) : @xmath261 or use the ones from eq . ( [ e3row ] ) : @xmath262 where the normalization constants @xmath263 , @xmath264 , @xmath265 , @xmath266 , @xmath267 and @xmath268 can easily be obtained from @xmath269 , @xmath270 and @xmath271 by permutation of the mass eigenvalues . all in all , one can write the matrix @xmath8 with 9 different parametrizations ( modulo a diagonal phase matrix ) , by permutations of the column vectors from eqs . ( [ e1row ] ) , ( [ e2row ] ) and ( [ e3row ] ) . of course , when one takes some special limits some of the parametrizations will reveal themselves less useful , as it is possible to find undetermined expressions of the type @xmath272 . for example in the parametrization shown in eq . ( [ e2rowpar ] ) this will happen if we take simultaneously the limits @xmath273 and @xmath274 . in that situation one simply chooses the parametrization with a smooth limit . it is easy to realize from eqs . ( [ e1rowpar ] ) , ( [ e2rowpar ] ) and ( [ e3rowpar ] ) that one can obtain very simple expressions for the absolute value of each element of w. by taking the real elements of these three parametrizations and squaring them we obtain , in terms of the eigenvalues and the parameters @xmath86 , @xmath92 , @xmath30 and @xmath27 : @xmath275 @xmath276 @xmath277 @xmath278 @xmath279 @xmath280 @xmath281 @xmath282 @xmath277 let s recall the notation for @xmath9 and @xmath10 : h_u= , h_d= . @xmath11 and @xmath12 are the unitary transformations diagonalizing @xmath9 and @xmath10 respectively . we want to find the parametrization of both @xmath11 and @xmath12 when the elements @xmath147 and @xmath148 vanish . the requirement for the cancellation of these elements is @xmath284 and @xmath285 , and with them we obtain in both sectors the simpler identities : @xmath286 from the above expressions we obtain the following simple parametrization of up and down quark mixing matrices , with @xmath66 , @xmath67 : @xmath287 and @xmath288 one can now check that in this ansatz , which basically shows that the structure of the quark mixing matrix is v__ckm= where @xmath75 , @xmath76 , @xmath77 and @xmath78 are given by c&= & v_tb + s&=&e^i _ b + c_&= & + c_&= & and @xmath79 and @xmath80 . the previous form for @xmath13 can then be put in the form given in eq.([vckm ] ) by pulling out two diagonal phase matrices @xmath73 and @xmath74 as given in the main text . we can also quickly compute the jarlskog invariant @xmath138 for this mixing matrix , for example by computing j&=&im(v_cbv^*_tbv^*_cdv_td ) + j&=&im(-c_^2s_^2|sc|^2 + c_c_s_s_|s|^2 e^i(_x-_y ) c^ * ) + j&= & c_c_s_s_|s|^2
the first normal stress difference ( ) and the microstructure in a dense sheared granular fluid of smooth inelastic hard - disks are probed using event - driven simulations . while the anisotropy in the second moment of fluctuation velocity , which is a burnett - order effect , is known to be the progenitor of normal stress differences in _ dilute _ granular fluids , we show here that the collisional anisotropies are responsible for the normal stress behaviour in the _ dense _ limit . as in the elastic hard - sphere fluids , remains _ positive _ ( if the stress is defined in the _ compressive _ sense ) for dilute and moderately dense flows , but becomes _ negative _ above a critical density , depending on the restitution coefficient . this sign - reversal of occurs due to the _ microstructural _ reorganization of the particles , which can be correlated with a preferred value of the _ average _ collision angle in the direction opposing the shear . we also report on the shear - induced _ crystal_-formation , signalling the onset of fluid - solid coexistence in dense granular fluids . different approaches to take into account the normal stress differences are discussed in the framework of the relaxation - type rheological models .
cond-mat0306689
in the last decade , a lot of research activity took place to unveil the properties of granular materials@xmath2 , primarily because of their industrial importance , but also due to their fascinating properties . this has unraveled many interesting and so far unresolved phenomena ( for example , clustering , size - segregation , avalanches , the coexistence of gas , liquid and solid , etc . ) . under highly excited conditions , granular materials behave as a fluid , with prominent _ non - newtonian _ properties , like the normal stress differences@xmath3 . while the normal stress differences are of _ infinitesimal _ magnitudes in a simple fluid ( e.g. air and water ) , they can be of the order of its isotropic pressure in a dilute granular gas@xmath4 . from the modelling viewpoint , the presence of such large normal - stress differences readily calls for higher - order constitutive models@xmath5 even at the _ minimal _ level . studying the non - newtonian behaviour is itself an important issue , since the normal stresses are known to be the _ progenitors _ of many interesting and unique flow - features ( e.g. rod - climbing or weissenberg - effect , die - swelling , secondary flows , etc.@xmath6 ) in non - newtonian fluids . also , normal stresses can support additional instability modes ( for example , in polymeric fluids and suspensions@xmath7 , which might , in turn , explain some flow - features of granular fluids . for example , particle - clustering@xmath8 has recently been explained from the instability - viewpoint using the standard newtonian model for the stress tensor@xmath9 the kinetic theory of jenkins & richman@xmath10 first showed that the _ anisotropy _ in the second moment of the fluctuation velocities , due to the inelasticity of particle collisions , is responsible for such normal stress behaviour . they predicted that the first normal stress difference ( defined as @xmath11 , where @xmath12 and @xmath13 are the streamwise and the transverse components of the stress deviator , respectively , and @xmath14 is the isotropic pressure , see section iib ) is maximum in the dilute limit , decreases in magnitude with density , and eventually approaches zero in the dense limit . goldhirsch & sela@xmath4 later showed that the normal stress differences appear only at the burnett - order - description of the chapman - enskog expansion of the boltzmann equation . their work has clearly established that the _ origin _ of this effect ( in the dilute limit ) is _ universal _ in both atomic and granular fluids , with inelasticity playing the role of a _ magnifier _ and thus making it a sizeable effect in granular fluids . while the _ source _ of the normal stress differences in the dilute limit has been elucidated both theoretically and by simulation , its _ dense _ counterpart has not received similar attention so far . this is an important limit since the _ onset _ of _ dilatancy _ ( volume expansion due to shear@xmath15 ) , crystallization , etc . occur in the dense regime , which in turn would influence the normal stress differences . previous hard - sphere simulations@xmath16 did look at the normal stress differences , but they did not probe the _ dense _ limit in a systematic way . these simulations showed that the first normal stress difference vanishes in the dense limit , in line with the theoretical predictions of jenkins & richman@xmath10 . on the other hand , the soft - sphere simulations of walton & braun@xmath17 , with frictional particles , showed that this quantity can change sign in the same limit . our work with smooth inelastic hard - disks unequivocally demonstrates that @xmath0 , indeed , changes its sign at some critical density in the dense regime , due to the sign - change of its collisional component at a critical density , which depends on the value of the coefficient of restitution ( @xmath18 ) . more importantly , we show that the origin of @xmath0 in the dense limit is distinctly different from that in a dilute granular gas . at the _ microstructural_-level , certain topological changes in the _ anisotropic _ structure of the collision - angle distribution with density are responsible for the observed sign - reversal of @xmath0 . we use the familiar smooth hard - disk model for an event - driven simulation@xmath19 of the uniform shear flow configuration , focussing mainly on the normal stress behaviour and the microstructure formation as functions of the density and inelasticity . the details of the simulation technique and the relevant macroscopic quantities are described in section [ sec : simulation ] . the simulation results on the first normal stress difference , the radial distribution function , the collision angle distribution and the crystalline - structure are presented in section [ sec : results ] . possible modelling approaches to incorporate the normal stress differences are discussed in section [ sec : nonnewtonian ] . in section [ sec : conclusion ] we summarize our findings , with suggestions for possible future work . we consider a collection of smooth inelastic hard - disks in a square box of size @xmath20 under uniform shear flow let @xmath21 and @xmath22 be the streamwise and transverse directions , respectively , with the origin of the coordinate - frame being positioned at the centre of the box . the snapshot of a typical simulation , with non - dimensional coordinates , is shown in fig . 1(@xmath23 ) . note that the dimensional quantities are denoted by tildes , and the reference length , time and velocity scales for non - dimensionalization will be specified later in this section . let the diameter and the mass of the particle be @xmath24 and @xmath25 , respectively . the pre- and post - collisional particle velocities of particle @xmath26 are denoted by @xmath27 and @xmath28 , respectively . hence , the velocity of particle @xmath29 relative to @xmath26 is @xmath30 . let @xmath31 be the unit vector directed from the center of particle @xmath29 to that of particle @xmath26 at contact . the pre- and post - collisional velocities are related by the expression : @xmath32 where @xmath18 is the coefficient of normal restitution , with @xmath33 ; note that we restrict ourselves to perfectly smooth particles . the expression for the collisional impulse is @xmath34 directed along @xmath35 . the system is periodic in @xmath21-direction , i.e. a particle crossing the left / right boundary re - enters the system through the opposite boundary at the same vertical position with unchanged velocities . to impose a uniform shear rate ( @xmath36 ) in the @xmath22-direction , the top and bottom image boxes , bounding the central box , are set in motion with velocities @xmath37 and @xmath38 , respectively , in the streamwise direction . this is the standard approach to attain the state of uniform shear flow ( usf ) by imparting momemtum transfer by shearing , originally introduced by lees & edwards@xmath39 . overall , this system represents an _ extended doubly - periodic _ system where the periodicity in the transverse direction is in the local lagrangian frame . in a typical simulation , the disks are initially placed randomly in the computational box , and the initial velocity field is composed of the uniform shear and a small gaussian random part . an event - driven algorithm is then used to update the system in time , the details of which may be found in alam & luding@xmath40 . to ascertain whether the system has reached the statistical steady - state , the time evolution of the average fluctuation kinetic energy ( ` granular ' energy , defined in the next section ) is monitored , see fig . 1(@xmath41 ) . due to the balance between the shear work and the collisional dissipation under homogeneous shear deformation , the granular energy attains a constant value at the steady state . depending on the value of the coefficient of restitution and the number of particles , it takes about thousand collisions per particle to reach such a statistical steady - state the lower the value of @xmath18 , the more quickly the system reaches the steady state and _ vice versa_. the simulation is then allowed to run for another @xmath42 collisions per particle to gather data to calculate the macroscopic quantities . a few longer runs ( 30000 collisions per particle ) were also checked , with no significant change on the measured quantities . another quantity which was simultaneously monitored , along with granular energy , is the linearity of the streamwise velocity profile across the couette gap , and we found that the calculated shear rate ( i.e. the slope of the velocity profile ) fluctuated around the imposed shear rate by at most @xmath43 at densities where crystallization is not evidenced . with @xmath20 , @xmath44 , @xmath45 , and @xmath25 , used as the reference length , time , velocity , and mass , respectively , the relevant dimensionless quantities are : @xmath46 where @xmath47 is the ` hydrodynamic ' velocity , @xmath48 the fluctuation ( peculiar ) velocity of particles , @xmath49 the material density of particles , @xmath50 the stress tensor , and @xmath51 the granular energy . the macroscopic stress , as measured in discrete particle simulations , is a byproduct of the particle - level mechanisms of momemtum transfer . as in the hard - core model of dense gases , the stress is the sum of its kinetic and collisional components . the former arises from the transport of momentum as the particles move through the system carrying their momentum , while the latter is due to the direct interparticle collisions . the homogeneity of the uniform shear flow allows us to calculate the stress by averaging it over the whole computational box@xmath52 . the stress , defined in the _ compressive _ sense , may be decomposed in the standard way : @xmath53 where @xmath14 is the pressure , @xmath54 the pressure deviator and @xmath55 the unit tensor . from the off - diagonal components of the pressure deviator , we can calculate the _ shear viscosity _ which relates the rate of strain to the shear stress : @xmath56 for the steady uniform shear flow , thus , the dimensionless shear viscosity can also be interpreted as the shear stress due to our adopted scaling , @xmath57 . the diagonal components of the pressure deviator can be non - zero , giving rise to normal stress differences . the first normal stress difference is defined as @xmath58 note that we have scaled this quantity by pressure to ascertain its relative magnitude with respect to pressure . for a standard newtonian fluid , @xmath59 and thus @xmath0 is an indicator of the _ non - newtonian _ character of the fluid . @xmath0 can be decomposed into kinetic and collisional parts : @xmath60 note that the _ sign _ of @xmath0 crucially depends on the convention used to define the stress tensor . for example , in the rheology literature , stress is typically defined in the _ tensile _ sense@xmath6 . a positive @xmath0 for the compressive case is equivalent to its negative value for the tensile case and vice versa . this point should be kept in mind while making any comparison with data in the rheology literature . from the trace of the kinetic stress tensor , @xmath61 , one can calculate the granular energy , @xmath62 , \label{eq : genergy1}\ ] ] which is a measure of the random motion of the particles with respect to the mean motion . there are two dimensionless control parameters : the volume fraction of particles ( @xmath63 ) and the coefficient of normal restitution ( @xmath18 ) . the shear rate is also a control parameter , however , due to normalization we have @xmath64 , and changing the value of @xmath65 does not influence the reported results ; the imposed shear rate @xmath66 is thus kept fixed at unity . the simulations are carried out for the whole range of solid volume fractions , varying from the dilute to the dense limit , over a large range of values for the coefficient of restitution ( @xmath67@xmath68 ) . for most of the simulations , the number of particles are fixed to @xmath69 , and increasing the value of @xmath70 by fourfold ( @xmath71 ) did not affect the reported quantities noticeably ; for example , the change in @xmath0 was about @xmath72 and @xmath73 at @xmath74 and @xmath75 , respectively , for a restitution coefficient of @xmath76 . we note here that the system - size dependence of the rheological quantities ( pressure and viscosity ) is known to be strong only for a small number of particles ( @xmath77)@xmath78 . for the typical simulation in fig . 1(@xmath23 ) , at steady - state , after @xmath79 collisions , the parameter values were @xmath80 , @xmath69 and @xmath81 . the variations of the granular energy @xmath51 and the calculated shear rate @xmath82 with time are shown in fig . 1(@xmath41 ) , along with corresponding initial variations in two insets . ( the data represent the instantaneous values of @xmath51 and @xmath82 sampled at a regular interval of @xmath83 collisions no time averaging is involved here . ) note that @xmath82 was computed by binning the system into @xmath84 equal - size bins in the transverse direction and taking averages over all particles in each bin . it is observed that the granular energy reaches its steady value ( @xmath85 ) quickly after the initial transients and the calculated shear rate fluctuates around its imposed value ( @xmath86 ) by about @xmath87 . the fluctuations in both @xmath51 and @xmath82 at steady - state are due to the finite - size of the system and diminish with increasing number of particles as @xmath88 . for detailed results on the quantities pressure , shear viscosity and granular energy , and for their comparison with kinetic theory predictions , we refer to our recent study@xmath89 . here , we will mainly focus on the behaviour of the first normal stress difference and its kinetic and collisional components . we also present results on the pair distribution function and the collision angle distribution to characterize microstructures . lastly we will present results on _ crystal_-formation at high densities , signalling the coexistence of fluid and solid , complementing recent results in non - sheared systems@xmath90 . figure 2(@xmath23 ) shows the variation of the first normal stress difference ( @xmath0 ) with density for two values of the coefficient of restitution . it is observed that @xmath0 is maximum at the dilute limit and decreases thereafter with @xmath63 . the overall variation of @xmath0 with density looks similar at other values of the coefficient of restitution , with a difference in the magnitude of @xmath0 . the inset in fig . 2(@xmath23 ) shows that @xmath0 decreases quite sharply in the dense limit and becomes negative at some density ( @xmath91 ) . increasing the value of the restitution coefficient decreases this critical density @xmath92 . the arrows on the left - ordinate indicate the asymptotic values of @xmath0 for a two - dimensional granular gas in the dilute limit@xmath4 : @xmath93 the anisotropy in the second moment of the fluctuation velocity is _ primarily _ responsible for the finite normal stress difference in the dilute limit@xmath94 and this shows up only at the burnett - order of the chapman - enskog expansion@xmath4 . we should mention here that the limit @xmath95 is singular and the normal stress difference survives even in the elastic limit as pointed out by goldhirsch & sela@xmath4 . the corresponding expression for @xmath0 in a molecular gas is : @xmath96 where @xmath97 is the mean free path and @xmath98 is the rms of the velocity fluctuations . however , because of its extremely small magnitude under normal conditions , the normal stress difference is not _ measurable _ in a molecular fluid . previous hard - sphere simulations of campbell and coworkers@xmath99 are in _ variance _ with our result in that they found @xmath100 as @xmath101 . however , the soft - sphere simulations of walton & braun@xmath17 , with frictional particles , support our observation that @xmath0 indeed undergoes a _ sign - reversal_. to better understand what is responsible for the sign - reversal of @xmath0 , we look at the kinetic and collisional components of the first normal stress difference . figure 2(@xmath41 ) shows the variations of @xmath102 and @xmath103 with density at @xmath81 . we observe that @xmath102 is maximum at the dilute limit and decreases monotonically to zero as @xmath63 approaches the packing limit . except for the dense limit , the overall behaviour of @xmath102 represents that of the total normal stress difference . the collisional component , @xmath103 , shows a _ non - monotonic _ variation with density : @xmath103 is zero in the dilute limit , increases with increasing @xmath63 , remains almost constant for intermediate densities , and then decays sharply in the dense limit . interestingly , @xmath103 becomes negative at some critical density ( @xmath91 ) beyond which the behaviour of @xmath103 mirrors that of @xmath0 ( see inset ) . thus the normal stress behaviour in the dense regime is clearly due to the anisotropy in the collisional stress . recall that the kinetic theory of jenkins & richman@xmath94 predicts that @xmath100 in the dense limit . the predictions of the revised enskog theory of santos _ et al._@xmath104 are in line with that of jenkins & richman , even though their kinetic model is claimed to be valid even in the crystalline - phase . since the source of normal stress differences in all these theories is linked to the anisotropy in the second moment of velocity fluctuations ( which vanishes as @xmath101 ) , they are unable to predict the correct behaviour of normal stresses in the dense limit . we would like to stress here that , as mentioned in the introduction , many fascinating non - newtonian effects@xmath6 are primarily determined by the first normal stress difference and its sign . for example , the rod - climbing effect can occur in a granular fluid if @xmath105 , _ i.e. _ only in the dense limit ( assuming that the second normal stress difference is negligible ) . furthermore , having a constitutive model which reproduces the correct sign of @xmath0 is also important , since it is well known in the rheology literature that the _ odd _ sign for @xmath0 leads to the _ instability _ of the rest state@xmath106 . the sign - reversal of @xmath103 can be succinctly presented as a phase - diagram in the ( @xmath107)-plane by plotting the zeros of @xmath103 as a function of the coefficient of restitution , see fig . below the solid line , @xmath103 is positive , and negative above it . also plotted in this figure is the line for the zeros of @xmath0 which , as expected , lies slightly above . thus , the first normal stress difference is zero along the solid line which may be called the _ symmetry - line_. it is observed that decreasing the coefficient of restitution increases the critical density ( @xmath92 ) at which @xmath103 changes sign . as we approach the elastic limit , @xmath92 depends strongly on the value of @xmath18 . we further note that as @xmath95 , @xmath108 which is well below the freezing - point density of a 2d hard - disk fluid , @xmath109@xmath90 . we need to mention here that the _ sign - change _ of @xmath0 is not uncommon in other non - newtonian fluids . for example , in non - brownian viscous suspensions , @xmath0 changes sign at high peclet number@xmath110 . however , the reason for this effect is quite different in granular fluids as we show below . to understand the microstructural mechanism for the origin of the first normal stress difference and its sign - reversal , here we probe several microstructural features of a dense granular fluid . typical snapshots of the system in the dense regime are shown in fig . 4 at four different densities with @xmath81 . note that the density for the subplot 4(@xmath111 ) is @xmath112 for which @xmath113 . looking at the corresponding distribution of granular energies ( not shown here for brevity ) , we could find signatures of clusters ( group of particles ) with lower energies surrounded by particles with higher energy . to understand the flow - microstructures and their energetics at such high densities , we need to probe the pair distribution function and similar measures for the structure of the packing . figure 5(@xmath23-@xmath114 ) shows the radial distribution function @xmath115 at four densities , with parameter values as in fig . 4 . the thin , dotted lines are data from a non - sheared , homogeneous , elastic system@xmath90 , whereas the thick , solid lines represent a sheared situation with rather strong dissipation @xmath81 . ( note that the elastic distribution function has been measured under non - sheared periodic boundary conditions ; also for @xmath18 somewhat smaller than unity , the same results were obtained as long as the system remains homogeneous . ) we observe that the weak difference at low density @xmath116 grows with increasing density , concerning two aspects : + ( i ) the peak value of contact in the sheared systems is always larger than that in a homogeneous system of the same density , and the difference increases strongly with density , another indicator for clustering@xmath117 . + ( ii ) the peaks and valleys , which allow to distinguish between different lattice structures , are different in the sheared case when compared to the non - sheared situation . in the former case , peaks at @xmath118 are observed , signalling shell - formation about any test - particle . in the latter case , the peaks at @xmath119 indicate a crystallization transition and the development of a triangular lattice . the peaks in the sheared situation , e.g. at @xmath120 , become sharper as the density is increased , but the one at @xmath121 is not well - defined even at @xmath122 ; higher order peaks are almost invisible , indicating long - range disorder due to the shearing in contrast to the long - range order that evolves in the non - sheared system . the comparison between the sheared and non - sheared cases suggest that the structure - formation is much slower in a sheared fluid which , in turn , implies that , as expected , the freezing - point density of the former would be larger than that of the latter . the splitting of the second - peak in the non - sheared case corresponds to the _ onset _ of freezing transition@xmath123 ; with increasing density this splitting becomes much more prominent , with similar structural - features appearing at the successively higher - order peaks . for the sheared system , however , we do not observe similar splitting , rather we see a _ sharp _ second - peak . this also occurs for a highly inelastic sheared fluid at a much lower density@xmath117 ; hence , a sharp second peak is a signature of short - range ordering due to the dissipative particle - clustering . the higher - order peaks in the sheared case become prominent only if we go beyond @xmath124 , which corresponds to the limit of perfect square - packing , but it is difficult to maintain homogeneous - shearing at such high densities , because the system splits into two parts , a dense , cold , crystalline area and a dilute , hot , fluid area see below for details . figure 6 shows the variations of the pressure and viscosity functions ( @xmath125 and @xmath126 , respectively ) with density at a restitution coefficient @xmath81 . we observe that both increase _ monotonically _ with density , much beyond @xmath127 , and we did not find the _ hysteretic _ van - der - walls loop in our pressure data upto @xmath124 , another indication that the crystallization is hindered / delayed by shear . these observations , together with our result that in the elastic limit @xmath0 changes sign at a much lower density ( @xmath128 , see fig . 3 ) than the corresponding freezing - density , suggest that the sign - reversal of @xmath0 is not related to the freezing - transition . associated with the _ sign - reversal _ of the first normal stress difference is a change in the relative magnitudes of the normal stress components ( @xmath129 and @xmath130 ) . subtle changes in the direction and magnitudes of the collisional - mode of momemtum transfer could influence the individual components of the stress tensor . in order to test this hypothesis , we focus on the collision angle distribution function , @xmath131 , which is defined such that @xmath132 is the probability of collisions occuring at an angle lying between @xmath133 and @xmath134 , with the angle @xmath133 being measured in the anticlockwise direction from the positive @xmath135-axis ( see fig . 13 ) . for a fluid in equilibrium , all collisions are equally likely , and hence @xmath131 is a uniform function of @xmath133 , _ i.e. _ @xmath136 . for a non - equilibrium system ( e.g. shear flow ) , however , preferred collisions are dictated by the nature of the external field , leading to an anisotropic distribution for @xmath131@xmath137 . following savage & jeffrey@xmath138 and campbell & brennen@xmath139 , an explicit expression for @xmath140 can be derived for the case of uniform shear flow as detailed in the appendix . note that the angular dependence of @xmath141 will be modified by both the density and the restitution coefficient . figure 7 shows the comparison of our simulation data on @xmath131 with the theoretical predictions of equation ( [ eqn_colang1 ] ) for two different values of the restitution coefficient at a density @xmath116 . it is observed that the probability of collisions is higher on the _ upstream_-faces of the colliding particles , i.e. for @xmath142 $ ] and @xmath143 $ ] ( i.e. the hatched - areas on the test - particle in fig . this is a consequence of the imposed shear - field which compresses the flow - structure along the @xmath144-direction and stretches it along the @xmath145-direction . regarding the comparison with theory , there is , clearly , a phase - difference between theory and simulation , and the overall agreement is only qualitative . note in fig . 7 that the probability of collisions on the upstream - faces increases further as the restitution coefficient increases . this , in turn , suggests that particle motion becomes more _ streamlined _ ( i.e. along the streamwise direction ) with increasing dissipation levels , which will naturally lead to a reduction in the transverse component of the fluctuation velocities of the particles . thus , the _ macroscopic _ manifestation of such _ microscopic _ streamlined - motion would be an increase in the magnitude of the kinetic component of the first normal stress difference ( @xmath102 ) . turning our attention to the range of densities where @xmath0 undergoes a sign - reversal , we show the collision angle distributions @xmath131 in fig . 8 as polar plots with @xmath81 ; the corresponding densities are as in the subplots of fig . it is observed that the _ anisotropic _ structure of @xmath131 gets further modified in this regime , with distinct peaks appearing near @xmath146 and @xmath147 ( see subplot @xmath41 ) . while the peak at @xmath146 corresponds to _ head - on _ collisions between particles in the same - layer , the one at @xmath148 clearly signals the _ onset _ of triangular - structure formation . another noteworthy point is that the collisions on the downstream - faces of the colliding particles are _ rare _ at these densities and hence @xmath131 can be approximated solely by its contributions from the second- and fourth - quadrants ( @xmath142 $ ] and @xmath149 $ ] , respectively ) . since the momentum transfer occurs mainly due to collisions in the dense regime , the stress tensor can be approximated by @xmath150 where @xmath35 is the unit vector joining the line of centres of the two colliding disks . assuming now that all the collisions would occur at some average collision angle @xmath151 so that @xmath152 , and recalling that @xmath131 is well represented in this regime by restricting @xmath133 only in the second- and fourth - quadrants , the expression for the first normal stress difference simplifies significantly to @xmath153_{\theta=\theta_{av } } c(\theta_{av}).\ ] ] it is trivial to check that @xmath59 at @xmath154 . from our simulation data , we have calculated @xmath151 by averaging @xmath131 over the second- and fourth - quadrants , whose variation with density is plotted in fig . 9 for two restitution coefficients . it is observed that @xmath151 crosses through @xmath155 ( i.e. @xmath144 ) at around the critical density @xmath156 for all restitution coefficients . for example , @xmath157 and @xmath158 at @xmath159 and @xmath160 , respectively , where @xmath0 changes sign . thus , the microstructural signature of the sign - reversal of @xmath0 is directly correlated with the _ average _ collision angle being greater or less than @xmath155 ( or @xmath144 ) . figure 10(@xmath23 ) shows a snapshot of the system at the steady state with parameter values being set to @xmath161 and @xmath76 . it is observed that a solid - layer coexists with two fluidized zones on either side of it . a closer look into the solid - layer reveals that the particles are arranged in a triangular - packing , representing a _ crystal _ , and thus we have a clear evidence for _ fluid - solid _ coexistence . the corresponding _ instantaneous _ streamwise velocity profile at @xmath162 ( i.e. the image - boxes have moved @xmath163 strain units from their original position ) is shown in fig . 10(@xmath41 ) ; the coarse - graining is done by binning the system into 20 equal - size bins in the transverse direction and then taking averages over all the particles in each bin . clearly , the shearing is _ inhomogeneous _ across the couette - gap : the crystal is aligned along the streamwise direction and hence we call it a _ layered_-crystal ; the shear - rate in the fluidized regimes on either side of the crystal is almost uniform . note that the _ asymmetric _ nature of the velocity profile also signals the breakdown of the lees - edwards boundary condition as a motor for the homogeneous shear . the formation and the time - evolution of this crystal can be ascertained from fig . 10(@xmath111 ) which shows the corresponding evolution of the streamwise velocity at early times . we observe that the crystal has fully formed at @xmath164 , and the velocity profile remains antisymmetric about @xmath165 till @xmath166 . with further time - evolution , however , the crystal does not remain stationary in the transverse direction , rather it moves slowly with particles diffusing across the fluid - solid interface . the overall life - time of this crystal is several orders of magnitude larger than the external time - scale @xmath167 , imposed by the shear . the corresponding collision - angle distribution @xmath131 in fig . 10(@xmath114 ) shows three distinct peaks at @xmath146 , @xmath168 and @xmath147 . note that the peak at @xmath169 does not exist in the fluid phase ( refer to figs . 8@xmath23 ) , and this provides evidence that the particles in the crystalline - phase are arranged in the triangular - packing structure . the large area of the crystalline phase as well as its relatively large value of solids fraction ( hence , small mean free time ) further suggest that most of the collisions occurred in the crystalline - phase . we should remark here that , at such high densities , the _ inelastic collapse_@xmath170 would eventually terminate the evolution of the system . we used the tc - model@xmath171 to avoid inelastic collapse , but could not altogether eliminate it within the crystal after some time . but the important point to note is that the crystalline - phase can be maintained for a long period of time ( @xmath172 strain units ) , and hence the reported results are not transient effects . analogous plots for a nearly elastic system ( @xmath173 ) are shown in fig . 11(@xmath23@xmath111 ) for the same density @xmath161 . the overall features are similar to that for @xmath76 , but the width of the crystalline zone is a little larger . by decreasing the dissipation - level to @xmath81 , we did not observe crystal formation , with other parameters being fixed ; by increasing the system - size to @xmath71 , however , we observed layered crystal at @xmath81 . thus , the formation of such layered - crystalline structure depends crucially on the system size and the dissipation level : _ the larger the system - size and/or the weaker the dissipation , the more susceptible the system is to crystallize_. note that even if we are well below the limit of perfect square - packing ( @xmath174 ) , the system could crystallize if the dissipation - levels are low ; for example , we observed layered crystalline structures at @xmath122 with @xmath173 and @xmath71 . decreasing the coefficient of restitution to @xmath76 , the flow - field remained homogeneous . thus , our layered crystalline structures appear to be tied to a _ long - wave _ instability of the elastic hard - sphere fluids@xmath175 . since our results are not driven by the inelastic dissipation , they are distinctly different from the layered shear - banding patterns , as predicted by the kinetic theory models@xmath176 , in granular couette flows . having said that we need to mention that such _ dissipative _ layering patterns ( i.e. those which become stronger with increasing dissipation levels@xmath177 ) were also found in simulations of a dilute sheared granular fluid@xmath178 , but the simulations were allowed to evolve from an _ unsheared _ initial configuration with a special kind of boundary condition . in contrast , all our simulations started from a uniform shear condition . in the present contribution , we have mainly focussed on the non - dissipative layering in the dense limit , and the related issues of dissipative layering are relegated to a future study . before moving further , we make a qualitative comparison with the earlier simulation work of campbell & brennen@xmath139 on the bounded couette flow of inelastic , frictional hard disks . they also reported similar layered - microstructure but due to the small system - size ( @xmath179 ) and boundary - effects ( they considered a shear - flow bounded by frictional walls ) , the distinct shear - band formation that we have reported is not evident in the snapshots of their simulations . nevertheless , we believe that our results are akin to that reported by campbell & brennen . one of the referees has drawn our attention to the recent work of campbell@xmath180 who probed the dense limit of a three - dimensional couette shear flow using soft - sphere simulations . he was able to maintain uniform shearing even at a density of @xmath181 , but beyond that he reported shear - band formation in that the flow - field degenerates into sheared and non - sheared zones in the gradient direction . this simply suggests that one can maintain uniform shearing in a three - dimensional geometry even beyond the analogous square - packing limit ( @xmath182 ) since the particles have now an additional degree of freedom , orthogonal to the shear plane , to rearrange themselves . moreover , most of campbell s simulations were done with @xmath183 particles ( equivalently , @xmath184 particles in 2d ) at a restitution coefficient of @xmath185 . with these parameter values , we did not find layered crystals at @xmath161 in two - dimensions ( see discussion in the next section ) . one of the referees has drawn our attention to the recent work of lutsko@xmath186 who studied the shear - induced ordering in a low - density ( @xmath187 ) elastic hard - sphere fluid . the earliest simulations of erpenbeck@xmath188 on 3d elastic hard - sphere fluids showed that at high shear rates the system breaks down into orderded ( solid ) and disordered ( fluid ) phases in the direction of the mean vorticity ( i.e. normal to the shear - plane ) . this induces a long - range two - dimensional ordering , called a _ string - phase _ , in the shear - plane . more importantly , such ordering occurs only for a range of shear - rates the lower the density , the larger this shear - rate interval . however , the latter work of evans and morriss@xmath189 showed that the disorder - order transition of erpenbeck arises due to the profile - biased - thermostat@xmath189 since the string - phase vanishes completely when a profile - unbiased - thermostat@xmath189 is used . thus , the shear - induced - ordering in elastic hard - sphere fluids appears to depend on the choice of the thermostat . note further that all the above works probed moderately dense systems only , well below the corresponding square - packing limit . now to compare these results with our observations in granular fluids , we first need to define an equivalent shear rate since the granular energy and the shear rate are dependent on each other for the latter system . from the energy balance equation , it is trivial to show that the granular energy has the following functional relation with the shear rate and the restitution coefficient : @xmath190 where @xmath191 is defined as the _ reduced shear rate_. in the quasielastic limit ( @xmath95 ) the reduced shear rate ( and hence the granular energy ) approaches infinity , which is equivalent to the high shear - rate limit of an elastic hard - sphere fluid . since we observed layered crystalline structures only in the quasielastic limit , we may thus conclude that such structures would also persist in elastic hard - disk fluids at large shear rates . it is interesting to ask whether the layered crystalline - structures of figs . 10 and 11 are , in any way , related to the concept of the reynold s dilatancy@xmath15 which is explained schematically in fig . the top two sketches depict the classical _ constant - load_@xmath192 shear - cell experiment ( in two - dimensions ) in which the material is subjected to a constant normal load . 12(@xmath23 ) corresponds to an ideal situation of maximum packing , with the particles being arranged in a triangular lattice ( @xmath193 ) . clearly , in this situation the top and the bottom plates of the shear cell will simply slide over the material , without deforming it . however , if one of the plates is allowed to move in the vertical direction and thereby allowing the particles to rearrange themselves , e.g. as in fig . 12(@xmath41 ) , the material can be deformed even homogeneously ( if @xmath194 ) . this is the _ shear - coupled volume change _ , commonly known as the reynold s dilatancy . note , however , that our simulations mimic _ constant - volume _ experiments , since the volume of the computational box remains fixed ( see figs . 12@xmath111 and 12@xmath114 ) ; but , of course , now the pressure ( i.e. normal load ) can vary in response to particle motions inside the shear cell . for this case also , the uniform shearing is possible if and only if the overall density remains below the square - packing limit ( @xmath195 ) . however , for densities above this value ( @xmath196 ) , the shearing can be started only if we allow the particles to rearrange themselves . this is possible if a part of the system becomes denser , allowing _ free _ volumes to the rest of the system which is nothing but reynold s dilatancy too . hence , we will end up with a crystalline - phase coexisting with a fluid phase , a typical example of which is shown schematically in fig . 12(@xmath114 ) . thus the phenomenon of reynold s dilatancy , for densities @xmath196 , would make the _ ordering _ transition , as depicted in figs . 10 and 11 , more prominent , and this effect would be much stronger in two - dimensions . here we attempt to describe the normal - stress behaviour of a granular fluid using the standard relaxation - type models . prior literature on the dense - gas kinetic theory , which forms the foundation of theoretical developments of granular fluids in the _ rapid - shear _ regime , indicates that such a stress relaxation mechanism does also exist in granular fluids@xmath197 . the relaxation - type models are routinely used to describe the non - newtonian behaviour of viscoelastic / viscoplastic materials , and hence might be apt for granular fluids in the dense limit as well . the recent work of zhang & rauenzahn@xmath198 clearly shows that such viscoelastic stress relaxation mechanism exists in dense granular flows . following a rigorous statistical mechanical procedure , they derived an evolution equation for the collisional stress tensor which boils down to a frame - indifferent viscoelastic model , with the jaumann derivative appearing directly without appealing to objectivity arguments . let us consider the viscoelastic relaxation approximation suggested recently by jin & slemrod@xmath199 to regularize the burnett order equations of sela & goldhirsch@xmath200 for a low - density granular fluid . their proposed equation for the pressure deviator , in our notation , is @xmath201 where @xmath202 here @xmath203 and @xmath204 are relaxation times , @xmath114 the dimensionality of the system , @xmath205 is the velocity gradient , @xmath206 the deviatoric part of the rate of strain tensor , @xmath207 the shear viscosity , @xmath208 the bulk viscosity and @xmath209 the identity tensor ; @xmath210 and @xmath211 are higher order terms , explicitly written down in jin & slemrod@xmath199 . note that both relaxation times are proportional to the ratio of the shear viscosity and the pressure , and hence proportional to the mean free time . in the limits of @xmath203 , @xmath212 and @xmath210 , @xmath213 , we recover the standard newtonian model for the stress tensor . neglecting the higher - order terms , an expression for the first normal stress difference can be obtained for the steady uniform shear flow : @xmath214 \left(\frac{\mu}{p}\right ) \;\;>\;\ ; 0.\ ] ] this quantity is always positive , as in our simulation results for dilute flows , if the two relaxation times are of the same order . it is important to note that the above evolution equation does not satisfy the _ principle of material frame indifference _ ( mfi ) which states that the constitutive laws should be _ invariant _ under rigid - rotation@xmath215 . the scalar field @xmath216 , the vector field @xmath217 and the tensor field @xmath218 are called _ frame - indifferent _ or _ objective _ if the following relations hold for all @xmath219 : @xmath220 where @xmath216 , @xmath217 & @xmath218 and @xmath221 , @xmath222 & @xmath223 are defined in two different frames @xmath224 and @xmath225 , respectively , and @xmath226 is a proper orthogonal tensor . here @xmath227 , with @xmath228 denoting the set of all frames obtainable from a given frame @xmath224 by observer transformations . that the stress - tensor in a granular gas is not a frame - independent quantity ( as in the hard - sphere gas@xmath229 ) is well - known . since the kinetic component of the first normal stress difference remains positive at all densities , the kinetic stress tensor can be modelled using a non - objective equation as discussed above . for the collisional stress tensor , one can postulate a similar evolution equation as in eqn . ( 11 ) , but @xmath203 must be multiplied by a factor which must change sign at the critical density . however , this equation would remain frame - dependent even in the dense limit . thus , if one has to recover the boltzmann - limit of relaxation - type equations , a frame - indifferent approach does not appear to work . a question now arises as to the possibility of modelling normal stress differences using the standard frame - indifferent relaxation type models . the simplest way is to use either the _ lower - convected _ or the _ upper - convected _ equations for the pressure deviator : @xmath230 respectively . here @xmath231 is an empirical constant , dependent on both the density and restitution coefficient . for both cases , the first normal stress difference is @xmath232 clearly , if @xmath231 is obtained from simulation , its sign - reversal would also correspond to the sign - reversal of @xmath0 . comparing the jin - slemrod equation with its corresponding frame - indifferent analog ( lower - convected model ) , we conclude that the loss of frame - indifference shows up as a _ sign - change _ of the first normal stress difference . it would be interesting to investigate whether one could relax the sela - goldhirsch equations@xmath233 using a frame - indifferent approach without violating the entropy inequality@xmath199 . similarly , one could postulate evolution equations using other objective derivatives . in this regard , the co - rotational jeffrey s model seems to be the ideal choice : @xmath234 with @xmath235 being the jaumann derivative@xmath6 . the corresponding first normal stress difference is @xmath236 which is positive / negative depending on whether @xmath203 is less / greater than @xmath204 . thus , the frame - indifferent relaxation models are able to predict positive and negative first normal stress differences . ( for the steady homogeneous shear flow , one can also model positive / negative normal stress differences by postulating a general orthonormal basis , generated by the nilpotent basis tensors , which satisfies the objectivity requirement ; for related issues , the reader is referred to goddard@xmath237 . ) we have probed the non - newtonian behaviour and the incipient crystalline - phase in a sheared , monodisperse , two - dimensional granular fluid . the standard event - driven technique is used to simulate a box of hard - disks under homogeneous shear deformation . the information about the stress tensor is obtained by decomposing it in the standard way : @xmath238 , where @xmath14 is the pressure and @xmath239 the pressure deviator . the non - newtonian behaviour is quantified in terms of the first normal stress difference : @xmath240 . the granular fluid is non - newtonian with a measurable first normal stress difference ( @xmath0 ) which is _ positive _ ( if the stress is defined in the _ compressive _ sense ) in the dilute limit . interestingly , however , @xmath0 _ changes _ from positive to negative at a critical density in the dense regime . by decomposing @xmath0 into the kinetic and collisional contributions , @xmath241 , we found that while @xmath102 is always positive and decays to zero in the dense limit , @xmath103 has a _ non - monotonic _ variation with density . in particular , @xmath103 increases from zero in the dilute limit as @xmath63 increases , reaches a maximum at some value of @xmath63 and then decreases , eventually becoming _ negative _ in the dense limit . the density at which @xmath242 ( @xmath243 ) depends crucially on the level of micro - scale dissipation ; in particular , @xmath92 increases as the coefficient of restitution decreases . we have constructed a phase - diagram in the ( @xmath107)-plane by identifying the regions where @xmath0 is positive / negative . we have shown that the _ origin _ of the first normal stress difference , in the dense limit , is tied to shear - induced collisional anisotropies . the underlying mechanism is distinctly different from that is known for a dilute granular gas@xmath244 where the anisotropy in the second moment of the fluctuation velcoity , which is a burnett - order effect , gives rise to normal stress differences . at the micro - level , the particles undergo reorganization as the dense - limit is approached . the signatures of microstructural - reorganization have been captured by probing the collision - angle distribution , @xmath131 , which is _ anisotropic _ due to the presence of the mean shear . in particular , we have found that the topology of the anisotropic - structure of @xmath131 changes , with collisions occurring at certain preferred angles on the upstream - faces of the colliding pairs . the _ sign - reversal _ of @xmath0 is correlated with a preferred value of the average collision angle , @xmath1 , averaged over the _ upstream_-faces of the colliding particles . the time evolution of the sheared granular fluid leads to _ crystallization _ in the dense limit , signalling the _ coexistence _ of fluid and solid . the particles are arranged in a triangular - packing inside the crystal , and it moves as a _ layer _ in the streamwise direction . the formation of such layered - crystalline structure depends crucially on the system size and the dissipation level : the larger the system - size and the weaker the dissipation , the more susceptible the system is to crystallize . this appears to be related to a long - wave instability@xmath175 of the elastic hard - sphere fluids . the present work clearly shows that the available kinetic - theory - based rheological models for granular fluids are not adequate to predict the behaviour of the first normal stress difference in the dense limit . certain microstructural - features , like the preferred distribution of collisions which eventually leads to crystal - formation , should be incorporated into the theory . at such high densities , many - body effects ( both positional and velocity correlations ) are important@xmath245 and the appropriate kinetic description is the bbgky - hierarchy@xmath246 . to incorporate the observed normal stress behaviour into the framework of plausible constitutive models , we showed that the standard frame - indifferent relaxation type models can be used to model both positive and negative first normal stress differences . in this regard , the two - parameter jeffrey s model@xmath247 appears to be the ideal choice ; however , we are unable to recover the corresponding boltzmann limit which is known to be non - objective . on the whole , we believe that a lot remains to be done for a better understanding of the dense - phase rheology of granular fluids even in the hard - sphere limit . m.a . acknowledges the financial support by the alexander von humboldt foundation and s.l . acknowledges the support of the deutsche forschungsgemeinschaft . m.a . is grateful to hans j. herrmann for discussions and hospitality at ica1 , and also acknowledges discussions with joe d. goddard on certain aspects of normal stress behaviour in granular fluids . we thank john f. brady for directing us to relevant references on normal stresses in suspensions . we also thank three anonymous referees for their constructive comments . here we are interested in the shear - induced anisotropy of the collision angle distribution of an inelastic hard - disk fluid . following savage & jeffrey@xmath138 , an expression for the collision angle distribution @xmath131 is derived , which is compared with the simulation data in iib.2 . to calculate the probablility of collisions at a specific angle @xmath133 , we focus on fig . 13 with two particles colliding at @xmath248 . note that @xmath133 is measured anticlockwise from the positive @xmath135-axis . for collisions to occur in a time @xmath249 , the center of particle @xmath26 must lie inside the volume @xmath250 , where @xmath251 is the relative velocity of the colliding pair . thus the expected total number of collisions ( per unit time and unit area ) with the line of centres @xmath35 lying between @xmath252 and @xmath253 is given by @xmath254 where @xmath255 is the two - particle distribution function which is defined so that @xmath256 @xmath257 is the number of pairs of particles such that the particle @xmath258 is located in an area element @xmath259 about @xmath260 with its velocity in the interval @xmath261 about @xmath262 while particle @xmath263 is located in an area element @xmath264 about @xmath265 with its velocity in the interval @xmath266 about @xmath267 . to progress further , we have to invoke the assumption of molecular chaos and hence the expected number of collisions is @xmath268 where @xmath269 is the pair - distribution function . for the steady uniform shear flow , @xmath269 is calculated from the relation@xmath270 : @xmath271 where @xmath272 is the _ contact _ value of the pair - distribution function and @xmath273 implies that the integration be carried out for impending collisions . as a first approximation , the single particle velocity distribution function @xmath274 is assumed to have the maxwellian - form : @xmath275,\ ] ] where @xmath51 is the granular temperature ( fluctuating kinetic energy ) and @xmath276 the usual boltzmann constant . now transforming the particle velocities ( @xmath262 , @xmath267 ) to their center - of - mass and relative velocities , equation ( [ eqn_pcorr1 ] ) can be integrated to arrive at the following explicit expression for the pair - distribution function@xmath138 : @xmath277,\ ] ] where @xmath278 is the complementary error function . using the above expression for the pair - distribution function and transforming in terms of polar coordinates ( @xmath279 ) , the integral for the normalized collision angle distribution yields@xmath139 @xmath280g(\theta),\ ] ] where @xmath281 is the angular pair - distribution function given by @xmath282\ ] ] and @xmath283 is a normalization constant . for the uniform shear flow , an expression for the granular temperature , can be obtained from the energy balance equation , by equating the energy production due to shear - work with the energy loss due to collisions : @xmath284 where @xmath285 is the shear viscosity and @xmath286 the collisional dissipation rate , with @xmath287 , \\ f_{\mathcal d}(\nu , e ) & = & \frac{4}{\sqrt{\pi}}(1-e^2)\nu^2g_c(\nu).\end{aligned}\ ] ] substituting this expression for @xmath51 , the normalized collision distribution function becomes @xmath288g(\theta ) \label{eqn_colang1}\ ] ] and the angular pair - distribution function @xmath289.\ ] ] it is clear that the angular dependence of @xmath141 is modified by both the inelastic dissipation and the density . herrmann , j .- hovi and s. luding , _ physics of dry granular media_ nato asi series e 350 , ( kluwer academic , dordrecht , 1998 ) . t. pschel and s. luding , _ granular gases _ , lecture notes in physics 764 , ( springer , berlin , 2001 ) . campbell , `` rapid granular flows , '' _ ann . fluid mech . _ , * 22 * , 57 ( 1990 ) . i. goldhirsch and n. sela , `` origin of normal stress differences in rapid granular flows , '' phys e. , * 54 * , 4458 ( 1996 ) . n. sela and i. goldhirsch , `` hydrodynamic equations for rapid flows of smooth inelastic spheres to burnett order , '' j. fluid mech . , * 361 * , 41 ( 1998 ) . j.d . goddard , `` dissipative materials as models of thixotropy and plasticity , '' j. non - newt . fluid mech . , * 14 * , 141 ( 1984 ) . bird , r.c . armstrong and o. hassager , _ dynamics of polymeric liquids _ , ( john wiley sons , london , 1977 ) . j. d. goddard , `` polymer fluid mechanics , '' adv . mech . , * 19 * , 143 ( 1977 ) . j.d . goddard and m. alam , `` shear - flow and material instabilities in particulate suspensions and granular media , '' particulate science and technology , * 17 * , 69 ( 1999 ) . p. coussot , q.d . nguyen , h.t . huynh and d. bonn , `` avalanche behavior in yield stress fluids , '' phys . lett . , * 88 * , 175501 ( 2002 ) . m.a . hopkins and m.y . louge , `` inelastic microstructure in rapid granular flows of smooth disks , '' phys . fluids a , * 3 * , 47 ( 1991 ) . i. goldhirsch and d. zanetti , `` clustering instability in dissipative gases , '' phys . rev . lett . , * 70 * , 1619 ( 1993 ) . s. luding and h.j . herrmann , `` cluster growth in freely cooling granular media , '' chaos , * 9 * , 673 ( 1999 ) . s. b. savage , `` instability of unbounded uniform granular shear flow , '' j. fluid mech . , * 241 * , 109 ( 1992 ) . m. alam and p.r . nott , `` stability of plane couette flow of a granular material , '' j. fluid mech . , * 377 * , 99 ( 1998 ) . jenkins and r.w . richman , `` plane simple shear of smooth inelastic circular disks : the anisotropy of the second moment in the dilute and dense limit , '' j. fluid mech . , * 192 * , 313 ( 1988 ) . goddard and a.k . didwania , `` computations of dilatancy and yield surfaces for assemblies of rigid frictional spheres , '' quart . , * 39 * , 145 ( 1997 ) . m. reiner , `` a mathematical theory of dilatancy , '' amer . , * 67 * , 350 ( 1945 ) . campbell and a. gong , `` the stress tensor in a two - dimensional granular shear flow , '' j. fluid mech . , * 164 * , 107 ( 1986 ) . o.r . walton and r.l . braun , `` viscosity , granular - temperature , and stress calculations for shearing assemblies of inelastic , frictional disks , '' j. rheol . , * 30 * , 949 ( 1986 ) . allen and d.j . tildesley , _ computer simulations of liquids _ , ( clarendon press , oxford , 1989 ) . a.w . lees and s.f . edwards , `` the computer study of transport processes under extreme conditions , '' j. phys . c , * 5 * , 1921 ( 1972 ) . m. alam and s. luding , `` how good is the equipartition assumption for the transport properties of a granular mixture ? '' granular matter , * 4 * , 139 ( 2002 ) . m. alam and s. luding , `` rheology of bidisperse granular mixtures via event - driven simulations , '' j. fluid mech . , * 476 * , 69 ( 2003 ) . liss and b.j . glasser , `` the influence of clusters on the stress in a sheared granular material , '' powder tech . , * 116 * , 116 ( 2001 ) . s. luding , `` global equation of state of two - dimensional hard sphere systems , '' phys . e. , * 63 * , 042201 - 1 ( 2001 ) . s. luding and o. strauss , `` the equation of state of polydisperse granular gases , '' in _ granular gases _ , lecture notes in physics 564 , t. pschel and s. luding ( eds . ) , springer , 2001 . richman , `` the source of second moment in dilute granular flows of highly inelastic spheres , '' j. rheol . , * 33 * , 1293 ( 1989 ) . a. santos , j.m . montanero , j.dufty and j.j . brey , `` kinetic model for the hard - sphere fluid and solid , '' phys . e , * 57 * , 1644 ( 1998 ) . r.l . fosdick and k.r . rajagopal , `` anomalous features in the model of second order fluids , '' arch . , * 70 * , 145 ( 1979 ) . j.f . brady and m. vivic , `` normal stresses in colloidal dispersions , '' j. rheol . , * 39 * , 545 ( 1995 ) . d.r . foss and j.f . brady , `` structure , diffusion and rheology of brownian suspensions by stokesian dynamics simulations , '' j. fluid mech . , * 407 * , 167 ( 2000 ) . s. luding , `` structures and non - equilibrium dynamics in granular media , '' comptes rendus academie des science , * 3 * , 153 ( 2002 ) . l. v. woodcock and c. a. angel , `` diffusivity of the hard - sphere model in the region of fluid metastability , '' phys . , * 47 * , 1129 ( 1981 ) . kirkwood , f.p . buff and m.s . green , `` the statistical mechanical theory of transport processes , '' j. chem . * 17 * , 988 ( 1949 ) . s.b . savage and d.j . jeffrey , `` the stress tensor in a granular flow at high shear rates , '' j. fluid mech . , * 110 * , 255 ( 1981 ) . campbell and c.e . brennen , `` computer simulation of granular shear flows , '' j. fluid mech . , * 151 * , 167 ( 1985 ) . jenkins and e. askari , `` boundary conditions for rapid granular flows : phase interfaces , '' j. fluid mech . , * 223 * , 497 ( 1991 ) . s. mcnamara and w.r . young , `` inelastic collapse in two dimensions , '' phys . e , * 50 * , r28 ( 1994 ) . s. luding and s. mcnamara , `` how to handle the inelastic collapse of a dissipative hard particle gas with the tc model ? '' granular matter , * 1 * , 113 ( 1998 ) . s. luding and a. goldshtein , `` hydrodynamic theory for dissipative hard spheres with multiparticle interactions , '' granular matter ( 2003 , in press ) . nott , p.r . , alam , m. , agrawal , k. , jackson , r. and sundaresan , s , `` the effect of boundaries on the plane couette flow of granular materials : a bifurcation analysis , '' j. fluid mech . , * 397 * , 203 ( 1999 ) tan , `` microstructures and macrostructures in rapid granular flows , '' ph.d . thesis , princeton ( 1995 ) . campbell `` granular shear flows at the elastic limit , '' j. fluid mech . , * 465 * , 261 ( 2002 ) . j. f. lutsko , `` molecular chaos , pair correlation , and shear - induced ordering of hard spheres , '' phys . lett . , * 77 * , 2225 ( 1996 ) . j. erpenbeck , `` shear viscosity of the hard sphere fluid via nonequilibrium molecular dynamics , '' phys . lett . , * 52 * , 1333 ( 1984 ) . d.j . evans and g.p . morriss , `` shear thickening and turbulence in simple fluids , '' phys . , * 56 * , 2172 ( 1986 ) savage and s. sayed , `` stresses developed by dry cohesionless granular materials in an annular shear cell , '' j. fluid mech . , * 142 * , 391 ( 1984 ) . h. grad , `` on the kinetic theory of rarefied gases , '' comm . pure and appl . , * 2 * , 331 ( 1949 ) . jenkins and m.w . richman , `` grad s 13-moment system for a dense gas of inelastic spheres , '' arch . , * 87 * , 355 ( 1985 ) . s. jin and m. slemrod , `` regularization of the burnett equations for rapid granular flows via relaxation , '' physica d , * 150 * , 207 ( 2001 ) . d.z . zhang and r.m . rauenzahn , `` a viscoelastic model for dense granular flows , '' j. rheol . , * 41 * , 1275 ( 1997 ) . d.z . zhang and r.m . rauenzahn , `` stress relaxation in dense and slow granular flows , '' j. rheol . , * 44 * , 1019 ( 2000 ) . e. ikenberry and c. truesdell , `` on the pressures and flux of energy in a gas according to maxwell s kinetic theory , '' j. rat . , * 5 * , 1 ( 1956 ) . i. mller , `` on the frame dependence of stress and heat flux , '' arch . , * 45 * , 241 ( 1972 ) . m. lee , j. dufty , j. montanero , a. santos and j. f. lutsko , `` long wavelength instability for uniform shear flow , '' phys . lett . , * 76 * , 2702 ( 1996 ) . r. soto , j. piasecki and m. mareschal , `` precollisional velocity correlations in a hard - disk fluid with dissipative collisions , '' phys . e , * 64 * , 031306 ( 2001 ) . s. hess , `` structure of liquids and solids in nonequilibrium , '' j. de physique , * 46 * ( c3 ) , 191 ( 1985 ) .
we use a large suite of carefully controlled full hydrodynamic simulations to study the ram pressure stripping of the hot gaseous halos of galaxies as they fall into massive groups and clusters . the sensitivity of the results to the orbit , total galaxy mass , and galaxy structural properties is explored . for typical structural and orbital parameters , we find that of the initial hot galactic halo gas can remain in place after 10 gyr . we propose a physically simple analytic model that describes the stripping seen in the simulations remarkably well . the model is analogous to the original formulation of gunn & gott ( 1972 ) , except that it is appropriate for the case of a spherical ( hot ) gas distribution ( as opposed to a face - on cold disk ) and takes into account that stripping is not instantaneous but occurs on a characteristic timescale . the model reproduces the results of the simulations to within at almost all times for all the orbits , mass ratios , and galaxy structural properties we have explored . the one exception involves unlikely systems where the orbit of the galaxy is highly non - radial and its mass exceeds about 10% of the group or cluster into which it is falling ( in which case the model under - predicts the stripping following pericentric passage ) . the proposed model has several interesting applications , including modelling the ram pressure stripping of both observed and cosmologically - simulated galaxies and as a way to improve current semi - analytic models of galaxy formation . one immediate consequence is that the colours and morphologies of satellite galaxies in groups and clusters will differ significantly from those predicted with the standard assumption of complete stripping of the hot coronae . [ firstpage ] hydrodynamics methods : n - body simulations galaxies : clusters : general galaxies : evolution galaxies : structure cosmology : theory
0710.0964
there are marked differences in the observed properties of the field and cluster galaxy populations . perhaps the best known difference is the larger fraction of galaxies that are ellipticals or s@xmath2s ( and the correspondingly lower spiral fraction ) in clusters relative to the field ( e.g. , dressler 1980 ; goto et al . not only are the morphologies of cluster galaxies different from those of field galaxies , but so too are a variety of their other observed properties , including colours ( e.g. , balogh et al . 2004 ; hogg et al . 2004 ) , star forming properties ( e.g. , poggianti et al . 1999 ; balogh et al . 2000 ; gomez et al . 2003 ) , and the distribution and total mass of their gaseous component ( e.g. , cayatte et al . 1994 ; solanes et al . these observed differences indicate that the dense environments of groups and clusters are somehow strongly modifying the properties of galaxies as they fall in . uncovering the physical mechanisms that give rise to the observed variation in galaxy properties has been an active topic of research over the past two or three decades ( e.g. , dressler 1984 ; sarazin 1988 ) . one of the most commonly mentioned processes is ram pressure stripping ( gunn & gott 1972 ) . here the gaseous component ( which can be composed of both cold atomic / molecular gas and a hot extended component ) of the orbiting galaxy is subjected to a wind due to its motion relative to the intracluster medium ( icm ) . the gas will be stripped if the wind is sufficiently strong to overcome the gravity of the galaxy . recently , direct observational evidence for the ram pressure stripping of galaxies in clusters has been provided by long ( up to tens of kpc ) tails of gas seen to be trailing behind several cluster galaxies ( e.g. , sakelliou et al . 2005 ; crowl et al . 2005 ; vollmer et al . 2005 ; sun & vikhlinin 2005 ; machacek et al . 2006 ; sun et al . such stripping could at least partially account for the differing properties of cluster and field galaxies . there have been numerous theoretical studies dedicated to calculating the effects of ram pressure stripping on galaxies using hydrodynamic simulations or semi - analytic models . the vast majority of these studies have focused on the stripping of cold gaseous disks with an emphasis on whether this can account for the observed lower star formation rates ( and redder colours ) of cluster spirals relative to their field counterparts ( e.g. , abadi et al . 1999 ; quilis et al . 2000 ; vollmer et al . 2001 ; okamoto & nagashima 2003 ; mayer et al . 2006 ; roediger et al . 2006 ; hester 2006 ; jachym et al . 2007 ; roediger & brggen 2006 ; 2007 ) . however , the stripping of extended _ hot _ gaseous halos of galaxies is only just beginning to be explored ( e.g. , kawata & mulchaey 2007 ) and has not yet been studied in a detailed and systematic way . the hot extended component is predicted to exist around most massive galaxies by semi - analytic models and cosmological simulations and is directly observable at x - ray wavelengths in the case of normal ellipticals . if the hot gaseous halo is completely stripped ( as is typically assumed ) , the only fuel available for star formation is that which resided in the cold component when the galaxy first fell into the cluster . ( this process of removing the supply of halo gas is sometimes referred to as `` strangulation '' or `` starvation '' . ) however , if the hot halo remains intact for some time it can , via radiative cooling losses , replenish the cold component and potentially significantly prolong star formation . this , in turn , would affect the colours and morphologies of cluster galaxies ( e.g. , larson et al . 1980 ; abadi et al . 1999 ; benson et al . 2000 ; balogh et al . 2000 ) . aspects of the stripping of the _ hot _ gaseous halos of galaxies in clusters have been considered in previous work ( e.g. , gisler 1976 ; sarazin 1979 ; takeda et al . mori & burkert ( 2000 ) studied the stripping of dwarf galaxies subject to a constant wind using two - dimensional simulations and found that the relatively shallow potential wells of these systems can not retain their hot gas component for long . however , these authors did not study more massive systems , such as normal ellipticals and spirals , where stripping of the hot ( @xmath3 k ) halo should be much less efficient due to their higher masses and deeper potential wells . [ indeed , a recent x - ray survey of massive galaxies in hot clusters by sun et al . ( 2007b ) has revealed that _ most _ of the galaxies have detectable hot gaseous halos . ] a few other studies have examined the stripping of more massive systems but not in the context described above . in particular , they have largely focused on the metal enrichment of the icm ( e.g. , schindler et al . 2005 ; kapferer et al . 2007 ) , the x - ray properties of the galaxies ( toniazzo & schindler 2001 ; acreman et al . 2003 ) or the generation of `` cold fronts '' ( e.g. , takizawa 2005 ; ascasibar & markevitch 2006 ) . in the present paper , we carry out a detailed study of the ram pressure stripping of the hot gaseous halos of galaxies as they fall into groups and clusters . this is performed using a large suite of controlled hydrodynamic simulations . unlike most previous studies , we use full three - dimensional ( 3d ) simulations in which the galaxies fall into a massive `` live '' group or cluster on realistic orbits . one important aim is to derive a physically simple and accurate description of the stripping seen in the simulations that can be easily employed in the modelling of observed or cosmologically - simulated galaxies . an additional motivation for deriving such a model is to improve the treatment of ram pressure stripping in semi - analytic models of galaxy formation . at present , these models typically assume that the hot gaseous halos of galaxies are stripped at the instant they cross the virial radius of the group or cluster . clearly , this is not a realistic assumption , especially in the case where the mass of the galaxy is not negligible compared to that of the group or cluster . such semi - analytic models tend to predict group and cluster galaxies whose colours are too red compared to observations ( e.g. , weinmann et al . 2006 ; baldry et al . 2006 ) . if the ram pressure stripping of the hot gaseous halos of cluster galaxies is not as ( maximally ) efficient as assumed by these models , the resulting galaxies would be bluer and perhaps in better agreement with observations . the present paper is structured as follows . in 2 , we present a discussion of our simulation setup and the results of convergence tests that demonstrate the robustness of our findings . in 3 , we first outline a simple analytic model for ram pressure stripping that is based on the original formulation of gun & gott ( 1972 ) but which is appropriate for spherically - symmetrical gas distributions ( as opposed to disks ) . we then compare this model to a wide variety of simulations and demonstrate that it provides an excellent match to the mass loss seen in the simulations . finally , in 4 , we summarise and discuss our findings . to study the ram pressure stripping of galaxies orbiting in massive groups and clusters , we make use of the public version of the parallel treesph code gadget-2 ( springel 2005 ) . by default , this code implements the entropy - conserving sph scheme of springel & hernquist ( 2002 ) . the procedure we use to set up our simulations is quite similar to that described in mccarthy et al . ( 2007a ) ( hereafter , m07 ) . we outline the basic procedure and note any relevant differences between our setup and that of m07 . in this study , ram pressure stripping is explored in two types of simulations . we refer to the first type as the `` uniform medium '' runs , where a galaxy is run through a uniform gaseous medium at constant velocity . in the second type of simulations ( the `` 2-system '' runs ) , the galaxies are placed on realistic orbits through a massive `` live '' galaxy group . in the uniform medium runs , the ram pressure to which the galaxy is exposed is effectively constant with time . furthermore , there is no external gravitational potential ( i.e. , due to a massive group or cluster ) to tidally distort or strip the galaxy . as a result , these simulations should provide a pure test of ram pressure stripping and should be easier to model than the second type of simulations . on the other hand , if the lessons learnt from modelling the uniform medium runs do not also generally apply to more realistic situations , such as those in the 2-system runs , they will be of little practical use . this is why we have elected to use both types of simulations to study this problem . the galaxies ( and the groups into which they fall , in the case of the 2-system runs ) are represented by spherically - symmetric systems composed of a realistic mixture of dark matter and diffuse baryons . the dark matter is assumed to follow the nfw distribution ( navarro et al . 1996 ; 1997 ) : @xmath4 where @xmath5 and @xmath6 here , @xmath7 is the radius within which the mean density is 200 times the critical density , @xmath8 , and @xmath9 . the only ` free ' parameter of the nfw profile is the scale radius , @xmath10 . the scale radius is often expressed in terms of a concentration parameter , @xmath11 . by default , we adopt the mean mass - concentration ( @xmath12 ) relation derived from the _ millennium simulation _ ( springel et al . 2005 ) by neto et al . ( 2007 ) . this relationship is similar to that derived previously by eke et al . ( 2001 ) . for simplicity , the diffuse baryons are assumed to initially trace the dark matter distribution , with the ratio of gas to total mass set to the universal ratio @xmath13 , where @xmath14 is the hubble constant in units of @xmath15 km s@xmath16 mpc@xmath16 . the other properties of the diffuse gas ( i.e. , temperature and pressure profiles ) are fixed by ensuring the gas is initially gravitationally bound and in hydrostatic equilibrium , @xmath17 while the assumption that the gas initially traces the dark matter is reasonable for the bulk of the baryons in massive groups and clusters ( e.g. , vikhlinin et al . 2006 ; mccarthy et al . 2007b ) , it is almost certainly not a very realistic approximation for relatively low - mass systems , such as galaxies . the reason , of course , is that non - gravitational physics , such as cooling and feedback due , for example , to supernovae and/or agn , which are neglected in our simulations , can significantly alter the properties of the gas in these systems . these processes are poorly understood and the properties of the gas will likely depend sensitively on the assumed feedback model . therefore , any distribution we select for the hot gaseous halo of the galaxies will be somewhat _ ad hoc_. the important point , however , is that one can use the simulations to develop a _ physical _ model for ram pressure stripping that can , with some confidence , be applied more generally . we argue that the analytic model developed below is just such a model . as will be demonstrated , tuning the model to match just one of our simulations results in very good agreement with all the other simulations , in spite of their widely varying physical conditions . ( dotted red curves ) and a group with mass @xmath18 ( solid blue curves ) . , width=317 ] the reader is referred to 2 of m07 for a detailed discussion of how we establish equilibrium configurations of dark matter and gas particles that follow an nfw distribution . in the case of the 2-system runs , the more massive system is set to have a total mass of @xmath18 , while the less massive systems have masses in the range @xmath19 ( i.e. , mass ratios from 50:1 to 10:1 ) . thus , the 2-system runs represent galaxies with masses comparable to or larger than a normal elliptical falling into a moderate mass group / low mass cluster . note that for galaxies within this mass range , the mean temperature of their gaseous halos ranges between @xmath20 k. in fig . 1 , we show the initial gas density and temperature profiles for the hot gaseous halo of one of the galaxies and for the icm of the @xmath21 group . the default gas particle mass , @xmath22 , is set to @xmath23 , while the default dark matter particle mass , @xmath24 , is set to @xmath25 . in the 2-system runs , these masses are fixed for both the group and the galaxy . this implies that the group is resolved with half a million gas and dark matter particles ( each ) within @xmath7 . the gravitational softening length for both the gas and dark matter particles is set to 5 kpc for all our simulations . ( we have experimented with different values of this parameter and find no significant differences in the results . ) a standard set of sph parameters is adopted ( e.g. , springel 2005 ) . the number of sph smoothing neighbour particles is set to 32 , the artificial viscosity @xmath26 parameter is set to 0.8 ( see 2.2 ) , and the courant timescale coefficient is set to 0.1 . the simulation data are output frequently , at intervals of 50 myr , and the simulations are run for a maximum duration of 10 gyr in the case of the 2-system runs or until a convergent result is achieved in the case of the uniform medium runs . the effects of ram pressure stripping are quantified by computing the mass of gas that remains gravitationally bound to the galaxy as a function of time . to determine which gas and dark matter particles are bound to the galaxy in any particular simulation output , we use the iterative method outlined in tormen et al . ( 1998 ) and hayashi et al . starting from the distribution of particles that were bound at the previous simulation output ( noting that all particles were bound initially ) , the potential , kinetic , and , in the case of the gas , the internal energies of each of the particles in the rest - frame of the galaxy are computed . we discard all particles for which the sum of the kinetic and internal energies exceeds the potential energy . the rest - frame of the bound structure is then recomputed , as are the energies of each particle , and any additional unbound particles are identified and discarded . this procedure is repeated until no further particles are identified as being unbound . furthermore , it is implicit that once a gas particle has been lost due to stripping it can not at a later time become gravitationally bound again ( i.e. , the mass of bound gas is necessarily a monotonically decreasing function of time ) . in this way , we are calculating a conservative lower limit to the mass of bound gas . to the initial mass of gas ( at @xmath27 ) versus time . in the top panel , we show the effect of raising the gas particle mass ( i.e. , lowering the particle number ) from our default gas mass resolution of @xmath28 . in the bottom panel , we show the effect of lowering the sph artificial viscosity parameter @xmath26 . , width=317 ] there are a variety of numerical issues that could potentially affect the simulations and hamper the development of a physical model for ram pressure stripping . perhaps of most concern is the effect of limited numerical resolution and , in the case of sph simulations , the role of the artificial viscosity term , which itself is resolution - dependent . the artificial viscosity , which is necessary in order for sph algorithms to capture shocks , acts like an excess pressure for the gas particles in their equation of motion and is therefore potentially relevant to our discussion of ram pressure stripping . we have investigated the effects of numerical resolution and artificial viscosity in our default 2-system run ( see 3.3.1 for a description of this run ) . the results are plotted in figure 2 . in the top panel , we show the effect of degrading the mass resolution of the gas particles ( the mass resolution of the dark matter particles is the same for all these runs ) on the ram pressure stripping of the galaxy . in the default case , there are @xmath29 bound gas particles inside @xmath7 of the galaxy initially . reassuringly , we find that lowering the number of particles does not significantly affect the resulting bound mass of gas as a function of time . this is the case even when the gas halo is represented initially by only 1000 particles . in fact , significant ( @xmath30 ) differences appear only when the initial gas particle number in the galaxy is lowered to a few hundred ( not shown ) . note that the top panel of fig . 2 implies that our results are not strongly sensitive to the artificial viscosity , since this is a resolution - dependent quantity . the bottom panel of fig . 2 adds further confidence that the results are robust . here , we experiment with lowering the @xmath26 parameter , which controls the effective ` strength ' of the artificial viscosity and is proportional to the excess pressure assigned to each gas particle in the equation of motion . lowering the value of @xmath26 from the default value of 0.8 has no significant consequences for the resulting bound mass of gas . this is the case even when the artificial viscosity is set to zero . we conclude that our ram pressure results are robust to our choice of resolution and artificial viscosity strength . it should be noted , however , that ram pressure is not the only mechanism by which gas can be stripped from galaxies as they orbit about groups and clusters . in particular , kelvin - helmholtz ( kh ) and rayleigh - taylor ( rt ) instabilities can potentially develop at the interface between the hot halo of the galaxy and the icm and eventually completely disrupt or destroy the gaseous halo of the galaxy . it is known that sph simulations tend to suppress such instabilities in the presence of large density gradients across the interface . this , in turn , will make the hot halo of a galaxy more resilient to stripping than it otherwise would have been . a good example of this can be found in agertz et al . ( 2007 ) , where a comparison between several eulerian grid - based codes ( which accurately follow the growth of these instabilities ) and several lagrangian sph codes is performed for an idealised case where a ` blob ' of gas moves through a uniform density medium . for example , their fig . 4 shows that , for one particular case , the grid - based codes all predict complete disruption of the blob at @xmath31 ( where @xmath32 is the kh timescale , i.e. , the time it takes kh instabilities to fully grow ) , whereas the sph codes predict that the blob should remain intact . with this in mind , one might conclude that sph simulations such as ours will overestimate the survivability of the hot halo of a galaxy . however , it is important to note that agertz et al . find that the grid - based and sph - based codes agree with each other rather well for @xmath33 ( see also appendix a of the present study ) . following the approach of mori & burkert ( 2000 ) ( see also nulsen 1982 ; murray et al . 1993 ; and mayer et al . 2006 ) , the kelvin - helmholtz timescale ( including the stabilising effects of gravity ) can be estimated as : @xmath34 where @xmath35 is the baryon fraction of the galaxy , @xmath36 is the total mass of the galaxy within the radius down to which the galaxy has been stripped by ram pressure , @xmath37 is the number density of hydrogen atoms in the icm , and @xmath38 is the velocity of the galaxy with respect to the icm . for our default 2-system run ( see 3.3.1 ) , for example , we estimate from eqn . ( 4 ) that the kelvin - helmholtz timescale at pericentre is approximately 4.5 gyr ( i.e. , which is comparable to the duration of our simulations ) . since most of the orbital period of the galaxy is spent far from pericentre , the value of @xmath32 will be substantially longer than this . note also that the timescale associated with the growth of rt instabilities is comparable to or exceeds @xmath32 . therefore , we do not expect kh or rt instability stripping to have important consequences for the results or conclusions of this study . we also point out that eqn . ( 4 ) neglects the possibly important effects of radiative cooling , physical viscosity , magnetic fields , etc . , all of which will tend to damp ( and possibly halt ) the growth of such instabilities in real cluster galaxies . finally , in order to dispel any lingering doubts that our adopted sph approach is unable to treat ram pressure stripping accurately , we have made a direct comparison of the predictions of the lagrangian sph code gadget-2 and the eulerian amr code flash for one of our uniform medium runs . this comparison is presented in appendix a and shows that there is excellent quantitative agreement between the results of the two codes . the study of ram pressure stripping of galaxies as they fall into groups and clusters dates back to the seminal paper of gunn & gott ( 1972 ) . using a static force argument , these authors derived a simple , physically - motivated condition for the instantaneous ram pressure stripping of a gaseous disk moving face - on through the icm . the gas will be stripped if the ram pressure , @xmath39 , defined as @xmath40 ( where @xmath41 is the density of the icm and @xmath38 is the speed of the galaxy with respect to the icm ) , exceeds the gravitational restoring force per unit area on the disk , which they derive as @xmath42 ( where @xmath43 and @xmath44 are the stellar and gaseous surface densities of the disk , respectively ) . we now seek to derive an analogous model for the ram pressure stripping of a spherically - symmetric gas distribution . at the outer edge ( radius @xmath45 ) of the gaseous halo of the galaxy . , width=317 ] since it is the least bound material , gas at the outer projected edge of the system will be stripped first ( see the schematic diagram in figure 3 ) . consider gas in a projected annulus between radii @xmath45 and @xmath46 . the projected area of this annulus , @xmath47 , is @xmath48 . therefore , the force due to ram pressure on this annulus is simply @xmath49 . the annulus of gas will be displaced in the direction opposite to @xmath38 ( which we will call the @xmath50 direction ) and will be stripped if the force due to the ram pressure exceeds the maximum gravitational restoring force in this direction . the maximum gravitational restoring force , @xmath51 , can be written approximately as @xmath52 , where @xmath53 is the maximum restoring acceleration in the @xmath50 direction and @xmath54 is the projected surface density of the gas in the annulus . therefore , the ram pressure stripping condition can be written as : @xmath55 if the gas density and total mass profiles of the galaxy can be represented by simple power laws , it is straightforward to evaluate the right - hand side of equation ( 5 ) . in the case of a singular isothermal sphere , for example , where @xmath56 and @xmath57 ( where @xmath58 is the _ total _ mass within radius @xmath59 ) , we find @xmath60 and @xmath61 . this leads to the following stripping condition @xmath62 for more general gas density and total mass profiles , the condition for ram pressure stripping may be expressed as @xmath63 where @xmath64 is a geometric constant of order unity which depends on the precise shape of the gas density and total mass profiles of the galaxy . we note that equation ( 7 ) is similar to the analytic stripping conditions derived previously by gisler ( 1976 ) and sarazin ( 1979 ) ( among others ) for elliptical galaxies . equation ( 7 ) implies that all the gas beyond the 3d radius @xmath65 where the ram pressure exceeds the gravitational restoring force per unit area ( which we will refer to as the stripping radius ) will be stripped . by assumption , the properties of both the gas and dark matter within the stripping radius are unmodified by the stripping . the left - hand side of eqn . ( 7 ) makes it clear that the ram pressure is , in general , a function of time ( i.e. , for non - circular orbits ) . below , we use the idealised uniform medium runs to test this simple analytic model . however , before doing so it is worth briefly discussing some of the assumptions of this simple model and their validity . firstly , the model neglects kh and rt instability stripping but , as we argued in 2.2 , we do not expect this to be an important omission . perhaps of more concern is that , by assuming that the properties of the system within the stripping radius do not change with time , the model implicitly neglects environmental effects such as tidal stripping and gravitational shock heating . in appendix b , we show , using a simple argument , that one expects ram pressure stripping to be more efficient than tidal stripping for cases where the mass of the galaxy is less than about 10% of the mass of the group . in other words , for galaxies with masses of less than about 10% of the group mass , tidal stripping is not expected to substantially modify the structure of the galaxy within its stripping radius . our 2-system runs involve only systems with mass ratios @xmath66 10:1 . the neglect of shock heating would naively appear to be a more serious omission , since the commonly - held picture of structure formation is that gas accreted by a massive system is shocked at the virial radius up to the virial temperature of the massive system . thus , one might expect the hot gas halo of the galaxy to be quickly shock heated and become unbound . however , high resolution simulations ( both cosmological and idealised ) do not confirm this picture . in particular , if the material being accreted is in small dense `` lumps '' ( e.g. , low - mass virialised systems , as in the present case ) , it can penetrate all the way to the core of the massive system without being significantly shocked ( e.g. , motl et al . 2004 ; murray & lin 2004 ; poole et al . 2006 ; m07 ; dekel & birnboim 2007 ) . in fact , most of the interaction energy is thermalised in the ambient medium of the more massive system ( the icm , in this case ) , while the accreted gas sinks to bottom of the potential well ( see m07 for a detailed discussion ) . however , m07 found that the fraction of the total energy that is thermalised in the gas of the less massive system ( the galaxy , in this case ) increases almost linearly with the ratio of the mass of the less massive system to the total mass of both systems . therefore , shock heating _ is _ expected to become important for cases where the mass of the galaxy is comparable to the mass of the group . our 2-system runs , however , only involve galaxies with masses lower than 10% of the mass of the group . we now explore the ram pressure stripping of galaxies as they move through a uniform density gaseous medium . for the uniform medium , we select densities that are typical of the group / cluster environment . the temperature of the medium is set such that its pressure equals that of the hot halo of the galaxy at its outer edge ( i.e. , the gaseous halo would be static if it were not moving with respect to the uniform medium ) . the galaxies are assigned velocities typical of systems orbiting in genuine groups and clusters ( i.e. , comparable to the circular velocity of the group or cluster ) . in fig . 4 we plot the bound mass of gas as a function of time for a small selection of the uniform medium runs we have performed and compare this with our proposed analytic model . we focus first on the @xmath67 curves from the simulations ( solid red curves ) . firstly , the @xmath67 curves in both panels asymptote to a particular value , as one would expect from the physical model proposed above where the ram pressure is effectively held constant with time but is low enough that not all of the gas should be stripped . in the bottom panel , where a galaxy is moved through media of two different densities but with velocities chosen such that the ram pressure is the same , the resulting @xmath67 curves are very similar . this unambiguously demonstrates that the mass loss is indeed due to ram pressure stripping . is run through a uniform gaseous medium of density @xmath68 at a velocity of 1000 km s@xmath16 . the solid red and dashed blue curves show the bound mass of gas and dark matter ( respectively ) in the simulation . in the bottom panel , the same galaxy is run through two different media : the thick red curve corresponds to the case where the background density is the same as in the top panel , but the velocity is 760 km s@xmath16 , while the thin red curve corresponds to the case where the velocity is 1000 km s@xmath16 but the density is a factor of @xmath69 times lower than in the top panel . thus , the ram pressure is the same for both cases in the bottom panel . in both the top and bottom panels the horizontal dotted lines correspond to the predictions of equation ( 7 ) for @xmath70 2 , 4 , 6 , 8 , and 10 ( bottom to top ) . the green dashed curve corresponds to equation ( 7 ) with @xmath71 but with a time delay factor that accounts for how long it takes the for galaxy to respond to ram pressure stripping ( i.e. , approximately a sound crossing time ) , as discussed in the text . , width=317 ] a comparison to the predictions of equation ( 7 ) ( horizontal dotted lines ) demonstrates that the asymptotic behaviour of the simulations is reproduced if @xmath72 . ( note that this is very similar to the analytic estimate of @xmath73 derived in 3.1 for an isothermal sphere . ) in fact , all the uniform medium runs we have performed yield a value of @xmath64 close to 2 . however , it is immediately apparent that the approximation of instantaneous stripping is not a particularly good one . for example , in the cases plotted in fig . 4 it takes @xmath74 gyr of stripping to reach a convergent value ( i.e. , to reach the 3d radius where the ram pressure equals the gravitational force per unit area ) . this `` time delay '' has been noted previously in studies of the stripping of cold disks ( e.g. , roediger & brggen 2006 ; 2007 ) and is expected on physical grounds ; the hot halo of the galaxy can only respond to changes in the local environment on a finite timescale . what is the relevant timescale ? if the galaxy is moving subsonically , a natural choice might be the sound crossing time ; i.e. , the time it takes for a pressure wave to cross the galaxy s hot halo . if the galaxy is moving supersonically , a better choice might be the time it takes a forward shock to propagate across the galaxy ( e.g. , nittmann et al . 1982 ; mori & burkert 2000 ) . ( although , as we noted above , shock heating of the hot gas of the galaxy is minimal in our simulations . ) alternatively , roediger & brggen ( 2007 ) estimate and use the timescale required for the ram pressure to accelerate the gas to the galaxy s escape velocity . we have experimented with including a time delay factor into the analytic model ( how we do this is described below ) that is set to either of these three timescales times . in practice , we find that use of either of these timescales leads to very similar results . this is not too surprising . the similarity between the sound and shock crossing times is due to the fact that , in the rest frame of the group , the galaxy is typically orbiting at transonic velocities ( i.e. , mach number @xmath74 ) . the similarity between the sound crossing time and the time required to accelerate gas to the galaxy s escape velocity is also not coincidental . since the galaxy s hot halo is in approximate hydrostatic equilibrium , the mean temperature of the gas will be close to the overall virial temperature of the galaxy ( which is dominated by the mass in dark matter ) and therefore the sound crossing time of the hot halo will be of order the dynamical time of the galaxy . consequently , if the force due to the ram pressure is of order the gravitational restoring force , as is the case for typical transonic velocities , the time it takes to accelerate the gas to the escape velocity will be comparable to the sound crossing time . note , however , that if the galaxy s motion is highly supersonic ( or if the gas is not in equilibrium ) one might expect differences between the three timescales . the present study does not consider this regime and instead focuses on the more physically relevant transonic regime where all three timescales are similar . below , we present results based on using the sound crossing time only . note that the simulation @xmath67 curves plotted in fig . 4 show that the mass loss proceeds with time more or less linearly until convergence is achieved . ( this is generally true of the 2-system runs presented below , as well . ) we therefore assume that the mass of gas stripped over some time interval @xmath75 is just the total mass of stripped gas inferred from the instantaneous assumption ( i.e. , the total gas mass external to the stripping radius ) scaled by the ratio @xmath76 , where @xmath77 is the characteristic timescale for ram pressure stripping ( i.e. , approximately the sound crossing time ) . for an appropriate comparison to the simulations , we set @xmath75 to the adopted simulation output time interval of 50 myr . the sound crossing time of the gaseous halo at any particular time is calculated as : @xmath78 where @xmath45 is the maximum radial extent of the bound galactic gas and @xmath79 is the local sound speed profile , which is given by @xmath80^{1/2}$ ] with @xmath81 . note that for an isothermal gas this leads to the familiar relation @xmath82 . in fact , the time it takes the gaseous halo to respond to changes in the local environment will only be comparable to the sound crossing time , not exactly equal to it . we therefore multiply this timescale by an adjustable coefficient @xmath83 ( which will be of order unity ) when computing how much mass can be stripped over a time interval ( i.e. , @xmath84 ) . ) at the outer edge of the gaseous halo of the galaxy plotted in the top panel of fig.4 , as a function of time . the solid squares represent the median of the 500 outermost gravitationally bound gas particles while the error bars represent the 25th and 75th percentiles . after @xmath85 the ram pressure and restoring force per unit area become comparable , which is why mass loss ceases after this time in fig . 4 . , width=317 ] the resulting model is plotted in fig . 4 . in this case @xmath64 has been fixed to @xmath86 to obtain agreement with the asymptotic @xmath67 behaviour of the simulated galaxies . a value of @xmath87 yields good agreement with the rate of decline of the bound gas mass seen early on in the simulations ( shown is the case corresponding to @xmath88 ) . it is worth bearing in mind that the analytic model uses only the _ initial _ radial profiles of the galaxy to compute the bound mass of gas as a function of time . the fact that the model matches the simulations and that the required values of @xmath64 and @xmath83 are of order unity is encouraging . as a further test of the analytic model , in fig . 5 we plot the ratio of ram pressure to the restoring force per unit area ( assuming @xmath71 ) at the outer edge of the gaseous halo of the simulated galaxy examined in the top panel of fig . 4 , as a function of time . this plot clearly demonstrates that at early times the ram pressure exceeds the gravitational restoring force per unit area , which is why stripping occurs . as shown in the top panel of fig . 4 , stripping continues until @xmath89 gyr and then stops rather abruptly . with fig . 5 we now clearly see the reason for this behaviour : the ram pressure no longer exceeds the restoring force per unit area at the outer edge of the bound halo after this time . in addition , we confirm that the maximum radial extent of the bound gas at @xmath90 gyr corresponds closely to the 3d radius where ram pressure equals the restoring force per unit area calculated from the _ initial _ gas distribution . this validates the basic assumptions of our analytic model , outlined in 3.1 . having calibrated the analytic model against the uniform medium simulations ( i.e. , @xmath64 and @xmath83 are now fixed ) , we now proceed to see whether or not this simple physical model can also account for the mass loss in the more realistic 2-system runs . and @xmath88 ) where stripping occurs on approximately a sound crossing time . the dotted curves are the predictions of equation ( 7 ) with @xmath70 2 , 4 , 6 , 8 , and 10 ( bottom to top ) under the assumption of instantaneous stripping . bottom panel : the ram pressure as a function of time as the galaxy orbits the group . the ram pressure has been normalised to the characteristic value of @xmath91 . for this particular orbit , which corresponds to the most common orbit of infalling substructure in cosmological simulations , pericentric ( apocentric ) passage occurs at @xmath92 gyr ( @xmath93 gyr ) . , width=317 ] the default 2-system run follows a massive galaxy with @xmath94 falling into a moderate mass group of @xmath18 ( implying a mass ratio of 25:1 ) . as noted in 2 , the concentration of these systems is set to match the mean mass - concentration of dark matter halos in the _ millennium simulation_. the 2-system runs are initialised such that the virial radii ( here defined as @xmath7 ) of the two systems are just barely touching . the adopted orbital parameters of the default run correspond to the most common orbit of infalling substructure measured in a large suite of cosmological simulations by benson ( 2005 ; see his fig . 2 ) . specifically , the initial relative radial velocity component , @xmath95 , is set to @xmath96 and the initial relative tangential component , @xmath97 is set to @xmath98 , where @xmath99 is the circular velocity of the group at @xmath7 . this corresponds to a total relative velocity of @xmath100 , which agrees well with the results of several other similar numerical studies ( e.g. , tormen 1997 ; vitvitska et al . 2002 ; wang et al . 2005 ) . in the following subsections , we experiment with varying the orbit , mass , and internal structure of the galaxy to test the generality of the analytic model . as in the uniform medium runs , the analytic model is supplied with the initial conditions of the galaxy ( i.e. , its gas and dark matter radial profiles ) and the magnitude of the ram pressure . in contrast to the uniform medium runs , however , the ram pressure is not constant with time . using the orbit from the simulations , along with the density profile of the group , @xmath101 is calculated and passed to the analytic model . the analytic model can then predict @xmath67 once the values of @xmath64 and @xmath83 have been selected . the mass loss curves for the default 2-system run are plotted in the top panel of fig . overall , the simple analytic model with @xmath71 , @xmath102 ( shown is @xmath103 ) and @xmath104 reproduces the mass loss seen in the default 2-system run very well . for example , both the simulations and the model show evidence for near convergence in @xmath67 at @xmath105 gyr , which corresponds to the ( first ) pericentric passage and , therefore , to the maximum ram pressure which the galaxy experiences along its orbit ( see the bottom panel of fig . 6 ) . the analytic model slightly underestimates the mass loss seen in the simulations at early times . this is a result of the fact that the hot halo of the galaxy is initially slightly over - pressurised with respect to the surrounding hot halo of the group . ( note that this was not the case for the uniform medium simulations plotted in fig . 4 . ) this leads to some expansion of the outer gas which , in turn , makes it more susceptible to stripping . since this effect is in general small and is an artifact of our idealised setup , we do not attempt to model it . while the analytic model with a time delay factor matches the simulations well , an instantaneous stripping model with @xmath106 ( represented by the second dotted curve from the bottom ) also performs well . however , even if the agreement is reasonable , this model is without physical justification and should not be expected to apply in situations that differ significantly from those of the default run . indeed , this is indicated by the results presented later in the paper ( c.f . fig . 8) . we also note that a significant fraction of the dark matter halo is also stripped , particularly near the first pericentric passage . this is not unexpected and is due to the tidal forces acting on the dark matter . we do not attempt to model the stripping of the dark matter , as there are already several published analytic studies which reproduce the dark matter stripping and tidal heating in simulations well ( e.g. , taylor & babul 2001 ; benson et al . instead , the analytic model proposed in 3.1 simply assumes that , within the stripping radius , the properties of the galaxy are unchanged from their initial state . thus , the dark matter halo is assumed to maintain its initial nfw configuration within this radius . in appendix b , we present a simple analytic argument that validates this assumption for systems where the mass of the galaxy is less than about 10% of the mass of the group . we have also directly computed the evolution of the tidal radius ( @xmath107 , defined in binney & tremaine 1987 ; see also appendix b ) of the galaxy in the simulations as a function of time . in fig . 7 we compare the tidal radius with the radial extent of the hot gaseous halo . the tidal radius shrinks at pericentre and then expands but at all times is safely larger than the gaseous halo by at least a factor of 2 . in fig . 6 we examined the ram pressure stripping of a galaxy on the most common orbit seen in cosmological simulations . we now experiment with varying the initial orbital parameters . this will have the effect of changing both the shape and normalisation of @xmath101 . we use fig . 2 of benson ( 2005 ) to select a range of cosmologically likely orbits ; the initial velocity of some orbits is dominated by the radial component while others have nearly circular motions initially is not a common one . we have simulated this atypical orbit just to see if the model breaks down for extreme cases . ] . we plot the mass loss curves for six such orbits in fig . 8 . the mass loss curves in fig . 8 exhibit a variety of behaviours . orbits that initially have a significant tangential component ( and have a total velocity of @xmath108 ) typically undergo only one pericentric passage over the course of 10 gyr . consequently , their associated @xmath67 curves tend to show only one period of significant decline . orbits that are predominantly radial , on the other hand , typically undergo two or more pericentric passages , with each successive passage bringing the galaxy closer to the centre of the group . in these cases we see two ( or more ) periods of significant decline in the bound mass of gas , as expected . in spite of the widely varying orbits , the simple analytic model with @xmath71 , @xmath102 ( shown is @xmath109 ) , and @xmath110 performs remarkably well in predicting the mass loss seen in the simulations . for all orbits and at all times the model predicts @xmath67 to within @xmath1 accuracy . finally , it is interesting to note that if the standard ( but unphysical ) instantaneous ram pressure stripping model were adopted , the implication would be that @xmath64 should vary as a function of the orbit . in particular , from fig . 8 , one would infer relatively low values of @xmath64 ( @xmath111 ) for more circular orbits and relatively high values of @xmath64 ( @xmath112 ) for more radial orbits . however , @xmath64 is a geometric constant that is not expected to depend on the orbit . this consideration provided one of the original motivations for us to explore models where the stripping is not instantaneous . as we have demonstrated , a fixed value of @xmath72 works well for all orbits when one takes into account the finite time required for stripping . we now investigate variations in total mass of the galaxy . this will mainly have the effect of changing the gravitational restoring force ( per unit area ) of the galaxy at all radii by a constant factor . as indicated by fig . 9 , the analytic model matches the higher mass ratio interactions ( lower galaxy masses ) well at all times but does less well for the low mass ratio 10:1 at late times . in particular , the analytic model predicts that there ought to be no further stripping following first pericentric passage while the simulations show evidence for further stripping . what is the origin of this behaviour ? and the default case corresponds to a mass ratio of 25:1 . ) the line types have the same meanings as in fig . , width=317 ] inspection of the 10:1 simulation reveals that the gaseous halo of the galaxy undergoes significant expansion at late times while the analytic model uses the initial gas distribution ( see fig . the expansion , in turn , makes the gas more susceptible to ram pressure stripping , and this accounts for the decline in the bound gas mass at late times . the physical reason for the expansion of the gaseous halo is as follows . at early times , the ram pressure exceeds the restoring force per unit area of the outer gas , which leads to stripping . this stripping proceeds until pericentric passage , when @xmath39 is largest . the remaining bound gaseous halo following pericentric passage is of higher mean density and pressure than the initial system , since all of the low density ( less bound ) material has been removed . following pericentre , the galaxy moves out to large group - centric radii , where the pressure and density of the icm are relatively low compared to pericentre . as a result , the gaseous halo of the galaxy becomes over - pressurised with respect to the local icm and begins to expand . this effect is larger in the case of more massive galaxies since they are more over - pressurised with respect to the icm . the expansion proceeds until approximately apocentre is reached , at which point the galaxy begins to move back into denser and higher pressure regions of the group . ( this effect is also responsible for the mild decline in bound gas mass for the highly tangential orbital case plotted in the top right - hand panel of fig . it is important to note that this over - pressurisation effect is not a numerical artifact , it is a real effect that should be experienced by massive galaxies with orbits that have large energies and tangential components . ) between the centres of the galaxy and group as a function of time . following pericentric passage , the hot halo of the galaxy is over - pressurized compared to the ambient icm and begins to expand . this expansion leads to further ram pressure stripping at late times . , width=317 ] modelling this effect may be possible with some effort . the expansion of the gaseous halo at late times is adiabatic , which greatly simplifies matters . one could therefore compute the radial properties of the gaseous halo as a function of time using the lagrangian entropy distribution of the gas and assuming hydrostatic equilibrium with an outer boundary condition that the pressure must match that of the ambient icm . however , this procedure is complicated by the fact that one must also know the distribution of the galaxy s dark matter halo out to the radius of maximum expansion . while the dark matter profile at small and intermediate radii is sufficiently similar to the initial distribution , this is not the case at very large radii . a proper treatment therefore requires that we factor in dark matter stripping and heating . this could potentially be achieved by combining our analytic ram pressure model with existing analytic models of dark matter stripping and heating ( e.g. , taylor & babul 2001 ; benson et al . however , this is beyond the scope of the present study and we leave it for future work . finally , we stress that the expansion effect just described is relevant to cases where both of the following are true : ( 1 ) the mass of the galaxy is greater than about 10% of the mass of the group ; and ( 2 ) the orbit has an appreciable tangential component and a large enough energy such that apocentre occurs at large group radius ( i.e. , comparable to the group virial radius ) . however , we expect that both of these conditions are rarely fulfilled simultaneously in real systems , as massive satellites tend preferentially to fall into groups and clusters on nearly radial orbits ( i.e. , along filaments ; see , e.g. , benson 2005 ) . . the line types have the same meanings as in fig . , width=317 ] finally , we experiment with varying the internal structure of the galaxy ( both its gas and dark matter ) by varying its initial concentration parameter , @xmath113 ( equivalently , its scale radius , @xmath10 ) . this will mainly have the effect of changing the shape of the radial profile of the restoring force ( per unit area ) . this test is motivated by the fact that in cosmological simulations there is a large degree of intrinsic scatter in the concentration parameter for a system of fixed mass ( e.g. , dolag et al . 2004 ; neto et al . note that changing the concentration can also mimic the addition of another mass component to the galaxy , such as a stellar component ( which we have neglected to include explicitly ) . 11 shows that the concentration has a significant effect on the amount of gas that the galaxy is able to retain as it orbits about the group . as expected , as the concentration is increased so too is the bound mass of gas . as in the previous experiments , the simple analytic model with @xmath71 , @xmath114 ( shown is @xmath88 ) , and @xmath115 matches the mass loss in the simulations very well . using a suite of carefully controlled 3d hydrodynamic simulations , we have investigated the ram pressure stripping of hot gas in the halos of galaxies as they fall into groups and clusters . we have proposed a physically simple analytic model that describes the stripping seen in the simulations remarkably well . this model is analogous to the original formulation of gunn & gott ( 1972 ) , except that it is appropriate for the case of a spherical gas distribution ( as opposed to a face - on disk ) and takes into account that stripping is not instantaneous but occurs on approximately a sound crossing time . the only pieces of information that the model requires are the initial conditions of the orbiting galaxy ( its gas and dark matter profiles ) , the density profile of the icm and the orbit [ the latter two are needed to calculate @xmath101 ] . the model contains two tunable coefficients that are of order unity . fixing these coefficients to match the stripping in just one of our idealised uniform medium simulations ( see 3.2 ) leads to excellent agreement with all our other simulations . with the exception of cases where the mass of the galaxy is greater than about 10% of the mass of the group and its orbit is highly non - radial , the analytic model reproduces the mass loss in the simulations to @xmath1 accuracy at all times and for all the orbits , galaxy masses , and galaxy concentrations that we have explored . for cases where the mass of the galaxy exceeds 10% of the mass of the group , it will likely be necessary to factor in the effects of tidal stripping and gravitational shock heating , which are neglected by our model . we re - iterate that the numerical simulations with which our analytic model has been calibrated have been demonstrated to be robust to the adopted resolution and artificial viscosity strength ( see 2.2 ) . furthermore , as we have demonstrated that kh ( and rt ) instability stripping is expected to be unimportant , sph codes should be fully capable of tackling the problem of hot halo gas stripping in galaxies orbiting in groups and clusters . a direct comparison between the results using the gadget-2 and flash hydrodynamic codes for one of our runs ( see appendix a ) confirms this conclusion . the model we have derived has a number of potentially interesting applications , including modelling observed satellite galaxies and satellite galaxies in cosmological simulations . one application that we are currently pursuing is the incorporation of our ram pressure stripping model into a semi - analytic model of galaxy formation . as mentioned in 1 , recent observations ( weinmann et al . 2006 ; baldry et al . 2006 ) have revealed that current semi - analytic models predict satellite galaxies whose colours are too red compared to the observed systems . the implementation of ram pressure stripping in these models is unrealistically efficient since , by assumption , the hot halo of the satellite galaxy is instantly transferred to the more massive system as soon as the satellite galaxy enters the massive system s virial radius . in reality , the hot gaseous halo of the galaxy will remain intact for a while . for example , for the most common orbital parameters , we find that between 20%-40% of the initial hot halo of the galaxy can remain in place even after 10 gyr of orbiting inside the group or cluster ( see fig . 9 ; note , however , that the quoted numbers could be sensitive to the adopted hot gas distribution of the galaxy ) . we note that these predictions are in qualitative agreement with recent _ chandra _ x - ray observations of massive galaxies orbiting in hot clusters by sun et al . ( 2007b ) , who find that most of the galaxies have detectable hot gaseous halos . depending on the efficiency of feedback ( e.g. , from supernovae winds ) in the semi - analytic models , radiative cooling of the remaining hot halo gas will replenish the cold gaseous component at the centre of the galaxy , which in turn will allow star formation to continue for some time . this will have the effect of making the colour of model satellite galaxies bluer and could resolve the discrepancy between semi - analytic models and observations ( font et al . , in prep ) . the authors thank the referee for useful suggestions that improved the paper and they thank simone weinmann , andrew benson , volker springel and frank van den bosch for helpful discussions . igm acknowledges support from a nserc postdoctoral fellowship . csf acknowledges a royal society wolfson research merit award . asf acknowledges support from a pparc postdoctoral fellowship . rgb acknowledges support from a pparc senior fellowship . mlb acknowledges support from a nserc discovery grant . this work was supported in part by a pparc rolling grant to durham university . some of the software used in this work was in part developed by the doe - supported asc / alliance center for astrophysical thermonuclear flashes at the university of chicago . 99 abadi , m. g. , moore , b. , & bower , r. g. 1999 , , 308 , 947 acreman , d. m. , stevens , i. r. , ponman , t. j. , & sakelliou , i. 2003 , , 341 , 1333 agertz , o. , et al . 2007 , , 380 , 963 ascasibar , y. , & markevitch , m. 2006 , , 650 , 102 baldry , i. k. , balogh , m. l. , bower , r. g. , glazebrook , k. , nichol , r. c. , bamford , s. p. , & budavari , t. 2006 , , 373 , 469 balogh , m. l. , navarro , j. f. , & morris , s. l. 2000 , , 540 , 113 balogh , m. l. , baldry , i. k. , nichol , r. , miller , c. , bower , r. , & glazebrook , k. 2004 , , 615 , l101 benson , a. j. 2005 , , 358 , 551 benson , a. j. , bower , r. g. , frenk , c. s. , & white , s. d. m. 2000 , , 314 , 557 benson , a. j. , lacey , c. g. , baugh , c. m. , cole , s. , & frenk , c. s. 2002 , , 333 , 156 binney , j. , & tremaine , s. 1987 , princeton , nj , princeton university press , 1987 , 747 , p. 233 cayatte , v. , kotanyi , c. , balkowski , c. , & van gorkom , j. h. 1994 , , 107 , 1003 crowl , h. h. , kenney , j. d. p. , van gorkom , j. h. , & vollmer , b. 2005 , , 130 , 65 dekel , a. , & birnboim , y. 2007 , arxiv e - prints , 707 , arxiv:0707.1214 dolag , k. , bartelmann , m. , perrotta , f. , baccigalupi , c. , moscardini , l. , meneghetti , m. , & tormen , g. 2004 , , 416 , 853 dressler , a. 1980 , , 236 , 351 dressler , a. 1984 , , 22 , 185 eke , v. r. , navarro , j. f. , & steinmetz , m. 2001 , , 554 , 114 fryxell , b. , et al . 2000 , , 131 , 273 gisler , g. r. 1976 , , 51 , 137 goto , t. , yamauchi , c. , fujita , y. , okamura , s. , sekiguchi , m. , smail , i. , bernardi , m. , & gomez , p. l. 2003 , , 346 , 601 gmez , p. l. , et al . 2003 , , 584 , 210 gunn , j. e. , & gott , j. r. i. 1972 , , 176 , 1 hayashi , e. , navarro , j. f. , taylor , j. e. , stadel , j. , & quinn , t. 2003 , , 584 , 541 hester , j. a. 2006 , , 647 , 910 hogg , d. w. , et al . 2004 , , 601 , l29 jchym , p. , palou , j. , kppen , j. , & combes , f. 2007 , , 472 , 5 kapferer , w. , et al . 2007 , , 466 , 813 kawata , d. , & mulchaey , j. s. 2007 , arxiv e - prints , 707 , arxiv:0707.3814 king , i. 1962 , , 67 , 471 larson , r. b. , tinsley , b. m. , & caldwell , c. n. 1980 , , 237 , 692 machacek , m. , jones , c. , forman , w. r. , & nulsen , p. 2006 , , 644 , 155 mayer , l. , mastropietro , c. , wadsley , j. , stadel , j. , & moore , b. 2006 , , 369 , 1021 mccarthy , i. g. , et al . 2007 , , 376 , 497 ( m07 ) mccarthy , i. g. , babul , a. , bower , r. g. , & balogh , m. l. 2007 , arxiv e - prints , 706 , arxiv:0706.2768 mori , m. , & burkert , a. 2000 , , 538 , 559 motl , p. m. , burns , j. o. , loken , c. , norman , m. l. , & bryan , g. 2004 , , 606 , 635 murray , s. d. , & lin , d. n. c. 2004 , , 615 , 586 murray , s. d. , white , s. d. m. , blondin , j. m. , & lin , d. n. c. 1993 , , 407 , 588 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1996 , , 462 , 563 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 neto , a. f. , et al . 2007 , arxiv e - prints , 706 , arxiv:0706.2919 nittmann , j. , falle , s. a. e. g. , & gaskell , p. h. 1982 , , 201 , 833 nulsen , p. e. j. 1982 , , 198 , 1007 okamoto , t. , & nagashima , m. 2003 , , 587 , 500 poggianti , b. m. , smail , i. , dressler , a. , couch , w. j. , barger , a. j. , butcher , h. , ellis , r. s. , & oemler , a. j. 1999 , , 518 , 576 poole , g. b. , fardal , m. a. , babul , a. , mccarthy , i. g. , quinn , t. , & wadsley , j. 2006 , , 373 , 881 quilis , v. , moore , b. , & bower , r. 2000 , science , 288 , 1617 roediger , e. , & brggen , m. 2006 , , 369 , 567 roediger , e. , & brggen , m. 2007 , , 380 , 1399 roediger , e. , brggen , m. , & hoeft , m. 2006 , , 371 , 609 sakelliou , i. , acreman , d. m. , hardcastle , m. j. , merrifield , m. r. , ponman , t. j. , & stevens , i. r. 2005 , , 360 , 1069 sarazin , c. l. 1979 , , 20 , l93 sarazin , c. l. 1988 , cambridge astrophysics series , cambridge : cambridge university press , 1988 , pg . 212 schindler , s. , et al . 2005 , , 435 , l25 solanes , j. m. , manrique , a. , garca - gmez , c. , gonzlez - casado , g. , giovanelli , r. , & haynes , m. p. 2001 , , 548 , 97 springel , v. 2005 , , 364 , 1105 springel , v. , & hernquist , l. 2002 , , 333 , 649 springel , v. , et al . 2005 , , 435 , 629 sun , m. , & vikhlinin , a. 2005 , , 621 , 718 sun , m. , donahue , m. , & voit , g. m. 2007 , arxiv e - prints , 706 , arxiv:0706.1220 sun , m. , jones , c. , forman , w. , vikhlinin , a. , donahue , m. , & voit , m. 2007 , , 657 , 197 takeda , h. , nulsen , p. e. j. , & fabian , a. c. 1984 , , 208 , 261 takizawa , m. 2005 , , 629 , 791 taylor , j. e. , & babul , a. 2001 , , 559 , 716 toniazzo , t. , & schindler , s. 2001 , , 325 , 509 tormen , g. 1997 , , 290 , 411 tormen , g. , diaferio , a. , & syer , d. 1998 , , 299 , 728 vikhlinin , a. , kravtsov , a. , forman , w. , jones , c. , markevitch , m. , murray , s. s. , & van speybroeck , l. 2006 , , 640 , 691 vitvitska , m. , klypin , a. a. , kravtsov , a. v. , wechsler , r. h. , primack , j. r. , & bullock , j. s. 2002 , , 581 , 799 vollmer , b. , cayatte , v. , balkowski , c. , & duschl , w. j. 2001 , , 561 , 708 vollmer , b. , braine , j. , combes , f. , & sofue , y. 2005 , , 441 , 473 wang , h. y. , jing , y. p. , mao , s. , & kang , x. 2005 , , 364 , 424 weinmann , s. m. , van den bosch , f. c. , yang , x. , mo , h. j. , croton , d. j. , & moore , b. 2006 , , 372 , 1161 here , we compare the results obtained using the lagrangian sph code gadget-2 ( springel 2005 ) with those obtained using the eulerian amr code flash ( fryxell et al . 2000 ) for one of the uniform medium runs ( specifically , the run presented in the top panel of fig . 4 ) . the characteristics of the gadget-2 simulation are given in 2.1 and 3.2 of the main text . for flash , we have tried three different versions of the same uniform medium run . in the first version , the galaxy is moved across a periodic box filled with a static background medium ( as in the gadget-2 simulation ) and the computational volume is resolved with a fixed @xmath117 base grid . this yields a spatial resolution comparable to that of the gadget-2 run . in the second version , we take advantage of the amr capability of flash , using a base grid of @xmath118 cells and allowing up to two levels of refinement . this significantly speeds up the calculation . finally , the third version is the same as the first version except that the galaxy is placed in the centre of the box and is assigned zero bulk velocity while the uniform background medium is assigned a velocity of @xmath119 km s@xmath16 . encouragingly , we find that all three versions of the flash run yield virtually identical results . below , we compare only the results of the first version with the results of the gadget-2 simulation . for all of the gadget-2 simulations presented in the main text , the bound mass of gas is determined by calculating the centre of mass of the galaxy , computing energies in this frame , throwing out unbound particles , recomputing the centre of mass , and so on until no further particles are identified as being unbound . under this iterative scenario , once a particle is stripped it can never be re - accreted . the @xmath67 curves are necessarily monotonically decreasing in this case . unfortunately , it is not trivial to implement this type of algorithm for the flash simulation since it is not a lagrangian code . instead , it is simply assumed that all of the dark matter remains gravitationally bound ( this is a good assumption , as indicated by the dashed blue curve in fig . we then use the dark matter halo to calculate the centre of mass of the galaxy and determine which of the gas cells in the box are gravitationally bound to this dark matter halo . under this scenario , gas that was once stripped can potentially be re - accreted . therefore , a direct comparison between the default gadget-2 result and flash result should be treated with caution . fortunately , however , it is straightforward to apply the same simplified bound mass algorithm used for the flash run to the gadget-2 run and we have done this . in fig . 12 , we compare the bound mass of gas as a function of time for the gadget-2 and flash runs . the plot demonstrates that when both runs are treated on an equal footing , using the same algorithm for computing the bound mass of gas , the agreement between them is superb . here , we present a simple argument that demonstrates that tidal stripping should only be relevant for cases where the mass of the galaxy exceeds @xmath120 of the mass of the group . where @xmath45 is 3d group - centric radius of the galaxy , @xmath58 is the total mass of the galaxy within radius @xmath59 , and @xmath122 is the total mass of the group within radius @xmath45 ( e.g. , king 1962 ) .
we derive a linearized kinetic equation for fermionic excitations with an ultrasoft momentum , , from the kadanoff - baym equation in a yukawa model and quantum electrodynamics ( qed ) at extremely high , where is the coupling constant . we show that this equation is equivalent to the self - consistent equation in the resummed perturbation theory used in the analysis of the fermion spectrum with the ultrasoft momentum . furthermore , we derive the equation that determines the-point function with external lines for a pair of fermions and bosons with ultrasoft momenta in qed .
1204.6532
relativistic plasmas at extremely high temperature @xmath1 such as electron and/or quark - gluon plasma are expected to be realized in the early universe and the experiments of heavy ion collisions at relativistic heavy ion collider and large hadron collider . these systems have multi - energy scale at weak coupling ( @xmath5 ) even at so high @xmath1 that all mass scale in the lagrangian is negligible ; e.g , typical energy of particles is of order @xmath1 , while the energy of the bosonic and fermionic collective excitations called plasmon @xcite and plasmino @xcite is of order @xmath6 , and the fermion damping rate in gauge theories is of order @xmath0 @xcite . these energy scales , @xmath1 , @xmath6 , and @xmath0 , are called hard , soft , and ultrasoft scales , respectively . unlike perturbation theory in the vacuum , the loop expansion at finite temperature does not correspond to the coupling nor @xmath7 expansions , and the scale dependent expansion is necessary ( fig . [ fig : intro ] ) . when the energy scale is @xmath6 , the one - loop approximation obtained by integrating out hard internal momenta , called hard thermal loop ( htl ) approximation @xcite , is reliable , and the approximation establishes the htl effective theory @xcite . if the energy is of order or much less than @xmath0 , some resummation is necessary @xcite . here let us focus on the analysis for the fermion propagator with an ultrasoft momentum . the resummation scheme used in the analysis of the fermion propagator consist of the following procedures @xcite : 1 . resumming the thermal mass and the decay width of the hard particles . 2 . summing all the ladder diagrams ( in gauge theory ) . by analyzing the fermion propagator using the resummed perturbation theory , a novel fermionic excitation was found @xcite . each perturbation scheme with soft and ultrasoft momenta can be interpreted as different kinetic equations . when the energy is soft , the effect of collisions is negligible and hence the collisionless kinetic equation called vlasov equation is valid @xcite . this equation corresponds to the htl approximation @xcite introduced above . when the energy is of order or less than @xmath0 , the effects of the interaction among the hard particles are no longer negligible in general , so it is necessary to take into account their effects . in fact , the analysis on the gluon self - energy with an ultrasoft momentum should be performed with the resummed perturbation theory including the effect of collisions , and it was done by constructing the linearized boltzmann equation @xcite . the computation of transport coefficients , whose energy scale is much smaller than @xmath0 , is also performed with the resummed perturbation theory @xcite , the linearized boltzmann equation @xcite , and the two - particle irreducible formalism @xcite . these methods produce the same result in the leading order of the coupling constant . using the correspondence between the perturbation theory and the linearized kinetic theory , the self - consistent equation in the resummed perturbation theory is interpreted with the language of the kinetic theory . in this paper , we derive a generalized and linearized kinetic equation for fermionic excitations with an ultrasoft momentum , which we will call `` off - diagonal '' kinetic equation later , in the yukawa model and qed , while the boltzmann equation discussed above is employed for bosonic excitations . our equation is systematically derived from the kadanoff - baym equation , and is equivalent to the self - consistent equation in the resummed perturbation theory @xcite used in the analysis of the fermion propagator . the derivation helps us to establish the foundation of the resummed perturbation scheme . the kinetic equation will also give us the kinetic interpretation of the resummation scheme . furthermore , we also discuss the procedure of analyzing the higher point functions not only two - point function of the fermion in qed . this paper is organized as follows : section [ sec : yukawa ] is devoted to the derivation of the generalized and linearized kinetic equation and the discussion on the kinetic interpretation of the self - consistent equation in the resummed perturbation theory in the yukawa model , which is the simplest fermion - boson system . in sec . [ sec : qed ] , a similar analyses in qed is done in the coulomb gauge . we make concluding remarks and summarize our results in sec . [ sec : summary ] . in appendix [ app : analyticity ] , we evaluate the structure of the induced source terms . we analyze the kadanoff - baym equation that would yield the linearized kinetic equation in the diagonal case in appendix [ app : diagonal ] . appendix [ app : k0 ] is devoted to showing that the temporal component of @xmath8 introduced in sec . [ sec : qed ] can be neglected . we show that the result obtained in sec . [ sec : qed ] is valid also in the temporal gauge , which is the gauge - fixing of the original paper @xcite of the resummed perturbation theory , in appendix [ app : temporal ] . in the complex @xmath9 plane.,scaledwidth=45.0% ] in this section , we derive a novel linearized kinetic equation from the kadanoff - baym equation in the yukawa model , which is the simplest model for fermion - boson systems . we will find the vertex correction is negligible , which makes the analysis simpler than that in gauge theories . next , we show that the kinetic equation is equivalent to the self - consistent equation in the resummed perturbation theory @xcite , and discuss the interpretation of the self - consistent equation using the correspondence to the kinetic theory . throughout this paper , we work in the closed - time - path formalism @xcite . we perform the derivation of the kinetic equation in a similar way used in @xcite by applying the gradient expansion to the kadanoff - baym equation @xcite and taking into account the interaction effect among the hard particles in the leading order . let us consider the following situation to analyze the fermionic ultrasoft excitation : before the initial time @xmath10 , the system is at equilibrium with a temperature @xmath1 . then , a ( anti- ) fermionic external source @xmath11 ( @xmath12 ) and a scalar external source @xmath13 are switched on . as a result , the system becomes nonequilibrium . we will consider the case that @xmath13 and @xmath12 vanish and @xmath11 is so weak that the system is very close to the equilibrium , i.e. , the linear response regime . concretely , we will retain only the terms in the linear order of the fermionic average field @xmath14 in the fermionic induced source , which will be introduced later . let us consider the generating functional in the closed time formalism @xcite , @xmath15=\int { \cal d}{\phi}\,{\cal d}\overline{\psi}\,{\cal d}\psi \,e^{is},\ ] ] with @xmath16-(j\phi+\overline{\psi}\eta+\overline{\eta}\psi)\bigr],\ ] ] where @xmath17 and @xmath18 are the scalar and the fermion fields . the space - time integral is defined as @xmath19 , where @xmath20 is the complex - time integral along the contour @xmath21 in fig . [ fig : contour ] . we will take @xmath22 and @xmath23 to factorize out the contribution from the path @xmath24 . the lagrangian in the yukawa model with the massless fermion and boson has the form , @xmath25= \frac{1}{2}(\partial^\mu\phi)^2+\overline{\psi}i({\ooalign{\hfil/\hfil\crcr$\partial$}}+i{g}\phi)\psi,\end{aligned}\ ] ] where we did not take into account the self - interaction of @xmath17 such as @xmath26 for simplicity . by performing an infinitesimal variation with respect to @xmath17 or @xmath18 in eq . ( [ eq : yukawa - z ] ) , we obtain the following equations of motion : @xmath27\varpsi(x)&= \eta(x)+{\eta_{\mathrm { ind}}}(x ) , \\ \label{eq : yukawa - eom - meanfield - boson } -\partial^2\varphi(x)-{g}{\overline{\varpsi } } \varpsi ( x)&=j(x)+{j_{\mathrm { ind}}}(x ) , \end{aligned}\ ] ] where @xmath28 ( @xmath29 ) is the expectation value of the scalar ( fermion ) field , and @xmath30\equiv{\ooalign{\hfil/\hfil\crcr$\partial$}}_x+i{g}\varphi(x)$ ] . here the expectation value for an operator @xmath31 is defined as @xmath32 @xmath33 ( @xmath34 ) is the fermionic ( scalar ) induced source , and the subscript @xmath35 denotes `` connected , '' i.e. , @xmath36 . by differentiating eq . ( [ eq : yukawa - eom - meanfield - fermion ] ) with respect to @xmath37 and eq . ( [ eq : yukawa - eom - meanfield - boson ] ) with respect to @xmath38 , we obtain @xmath39 k(x , y)-gd(x , y)\varpsi(x)=i\frac{\delta { \eta_{\mathrm { ind}}}(x)}{\delta j(y)},\\ \notag & \partial^2_xk(y , x)-{g}(\overline{\varpsi}(x)\langle\psi(y)\psi(x)\rangle_c+s(y , x)\varpsi(x))\\ \label{eq : yukawa - fluctuation - boson } & \quad = i\frac{\delta { j_{\mathrm { ind}}}(x)}{\delta \overline{\eta}(y)}.\end{aligned}\ ] ] here we have introduced the following propagators : @xmath40 where @xmath41 means the path ordering on the complex - time path @xmath42 ; explicitly , @xmath43 with @xmath44 and @xmath45 being the step - function along the path @xmath42 . in the approximations introduced later , we can see that @xmath46 and @xmath47 coincide , which can be checked by @xmath48 \simeq0 $ ] . for this reason , we simply write these two functions as @xmath49 from now on . we call @xmath49 `` _ off - diagonal propagator _ , '' which mixes the fermion and boson , while we call @xmath50 and @xmath51 `` _ diagonal propagators_. '' as will be seen in sec . [ ssc : yukawa - correspond ] , in the calculation of the ultrasoft fermion self - energy , the off - diagonal propagator is more relevant than other diagonal ones . by setting @xmath52 and @xmath53 in eqs . ( [ eq : yukawa - fluctuation - fermion ] ) and ( [ eq : yukawa - fluctuation - boson ] ) , we obtain @xmath54 k(x , y)-{g}d^<(x , y)\varpsi(x ) = i\frac{\delta { \eta_{\mathrm { ind}}}(x)}{\delta j(y)},\\ \notag & -\partial^2_yk(x , y)+{g}(\overline{\varpsi}(y)\langle \psi(y)\psi(x)\rangle_c+s^<(x , y)\varpsi(y))\\ \label{eq : yukawa - boson1 } & \quad = i\frac{\delta { j_{\mathrm { ind}}}(y)}{\delta \overline{\eta}(x)}.\end{aligned}\ ] ] here we have interchanged @xmath55 and @xmath56 in the second equation . let us evaluate the right - hand side of eqs . ( [ eq : yukawa - fermion1 ] ) and ( [ eq : yukawa - boson1 ] ) using the chain rule : @xmath57 here we have dropped @xmath58 and @xmath59 since they contain more than one fermionic average field . we have also used @xmath60 and @xmath61 , where @xmath62 ( @xmath63 ) is the fermion ( scalar ) self - energy @xcite . we also introduced the _ off - diagonal _ self - energy , @xmath64 . the self - energies are decomposed for arbitrary @xmath9 and @xmath65 on the time path @xmath42 : @xmath66 we have not taken into account contact terms , which is negligible in the leading order as will be seen later . from eqs . ( [ eq : app - ana-1 ] ) , ( [ eq : app - ana-2 ] ) , ( [ eq : app - ana-4 ] ) , and ( [ eq : app - ana-5 ] ) in the appendix , we can rewrite eqs . ( [ eq : yukawa - fermion1 ] ) and ( [ eq : yukawa - boson1 ] ) in terms of real time integral instead of that on the complex - time - path : @xmath67 k(x , y)-{g}d^<(x , y)\varpsi(x)\\ \label{eq : yukawa - fermion3 } & \quad=\int ^\infty_{-\infty}d^4z ( \varsigma^{r}(x , z ) k(z , y)+ { \varxi } ^{r}(x , z)d^<(z , y)),\\ \notag & -\partial^2_yk(x , y)+{g}s^<(x , y)\varpsi(y)\\ \label{eq : yukawa - boson3 } & \quad=\int ^\infty_{-\infty } d^4z ( \varpi^{a}(z , y ) k(x , z)- s^<(x , z ) { \varxi } ^{r}(z , y)),\end{aligned}\ ] ] where the retarded fermion ( advanced scalar ) self - energy @xmath68 $ ] ( @xmath69 $ ] ) and the retarded off - diagonal self - energy @xmath70 $ ] have been introduced , and the limits @xmath71 and @xmath72 have been taken . here we dropped @xmath73 because @xmath74 contains more than one @xmath14 . equations ( [ eq : yukawa - fermion3 ] ) and ( [ eq : yukawa - boson3 ] ) are the kadanoff - baym equations from which the kinetic equation is derived . let us introduce the wigner transformation to derive the kinetic equation , which is defined as @xmath75 where @xmath76 , @xmath77 , and @xmath78 is an arbitrary function . after performing the wigner transformation , eqs . ( [ eq : yukawa - fermion3 ] ) and ( [ eq : yukawa - boson3 ] ) become @xmath79 here we have used the following transformation law under the wigner transformation , @xmath80 where @xmath81 is the poisson bracket , and neglected higher - order terms that contain @xmath82 since we focus on the case that the inhomogeneity of the average field is @xmath83 , while a typical magnitude of @xmath84 is of order @xmath1 . this expansion is called gradient expansion @xcite . we retained the second terms in the left - hand sides of eqs . ( [ eq : yukawa - fermion4 ] ) and ( [ eq : yukawa - boson4 ] ) because the first terms , which seem to be the leading terms in the gradient expansion , will cancel out in the next manipulation . by multiplying eq . ( [ eq : yukawa - fermion4 ] ) by @xmath85 , adding eq . ( [ eq : yukawa - boson4 ] ) , and setting @xmath86 , we get @xmath87 here we have introduced @xmath88 . we have neglected higher order terms of order @xmath89 and @xmath90 . in the leading order , the coupling dependence in @xmath91 and @xmath92 is negligible , so that @xmath91 and @xmath92 are replaced by the propagators at equilibrium and free limit ( @xmath93 ) : @xmath94 where @xmath95 is the free spectral function given by @xmath96 and @xmath97 ( @xmath98 ) is the boson ( fermion ) distribution function at equilibrium . we note that though the massless condition @xmath99 appears in eq . ( [ eq : spectralfunction ] ) in the present approximation , @xmath100 is expected to be of order @xmath101 if one takes into account the interaction at equilibrium . for this reason , we will use the order estimate @xmath102 . we also note that @xmath103 can not be replaced by that at equilibrium since @xmath103 vanishes at equilibrium . we see that @xmath100 terms in the left - hand side of eq . ( [ eq : yukawa - a ] ) were canceled out and @xmath104 term remains . thus , we can neglect the terms much smaller than @xmath105 in the calculation of the leading order . following this line , the diagonal self - energies are replaced by those at equilibrium in the leading order , whose diagrams are shown in figs . [ fig : fermion - selfenergy ] and [ fig : boson - selfenergy ] : @xmath106 where @xmath107 and @xmath108 are asymptotic thermal masses for the fermion and the scalar boson in the leading order @xcite . note that the imaginary parts of the self - energies @xmath109 and momentum dependence are negligible due to higher order of the coupling constant . we have used the on - shell condition , @xmath110 , which will be verified later . the same line as the diagonal self - energies case justifies substituting the off - diagonal self - energy in the leading order , shown in fig . [ fig : vertex - correction ] diagrammatically . the expression is given by @xmath111 here @xmath112 is the free fermion retarded propagator at equilibrium . we note that the self - energies can not be neglected unlike @xmath113 case , @xmath114 , @xmath115 . ] @xcite , because @xmath116 , @xmath114 , @xmath117 is the same order as @xmath118 . using these expressions , eq . ( [ eq : yukawa - a ] ) becomes @xmath119 where @xmath120 . we note that @xmath121 becomes finite only when @xmath99 because of @xmath122 in the right - hand side . we also note that @xmath123 , which is confirmed by multiplying eq . ( [ eq : yukawa - result-1 ] ) by @xmath124 from the left . this property makes the vertex correction term , @xmath125 , negligible , which corresponds to the fact that there is no vertex correction in the analysis using the resummed perturbation theory @xcite . thus we get @xmath126 the schematic figure of @xmath121 is depicted in fig . [ fig : yukawa - k ] . the solid ( dashed ) line with the blob stands for the resummed fermion ( boson ) propagator . in the leading order . the solid line is the fermion propagator and the dashed line is the scalar ( the case of the yukawa model ) or photon propagator ( the case of qed ) . we note that the photon propagator in the case of qed should be replaced with the htl - resummed one @xcite in this figure.,scaledwidth=25.0% ] in the leading order . the notations are the same as fig . [ fig : fermion - selfenergy].,scaledwidth=30.0% ] in the leading order . the propagator that is composed of the solid line and the dashed line with the black blob is the off - diagonal propagator . the other notations are the same as fig . [ fig : fermion - selfenergy].,scaledwidth=27.0% ] in the yukawa model in the leading order in the linear response regime . the solid ( dashed ) line with black blob is the resummed fermion ( boson ) propagator that contains the information on the fermion ( boson ) self - energy @xmath62 ( @xmath63 ) . the gray blob represents the fermionic average field @xmath14 . , scaledwidth=45.0% ] in the leading order in the yukawa model . this diagram is obtained by connecting two ends in fig . [ fig : yukawa - k ] . by truncating @xmath14 , we obtain the diagram expressing @xmath62 in the resummed one - loop analysis @xcite.,scaledwidth=30.0% ] by introducing the `` _ off - diagonal density matrix _ '' @xmath127 defined as @xmath128 , we arrive at the following generalized and linearized kinetic equation from eq . ( [ eq : yukawa - result ] ) : @xmath129 where @xmath130 and @xmath131 . we have to note that this equation is not a usual kinetic equation because @xmath132 can not be interpreted as a distribution function since it is the propagator between different particles in the fermionic background . nevertheless , we call this equation `` _ generalized kinetic equation _ '' because we can obtain the boltzmann equation if we analyze the time - evolution of the diagonal propagator instead of the off - diagonal one @xcite . in fact , eq . ( [ eq : yukawa - kineticeq ] ) has the following points similar to the boltzmann equation : * the particle is on - shell ( @xmath110 ) . * the equation has the common structure as the boltzmann equation : both have the non - interacting part , the interaction part between the hard particle and the average ultrasoft field , and the interaction part among the hard particles , which correspond to the drift term , the force term , and the collision term in the boltzmann equation , respectively . when @xmath133 , @xmath134 is negligible , and eq . ( [ eq : yukawa - kineticeq ] ) becomes the counterpart of the vlasov equation @xcite . let us recapitulate the interpretations of each term in eq . ( [ eq : yukawa - kineticeq ] ) except for the @xmath134 term . the first term in the left - hand side describes the time - evolution of @xmath135 in the free limit ( @xmath93 ) , so this term corresponds to the drift term in the boltzmann equation . on the other hand , the term in the right - hand side expresses the effect from the average fermionic field . hence this term corresponds to the force term in the boltzmann equation . @xmath134 term has no counterpart in the usual boltzmann equation , which describes the time - evolution of the diagonal propagators , @xmath136 and @xmath137 . therefore we can not obtain kinetic interpretation of that term in the usual sense . now let us discuss the origin of the @xmath134 term . the origin of the term is @xmath138 in eq . ( [ eq : yukawa - a ] ) . the real parts of those terms are of order @xmath101 while the imaginary parts are @xmath139 , so the contribution in the leading order comes from the real parts . the difference of the real parts of the diagonal self - energies expresses the difference of the dispersion relations of the scalar and the fermion , so we call the @xmath134 term `` _ mass difference term_. '' we note that the `` mass '' here is not the bare one but dynamically generated one thorough the interaction among the hard particles . here let us see the reason why the mass difference term does not have its counterpart in the diagonal case . in the case of the kinetic equation on @xmath136 , the corresponding term to the mass difference becomes @xmath140 instead of @xmath138 , so that the real part is canceled out ( see appendix [ app : diagonal ] for the detailed derivation ) . we note that this term , which is pure imaginary as a result of the cancellation of the real part , is of order @xmath141 and thus negligible since @xmath142 . we also see that the terms coming from the self - energy have complicated form in the diagonal case , while they are reduced to the simple form , @xmath143 term in eq . ( [ eq : yukawa - a ] ) , in the off - diagonal case in the linear response region . this difference comes from the following two facts : 1 . since we linearize the equation in terms of the deviation from the equilibrium state , the terms containing @xmath144 and @xmath145 in eq . ( [ eq : app - diagonal - kinetic ] ) do not have its counter parts in the off - diagonal case [ @xmath121 vanishes at equilibrium . ] 2 . the structure of the right - hand sides of eqs . ( [ eq : yukawa - fermion4 ] ) and ( [ eq : yukawa - boson4 ] ) after neglecting the vertex correction terms are simpler than those of eqs . ( [ eq : yukawa - diagonal-1 ] ) and ( [ eq : yukawa - diagonal-2 ] ) . it is because @xmath146 . here , let us show the relation between our kinetic equation , eq . ( [ eq : yukawa - result ] ) , and the self - consistent equation in the resumed perturbation theory in ref . @xcite . for this purpose , we evaluate the retarded self - energy of the fermion , @xmath147 , which can be obtained from the linear response theory @xcite in momentum space as @xmath148 on the other hand , @xmath149 can be written as @xmath150 in our theory . thus , the induced source is obtained from eq . ( [ eq : yukawa - result ] ) as @xmath151 by performing the fourier transformation , @xmath152 we obtain @xmath153 comparing eq . ( [ eq : yukawa - fermion - induced ] ) with eq . ( [ eq : yukawa - fermion - induced2 ] ) , we obtain the self - energy , @xmath154 this expression coincides with that of the retarded fermion self - energy in ref . @xcite except for the absence of the decay widths of the hard particles in the denominator . as mentioned in the previous subsection , the decay widths of order @xmath141 is neglected when the external momentum is of order @xmath155 ; one can include them by taken into account the imaginary part of eq . ( [ eq : yukawa - result ] ) if one is interested in the decay width . the diagrammatic representation of the fermion retarded self - energy in our formalism is the same as that in the resummed perturbation theory @xcite , which is explained as follows : the off - diagonal density matrix @xmath135 , which follows the generalized kinetic equation , is represented by fig . [ fig : yukawa - k ] . @xmath156 , shown in fig . [ fig : yukawa - induced - source ] , is diagrammatically obtained by connecting the ends of fermion and boson propagators in the right - hand side of fig . [ fig : yukawa - k ] . this diagram is the resummed one - loop diagram appeared in @xcite itself except for the fermion average field @xmath14 . first we introduce the background field method , which is useful to construct the equations for the average fields and the kadanoff - baym equation in a gauge - covariant form . next , we derive the generalized kinetic equation in the linear response regime adopting the coulomb gauge fixing , in which the transversality of the free photon propagator simplifies the analysis . after the derivation , we show the equivalence between the generalized kinetic equation and the resummed perturbation theory @xcite , and discuss the interpretation of the terms in the kinetic equation . we also check that the ward - takahashi identity for the self - energy , consequence of the @xmath157 gauge symmetry , is satisfied in our formalism . finally we discuss how to compute the higher - point - vertex function whose external momenta are all ultrasoft , and make an order estimate of it at the weak coupling regime . in the derivation of the average field equation and the kadanoff - baym equation in qed , it is convenient to formulate them in a covariant form under gauge transformations . for this purpose , we employ the background field gauge method @xcite . in this method , the following generating functional is employed : @xmath158=\int [ { \cal d}{a}][{\cal d}\overline{\psi}][{\cal d}\psi]e^ { is } , \label{eq : generatingfunctionalqed } \end{split}\end{aligned}\ ] ] with @xmath159+{\cal { l}}_{\mathrm { gf}}\\ & \quad-(j^\mu a_\mu+\overline{\psi}\eta+\overline{\eta}\psi)\bigr ] , \end{split}\ ] ] where we dropped the ghost term , which is not coupled with the other fields . @xmath160 and @xmath161 are vector fields , @xmath14 ( @xmath162 ) and @xmath18 ( @xmath163 ) are ( anti- ) spinor fields , and @xmath164 is the external current , respectively . the lagrangian of qed has the form , @xmath165=-\frac{1}{4}f^{\mu\nu}[a]f_{\mu\nu}[a ] + i\overline{\psi}{\ooalign{\hfil/\hfil\crcr$d$}}[a]\psi,\end{aligned}\ ] ] where @xmath166\equiv \partial_\mu a_\nu-\partial_\nu a_\mu $ ] is the field strength and @xmath167\equiv \partial_\mu + i{g}a_\mu$ ] is the covariant derivative . we have used @xmath168 as a coupling constant of qed instead of the standard notation @xmath169 to make it clear that the same order counting appears as that in the yukawa model . in the background field method , the fields in the lagrangian are decomposed to the classical field , identified as the average fields later , and fluctuations in eq . ( [ eq : generatingfunctionalqed ] ) . the external sources are chosen to be coupled to @xmath161 , @xmath18 , and @xmath163 , but not to @xmath160 , @xmath14 , and @xmath162 . we impose the following conditions : @xmath170 which implies that @xmath171 and @xmath14 ( @xmath162 ) can be interpreted as the average parts of the photon and ( anti ) electron field , respectively , and @xmath172 coincides with the effective action @xcite . in the background gauge field method , the gauge - fixing term is chosen to be a functional of @xmath173 such as @xmath174)^2}{2},\end{aligned}\ ] ] where @xmath175 $ ] is the gauge - fixing function and @xmath176 is the gauge - fixing parameter . although @xmath173 is fixed by the gauge fixing term , the generating functional , eq . ( [ eq : generatingfunctionalqed ] ) , is invariant under the background field gauge transformations defined by @xmath177 where @xmath178 $ ] . since the fluctuations covariantly transform under eq . ( [ eq : qed - gauge - trans ] ) , the propagators also covariantly transform as @xmath179 also , the wigner transformed off - diagonal propagator in the leading order of @xmath168 is covariant , which can be confirmed by performing the gradient expansion @xcite : @xmath180 which implies that the kadanoff - baym equation covariantly transforms with respect to the background gauge transformations as will be seen in the next subsection . we note that , apart from the covariance with respect to the background field gauge transformation , the gauge - fixing dependence , which will be confirmed only in the coulomb gauge and temporal gauge in this paper , should be analyzed . we work in the coulomb gauge - fixing condition because this gauge fixing makes the analysis simple owing to the transversality of the free photon propagator . the gauge - fixing condition is @xmath175=\partial_i a^i$ ] and @xmath181 , which constrain the off - diagonal propagator as @xmath182 the equations of motion for the average fields are given by @xmath183\varpsi(x)&=\eta(x)+{\eta_{\mathrm { ind}}}(x),\\ \notag \partial^2a^\mu(x)-\partial^\mu\partial^\nu a_\nu(x)&-{g}\overline{\varpsi}(x)\gamma^\mu\varpsi(x)\\ \label{eq : qed - meanfield - boson } & = j^\mu(x)+{j_{\mathrm { ind}}}^\mu(x).\end{aligned}\ ] ] here we have used eq . ( [ eq : qed - condition ] ) , and the induced fermionic source and the induced current are defined as @xmath184 which transform @xmath185 and @xmath186 under the background gauge transformations . therefore , eqs . ( [ eq : qed - meanfield - fermion ] ) and ( [ eq : qed - meanfield - boson ] ) transform covariantly with respect to the background gauge transformation . the equations for the propagators are given by @xmath187 where @xmath188 and @xmath189 are the photon and the off - diagonal self - energies , respectively . here we set @xmath52 and @xmath53 . the wigner transformed equations read @xmath190 where eq . ( [ eq : qed - coulomb - condition ] ) , the estimation @xmath191 , shown in appendix [ app : k0 ] , and the gradient expansion have been used . multiplying eq . ( [ eq : qed - wigner - fermion ] ) by @xmath192 $ ] and eq . ( [ eq : qed - coulomb - wigner - boson ] ) by @xmath193 defined below , and subtracting the latter from the former , we obtain @xmath194 where we have introduced the projection operator into the transverse component , @xmath195 , and @xmath196 , and used @xmath197 . one can show that the background fields and the coupling dependences in the diagonal propagators are weak and @xmath198 in the diagonal kinetic equation : @xmath199 from eq . ( [ eq : app - diagonal - kinetic - qed ] ) . similarly , @xmath200 . these order estimates imply that @xmath160 contribution from @xmath201 and @xmath202 in the right - hand side of eq . ( [ eq : qed - coulomb - k ] ) is much smaller than that from the left - hand side of that equation , by using the order estimate @xmath203 . ] , so that we can replace the electron and the photon propagator , which is given as follows , by that in the free limit at equilibrium : @xmath204 the diagonal self - energies at on - shell in the leading order are given by @xmath205 where @xmath206 and @xmath207 are the asymptotic thermal masses of the electron and the photon in the leading order . @xmath208 is the damping rate of the hard electron @xcite , which is much larger than the case of the yukawa model . unlike in the case of yukawa model , the damping rate of the electron @xmath209 can not be neglected because @xmath210 , while the photon damping rate of order @xmath211 can be neglected @xcite . we note that the longitudinal part of the photon self - energy does not contribute because the projection operator @xmath212 is multiplied . the off - diagonal self - energy in the leading order has the following expression , which has a similar form to that in the yukawa model : @xmath213 the off - diagonal self - energy in @xmath214 has to be retained in the case of qed unlike the yukawa model since there is no special suppression mechanism . by substituting these expressions in eq . ( [ eq : qed - coulomb - k ] ) , we obtain @xmath215 here we have used @xmath216 and introduced @xmath217 . in this gauge - fixing condition , it is apparent that only the transverse component of the thermal photon contributes to @xmath218 because of the projection operator @xmath219 appearing in the right - hand side of eq . ( [ eq : qed - coulomb - result - k ] ) . we note that this equation transforms covariantly with respect to the background gauge transformation from eq . ( [ eq : qed - k - transform ] ) . from eq . ( [ eq : qed - coulomb - result - k ] ) , we write @xmath218 in terms of the off - diagonal self - energy for later use : @xmath220 the diagrammatic representation of this equation is shown in fig . [ fig : qed - k ] . next , we derive the linearized kinetic equation . multiplying eq . ( [ eq : qed - coulomb - result - k ] ) by @xmath221 from the left , we obtain @xmath222 we decompose @xmath223 into positive and negative energy terms as @xmath224 $ ] , so that we arrive at the kinetic equation from eq . ( [ eq : qed - coulomb - result - kslash ] ) : @xmath225\varpsi(x)\\ & \qquad-{g}^2\gamma_{i}{\ooalign{\hfil/\hfil\crcr$v$}}[{n_f}(|{{\mathbf{k}}}|)+{n_b}(|{{\mathbf{k}}}|)]p^{\nu i } _ { \mathrm t}(v)\\ & \qquad\times\sum_{s=\pm}\int\frac{d^3{{\mathbf{k}}}'}{(2\pi)^3}\frac{1}{2|{{\mathbf{k}}}'|}\frac{s|{{\mathbf{k}}}|v^\alpha\gamma_\nu\pm |{{\mathbf{k}}}'|v _ { \nu}'\gamma^\alpha}{|{{\mathbf{k}}}| |{{\mathbf{k}}}'|v\cdot v'}\varlambda_{s \alpha}({{\mathbf{k}}}',x ) , \end{split } \end{aligned}\ ] ] where we have introduced @xmath226 . there are two terms that do not appear in the yukawa model analyzed in sec . [ sec : yukawa ] . one is the last term in the right - hand side . because the bare vertex term , which contains @xmath227 and is in the right - hand side , is interpreted as the counterpart of the force term in the diagonal case @xcite , that vertex correction term acts like `` _ the correction to the force term _ , '' at least in the linear response regime . note that this term mixes the positive and negative energy modes , unlike the case of the yukawa model . the other is the third term in the left - hand side . this term has a similar form to the collision term in the relaxation time approximation of the diagonal case , i.e. , pure imaginary constant ( @xmath228 ) times @xmath135 . for this reason , we call this term `` _ collision term_. '' we note that this term is negligible in the case of the yukawa model as shown in the previous section . in the diagonal case @xcite , the collision term contains momentum integral for the diagonal density matrix in eq . ( [ eq : app - diagonal - kinetic - qed ] ) ] . in contrast , in the off - diagonal case , such term does not survive in the linearized equation . as a result , the collision term in the off - diagonal kinetic equation has a similar form to that in the relaxation time approximation . we emphasize that the off - diagonal self - energy is not negligible in the off - diagonal kinetic equation , while negligible in the diagonal one because we neglect the higher order terms in @xmath14 . this fact makes the correction to the force term , which is absent in the diagonal case , appears in eq . ( [ eq : qed - kineticeq ] ) . as we discussed in sec . [ ssc : yukawa - kinetic ] , both of the usual boltzmann equation and our generalized and linearized kinetic equation are composed of the non - interacting part , the interaction part between the hard particle and the average ultrasoft field , and the interaction part among the hard particles . which part is the counterpart of the mass difference term , the collision term , and the correction to the force term ? because the mass difference and the collision term come from the self - energies at equilibrium , they correspond to the interaction part among the hard particle . the correction to the force term is a part of the interaction part between the hard particle and the average ultrasoft field . .the correspondence between the resummed perturbation theory and the generalized and linearized kinetic equation . [ cols="^,^",options="header " , ] [ tab : ] here , let us show the equivalence between eq . ( [ eq : qed - coulomb - result - k ] ) and the self - consistent equation in the resummed perturbation theory @xcite . to this end , we rewrite eq . ( [ eq : qed - vertex - correction ] ) in terms of the off - diagonal self - energy using eq . ( [ eq : qed - coulomb - k - vertex ] ) : @xmath229}{2ik'\cdot \partial_x + 2i{\zeta_e}k'^{0}+\delta m^2}\\ & \quad \times{\tilde{\vargamma}}^\alpha(k',x ) . \end{split}\end{aligned}\ ] ] here we set @xmath230 . by performing the fourier transformation , eq . ( [ eq : fourier - p ] ) , we get @xmath231\vargamma^\alpha(k',p)}{2k'\cdot p + 2i{\zeta_e}k'^{0}+\delta m^2 } , \end{split}\end{aligned}\ ] ] where @xmath232 . we note that @xmath233 is the vertex function introduced in @xcite whose momenta are hard and ultrasoft . equation ( [ eq : qed - bs ] ) is none other than the integral equation appearing in @xcite . the retarded fermion self - energy is also written in terms of the vertex function : from eq . ( [ eq : qed - coulomb - k - vertex ] ) , we arrive at @xmath234}{2k\cdot p + 2i{\zeta_e}k^0 + \delta m^2}\\ & \quad\times \vargamma^\nu(k , p ) . \end{split}\end{aligned}\ ] ] this expression equals to that of the fermion retarded self - energy in the resummed perturbation theory ) is analytically solved for the energy region @xmath235 @xcite . ] thus we see that eq . ( [ eq : qed - coulomb - result - k ] ) , derived in the non - equilibrium state in a linear response regime from the kadanoff - baym equation , is equivalent to the self - consistent equation in the resummed perturbation theory , which is constructed in the thermal equilibrium state . here we discuss the correspondence between each scheme in the resummed perturbation theory @xcite and each term in the kinetic equation . as in the yukawa model , the resummation of the thermal mass difference in the resummed perturbation theory corresponds to the mass difference term in the kinetic equation . the decay width corresponds to the collision term since both of them contains the damping rate of the hard electron , @xmath209 . the ladder summation in the resummed perturbation theory @xcite is caused by the correction to the force term in the kinetic equation . thus , the ladder summation corresponds to the correction to the force term . these interpretations are summarized in the table . [ tab : ] . the fermionic induced source @xmath149 generates the higher point function , not only the fermion self - energy . in this subsection , we derive the self - consistent equation determining the @xmath3-point function whose external lines consist of two fermions ( @xmath14 ) and @xmath4 bosons ( @xmath160 ) with ultrasoft external momenta , and make an order estimate of the quantity . for example in the case of @xmath236 , we obtain the correction to the bare three - point function , @xmath237 , from @xmath149 @xcite : @xmath238 here we have expanded @xmath239 around @xmath230 : @xmath240 where @xmath241 contains one @xmath14 while @xmath242 contains one @xmath14 and one @xmath160 . @xmath242 can be obtained by the following way . collecting terms that contain one @xmath160 in eq . ( [ eq : qed - coulomb - result - k ] ) , we obtain @xmath243\\ & \qquad\times \int\frac{d^4 k'}{(2\pi)^4}\frac{k^\alpha \gamma_\nu+\gamma^\alpha k'_\nu}{k\cdot k ' } \delta k_\alpha(k',x ) . \label{eq : deltaki } \end{split}\end{aligned}\ ] ] since @xmath241 is determined by setting @xmath230 in eq . ( [ eq : qed - coulomb - result - k ] ) , this equation is closed and @xmath242 can be determined . let us estimate the order of @xmath242 . from eq . ( [ eq : qed - coulomb - result - k ] ) and @xmath244 , we find @xmath245 . therefore , the vertex correction is estimated as @xmath246 , which is much larger than the bare vertex , @xmath247 . similar order estimate for the @xmath3-point function with @xmath248 can be done with the same procedure ; as a result , we find that the order of the @xmath3-point - vertex function is @xmath249 . let us see that the off - diagonal self - energy given in eq . ( [ eq : qed - vertex - correction ] ) satisfies the ward - takahashi ( wt ) identity . from eq . ( [ eq : qed - vertex - correction ] ) , we get @xmath250 here we have used @xmath251 , which can be confirmed by multiplying eq . ( [ eq : qed - coulomb - result - k ] ) by @xmath124 from the left . this equation generates the wt identity derived in ref . @xcite by setting @xmath230 . the wt identity implies that the vertex correction is not negligible because the identity relates the vertex correction to the fermion self - energy , which is much larger than the inverse of the fermion propagator with an ultrasoft momentum . in the yukawa model , the wt identity associated with gauge symmetries is absent , so the smallness of the vertex correction is not in contradiction with any identity . equation ( [ eq : qed - wt ] ) can be derived from the conservation law of the induced current , @xmath252 by differentiating eq . ( [ eq : conservationlaw ] ) with respect to @xmath253 , we obtain @xmath254 here @xmath255 is the delta function defined along the contour @xmath42 . by multiplying this equation by @xmath256 and taking only the leading - order terms , we find @xmath257 this is nothing but eq . ( [ eq : qed - wt ] ) . the detail of the derivation of eq . ( [ eq : qed - wt2 ] ) is shown in the appendix [ app : analyticity ] . we derived the novel generalized and linearized kinetic equation with an ultrasoft momentum from the kadanoff - baym equation in the yukawa model and qed . our kinetic equation is equivalent to the self - consistent equation in the resummed perturbation theory @xcite used in the analysis of the fermion propagator . this derivation helps us to establish the foundation of the resummed perturbation theory . we gave the kinetic interpretation of the resummation scheme ( table . [ tab : ] ) . furthermore , we made an order estimate of the higher - point function with ultrasoft external momenta , and obtained the equation determining the vertex correction in qed . in our analysis , the difference of the masses and the vertex correction , which reflects the fact that we analyze a process that changes the type of particle , plays an important role . this fact suggests that these quantities may play an important role in the analysis of other ultrasoft quantity such as the flavor diffusion constant in qcd , where the masses of @xmath258 , @xmath259 and @xmath260 are different . it is natural to ask whether such kinetic equation is derived in qcd because the investigation of the qcd at high temperature is quite interesting since it is relevant to analysis of quark - gluon plasma @xcite , which is realized in heavy ion collision experiments . we hope to report the analysis in qcd elsewhere @xcite . this work was supported by the grant - in - aid for the global coe program `` the next generation of physics , spun from universality and emergence '' from the ministry of education , culture , sports , science and technology ( mext ) of japan . . is supported by a grant - in - aid for scientific research ( no.23340067 ) from mext . in this appendix , we derive the right - hand sides of eqs . ( [ eq : yukawa - fermion3 ] ) , ( [ eq : yukawa - boson3 ] ) , and ( [ eq : qed - wt2 ] ) . first let us derive the right - hand sides of eqs . ( [ eq : yukawa - fermion3 ] ) and ( [ eq : yukawa - boson3 ] ) : @xmath261 in the last line we have taken @xmath262 . we used the fact that the term integrated on @xmath24 becomes negligible in this limit @xcite . in the same way , we get @xmath263 next , we evaluate the off - diagonal self - energy term . the vertex correction term of eq . ( [ eq : yukawa - fermion2 ] ) becomes @xmath264 where the advanced boson propagator is @xmath265 $ ] . here we stop and discuss the structure of the off - diagonal self - energy in the leading order in the @xmath266 space . the off - diagonal self - energy in the leading order and linear response regime is given by @xmath267 where @xmath268 is the free fermion propagator at equilibrium . thus , the components of @xmath269 are given by @xmath270 by performing the wigner transformation , we get @xmath271 with @xmath272 . we see that since @xmath273 contains @xmath274 and we focus on the on - shell case @xmath110 , which are confirmed from the analysis in sec . [ ssc : yukawa - derivation ] , @xmath275 . for this reason , @xmath276 , which implies @xmath277 , so the only nonzero function of the off - diagonal self - energy appearing at wigner - transformed eq . ( [ eq : app - ana-3 ] ) is @xmath278 . therefore , we drop the second term in eq . ( [ eq : app - ana-3 ] ) because that term becomes negligible after the wigner transformation , and hence the equation becomes @xmath279 in the same way , we get @xmath280 the calculation in qed can be performed in the same way . finally we derive eq . ( [ eq : qed - wt2 ] ) . by multiplying eq . ( [ eq : qed - wt - pre ] ) by @xmath256 , we get @xmath281 here we have set @xmath9 , @xmath282 and neglected the sub - leading terms . the first term in the left - hand side has the same order of magnitude as the hard fermion self - energy @xmath283 times @xmath227 , so that term is negligible . thus the left - hand side becomes @xmath156 . the right - hand side becomes @xmath284 we see that the first term in the last line is negligible because of the on - shell condition . thus we obtain eq . ( [ eq : qed - wt2 ] ) if we remember that @xmath278 . we derive the equation that corresponds to eq . ( [ eq : yukawa - result ] ) in the diagonal case in this appendix . the equation governing the propagator of the fermion is as follows : @xmath285 s(x , y)+i{g}(k(y , x))^\dagger\gamma^0\varpsi(x)\\ & \quad=\delta^{c(4)}(x - y)+i\frac{\delta { \eta_{\mathrm { ind}}}(x)}{\delta \eta(y)}. \end{split}\end{aligned}\ ] ] since the second term in the left - hand side contains two @xmath14 , we neglect that term . let us calculate the right - hand side . we set @xmath52 and @xmath53 . since the vertex correction term , which contains more than one @xmath14 , is negligible , we obtain @xmath286 by taking the limit @xmath262 , we get @xmath287 here we have introduced the advanced fermion propagator , @xmath288 . by performing the wigner transformation , we get @xmath289 s^<(k , x ) \\ & \quad =- i(\varsigma^{r}(k , x)s^<(k , x)+\varsigma^<(k , x)s^{a}(k , x ) ) . \end{split}\end{aligned}\ ] ] the following equation is derived from the conjugate of eq . ( [ eq : yukawa - diagonal-1 ] ) by using @xmath290 : @xmath291\\ & \quad = i(s^{r}(k , x)\varsigma^<(k , x)+s^<(k , x)\varsigma^{a}(k , x ) ) . \end{split}\end{aligned}\ ] ] here we have introduced the retarded fermion propagator , @xmath292 , and the advanced fermion self - energy , @xmath293 , which are defined as follows : @xmath294 by multiplying eq . ( [ eq : yukawa - diagonal-1 ] ) ( eq . ( [ eq : yukawa - diagonal-2 ] ) ) by @xmath295 @xmath296 from the left ( right ) , we get @xmath297 by subtracting eq . ( [ eq : yukawa - diagonal-3 ] ) from eq . ( [ eq : yukawa - diagonal-4 ] ) , we get @xmath298 \\ & \qquad + i{g}(\partial^\nu_x\varphi(x))\{{\ooalign{\hfil/\hfil\crcr$k$}},\partial_{k\nu}s^<(k , x)\}\\ & \qquad - i{g}\varphi(x)\{\gamma_\mu , \partial^\mu_x s^<(k , x)\}\\ & \quad={\ooalign{\hfil/\hfil\crcr$k$}}\varsigma^<(k , x)s^{a}(k , x)-s^{r}(k , x)\varsigma^<(k , x){\ooalign{\hfil/\hfil\crcr$k$ } } . \end{split}\end{aligned}\ ] ] here we linearize this equation . by introducing @xmath299 and @xmath300 , we arrive at the following equation : @xmath301 similarly , we obtain the following equation in qed : @xmath302\\ & \quad={\ooalign{\hfil/\hfil\crcr$k$}}\delta \varsigma^<(k , x)s^{0 { a}}(k)-s^{0 { r}}(k)\delta\varsigma^<(k , x){\ooalign{\hfil/\hfil\crcr$k$ } } \\ & \qquad+{\ooalign{\hfil/\hfil\crcr$k$}}\varsigma^{{\text { ( eq ) } } < } ( k)\delta s^{a}(k , x)-\delta s^{r}(k , x)\varsigma^{{\text { ( eq ) } } < } ( k){\ooalign{\hfil/\hfil\crcr$k$ } } . \end{split}\end{aligned}\ ] ] here we show that @xmath303 is negligible compared with the spatial components in the coulomb gauge . for simplicity , we set @xmath230 . we get the following equation by multiplying eq . ( [ eq : qed - wigner - fermion ] ) by @xmath304 from the left , subtracting eq . ( [ eq : qed - coulomb - wigner - boson ] ) from the quantity obtained above , and setting @xmath305 : @xmath306 since @xmath307 , we see that @xmath308 we show that the equation determining @xmath309 in the temporal gauge is the same as that in the coulomb gauge , eq . ( [ eq : qed - coulomb - result - k ] ) , in this appendix because the resummed perturbation theory was first proposed in the temporal gauge @xcite . the gauge - fixing condition in the temporal gauge is @xmath175=a^0 $ ] and @xmath181 . this condition is equivalent to the constraint @xmath310 . because of this constraint , we have @xmath311 the equations governing @xmath312 are @xmath313 from these equations , we obtain @xmath314 here let us evaluate @xmath315 , which is the longitudinal component of @xmath218 . by multiplying eq . ( [ eq : qed - temporal - k ] ) by @xmath316 , we get @xmath317 here we have neglected the terms that are of order @xmath318 . we see that @xmath319 and thus the longitudinal component of @xmath320 , @xmath321 , is negligible compared with the transverse component of @xmath320 . we note that @xmath322 , which is the result of the gauge - fixing condition , and @xmath323 are valid also in the coulomb gauge in the leading order . furthermore , also the free photon propagator at equilibrium is the same as that in the coulomb gauge . thus , we can obtain eq . ( [ eq : qed - coulomb - result - k ] ) in the same way as in the sec . [ ssc : derivation ] . r. d. pisarski , phys . lett . * 63 * , 1129 ( 1989 ) ; j. p. blaizot and e. iancu , _ ibid_. * 76 * , 3080 ( 1996 ) ; phys . rev . d * 55 * , 973 ( 1997 ) ; see also v. v. lebedev and a. v. smilga , phys . b * 253 * , 231 ( 1991 ) ; physica a * 181 * , 187 ( 1992 ) . j. p. blaizot and e. iancu , nucl . b * 557 * , 183 ( 1999 ) ; * 570 * , 326 ( 2000 ) ; see also d. bodeker , phys . b * 426 * , 351 ( 1998 ) ; nucl . b * 559 * , 502 ( 1999 ) ; * 566 * , 402 ( 2000 ) ; phys . b * 516 * , 175 ( 2001 ) ; nucl . b * 647 * , 512 ( 2002 ) . v. v. lebedev and a. v. smilga , nucl . b * 318 * , 669 ( 1989 ) . s. jeon , phys . d * 52 * , 3591 ( 1995 ) ; e. wang and u. w. heinz , phys . b * 471 * , 208 ( 1999 ) ; m. e. carrington , d. f. hou and r. kobes , phys . d * 62 * , 025010 ( 2000 ) ; m. a. valle basagoiti , _ ibid_. * 66 * , 045005 ( 2002 ) ; e. wang and u. w. heinz , _ ibid_. * 67 * , 025022 ( 2003 ) ; j. s. gagnon and s. jeon , _ ibid_. * 75 * , 025014 ( 2007 ) [ erratum - ibid . d * 76 * , 089902 ( 2007 ) ] ; * 76 * , 105019 ( 2007 ) . b. s. de witt , phys . rev . * 162 * , 1195 ( 1967 ) , 1239 ( 1967 ) ; phys . rep . * 19c * , 295 ( 1975 ) ; l. f. abbott , nucl . b * 185 * , 189 ( 1981 ) ; k. a. meissner , acta phys . b * 17 * ( 1986 ) 409 ; t. h. hansson and i. zahed , phys . lett . * 58 * , 2397 ( 1987 ) ; nucl . b * 292 * , 725 ( 1987 ) ; as a review article , see l. f. abbott , acta phys . b * 13 * , 33 ( 1982 ) . j. i. kapusta , p. lichard and d. seibert , phys . d * 44 * , 2774 ( 1991 ) [ erratum - ibid . d * 47 * , 4171 ( 1993 ) ] ; r. baier , h. nakkagawa , a. niegawa and k. redlich , z. phys . c * 53 * , 433 ( 1992 ) ; p. aurenche , f. gelis and h. zaraket , phys . d * 61 * , 116001 ( 2000 ) ; phys . d * 62 * , 096012 ( 2000 ) ; p. b. arnold , g. d. moore and l. g. yaffe , jhep * 0111 * , 057 ( 2001 ) ; jhep * 0112 * , 009 ( 2001 ) ; jhep * 0206 * , 030 ( 2002 ) .
monte carlo results for the low - lying glueball spectrum using an improved , anisotropic action are presented . ten simulations at lattice spacings ranging from 0.2 to 0.4 fm and two different anisotropies have been performed in order demonstrate the advantages of using coarse , anisotropic lattices to calculate glueball masses . our determinations of the tensor ( ) and pseudovector ( ) glueball masses are more accurate than previous wilson action calculations . epsf
hep-lat9704011
numerical simulations of gluons on a space - time lattice provide at present the most reliable means of studying glueballs . glueball correlation functions are , however , notoriously difficult quantities to measure in monte carlo simulations : since the masses of these states are rather high and their creation operators have large vacuum fluctuations , the signal - to - noise ratio falls extremely rapidly as the separation between the source and sink is increased . because of this , reliable studies of glueballs on fine lattices have required prohibitively large computer resources . thus , the development of more efficient simulation techniques in lattice qcd is crucial to establishing a detailed description of glueballs and their interactions . the objective of this work is to examine the effectiveness of using an improved , anisotropic lattice action to reduce the computational effort needed to determine the glueball spectrum in quenched qcd . improved actions allow access to continuum physics on coarser lattices than possible using the simple wilson discretization . coarse lattice simulations are more efficient for several reasons : for a given physical volume , much fewer lattice sites are needed ; the alleviation of critical slowing down permits the faster generation of statistically independent gauge - field configurations ; glueball operator smearing is faster due to the decreased number of links and a decrease in the number of smearing iterations required ; glueball wavefunctions extend over much fewer lattice sites on a coarse lattice , making the variational technique far more effective when using a feasible number ( a dozen or so ) of basis operators . however , for glueball mass calculations , the coarseness of the temporal lattice spacing is a severe drawback . as the masses in lattice units of the states of interest are so large , the number of correlator time intervals which can be measured is reduced greatly @xcite . a straightforward solution to this problem which preserves the computational advantages of coarse lattices@xcite is to make use of anisotropic lattices in which the temporal spacing is much smaller than that in the spatial directions . this enables us to exploit the enhanced signal - to - noise of the correlation functions at smaller temporal separations . a natural scale for the temporal lattice spacing should be the inverse of the energy of the states of interest ; thus , for glueballs , a temporal cut - off larger than 1.5 gev allows resolution from accessible statistics of the correlator over a few time - slices . meanwhile , the scale for the spatial lattice should be set by the size of the wavefunction of the state ; a spatial grid separation in the range @xmath2 fm would seem reasonable . since we propose to use lattices in which the temporal lattice spacing is small , improvement of the discretization in this direction is not needed . thus , a lattice action which couples only nearest - neighbor time - slices can be used . the transfer matrix corresponding to such an action is hermitian and positive definite ; all of our effective masses must converge to their plateau values monotonically from above . this ensures the validity of variational techniques which minimize the effective masses at small temporal separations . such techniques are very effective in diminishing the excited - state contributions to the glueball correlation functions and are crucial for efficient extraction of ground - state masses . in this paper , we demonstrate the increased efficiency of glueball simulations using these actions on anisotropic lattices . we present results for the masses of three of the lighter su(3 ) glueball states , the scalar ( @xmath3 ) , the tensor ( @xmath0 ) , and the pseudovector ( @xmath1 ) . the masses of the first excited states in the scalar and tensor channels were also examined . ten simulations at lattice spacings ranging from 0.2 to 0.4 fm were performed , enabling reliable extrapolations to the continuum limit ( although the mass of the scalar glueball was somewhat problematic ) . the results are compared to previous simulation data obtained using the wilson action and we find that more accurate determinations of the tensor and pseudovector glueball masses have been achieved . a comparison of efficiencies is also made . lastly , finite volume effects are shown to be small . the new action used in our simulations is described in sec . [ sec : action ] . the details of the glueball simulations , including the construction of the glueball operators , the generation of the gauge - field configurations , and the analysis of the monte carlo data , are given in sec . [ sec : simulation ] . the hadronic scale @xmath4 is used to relate our results at different values of the coupling @xmath5 and the aspect ratio @xmath6 . the determination of this scale in terms of the lattice spacing using the static potential is outlined in sec . [ sec : r0as ] . [ sec : resdiscuss ] contains our results and discussion : the glueball mass measurements are presented in detail ; finite volume effects are studied ; extrapolations of the masses at finite spacing to the continuum limit are undertaken ; the conversion of our results into physical units is described ; and a comparison of efficiencies with wilson action simulations is made . our conclusions are given in sec . [ sec : conclusions ] , along with an outline of future work . our glueball mass determinations rely on numerical simulations of glueballs on a euclidean space - time lattice with spatial and temporal spacings @xmath7 and @xmath8 , respectively . the improved gluonic action used in this study is given by @xcite @xmath9 where @xmath10 , @xmath11 is the qcd coupling , @xmath12 and @xmath13 are mean link renormalization parameters , @xmath6 is the aspect ratio ( @xmath14 at tree level in perturbation theory ) , and @xmath15 , with @xmath16 denoting the path - ordered product of link variables along a closed contour @xmath17 on the lattice . @xmath18 includes the sum over all spatial plaquettes on the lattice , @xmath19 indicates the temporal plaquettes , @xmath20 denotes the product of link variables about planar @xmath21 spatial rectangular loops , and @xmath22 refers to the short temporal rectangles ( one temporal link , two spatial ) . explicitly , @xmath23,\\ \omega_{\rm tp } & = & \sum_x\sum_{i } \textstyle{\frac{1}{3}}{\rm re tr } \left[1-u_t(x)u_i(x\!+\!\hat t)u^\dagger_t(x\!+\!\hat\imath)u^\dagger_i(x ) \right],\\ \omega_{\rm sr } & = & \sum_x\sum_{i\neq j } \textstyle{\frac{1}{3}}{\rm re tr } \left[1-u_i(x)u_i(x\!+\!\hat\imath)u_j(x\!+\!2\hat\imath ) u^\dagger_i(x\!+\!\hat\imath\!+\!\hat\jmath ) u^\dagger_i(x\!+\!\hat\jmath ) u^\dagger_j(x)\right],\\ \omega_{\rm str } & = & \sum_x\sum_{i } \textstyle{\frac{1}{3}}{\rm re tr } \left[1-u_i(x)u_i(x\!+\!\hat\imath)u_t(x\!+\!2\hat\imath ) u^\dagger_i(x\!+\!\hat\imath\!+\!\hat t ) u^\dagger_i(x\!+\!\hat t ) u^\dagger_t(x)\right],\end{aligned}\ ] ] where @xmath24 labels the sites of the lattice , @xmath25 are spatial indices , and @xmath26 is the parallel transport matrix in the gluon field from site @xmath24 to @xmath27 . this action , intended for use with @xmath28 , has @xmath29 discretization errors . the @xmath30 errors can be removed by the addition of counterterms which couple next - nearest - neighbor time - slices , but this introduces spurious high - energy modes which can cause considerable problems for our glueball simulations . these unphysical states appear in perturbation theory as additional poles in the gluon propagator . their detrimental effects on the glueball correlation functions have been previously demonstrated@xcite . although these spurious states do not affect the asymptotic behavior of the glueball correlators , they do appreciably change the correlators at short temporal separations and can seriously hinder attempts to reduce excited - state contamination to hasten the onset of asymptotic behavior . since our glueball mass measurements rely heavily on the reduction of such excited - state contributions to the glueball correlation functions , the use of an action which is free of spurious lattice modes is crucial . the action given in eq . [ eq : impaction ] couples only link variables on neighboring time - slices , which ensures that all of our effective masses converge to their plateau values monotonically from above and so validates the variational techniques employed . it is now known that perturbation theory by itself does not reliably determine the couplings in an improved action in lattice gauge theory . hence , the interaction strengths in @xmath31 have been determined using a judicious combination of perturbation theory and mean field theory . mean field theory is introduced by separately renormalizing the spatial and temporal link variables : @xmath32 and @xmath33 , where @xmath12 and @xmath13 denote the renormalization factors for the spatial and temporal links , respectively . the mean - link parameters @xmath13 and @xmath12 are best determined by guessing input values for use in the action , measuring the mean links in landau gauge in a simulation , then readjusting the input values accordingly and tuning until the input values match the measured values . the determination of these renormalization factors is described in more detail in refs . however , when @xmath8 is significantly smaller than @xmath7 , we expect the mean temporal link @xmath13 to be very close to unity since @xmath34 in perturbation theory . hence , to simplify matters , we set @xmath35 . we introduce further simplifications by using a convenient and gauge - invariant definition for @xmath12 in terms of the mean spatial plaquette given by @xmath36 , where @xmath37 denotes the spatial plaquette . this eliminates the need for gauge fixing , yields values for @xmath12 which differ from those found using the landau gauge definition by only a few per cent , and significantly speeds up the tuning process . at finite coupling @xmath11 , the anisotropy @xmath38 is renormalized away from its input value @xmath6 . measurements of this renormalization have been made using the static potential extracted from correlations along the different spatial and temporal axes of the lattice @xcite . without mean - link improvement , this renormalization can be as large as @xmath39 . when the action includes mean - link corrections , this renormalization is found to be small , typically a few per cent . we used @xmath40 in all of our calculations , accepting the small radiative corrections to the anisotropy as finite lattice spacing errors , which vanish in the continuum limit . glueballs may be labeled by their total ( integral ) spin @xmath41 and their symmetries under spatial inversion and charge conjugation . however , on a cubic lattice , glueballs are characterized by their transformation properties under the cubic point group , combined with parity and charge conjugation operations . the cubic group , @xmath42 , has 24 elements that fall into five conjugacy classes , and thus , the dimensions of the irreducible representations ( irreps ) are 1 , 1 , 2 , 3 and 3 . these irreps are labeled @xmath43 , @xmath44 , @xmath45 , @xmath46 , and @xmath47 , respectively . including parity and charge conjugation symmetry operations , there are 20 irreps ( labeled by @xmath48 , where @xmath41 now denotes an irrep of @xmath42 ) . in this study , four of the irreps which generate light ( @xmath49 gev ) glueball states were simulated : the @xmath50 , @xmath51 , @xmath52 , and @xmath53 channels . of particular interest are the @xmath51 and @xmath52 irreps whose combined five rows correspond to the five polarization states of the tensor ( @xmath0 ) glueball which become degenerate as continuum rotational invariance is restored . this then gives information on the magnitude of lattice artifacts at finite cut - off . the mass of a glueball , @xmath54 , having a given @xmath48 can be extracted from the large-@xmath55 behavior of a correlation function @xmath56 , where @xmath57 denotes the lattice irrep corresponding to the @xmath48 of interest and @xmath58 is a gauge - invariant , translationally - invariant , vacuum - subtracted operator capable of creating a glueball out of the qcd vacuum @xmath59 . as the temporal separation @xmath55 becomes large , this correlator tends to a single decaying exponential @xmath60 , where @xmath61 is the mass of the lowest - lying glueball which can be created by the operator @xmath62 . in order to extract @xmath61 , the correlator @xmath63 must be determined for @xmath55 sufficiently large that @xmath63 is well approximated by its asymptotic form . however , the signal - to - noise ratio in any monte carlo determination of @xmath63 falls exponentially fast with respect to @xmath55 . thus , it is crucial to use a glueball operator for which @xmath63 attains its asymptotic form as quickly as possible . if @xmath64 denotes the glueball state of interest , this means that we must choose an operator for which the overlap @xmath65^{1/2}$ ] is as near to unity as possible . for such an operator , the signal - to - noise ratio is also optimal@xcite . in order to construct such operators , we exploited the smearing @xcite and variational techniques which have been used with success in earlier wilson action simulations . in each of the @xmath48 channels of interest , glueball operators were constructed on each time - slice in a sequence of three steps . first , smeared links @xmath66 and fuzzy superlinks @xmath67 were formed . secondly , a set of basic operators @xmath68 were constructed using linear combinations of gauge - invariant , path - ordered products of the @xmath66 and @xmath67 matrices about various closed spatial loops ; each such linear combination was designed to be invariant under spatial translations and to transform irreducibly under the symmetry operations of the cubic point group according to the irrep of interest . lastly , the glueball operators @xmath69 were formed from linear combinations of the basic operators , @xmath70 , where the coefficients @xmath71 were determined using the variational method . each of these three steps is described below . operators constructed out of smeared links and fuzzy superlinks have dramatically reduced mixings with the high frequency modes of the theory . thus , the use of spatially - smoothed links is an important part of reducing excited - state contamination in the glueball correlation functions . two smoothing procedures were used : a single - link procedure and a double - link procedure . in the single - link procedure , every spatial link @xmath72 on the lattice is replaced by itself plus a sum of its four neighboring ( spatial ) staples , projected back into su(3 ) : @xmath73 where @xmath74 denotes the projection @xcite into su(3 ) . here , we denote this mapping of the spatial link matrices into the smeared link variables by @xmath75 . in the double - link procedure , new superlinks @xmath76 of length @xmath77 are built using neighboring staples which connect sites separated by a distance twice that of the length of the source link variables : @xmath78 and we denote this mapping by @xmath79 . both procedures can be applied recursively ; smeared links can be smeared again and fuzzy links of increasing length @xmath80 can be constructed . a _ smoothing scheme _ , @xmath81 , is defined as a composition of single - link mappings and double - link mappings . six different smoothing schemes were used . the simplest scheme used was the composition of two single - link smearings : @xmath82 . to simplify notation , we write this as @xmath83 . we also used the compositions of four and six single - link mappings : @xmath84 and @xmath85 . in the other three smoothing schemes , the application of several single - link smearings , followed by one final iteration of double - link fuzzing was used : @xmath86 , @xmath87 , and @xmath88 . only one iteration of the fuzzing procedure which results in links connecting sites separated by @xmath77 was found to be useful for the range of coarse @xmath7 values explored here . for the finer lattices ( @xmath89 and @xmath90 ) , an extra four initial iterations of single - link smearing were used in all six smoothing schemes to enhance ground - state overlap . to simplify matters , the same values for the two parameters @xmath91 and @xmath92 were used in all smearing and fuzzing iterations . these values were chosen to minimize excited - state contamination in the glueball correlation functions . a crude optimization was done in a set of low statistics runs and the optimal values @xmath93 and @xmath94 were then used in all the glueball simulations . the second step in the construction of our glueball operators was the formation of a set of basic operators @xmath68 using linear combinations of gauge - invariant , path - ordered products of the @xmath66 and @xmath67 matrices about various closed spatial loops . combinations which were hermitian , invariant under spatial translations , and transformed irreducibly under the operations of the cubic point group according to the irrep of interest were constructed . for a more detailed exposition of this construction , see ref . @xcite . in each channel , a large set of prototypes were programmed , and a short simulation was then performed to determine the coefficients of each operator in the variational ground state . in each channel , the four operators with the highest of these contributions were then chosen for use in the production runs . the paths in this optimal set are illustrated in fig . [ glueops ] . in the glueball simulations , these wilson loops were measured on the link variables from the six smoothing schemes , yielding a total of @xmath95 basic operators @xmath68 in each of the four channels . finally , @xmath69 was formed from a linear combination of the basic operators , @xmath96 . the coefficients @xmath71 were determined using the variational method . first , the @xmath97 correlation matrix was computed in the glueball simulations : @xmath98 where @xmath99 denotes a vacuum - subtracted operator @xmath100 . note that @xmath101 is independent of @xmath55 . the coefficients @xmath71 were then determined by minimizing the effective mass @xmath102,\ ] ] where the time separation for optimization was fixed in all cases to @xmath103 . let @xmath104 denote a column vector whose elements are the optimal values of the coefficients @xmath71 . then requiring @xmath105 for all @xmath106 yields an eigenvalue equation : @xmath107 the eigenvector @xmath108 corresponding to the largest eigenvalue @xmath109 then yields the coefficients @xmath110 for the operator @xmath111 which best overlaps the lowest - lying glueball @xmath54 in the channel of interest . operators which overlap excited glueball states can also be constructed using the other eigenvectors of eq . [ eq : variation ] . in particular , the operator @xmath112 expected to best overlap the first - excited glueball state @xmath113 was obtained from the eigenvector corresponding to the second largest eigenvalue of eq . [ eq : variation ] . the elements of the correlator matrix given in eq . [ eq : gluecorr ] were estimated using the monte carlo method . ten separate glueball simulations were performed on dec alpha - workstations . configuration ensembles were generated using both cabibbo - marinari ( cm ) pseudo - heatbath and @xmath114 sub - group over - relaxation ( or ) methods . link variables were updated in serial order on the lattice . we define a _ compound sweep _ as one cm updating sweep followed by three or sweeps . in the glueball simulations , three compound sweeps were performed between measurements , and the measurements were averaged into bins of 100 in order to reduce data storage requirements ( except for the @xmath115 , @xmath116 run in which 40 configurations were included in each bin ) . in all ten simulations , 100 bins were obtained . our ensembles were tested for residual autocorrelations during the analysis phase by over - binning by factors of two and four ; in all cases , the statistical error estimates remained unchanged . values for the mean link parameter @xmath12 were determined self - consistently as previously described . this tuning procedure required a minimal amount of computational effort and provided thermalized configurations for later computations . the improved action simulation parameters used are given in table [ tab : runparams ] . for the data - fitting phase , the large @xmath117 correlator matrices in each channel were reduced using the coefficients @xmath108 and @xmath118 to smaller @xmath119 matrices @xmath120 for @xmath121 : @xmath122 the ground - state correlator @xmath123 was fit for @xmath124 using a single exponential @xmath125 where @xmath126 was the temporal extent of the periodic lattice , to obtain an estimate of the mass @xmath61 ( in terms of @xmath127 ) of the lowest - lying glueball in each channel . to determine the mass @xmath128 of the first - excited glueball and another estimate of @xmath61 , the @xmath119 correlator was also fit for @xmath124 using the form @xmath129 various fit regions @xmath130 to @xmath131 were used in order to check for consistency in the extracted values for the masses . best fit values were obtained using the correlated @xmath132 method . error estimates were calculated using a @xmath133point bootstrap procedure ; in all cases , error estimates were very close to being symmetric about the central best - fit values and were thus averaged to simplify presentation . in order to convert the glueball masses as measured in our simulations into physical units , we must set the scale by determining the lattice spacing @xmath8 for each @xmath5 and @xmath6 we consider . to do this , we must first choose one physical quantity to use as a reference . this reference quantity must then be measured on the lattice in terms of @xmath8 . the experimentally - known value for the reference quantity is then used to extract the lattice spacing . a quantity which can be easily and accurately determined both experimentally and in numerical simulations is an ideal choice for such a reference . the mass of a low - lying particle is typically used for setting the scale . in our case , however , there are no unambiguous experimental determinations of the glueball masses , so instead , we must look for another purely gluonic quantity . the hadronic scale parameter @xmath4 defined in terms of the force between static quarks by @xmath134_{r = r_0}=1.65 $ ] , where @xmath135 is the static - quark potential , is an attractive possibility . it can be measured very accurately on the lattice . the advantages in using @xmath4 to set the scale have been enumerated in ref . @xcite . from phenomenological potential models , one finds @xmath136 fm . a disadvantage in using @xmath4 is that its physical value must be deduced indirectly from experiment , and there is some ambiguity in doing this , as will be discussed below . however , in the absence of a better gluonic reference , we have chosen @xmath4 to set the scale . in this section , we outline the determination of @xmath4 in terms of @xmath7 . in order to determine @xmath4 in terms of the lattice spacing , we need accurate measurements of the static - quark potential . we extracted @xmath135 for various spatial separations @xmath137 , both on and off the axes of the lattice , from the expectation values of wilson loops @xmath138 in the standard manner : @xmath139 + { \rm excited\ state\ contributions}.\ ] ] in the monte carlo evaluation of the wilson loops , measurements were taken after every four compound sweeps ( as defined in sec . [ sec : simulation ] ) . the measurements of the wilson loops were done independently of the glueball mass studies using separate ensembles of configurations . to minimize contamination from excited states , the wilson loops were constructed from iteratively smeared spatial links . the single - link smearing method described previously was used . a given smearing scheme is specified not only by the parameter @xmath91 , but also by the total number of smearing iterations , denoted by @xmath140 . two different choices of the smearing parameter were used in all cases : one smearing was chosen to work well for small @xmath141 , the other to work well for large @xmath142 . separate measurements for each smearing were taken ; cross correlations were not determined . the statistical noise in the evaluation of @xmath138 was reduced dramatically , especially for large temporal separations , by constructing the wilson loops , whenever possible , from thermally - averaged temporal links @xcite . the thermal averaging was accomplished using the cabibbo - marinari pseudo - heatbath method ( 40 updates ) . other relevant run parameters are given in table [ potrun ] . the values of the potential @xmath135 were extracted from the wilson loop measurements by fitting @xmath138 to the exponential form @xmath143 $ ] in the range @xmath144 , for each @xmath137 . the plateau region from @xmath130 to @xmath131 was chosen separately for each @xmath137 in order to minimize the uncertainty in the extracted values for @xmath135 while maintaining a good quality @xmath145 of fit . best fit values were determined using the standard @xmath146-square test , taking into account temporal correlations among the @xmath138 . the covariance matrix in @xmath132 was determined using the jack - knife procedure , and estimates for the uncertainties in the extracted values for @xmath135 were computed using the bootstrap method . binning of the data was done as a crude check that our measurements were statistically independent . the results of a typical fit are shown in fig . [ veffmass ] , which is an effective mass plot for @xmath135 for @xmath147 . the effective mass for @xmath135 is a function of @xmath55 , defined as @xmath148 $ ] , which tends to the true mass as @xmath55 becomes large . once a suitable plateau region in the effective mass was established for each @xmath135 , the hadronic scale @xmath149 could be determined . we found that the on - axis potential @xmath135 for the range of @xmath137 values studied here using coarse lattices fit a coulomb plus linear form @xmath150 very well ( with qualities of fit ranging from @xmath151 to @xmath152 ) . we , therefore , used this form to interpolate @xmath135 and the force between static quarks . simultaneous fits of the wilson loops for the on - axis potential to the form @xmath153 $ ] were done , taking into account all correlations among the @xmath154 for both different @xmath55 and @xmath142 . different regions in @xmath55 were used for different @xmath142 values ; the plateau regions determined previously were used . only the on - axis potentials were used ; this prevented the covariance matrix in the @xmath132 to be minimized from getting too large . this covariance matrix was evaluated using the jack - knife method ; uncertainties in the fit parameters @xmath155 , @xmath156 , @xmath157 , and @xmath158 were obtained using the bootstrap method . once we had an ensemble of bootstrap estimates for these fit parameters , the ratio @xmath149 and its bootstrap uncertainty were then determined using @xmath159 note that to compute @xmath149 , we need the ratio @xmath38 since our fits yielded estimates of @xmath160 only . we used the input value @xmath6 since we know that its renormalization is small . results for @xmath149 are given in table [ r0values ] . using the results in table [ r0values ] , we can now express all energies measured in simulations in terms of @xmath4 . for example , in fig . [ potential ] , we show the potential , including off - axis inter - quark separations , expressed in terms of @xmath4 . lattice spacing errors are seen to be small . to allow clear resolution of the scaling properties of the low - lying glueball masses in the improved action , two sets of simulations were performed at two different anisotropies : six lattice spacings for an aspect ratio @xmath116 and four spacings for @xmath161 were studied . the input parameters used in these simulations are given in table [ tab : runparams ] . the results of fitting the variationally - optimized correlators @xmath63 to the functions given in eqs . [ eq : ggcorrelator ] and [ eq:2by2correlator ] are summarized in tables [ tab : emone]-[tab : emten ] . effective mass plots for the two smallest-@xmath7 simulations are presented in figs . [ fig : b26-r3-effmass - a1pp]@xmath162[fig : b26-r3-effmass - t2pp ] and figs . [ fig : b24-r5-effmass - a1pp]@xmath162[fig : b24-r5-effmass - t2pp ] for @xmath116 and @xmath161 , respectively . for each channel in each of the ten simulations , it was possible to find a fit region @xmath163 in which the correlation function was well described by its asymptotic form as indicated by the quality of fit . in other words , convincing plateaux were observed in all effective masses . the most impressive plateau , observed in the @xmath50 channel for @xmath164 and @xmath161 , spanned ten time - slices . in most cases , the onset of the plateau occurred when the source and sink operators were separated by only one time - step . the overlaps with the lowest - lying states were also found to be extremely good , better than @xmath165 in most cases and often consistent with unity . this clearly demonstrates the effectiveness of the link - smearing and variational techniques in diminishing excited - state contamination . fits using @xmath166 or @xmath167 were also done to check for agreement with the @xmath168 results . at time separations for which the ground state could be reliably observed , the off - diagonal elements of the reduced correlation matrices @xmath120 were found to be consistent with zero within statistical uncertainty . this suggests that the link - smearing , variational method also gives an excellent construction of the first - excited state in each channel . our best estimates for the glueball masses in terms of @xmath127 are indicated in boldface in each of the tables [ tab : emone]-[tab : emten ] . these estimates are summarized in table [ tab : final ] . masses for the first - excited states are also indicated in the @xmath169 fits listed in the tables [ tab : emone]-[tab : emten ] . in this work , we were concerned with the magnitude of discretization errors in the glueball mass determinations from coarse lattice simulations using an improved action . in order to evaluate these errors , we had to eliminate uncertainties from all other sources . the increased efficiency of simulations on coarse , anisotropic lattices allowed us to reduce statistical errors to the acceptable level of about @xmath170 . the only remaining source of uncertainty we had to address was finite volume . the masses of particles confined in a small box with periodic boundary conditions can differ appreciably from their infinite volume values ; finite volume effects can also induce a splitting in the masses of the @xmath45 and @xmath47 tensor polarizations . finite volume effects on the scalar glueball mass have been analyzed before@xcite , but the effects on the tensor and the pseudovector are less well known . in order to ascertain the effects on our glueball masses of simulating in a finite volume , four extra simulations were performed for @xmath164 , @xmath116 using lattices of spatial extent @xmath171 and @xmath172 . the temporal extent was held fixed at 24 grid points . for each of these volumes , the mean - field renormalization parameter @xmath12 was recalculated . the @xmath173 lattice was the only simulation that required any change in this parameter , and in this case , the effect was small ; @xmath12 increased by only @xmath174 . the results from the @xmath175 and @xmath176 runs for the glueball masses in terms of @xmath127 are given in table [ tab : finvolvals ] . note that the results from the @xmath177 lattice differ very little from those from the @xmath178 lattice , suggesting that our lattice volumes are sufficiently large to ensure that finite volume errors are negligible . for the @xmath173 lattice , no plateaux in the effective masses for the @xmath50 , @xmath51 and @xmath52 channels were observed ; the mass in the @xmath53 channel was found to be @xmath179 . the operators used in these runs were the same as those constructed for the large volume runs and thus , were not optimized to give large overlaps with the light torelon states present in small volumes . it is likely that this effect was responsible for the poor overlap of our operators with the scalar and tensor states on the @xmath173 lattice . the properties , such as the mass , of a glueball confined in a small box with periodic boundary conditions differ from those in an infinite volume . the modification of the mass of a particle due to finite volume effects has been estimated in ref . @xcite : @xmath180 , \label{eq : luscherfinvol}\ ] ] where @xmath181 is the dimensionless length scale @xmath182 , @xmath183 is the infinite - volume mass of the scalar glueball , and @xmath184 is related to the strength of an effective triple scalar glueball interaction vertex . the mass shift given in eq . [ eq : luscherfinvol ] is valid for sufficiently large @xmath181 and arises from the exchange of scalar glueballs across the periodic boundaries of the lattice . finite volume errors in our glueball masses measured on an @xmath178 lattice at @xmath164 , @xmath116 ( where @xmath185 fm , similar to the volumes used in the other nine simulations ) can be estimated by fitting the form given in eq . [ eq : luscherfinvol ] to the masses in table [ tab : finvolvals ] . let @xmath186 and @xmath187 , then @xmath188 . the @xmath50 fit was done first using the function @xmath189 , where @xmath190 and @xmath191 were the fitting parameters . the best fit value for @xmath190 was then used in the fits to the results for the other irreps ; to simplify matters , the uncertainty in @xmath190 was neglected in these fits . the @xmath53 fit also included the energy estimate extracted from the @xmath192 simulation . the results of these fits are summarized in table [ tab : finvolfits ] ; the estimates of the finite volume errors are listed in the final column of this table and are given by @xmath193 using eq . [ eq : luscherfinvol ] . in all cases , these errors were insignificant compared to statistical errors ; this means , for example , that any differences between the large - volume masses in the @xmath47 and @xmath45 channels must be due purely to discretization errors . it is interesting to note that our estimate of @xmath191 agrees well with the value @xmath194 found in ref . @xcite . the glueball mass estimates in terms of @xmath127 were combined with the determinations of the hadronic scale @xmath149 . the results are shown in figs . [ fig : scaling ] , [ fig : tensor ] , and [ fig : scalarscaling ] . in these figures , the dimensionless product of @xmath4 and the glueball mass estimates are shown as functions of @xmath195 . solid symbols indicate results from the @xmath116 simulations , while open symbols are used for the results from the @xmath161 runs . in fig . [ fig : scaling ] , the lowest - lying masses in each of the channels @xmath50 , @xmath51 , @xmath52 , and @xmath53 are compared with results from small-@xmath7 wilson action simulations . the lowest - lying and first - excited masses in the @xmath51 and @xmath52 channels are shown in fig . [ fig : tensor ] , and the ground - state and first - excited state in the @xmath50 channel are depicted in fig . [ fig : scalarscaling ] . to extract physical predictions ( for the pure - gauge theory ) , the curves in these plots must be extrapolated to the continuum limit @xmath196 . discretization errors are given by the deviations of the finite-@xmath7 results from these limiting values . the lowest - lying states in the @xmath51 and @xmath52 channels correspond to the five polarizations of the tensor @xmath0 glueball in the continuum . differences between the @xmath51 and @xmath52 masses are a measure of violations of rotational symmetry due to finite spacing artifacts . in fig . [ fig : scaling ] , such violations are seen to be small for our less coarse lattices and become appreciable as the spacing gets very large . discretization errors in the @xmath52 exceed those of the @xmath51 ; on our coarsest lattices , finite spacing errors are only a few per cent for the @xmath51 channel , but about @xmath197 in the @xmath52 channel . in the @xmath51 channel , the @xmath116 results differ very little from those using the higher aspect ratio @xmath161 , suggesting that the @xmath30 errors are negligible . however , small differences between the results from the two anisotropies are visible in the @xmath52 channel . one expects that @xmath30 errors will decrease as @xmath6 is increased . since the @xmath52 discretization errors are slightly larger for the @xmath161 runs , @xmath30 errors can account for this difference only if such errors offset the @xmath198 errors . the leading discretization errors in the tensor glueball masses are expected to be @xmath199 . however , we have already argued that the results in fig . [ fig : scaling ] imply that the @xmath30 errors are negligible . since the action included mean - field correction factors , we also expected that @xmath198 errors would dominate over @xmath200 errors and in our continuum limit extrapolations , we assumed that this was true unless the fit provided compelling evidence to the contrary . although we expected the leading discretization errors to be @xmath198 , the following three functions were used in our continuum limit extrapolations : @xmath201 where @xmath202 , @xmath203 and @xmath204 are best - fit parameters . the results of these fits are given in table [ tab : extrap-3 ] for the @xmath116 data and table [ tab : extrap-5 ] for the @xmath161 simulations . comparing the values of @xmath205 , one sees that the fitting function @xmath206 was preferred for both the @xmath116 and @xmath161 results , although only marginally so for @xmath116 . given this fact and our expectation that @xmath206 should best describe the leading discretization effects , we took @xmath204 from the @xmath206 fits as our continuum limit estimates ( indicated in boldface in tables [ tab : extrap-3 ] and [ tab : extrap-5 ] ) . these four estimates are in very good agreement not only with one another , but also with the wilson action estimates . these fits using @xmath206 are shown in fig . [ fig : scaling ] . for our final estimate of the tensor glueball mass , we performed a simultaneous fit with the four data sets ( two irreps and two anisotropies ) using four separate @xmath206 functions but constraining the intercept parameter @xmath204 to be the same for all four fitting functions . this yielded @xmath207 with @xmath208 , in agreement with the wilson action estimate @xmath209 , obtained by fitting all of the wilson action measurements shown in fig . [ fig : scaling ] to @xmath210 . we also examined the discretization errors in the masses of the first - excited glueball states in the @xmath51 and @xmath52 channels . these are shown in fig . [ fig : tensor ] . there are several reasons for interpreting these data as different polarizations of a spin - two excited state : the two irreps extrapolate to the same continuum limit value ; if the @xmath52 state were spin three , then there would be a degeneracy with the @xmath211 and @xmath212 channels and this was not observed ( these results will be presented elsewhere ) ; if they were polarizations of a spin - four state , then again , a similar level must also be found in the @xmath211 channel . the degeneracy between the two irreps and the weak finite - volume dependence of their energies also makes an interpretation of this state as a torelon pair or a two - scalar - glueball scattering state unlikely , although the mass of this level lies close to twice the mass of the scalar glueball . continuum limit extrapolations were performed using the three functions of eqs . [ eq : fitconst]-[eq : fita4 ] ; measurements from the two largest @xmath7 spacings for @xmath116 and the single largest spacing for @xmath161 were excluded from these fits . the results are given in tables [ tab : extrap-3 ] and [ tab : extrap-5 ] . again , we expected @xmath206 to provide the most reliable extrapolations to the @xmath213 limit ; this was confirmed by the fact that @xmath206 yielded @xmath51 and @xmath52 continuum limits in best agreement with each other . differences found between the @xmath116 and @xmath161 extrapolations were not statistically significant . the fits using @xmath206 are shown in fig . [ fig : tensor ] . our final determination of the mass of the excited tensor glueball , obtained from a constrained set of four @xmath206 , similar to the ground - state extraction , was @xmath214 , where @xmath215 . this mass has not been reliably determined in any previous simulations . finite spacing errors in the mass of the @xmath53 pseudovector glueball were also studied . these were found to be small and are shown in fig . [ fig : scaling ] . the results from the different anisotropies are in good agreement . extrapolations to the @xmath213 limit were done using the three functions @xmath210 , @xmath216 , and @xmath206 ; the results of these fits are summarized in tables [ tab : extrap-3 ] and [ tab : extrap-5 ] . the continuum limits obtained from fits to the @xmath116 and @xmath161 data agreed only for the constant fit form @xmath210 . the fits to @xmath216 and @xmath206 yielded slope parameters ( @xmath202 and @xmath203 ) with large relative errors and opposite signs for the different anisotropies . hence , the function @xmath210 was used to extrapolate to the continuum limit . due to the very good agreement between the @xmath116 and @xmath161 results , all ten data points were used in our extrapolation fit . our estimate from this fit ( shown in fig . [ fig : scaling ] ) was @xmath217 , where @xmath218 , in agreement with the extrapolation @xmath219 using @xmath210 of the wilson action results shown in fig . [ fig : scaling ] . in contrast to the tensor and pseudovector , the scalar glueball mass showed significant finite spacing errors ( see fig . [ fig : scaling ] ) , even for our less coarse lattices . as @xmath7 was increased , the scalar mass first decreased , reached a minimum near @xmath220 , then gradually increased . near the minimum , the mass was about @xmath221 lower than estimates of the continuum limit from small-@xmath7 wilson action computations ; a @xmath222 discretization error was observed in the result from our smallest @xmath7 simulation . although the magnitudes of these errors were significant , they were smaller than those obtained using the wilson action by a factor of two . in order to extrapolate to the continuum limit , an appropriate fitting function was needed . the leading discretization errors were expected to be @xmath199 . however , there were no distinguishable differences between the @xmath116 and @xmath161 results , suggesting that the @xmath30 errors were negligible , leaving us to consider @xmath223 effects . by inspection , one sees that the fitting form @xmath206 , which neglects one - loop @xmath200 effects , can not describe the data , in contrast to the data for the tensor and pseudovector glueballs . as @xmath213 , we expect the coupling @xmath224 to vanish as @xmath225 , where @xmath226 is an appropriate scale parameter . hence , we were led to consider the following four - parameter fitting function : @xmath227 } + c_4\ ( a_s / r_0)^4 . \label{eq : scalarfit2}\ ] ] however , it was not known how reliably the leading perturbative behavior of @xmath224 would describe the true cut - off dependence of the coupling over the large range of spacings considered . taking this into account and inspecting the behavior of the actual data , we decided to also consider the following simpler quadratic form : @xmath228 both of these functions were fit to the mass measurements from all ten simulations ; the results of these fits are summarized in table [ tab : scalarextrap ] . the function @xmath229 yielded a slightly better fit and a continuum limit for the scalar glueball mass of @xmath230 . this fit is shown in fig . [ fig : scaling ] . an extrapolation of existing wilson action data using @xmath216 yielded @xmath231 . given the quality of the scalar glueball mass estimates using the wilson action , this slight discrepancy raises doubts concerning the reliability of the extrapolation using @xmath229 ; mass estimates using the improved action for a few values of @xmath7 smaller than those considered here would be needed to resolve this discrepancy . one explanation for the @xmath222 discretization errors in the scalar glueball mass is that the scalar glueball is extremely small . however , there is evidence@xcite that the presence of a critical endpoint of a line of phase transitions ( not corresponding to any physical transition found in qcd ) in the fundamental - adjoint coupling plane is responsible for lowering the scalar glueball mass near the crossover region in the wilson action . it is possible that the scalar glueball mass in the improved action used here may be similarly influenced . if so , the fact that this effect appears to be less pronounced for this action suggests the possible existence of other perturbatively - improved actions in which the scalar glueball mass is even less affected by scaling violations . we are currently searching for such actions . discretization errors in the mass of the first - excited state in the @xmath50 channel were also found to be significant , as shown in fig . [ fig : scalarscaling ] . the mass of this state is nearly twice that of the lowest - lying scalar glueball , suggesting that this state may simply be two glueballs . given the significant discretization errors in the single glueball scaling data , one would expect similar systematic errors in the two glueball state . the absence of any level of similar mass in all other channels justifies the spin - zero interpretation of this state . considering the difficulties encountered in extrapolating the lowest - lying scalar to the continuum limit , we made no serious attempt to determine the continuum limit of this first - excited state . however , the result of a fit using @xmath232 is included in table [ tab : scalarextrap ] . the possible interpretation of this level as a two glueball system might be strengthened by a more precise finite - volume study . in order to convert our glueball mass computations into physical units , we must specify the value of the hadronic scale . the hadronic scale @xmath4 has a precise definition in terms of the static - quark potential . however , the static - quark potential can not be directly measured in an experiment ; it must be deduced indirectly from other observables . we decided to use a variety of different physical quantities to deduce @xmath4 . in table [ hadronictable ] , estimates of @xmath233 using the results from various quenched lattice simulations are shown . for each computation , the quantity used to set the lattice spacing , such as the mass of the @xmath234 or the @xmath235 splitting in heavy quarkonia , is indicated . the determination of @xmath233 from @xmath236 was accomplished using values of @xmath237 given in ref . @xcite for the wilson gluonic action at various values of @xmath5 , interpolating where necessary . note that due to quenching effects , @xmath4 varies with the quantity used to set the scale . the entries in the last column of table [ hadronictable ] can not be considered as different measurements of a single quantity and thus , strictly speaking , their weighted average has no statistical meaning . the last column of the table is meant to illustrate the range of values one obtains for @xmath233 when using various scale setting quantities . we expect that the value of @xmath233 appropriate for the low - lying glueballs should lie somewhere within this range . the simple average @xmath238 mev of the determinations in table [ hadronictable ] was taken as our estimate of the hadronic scale . for our final continuum mass estimates of the tensor glueball , we found @xmath239 mev ( where the first error is statistical and the second is from uncertainties in the determination of @xmath4 ) . it is interesting to note that our mass estimate lies within @xmath240 of the mass of the @xmath241 resonance @xcite , reported to have quantum numbers @xmath242 . in order to make a direct comparison with experiment , however , corrections to our result from light quark effects and mixings with nearby conventional mesons must be taken into account . our estimate of the mass of the first - excited glueball in the tensor channel was @xmath243 mev ; for the pseudovector state , we found a mass of @xmath244 mev . our estimate from the fit using @xmath229 for the mass of the scalar glueball was @xmath245 mev ; however , we regard the continuum limit extrapolation for this state as being less reliable than those for the other glueballs . a quantitative comparison of efficiencies is difficult to make . there are many factors which affect the overall efficiency of a monte carlo simulation . certainly , the number of link updates is an important factor . the speed of an update is , of course , platform- , action- , and algorithm - dependent . on the dec alpha - workstations we used , a cm update using an improved action required twice as much time as for the wilson action ; the improved - action or updating time was three times longer . critical slowing down and thermalization are also contributing factors , but a crucial issue is the reduction of excited - state contaminations in the glueball correlators . in our coarse lattice simulations , we found that current methods for constructing good glueball operators were very effective in hastening the onset of plateaux in the effective masses . given these difficulties in assessing the efficiency of a glueball simulation , we decided to make our comparisons based simply on the number of link updates @xmath246 and the fractional error @xmath247 attained in the final mass estimates . since the error in a monte carlo estimate decreases with the number of measurements @xmath248 as @xmath249 , we expect that the reciprocal product of the number of link updates and the square of the fractional error is approximately proportional to the efficiency of a simulation ; we denote this quantity by @xmath250 . an interesting comparison to make is between simulations at a small lattice spacing , such as @xmath251 fm , using the wilson action and improved - action simulations at a spacing @xmath252 fm . such a comparison is relevant because the discretization errors in these cases are of comparable magnitude , excepting the scalar glueball mass . for the @xmath253 wilson action run in ref . @xcite by the gf11 collaboration using gauge invariant glueball operators , a total of @xmath254 link updates were performed , an error of @xmath255 was achieved in the scalar glueball mass , and a fractional error of @xmath256 was obtained for the tensor mass . using @xmath257 fm and @xmath258 , the lattice volume in this simulation was @xmath259 . for the same value of @xmath5 in ref . @xcite by the ukqcd collaboration , @xmath260 link updates were made and fractional errors @xmath261 , @xmath262 , and @xmath263 were obtained for the scalar , tensor , and pseudovector masses , respectively . the lattice volume was @xmath264 for this simulation . in our @xmath115 , @xmath116 run , @xmath265 link updates were performed and @xmath266 , @xmath267 , and @xmath267 errors in the scalar , tensor , and pseudovector masses were achieved . our lattice volume was @xmath268 since @xmath269 . for our @xmath164 , @xmath161 run , the number of link updates was @xmath270 and the errors obtained in the scalar , tensor , and pseudovector masses were @xmath267 , @xmath271 , and @xmath272 , respectively . using @xmath273 , our lattice volume for this run was @xmath274 . thus , the ratios of the @xmath275 values for our @xmath115 and @xmath164 simulations to those of the gf11 run were @xmath276 and @xmath277 , respectively , for the scalar glueball mass , and @xmath278 and @xmath279 for the tensor glueball mass . the ratios of the @xmath115 and @xmath164 @xmath275-values to those of the ukqcd run were @xmath280 and @xmath281 , respectively , for the scalar mass , @xmath282 and @xmath283 for the tensor , and @xmath284 and @xmath285 for the pseudovector . considering our ten simulations together , a total of @xmath286 link updates were performed . for the wilson action simulations of refs . @xcite , an estimated @xmath287 and @xmath288 link updates were required , respectively , to generate continuum limit results whose statistical uncertainties were larger ( for the tensor and pseudovector states ) than those quoted here : the statistical error on our estimate for @xmath289 was about five times smaller than that from the extrapolation of the wilson action results , and for @xmath290 , the uncertainty was twenty times smaller , implying that about @xmath291 times greater statistics would be required for similar accuracy . thus , in total , the anisotropic lattice simulations were certainly more than 1000 times more efficient . the above discussion illustrates the computational advantages of extracting non - scalar glueball masses from simulations on coarse , anisotropic lattices ( @xmath252 fm ) using an improved action instead of lattices for which @xmath251 fm . the excellent overlaps achieved from our variational calculations demonstrate another advantage of simulating on coarse spatial lattices . the glueball wavefunctions extend over only a few points of the lattice when @xmath292 fm . thus , variational calculations using a feasible number of basis functions ( a dozen or so ) can yield very good approximations to the glueball wavefunctions . this will be especially important for future decay constant calculations and determinations of mixings with non - glueball states . these advantages have already enabled us to study the more massive glueball states which have yet to be simulated reliably using the wilson action ; these results will be reported elsewhere . these advantages are less clear for the scalar glueball mass due to the presence of @xmath222 discretization errors at @xmath252 fm . using the action of eq . [ eq : impaction ] , simulations at one or more lattice spacings smaller than 0.2 fm would be needed to firmly establish the continuum limit . a more attractive approach would be to use an action for which discretization errors in the scalar glueball mass at @xmath252 fm are negligibly small . the search for such an action is currently underway . we have demonstrated the advantages of using anisotropic lattices and an improved gluonic action for simulating glueballs . ten simulations at lattice spacings ranging from 0.2 to 0.4 fm were performed using dec alpha - workstations , and the results were extrapolated to the continuum limit . results for the masses of the scalar ( @xmath3 ) , the tensor ( @xmath0 ) , and the pseudovector ( @xmath1 ) glueballs in @xmath293 pure - gauge theory were presented in terms of the hadronic scale @xmath233 . the continuum limits for the tensor and pseudovector glueball masses were obtained with uncertainties of less than @xmath170 , significantly improving upon previous estimates from wilson action simulations carried out with the aid of super - computer resources . extrapolation of the scalar glueball mass to the continuum limit was hampered by uncertainties in choosing the fitting function and discretization errors which were @xmath294 even for our smallest lattice spacings ; although uncomfortably large , these finite - spacing errors were half as large as those obtained using the wilson action . finite volume errors in our results were shown to be negligible . the masses of the first excited states in the scalar and tensor channels were also examined . our results show that spatially - coarse , anisotropic lattice simulations are an effective means of studying gluonic systems . the techniques exploited here are sufficiently powerful to overcome the difficulties which plague monte carlo calculations involving gluonic excitations . these methods should be useful for studying the spectrum of heavier glueball states . data for the masses of all twenty lattice irreps of the cubic group ( including parity and charge conjugation ) are currently being accumulated in order to survey the spectrum of @xmath293 glueball states below 4 gev comprehensively . we shall report on these results in the near future . we also plan to use the techniques outlined in this paper to determine various glueball matrix elements and decay strengths , to investigate the mixings of glueballs with conventional hadronic states , and to study mesonic states containing excited glue ( the so - called hybrid mesons ) . the size of discretization errors in the scalar glueball mass was the only disappointing aspect of this work ; we are currently investigating a new class of lattice actions with the hope of reducing these lattice artifacts for @xmath252 fm to the level of a few per cent . we would like to thank peter lepage , julius kuti , chris michael , terrence draper , and keh - fei liu for helpful discussions . this work was supported by the u.s . doe , grant no . de - fg03 - 90er40546 . mp is grateful to the university of kentucky center for computational sciences for financial support . c. morningstar and m. peardon , nucl . b ( proc . suppl . ) * 47 * , 258 ( 1996 ) . c. morningstar , nucl . . suppl . ) * 53 * , 914 ( 1996 ) . c. morningstar and m. peardon , nucl . b ( proc . suppl . ) * 53 * , 917 ( 1997 ) . m. alford _ et al . , _ in preparation . f. brandstaeter , a. kronfeld , and g. schierholz , nucl . phys . * b345 * , 709 ( 1990 ) . m. albanese _ et al . , _ phys . b * 192 * , 163 ( 1987 ) . m. teper , phys . b * 183 * , 345 ( 1986 ) . y. liang _ et al . _ , phys . b * 307 * , 375 ( 1993 ) . b. berg and a. billoire , nucl . * b221 * , 109 ( 1983 ) . r. sommer , nucl . phys . * b411 * , 839 ( 1994 ) . g. parisi _ et al . _ , phys . * 128b * , 418 ( 1983 ) . g. schierholz , nucl . b ( proc . suppl . ) * 9 * , 244 ( 1989 ) . m. lscher , commun . math . phys . * 104 * , 177 ( 1986 ) . u. heller , phys . b * 362 * , 123 ( 1995 ) . a. patel _ et al . , _ phys . * 57 * , 1288 ( 1986 ) . bali and k.s . schilling , int . j. mod . * c4 * , 1167 ( 1993 ) and private communication ; h. wittig , nucl . b ( proc . suppl . ) * 42 * , 288 ( 1995 ) . barnett _ et al . _ , phys . d * 54 * , 1 ( 1996 ) . et al . , _ phys . lett . * 56 * , 107 ( 1986 ) . h. chen _ et al . _ , nucl . b ( proc . suppl . ) * 34 * , 357 ( 1994 ) . bali _ et al . _ , phys . b * 309 * , 378 ( 1993 ) . p. de forcrand and k - f . liu , phys . lett . * 69 * , 245 ( 1992 ) . c. michael and m. teper , nucl . * b314 * , 347 ( 1989 ) . forcrand _ et al . _ , phys . * 160b * , 137 ( 1985 ) . c. davies _ et al . , _ phys . d * 50 * , 6963 ( 1994 ) . c. davies _ et al . , _ phys . rev . d * 52 * , 6519 ( 1995 ) . t. bhattacharya _ et al . , _ phys . d * 53 * , 6486 ( 1996 ) . f. butler _ et al . , _ nucl . phys . * b430 * , 179 ( 1994 ) . s. aoki _ et al . , _ nucl . b ( proc . suppl . ) * 53 * , 355 ( 1997 ) . s. aoki _ et al . , _ nucl . b ( proc . suppl . ) * 53 * , 349 ( 1997 ) . el - khadra , nucl . b ( proc . suppl . ) * 34 * , 141 ( 1994 ) . c. bernard , j. labrenz , and a. soni , phys . d * 49 * , 2536 ( 1994 ) . allton _ et al . , _ phys . b * 284 * , 377 ( 1992 ) . ( 30,20 ) ( 10.00 , 7.00 ) ( 2 , 1 ) 8 ( 18.00 , 11.00 ) ( 1 , 0)10 ( 20.00 , 7.00 ) ( 2 , 1 ) 8 ( 14.00 , 7.00 ) ( 1 , 0 ) 6 ( 2.00 , 3.00 ) ( 2 , 1 ) 8 ( 2.00 , 3.00 ) ( 0 , 1)10 ( 2.00 , 13.00 ) ( 2 , 1 ) 8 ( 10.00 , 11.00 ) ( 0 , 1)6 ( 10.00 , 11.00 ) ( 1,-1)4 ( 30,20 ) ( 10.00 , 7.00 ) ( 2 , 1 ) 8 ( 18.00 , 11.00 ) ( 1 , 0)10 ( 20.00 , 7.00 ) ( 2 , 1 ) 8 ( 14.00 , 7.00 ) ( 1 , 0 ) 6 ( 2.00 , 3.00 ) ( 2 , 1 ) 8 ( 2.00 , 3.00 ) ( 0 , 1)10 ( 2.00 , 13.00 ) ( 2 , 1 ) 8 ( 10.00 , 11.00 ) ( 0 , 1)6 ( 10.00 , 11.00 ) ( 1,-1)4 ( 30,20 ) ( 14.00 , 7.00 ) ( 1 , 0)10 ( 16.00 , 3.00 ) ( 2 , 1 ) 8 ( 6.00 , 3.00 ) ( 1 , 0)10 ( 6.00 , 3.00 ) ( 0 , 1)9 ( 5.00 , 12.50 ) ( 2 , 1)9 ( 14.00 , 7.00 ) ( 0 , 1)10 ( 13.00 , 6.00 ) ( 1 , 0)11 ( 16.00 , 2.00 ) ( 2 , 1 ) 8 ( 5.00 , 2.00 ) ( 1 , 0)11 ( 5.00 , 2.00 ) ( 0 , 1)10.5 ( 6.00 , 12.00 ) ( 2 , 1 ) 7 ( 13.00 , 6.00 ) ( 0 , 1)9.5 ( 30,20 ) ( 10.00 , 7.00 ) ( 2 , 1 ) 8 ( 18.00 , 11.00 ) ( 1 , 0)10 ( 20.00 , 7.00 ) ( 2 , 1 ) 8 ( 14.00 , 7.00 ) ( 1 , 0 ) 6 ( 2.00 , 3.00 ) ( 2 , 1 ) 8 ( 2.00 , 3.00 ) ( 0 , 1)10 ( 2.00 , 13.00 ) ( 2 , 1 ) 8 ( 10.00 , 11.00 ) ( 0 , 1)6 ( 10.00 , 11.00 ) ( 1,-1)4 ( 30,20 ) ( 14.00 , 9.00 ) ( 2 , 1 ) 4 ( 18.00 , 11.00 ) ( 1 , 0)10 ( 20.00 , 7.00 ) ( 2 , 1 ) 8 ( 10.00 , 7.00 ) ( 1 , 0)10 ( 2.00 , 3.00 ) ( 2 , 1 ) 8 ( 2.00 , 3.00 ) ( 0 , 1)10 ( 2.00 , 13.00 ) ( 2 , 1 ) 8 ( 10.00 , 11.00 ) ( 0 , 1)6 ( 14.00 , 9.00)(-2 , 1)4
we use a relativistic ray - tracing code to calculate the light curves observed from a global general relativistic magneto - hydrodynamic simulation of an accretion flow onto a schwarzschild black hole . we apply three basic emission models to sample different properties of the time - dependent accretion disk . with one of these models , which assumes thermal blackbody emission and free - free absorption , we can predict qualitative features of the high - frequency power spectrum from stellar - mass black holes in the `` thermal dominant '' state . the simulated power spectrum is characterized by a power law of index and total rms fractional variance of above 10 hz . for each emission model , we find that the variability amplitude should increase with increasing inclination angle . on the basis of a newly - developed formalism for quantifying the significance of quasi - periodic oscillations ( qpos ) in simulation data , we find that these simulations are able to identify any such features with ( rms / mean ) amplitudes near the orbital frequency at the inner - most stable orbit . initial results indicate the existence of transient qpo peaks with frequency ratios of nearly 2:3 at a confidence limit , but they are not generic features because at any given time they are seen only from certain observer directions . additionally , we present detailed analysis of the azimuthal structure of the accretion disk and the evolution of density perturbations in the inner disk . these `` hot spot '' structures appear to be roughly self - similar over a range of disk radii , with a single characteristic size and , and typical lifetimes .
astro-ph0606615
our understanding of accretion onto black holes has grown substantially in recent years . correlated magneto - hydrodynamic ( mhd ) turbulence , stirred by an underlying magneto - rotational instability , has now been well - established as the fundamental mechanism of angular momentum transfer in accretion disks @xcite . with that achievement in hand , detailed studies of global accretion disk dynamics have been undertaken via large - scale numerical simulations in both the pseudo - newtonian @xcite and general relativistic @xcite frameworks . at the same time , we have also come into possession of a wealth of observational data . high signal - to - noise spectra of active galactic nuclei ( agn ; see e.g. , the sdss composite : @xcite ) as well as of black hole binaries in a variety of spectral states are now available @xcite ; so , too , are detailed light curves and power - density spectra , particularly of black hole binaries @xcite , but also for seyfert 1 galaxies and other agn @xcite . fluorescent fe k@xmath7 profiles have been used to infer detailed diagnostics of the disk surface in its innermost portions @xcite . one particularly exciting area of research has been the discovery of high - frequency quasi - periodic oscillations ( qpos ) in accreting black hole binaries @xcite . in a growing number of sources , these qpos appear in pairs with integer frequency ratios of 2:3 @xcite . the high frequencies of these qpos ( near the frequency of the inner - most stable circular orbit ; isco ) suggests a strongly relativistic origin , but there exists a very broad range of theoretical models , none of which at this point appear overwhelmingly convincing @xcite . at the same time , there have not yet been any clear identifications of qpos in the mhd simulations , much less pairs of qpos with integer frequency ratios , making it difficult to choose one imperfect theoretical model over another . without a physically consistent way of associating light output with numerical simulations of accretion dynamics , we can not directly compare the predictions of our best theoretical disk models with any spectral or timing observations . therefore it is the goal of this paper to move toward the objective of linking dynamical simulations to observational diagnostics . we will do so by applying several phenomenological models to detailed simulation data in order to predict the light that would be produced . from these models , we generate images and light curves for accretion disks as they might be seen by distant observers . statistical analysis of this derived data will lead us to several interesting generalizations about the nature of variability in the light output of accreting black holes . although the particular models we employ here for the emissivity and opacity of disks are not entirely realistic , the formalism we create can be readily applied to more realistic models in the future . to be specific , in this paper we will combine the results from general relativistic ( gr ) mhd simulations similar to those of @xcite with the ray - tracing and radiative transfer calculations described in @xcite and @xcite to produce light curves and power spectra of accreting black holes . this coupling of ray - tracing with the output data from an independent simulation is often referred to as a `` post - processor , '' a technique commonly used in other branches of physics , particularly when attempting to compare a simulation directly with experiment . the ray - tracing post - processor described in this paper is closely related to a number of radiative transfer codes used in modeling laboratory plasma physics experiments ( see , e.g. @xcite ) . while astrophysicists in general do not have the luxury of being able to control ( or in many cases even determine ) the initial conditions of their experiments , simulations and observational data are achieving levels of quality that will soon prove such post - processors invaluable . one successful post - processor application in astrophysics ( for mhd simulations no less ! ) has been the production of synthetic observations of jets from radio galaxies , which are large enough to have detailed features resolved spatially @xcite . even without imaging observations of accretion disks , the results of the post - processor analysis are very useful in understanding the structure and evolution of the mhd disks , as well as giving valuable insights into x - ray timing measurements of accreting black holes , particularly _ rxte _ observations of binary black holes . in this paper we focus primarily on emission models that may track the disk s thermal emission . in future work , we plan to develop other models that will perform the analogous job for coronal emission processes . the outline and major results of this paper are as follows : we begin in section [ model ] with a detailed description of the mhd simulations and the relativistic radiative transfer equation . we describe three different emission models ( optically thin line , optically thick line , and thermal blackbody ) , each useful as a different diagnostic of the accretion disk structure . for the thermal emission model , we describe a method to convert from the dimensionless `` code units '' of the simulation to more useful cgs units that appear in the absorption and emission coefficients in the radiation transport . in section [ results ] we apply these methods to a single mhd simulation of an accretion flow around a schwarzschild black hole and present light curves and power spectra for the three different emission models and a sample of observer inclination angles . we estimate our sensitivity to detecting high - frequency qpos in the simulated light curves and discuss the significance and robustness of a transient pair of qpos with a 2:3 frequency ratio . to gain a deeper understanding of the power spectra and physical mechanism behind the qpos , in section [ az_structure ] we develop a number of new tools to analyze the azimuthal structure of hot spot perturbations in the turbulent magnetized disk . for a range of radii throughout the disk , we find the characteristic hot spot shape to be self - similar with constant extent in azimuth and radius . furthermore , we find the hot spot lifetimes to be distributed exponentially , much like the simple model of @xcite , with a typical lifetime proportional to the orbital period ( i.e. the dynamical time ) at that radius . finally , in section [ discussion ] , we discuss the implications of these results and outline possibilities for more direct comparison with observations . our basic data come from a single numerical simulation of accretion onto a schwarzschild black hole that uses the fully general relativistic 3-dimensional mhd code described in @xcite . this code follows the dynamical evolution of magnetized matter in the ideal mhd approximation assuming that the fluid s internal energy evolves adiabatically except for the creation of entropy in shocks . magnetic field evolution utilizes the constrained transport algorithm @xcite . the calculation is done in boyer - lindquist coordinates , so the grid s inner radial boundary is placed just outside the event horizon ( @xmath8 ; throughout this paper , unless stated otherwise , we adopt units in which @xmath9 ) . the outer radial boundary is at @xmath10 . at both radial boundaries , matter can leave the grid , but not enter it . to avoid the ( flat - space ) coordinate singularity at the polar axis , there are reflecting boundary conditions at polar angles @xmath11 and @xmath12 . the full range of azimuthal angles is included in the grid , with , of course , periodic boundary conditions linking @xmath13 and @xmath14 . the @xmath14 range in @xmath15 is divided into 256 equal - sized zones . however , the 192 zones in both @xmath16 and @xmath17 are concentrated where we expect the sharpest gradients : near the equatorial plane and at small radius ( see @xcite for details ) . in the initial state , all the matter is contained within a hydrostatic torus centered on the equatorial plane with pressure maximum at @xmath18 and inner edge at @xmath19 . it possesses a purely poloidal and axisymmetric magnetic field whose fieldlines follow isodensity contours ; the plasma initially has a mean @xmath20 of 100 . outside the torus , the density is fixed at @xmath21 times the density at the pressure maximum . as found in many simulations before ( e.g. , @xcite ) , within a few orbits after the beginning of the simulation , the magneto - rotational instability grows to nonlinear amplitude and magnetic stresses redistribute angular momentum throughout the matter . the result is a disk of moderate thickness ( @xmath22 ) and nearly - keplerian angular momentum distribution accreting onto the central black hole . over the entire simulation run time @xmath23 , at @xmath24 intervals in coordinate time , we write out full 3-dimensional snapshots of the major hydrodynamic variables : the density , pressure , and fluid 4-velocity ( in practice we do not actually save data from the first @xmath25 , the time when the early transient features relax ) . in figure [ plotone ] we plot a selection of time- and azimuth - averaged disk variables from the simulation data ( solid curves ) , and compare them with a steady - state novikov - thorne disk ( dashed curves ) . the conversion from code units to cgs units follows the proceedure described in section [ emission_models ] , assuming an accretion rate of @xmath26 onto a black hole of mass @xmath27 . additionally , for the steady - state disk , we have added a small torque at the inner - most stable circular orbit ( isco ) . following @xcite , we have parameterized this torque by the extra efficiency it adds to the disk through increased dissipation just outside the isco . specifically , in figure [ plotone ] we use @xmath28 ( chosen to best match the disk height from the simulations ; see fig . [ plotone]d ) , while the no - torque efficiency for a schwarzschild black hole is @xmath29 . the density , temperature , pressure , and radial velocity shown in figure [ plotone ] are mass - weighted averages , to provide a focus on the most `` disk - like '' region of the mhd simulation near the midplane . outside @xmath30 , the simulation can not really be thought of as an accretion disk , since there is actually net mass flow outwards in that region , the result of the initial torus expanding as angular momentum is transferred from the inner to the outer disk . in the inner - most regions ( @xmath31 ) , the simulation diverges from the novikov - thorne disk as the gas acquires a larger inward radial velocity . to maintain mass conservation , this requires a significant decrease of the surface density compared to the novikov - thorne disk , as is evident from figures [ plotone]a and [ plotone]e . we should note that it may be somewhat coincidental that the agreement between the simulations and analytic model is so good , as they use rather different equations of state . the simulations include only gas pressure , and ignore any loss of internal energy to radiation , while the novikov - thorne model is based on a radiation - pressure dominated disk that balances energy generation by means of viscous dissipation with energy loss by radiation flux from the disk surface . in fact , using significantly different values of @xmath32 would change the scale height of the novikov - thorne disk , while only modifying the cgs scale of the density and pressure of the simulation data and leaving the disk thickness unchanged . the ray - tracing method used in this analysis can be broken up into two independent pieces : ( i ) the calculation of photon geodesic trajectories through the black hole spacetime , and ( ii ) the integration of the relativistic radiative transfer equation along these geodesic paths . the first step , which is carried out using a hamiltonian formulation for the equations of motion , is described in detail in @xcite for the general kerr metric with boyer - lindquist coordinates . the only major difference between our current approach and the methods used in that paper is the manner in which the photon positions and momenta are tabulated . in @xcite , the photons passed through a disk of finite thickness , crossing each surface of constant coordinate @xmath16 exactly once ( see fig . 1 of that paper ) . while computationally convenient , this approach limited the range of viewer inclination angles ( fully edge - on views were impossible as the observer at infinity would be embedded inside the computational grid and photon intersections with the @xmath16 surfaces would occur near the observer and not the accretion disk ) . furthermore , since each photon path would terminate after passing through all the @xmath16-surfaces once , we could not produce multiple images via strong gravitational lensing . we have now solved these problems by tabulating the photon paths according to coordinate time @xmath33 ( the independent variable in our hamiltonian equations of motion ) instead of latitude @xmath16 . for each photon , we record the position and momentum at events equally spaced in @xmath33 over a section of the trajectory centered on the point of closest approach ( minimum @xmath17 ) to the black hole ( this value of @xmath17 will always be unique for null geodesics ) . figure [ plottwo ] shows a schematic of this approach for a schwarzschild black hole , but the discussion below can be applied equally well to the general case of a spinning black hole . this method allows for arbitrary viewing orientations and the modeling of multiply lensed images . instead of recording the coordinate momenta @xmath34 , at each point along the path we convert to the locally flat frame of a zero - angular momentum observer ( zamo ) , denoted by hatted indices : @xmath35 . working in the zamo frame simplifies the solution of the radiative transfer equation , described below . from the tabulated set of photon positions and momenta we obtain a series of null vectors @xmath36 parallel to @xmath37 , defined in the zamo frames at space - time coordinates @xmath38 the fluid properties at these points are interpolated from the fixed computational grid that holds the data output from the mhd simulation . specifically , we record the fluid 4-velocity @xmath39 , also defined in the zamo frame , and the gas density and pressure as measured in the fluid frame . as described below in section [ emission_models ] , the code units of density and pressure are converted to cgs units and the temperature is derived assuming a radiation pressure - dominated gas . with all these pieces in place , the problem is reduced to solving the classical radiative transfer equation through a relativistic fluid with ( locally ) uniform density , temperature , and velocity . we first write the nonrelativistic radiative transfer equation as @xmath40 where @xmath41 is the differential path length and @xmath42 , @xmath43 , and @xmath44 are respectively the spectral intensity , emissivity , and absorption coefficient of the fluid medium at a frequency @xmath45 . the absorption coefficient is related to the opacity @xmath46 through the density @xmath47 : @xmath48 . in this form , true emission and absorption are included , but not electron scattering . the inclusion of scattering adds an integral over @xmath49 to the rhs of equation ( [ rad_trans_eq ] ) , greatly complicating its solution . defining the optical depth @xmath50 through @xmath51 the transfer equation can be written as @xmath52 where the source function is defined as @xmath53 . over regions of constant source function @xmath54 , equation ( [ rad_trans_eq2 ] ) has the simple solution @xmath55.\ ] ] both @xmath42 and @xmath54 have the same properties under lorentz transformations , namely @xmath56 and @xmath57 are both invariant . another lorentz invariant is the optical depth , since the fraction of photons passing through a finite medium is given by @xmath58 , which is just a number , and thus the same in any reference frame . following @xcite , we can calculate the absorption coefficient in a relativistic medium by considering a small volume of matter flowing in the @xmath59 direction with respect to the lab frame . since the motion is in the @xmath60 direction , the fluid volume thickness @xmath61 in the @xmath62-dimension is the same in the lab ( unprimed ) and fluid ( primed ) frames . for a photon propagating at angle @xmath16 with respect to the fluid velocity in the lab frame , the total optical depth in the @xmath62-dimension can be written @xmath63 since @xmath64 is proportional to the @xmath65 component of the photon 4-momentum , it must be the same in both frames because the boost is in a perpendicular direction . thus @xmath64 is another lorentz invariant , and so is @xmath66 . from the definition of the source function @xmath67 , we find that @xmath68 is also lorentz invariant , or @xmath69 in the zamo frame ( here it can be thought of as the lab frame ) , the spatial path length of the photon trajectory is given by @xmath70 . the photon s 4-momentum and the 4-velocity of the fluid in the zamo basis give the angles @xmath16 and @xmath71 , and the absorption @xmath72 and emissivity @xmath73 are given in the fluid frame by the specific emission model being used . now we have enough information to solve the radiative transfer equation in a relativistic flow : @xmath74 the special relativistic redshift between the photons in the zamo frame and fluid frame is @xmath75 where @xmath76 as usual . the above analysis , while quite useful for special relativistic flows in the locally flat zamo basis , ignores all general relativistic effects of curved spacetime around the black hole . to include these effects , we need only consider the invariant @xmath56 along the geodesic path of the photons . this is particularly straightforward from a computational point of view , where the spectrum is stored as a discrete array @xmath77 , evaluated at the frequencies @xmath78 . these frequencies are redshifted from one point on the trajectory to the next due solely to gravitational effects : let @xmath79 be the 4-velocity of a zamo at position @xmath38 . we can define the inner product with the photon 4-momentum as @xmath80 then the array of frequencies is redshifted along the photon path according to @xmath81 similarly , the spectral intensity defined at each frequency point scales as @xmath82 while these methods are well - suited for our implementation of the ray - tracing code , it should be noted that purely covariant approaches also exist for solving the radiative transfer equation in curved spacetime @xcite . as in @xcite , here we consider three different emission models for the disk . the simplest model , optically thin line emission , assumes that the gas is emitting monochromatic radiation isotropically in the fluid rest frame . the emissivity is proportional to rest mass density , and there is no absorption : @xmath83 and @xmath84 . while not entirely realistic ( geometrically thin accretion disks are generally optically thick ) , this model is extremely useful as a diagnostic tool . the observed photons serve as tracers for the disk s surface density , relativistic beaming , and the gravitational redshift of the fluid near the black hole . the second model considered is one of optically thick line emission and absorption , with both @xmath44 and @xmath43 proportional to @xmath85 . just as the optically thin model served as a diagnostic of surface density perturbations , the optically thick model effectively maps out the doppler and gravitational red - shifts of the disk s photosphere . in this regard , it can be thought of as a model for the fluorescent iron line seen in many active galactic nuclei and galactic black holes ( modulo a specified emissivity profile in @xmath17 ) . however , unlike the many popular thin - disk models such as @xcite , our optically thick model samples the upper layers of a vertically extended disk and corona , somewhat complicating the simple iron line profiles predicted by a flat disk . the third emission model combines a kramers opacity law with its thermodynamic inverse emissivity : @xmath86 and @xmath87 giving the classical blackbody distribution in an optically thick region of constant temperature : @xmath88 where @xmath89 . we believe this final model gives the most accurate description of the `` thermal dominant , '' or `` high - soft '' state for x - ray binaries . for the thermal emission model , we must first convert the hydrodynamic variables as provided by the mhd simulations in code units to more useful physical ( cgs ) units of density , temperature and pressure ( as described in @xcite , the fluid variables used in the code are the transport velocity @xmath90 , matter density @xmath47 , pressure @xmath91 , and lorentz factor @xmath92 ) . to do so requires choosing a few specific model parameters . first , we set the overall time and distance scales of the problem by picking a mass for the central black hole . for the boyer - lindquist coordinates used in the simulation , this sets [ cgs_units ] @xmath93 @xmath94 where @xmath95 is the black hole mass in units of @xmath96 . to convert code rest - mass density to g/@xmath97 , we need to set an average mass accretion rate @xmath98 in cgs units : @xmath99 we then use the code - calculated accretion rate @xmath100 to get @xmath101 where the eddington luminosity is @xmath102 erg s@xmath103 , and the efficiency @xmath104 gives the bolometric luminosity through @xmath105 . the novikov - thorne model predicts @xmath106 for the schwarzschild black hole , and the non - zero torque at the isco provides an additional efficiency factor of @xmath28 . the cgs pressure and temperature of the gas are determined by assuming a radiation - pressure dominated equation of state , which is quite reasonable for the inner regions of a black hole accretion disk with an appreciable fraction of the eddington luminosity . for such a gas , the isothermal sound speed is @xmath107 , giving @xmath108 from which the temperature is easily derived : @xmath109 where @xmath110 is simply the radiation density constant ( @xmath111 erg @xmath112 @xmath113 ) . as mentioned above in section [ grmhd_data ] , this method of determining the gas temperature is not entirely consistent with the assumptions made in the mhd simulation . namely , the code assumes a gas pressure - dominated adiabatic equation of state , with no explicit radiation pressure and the gas entropy increased only by shock heating . yet perhaps this is a case of the `` ends justifying the means : '' even though we assume a radiation - dominated pressure to derive the temperature from a calculation that actually ignores radiation pressure , the resulting values are actually in close agreement with the analytic predictions of a standard alpha - disk for the same black hole mass and accretion rate . alternatively , instead of using the mass accretion rate to determine the code - to - cgs conversion factors , we could attempt to match any of the other novikov - thorne disk variables , such as surface density , central pressure , or temperature . these all give somewhat different scaling factors for the density , but we find the behavior of the light curves for the thermal emission model are ultimately very similar in all cases . as a demonstration of how these different models can be used to probe the structure of the accretion disk , we show in figure [ plotthree ] a snapshot of the inner disk , as seen by observers at inclinations of @xmath114 and @xmath115 ( @xmath116 is edge - on ) for the three emission models . for these figures and throughout the majority of this paper , we include only emission from the inner region of the accretion flow with @xmath117 , where the gas behaves most like a classical accretion disk . in all panels , the observed intensity is represented by a color - coded logarithmic scale normalized to the peak intensity of that frame . in the high - inclination frames , the gas is moving towards the observer on the left - hand side of the disk , i.e. in the positive @xmath118 direction . in figures [ plotthree]a and [ plotthree]b , the optically thin nature of the gas is clear from the einstein ring seen in the inner - most region of the disk : this ring is formed by the photons that pass around the black hole and then through the disk and to the observer . with the exception of the emission from the einstein ring , the intensity clearly falls off with decreasing radius as the surface density decreases inward with radius and the gravitational redshift further reduces the observed emission . in figure [ plotthree]b , the special relativistic beaming of photons is clear from the higher intensity seen from the gas moving towards the observer on the left . in figures [ plotthree]c and [ plotthree]d , we show the optically thick line emission model , demonstrating the nearly uniform , monochromatic emission from the face - on disk . for the face - on view , the only variations in intensity come from the gravitational redshift and the transverse doppler shift of gas orbiting more rapidly in the inner - most disk . small local pertubations are also caused by the doppler shift due to turbulent motion of the photosphere perpendicular to the plane of the disk . as in figure [ plotthree]b , the edge - on view in figure [ plotthree]d shows the beaming of photons towards the observer , but produced from the upper layers of the optically thick disk and surrounding atmosphere . finally , in figures [ plotthree]e and [ plotthree]f , we show the image of the disk using the thermal emission and absorption model . unlike the optically thin model where emission traces surface density , here the emission is greatest in the inner disk where the temperature is highest . also , the non - linear dependence of emission and absorption on the hydrodynamic variables produces larger amplitude modulations , clearly seen in the structure of the disk . because we include emission and absorption only inside of @xmath18 , the interior of the disk can be seen in figure [ plotthree]f as a bright band at the outer edge . while this effect is not physical ( in a real accretion disk it would be blocked by the outer portion of the disk ) , it nonetheless serves as a useful imaging device and we believe it does not significantly affect the qualitative behavior of the light curve . one piece of evidence in support of this claim is the lack of any significant excess power at the orbital frequency of @xmath18 in the high - inclination light curves . just as the optically thin emission model can be used to track surface density in the disk , the thermal model is able to measure the local density and temperature at different depths of the disk . in this way , the ray - tracing analysis provides us with both _ literal _ and _ figurative _ x - ray vision of the disk . the regions of highest temperature ( typically closer to the disk midplane ) produce photons in the high - energy tail that can pass directly through the disk to the observer ( @xmath119 for @xmath120 ) . conversely , since the disk is optically thick to the low - energy rayleigh - jeans tail of the thermal spectrum , the low - energy photon emission is a direct measurement of the temperature at the surface of the disk , where the intensity is proportional to @xmath121 . this 3-dimensional imaging method is demonstrated in figure [ plotfour ] for the same disk snapshot shown in figure [ plotthree]e , now divided into three different energy bands : 1 , 10 , and 100 kev . the thermal emission model has an inner disk temperature @xmath122 kev , so the three plots sample the low- and high - energy tails of the spectrum and the thermal peak . as described above , the 1 kev emission shown in figure [ plotfour]a probes the surface temperature of the disk , highlighting the inner - most region where the gas is hottest . the peak of the thermal spectrum near @xmath123 kev ( fig.[plotfour]b ) of course produces the majority of disk emission and thus more closely resembles the integrated emission shown in figure [ plotthree]e . since the disk is actually optically thin to the high - energy photons at 100 kev , they can be used to trace the regions of greatest temperature and over - density near the midplane ( fig . [ plotfour]c ) . we find that these regions are distributed nearly uniformly in radius throughout the disk . merging the ray - tracing methods described above with the time - dependent output from the mhd simulations , we are able to produce simulated light curves of the accretion disk . for each of the three emission models , we plot in figure [ plotfive ] the light curves for three different inclination angles . a total of 870 data frames have been sampled , spaced at coordinate time steps of @xmath124 , corresponding to a total simulation time of just over 200 ms for a @xmath27 black hole , or about fifty orbits at the isco . this is a remarkably small amount of `` real '' time for a simulation that takes hundreds of hours to run on a major supercomputer . nonetheless , we believe it is still sufficiently long enough to characterize the broad - band features in the power spectrum , and additionally , as we show below , to provide interesting constraints on the possible presence of high - frequency qpos . each of the light curves in figure [ plotfive ] has been normalized and , for inclinations @xmath125 and @xmath126 , shifted vertically in order to facilitate comparison . additionally , we have applied a `` pre - whitening '' filter to each light curve , removing linear secular trends and fixing @xmath127 in order to avoid spurious high - frequency noise in the fourier power spectra . for the optically thin line emission model , shown in figure [ plotfive]a , it is clear that the same large - amplitude , low - frequency features appear in phase for all inclinations . these features most closely represent the total mass within @xmath117 , a variable that is primarily driven by the mass flux in the outer regions of the disk , and of course is independent of inclination angle . the higher - frequency variability is due to regions of overdensity moving through the gravitational potential and also getting beamed towards and away from the observer for higher inclinations . thus the short - term variability is greater for inclinations of @xmath128 and @xmath115 , which is evident from inspection of figure [ plotfive]a . for the other emission models , the low - frequency correlation between different inclinations is still present , but weaker than in the optically thin case . this is quite reasonable , as the optically thick models are more sensitive to localized perturbations that appear different to observers at different inclinations . similarly , there is clearly more fractional variation on short time scales , again due to the increased importance of local perturbations : if a hot spot rises to the top of the disk and then disappears again on a dynamical time scale , the effect will be greater for an optically thick disk than a transparent one . furthermore , the light curves from different emission models do not appear to be correlated with each other , confirming the presumption that they probe fundamentally different properties of the disk . moving beyond simple visual inspection of the light curves , we plot in figure [ plotsix ] the power spectral density , in units of @xmath129 , for the same light curves plotted in figure [ plotfive ] . as suggested by the discussion above , we find significantly more high - frequency power in the optically thick and thermal emission models . within each model , we find more high - frequency power for the systems with higher inclinations . the total power for each light curve is listed in table [ table_power ] , as well as the amount of high - frequency @xmath130 power . in all the plots shown in figure [ plotsix ] , the power spectra can be described by a steep power - law in frequency . there is some evidence of a flatter slope below @xmath131 hz , but this is quite possibly a spurious effect due to the short duration of the simulation , so that we have poor statistical sampling of low - frequency features in the light curves . we have listed in table [ table_power ] the best - fit power - law indices for each power spectrum above 50 hz , mostly in the range @xmath132 . while the spectra are somewhat flatter for the optically thick and thermal models , all appear steeper than those seen in _ rxte _ spectra , which are more typically around @xmath133 @xcite ( however it should be noted that the data become increasingly poor above @xmath134 hz for most observations in the thermal state ) . at the same time , the simulations of @xcite , where emission traces the vertically integrated stress , produce power spectra with slopes of @xmath135 . in all cases the slope appears to be independent of the inclination angle , and a function only of the emission model . for each emission model , the total rms amplitude of the light curve fluctuations increases with inclination , a result that was also found by @xcite . in fact , this may be the most robust measure with which to compare the simulated light curves with observations . for a broad range of possible emission mechanisms , we find that the mhd turbulent disks consistently appear more variable when viewed at high inclinations , a prediction that should be testable with archived _ rxte _ data . as shown in @xcite , the power spectrum from a single radius ( i.e. annulus of the disk ) should show a clear break frequency at roughly the orbital frequency at that radius . following that approach , we show in figure [ plotseven ] the power spectra from five different radii in the disk . again we use the thermal emission model and a face - on inclination @xmath114 so as not to introduce spurious features from the direct emission from the disk interior ( artificially visible due to cutting off absorption outside the given annulus ) . the radii sampled in figure [ plotseven ] have @xmath136 , @xmath137 , @xmath138 , @xmath139 , and @xmath140 . there is some visible evidence for broken power laws , with lower break frequencies for annuli at larger radius , but again the short simulation time severely limits our confidence in the behavior of the power spectra at low frequencies . more statistically significant is the difference between the power - law slopes at high frequency , where the inner regions of the disk clearly have flatter power spectra . the inner regions also produce much more total power , dominating the overall variance in the integrated flux . these results are summarized in table [ annuli_power ] , which lists the best - fit power - law indices above and below 100 hz , as well as the integrated power for each annulus . even if we were to include emission only from the inner - most regions of the disk , the corresponding power spectra would still be significantly steeper than those seen in the _ rxte _ observations . the discrepancy with the the @xmath141 result of @xcite should also be investigated in future work in order to understand the relative importance of the simulation details compared to the post - processor emission models in producing a given power - law slope . in fact , these discrepancies provide one of the strongest arguments demonstrating the need of an accurate radiation transfer post - processor for the mhd simulations . by looking solely at the variation of the code variables , as many previous works have done , one can not hope to reproduce the light curves and power spectra that are actually observed from the disk . to agree quantitatively with the data will require either new physics in the mhd simulations or new emission models for the post - processors , presumably with an increased focus on the emission from the corona . additionally , we know electron scattering plays an important role in the radiation transport for hot accretion disks , and therefore geodesic photon paths are not completely realistic . as yet further motivation for understanding the emission processes that produce these power spectra , recent observations of grs 1915 + 105 actually show a much steeper power - law , with @xmath142 @xcite . however , those observations were taken during a relatively hard state ( as opposed to the soft thermal state ) , further complicating direct comparison with our emission models . from visual inspection of the power spectra in figure [ plotsix ] , no clear features ( e.g. peaks , power - law breaks , etc . ) are immediately evident . however , the short simulation time makes inherent statistical fluctuations unavoidable , which could either hide true features or falsely imitate absent ones . we would therefore like to derive a more quantitative understanding of our sensitivity to timing features in the simulated light curves . from the power spectra plotted in figure [ plotsix]c , we can derive an estimate for the theoretical upper limits on the possible existence of high frequency qpos in the mhd simulations . with the high spatial resolution of our ray - tracing post - processor , we can safely assume the power - law behavior of the power spectrum is due to the inherent turbulence of the disk , as opposed to any sort of `` detector '' or finite sampling statistics . further assuming that the power in neighboring frequency bins is uncorrelated , we can treat the variations ( jitter ) in the power spectrum as behaving locally like poisson variants . in the limit of an infinite length simulation , the frequency bins would become infinitesimally small , and thus could be averaged to get an arbitrarily smooth power spectrum , i.e. , the `` inherent '' turbulent power , which we call @xmath143 , modeled as a simple power - law : @xmath144 then the probability distribution function for the power at a given frequency is given by that of a poisson variant : @xmath145 which has mean @xmath146 and variance @xmath147 . we have checked this relation empirically for the power spectra in figure [ plotsix ] and find that the measured variations in @xmath148 do indeed behave according to equation ( [ fpnu ] ) . for this distribution , the observed power in a single frequency bin is expected to be @xmath143 , plus or minus 100% ( hence the critical importance of frequency binning ) . to identify a potential qpo , we would integrate over some range in frequency @xmath149 to @xmath150 . for a simulation duration of time @xmath151 , the number of frequency bins in this range is @xmath152 assuming @xmath143 is roughly constant over this range ( i.e.@xmath153 ) , the total power @xmath154 in the @xmath155 frequency bins should have a distribution than is approximately normal ( central limit theorem for large @xmath155 ) : @xmath156.\ ] ] a potential qpo in the power spectrum would appear as a small amount of excess power above the `` background noise '' @xmath143 . a qpo centered around @xmath157 with a fwhm of @xmath158 and a lorentzian profile @xmath159 has total ( rms / mean ) amplitude of @xmath160 for our significance estimate , we will consider only the power within @xmath161 , which has an rms departure from the background power - law of @xmath162 . from equation ( [ f_ptot ] ) , we see that the total measured power @xmath163 can be used to give a `` 1-sigma '' confidence limit on the qpo power by @xmath164 where @xmath165 is the size of the frequency bin in hertz . more generally , at the `` n - sigma '' significance level , we can rule out any qpo with rms amplitude greater than @xmath166^{1/2}\nonumber\\ & = & \left[2n_\sigma d\nu\sqrt{n_{\rm bin}\sigma^2}\right]^{1/2 } = \left[n_\sigma p_0\nu_0^{-\gamma}\sqrt{\frac{8\delta\nu}{t_{\rm sim}}}\right]^{1/2}.\end{aligned}\ ] ] in terms of the oscillator quality factor @xmath167 , the amplitude limit can be written @xmath168 plugging in some typical numbers from our simulations ( e.g. , the power spectra in fig . [ plotsix]c ) , we have @xmath169 , @xmath170 , and @xmath171 msec . for a typical high - frequency qpo with @xmath172 hz and @xmath173 hz ( @xmath174 ; @xcite ) , we should thus be able to rule out the presence of a qpo with rms amplitude 1% at the 3-sigma significance level . a number of important relationships can be seen from equation ( [ a_significance ] ) . first , the dependence on @xmath157 shows that due to the steep power - law behavior of the background turbulence , with the current simulations it is very difficult to detect or rule out qpos at _ low _ frequencies . for example , a similar qpo with quality factor @xmath174 at @xmath175 hz would require an amplitude of @xmath176 to be detected at the same significance level . narrower qpos ( large @xmath177 , small @xmath178 ) are easier to detect , but not by much , due to the @xmath179 dependence in equation ( [ a_significance ] ) . similarly , a much longer simulation would also not improve sensitivity that much , due to the same weak dependence on @xmath151 . the discussion above provides a simple way to estimate _ a priori _ the amplitude of any potential qpo at a given significance limit . turning the problem around , and without assuming anything about the shape or amplitude of the qpo , one can simply calculate the significance of any power excess found in a localized group of @xmath155 frequency bins . in this approach , we set a fixed value for the quality factor @xmath177 , which in turn gives @xmath180 and thus @xmath181 , from which the significance of any region of excess ( or deficit ) power can be determined by @xmath182 note that this definition of @xmath183 allows for negative significance values ( i.e. deficit power ) and has a mean value of @xmath184 . if the distribution of power in individual frequency bins follows an exponential distribution , which is what we find for our power spectra , the distribution of @xmath183 will have a chi - squared distribution with @xmath185 degrees of freedom . for large values of @xmath155 , this approaches a normal distribution with variance @xmath186 . in figure [ ploteight ] we plot a collection of these significance curves for the thermal emission model with a range of inclination angles and orientations , fixing @xmath187 . for each power spectrum , we determine the power - law background @xmath143 with a least - squares fit to the simulated data . while not immediately obvious in the original power spectra , a number of interesting features appear in the significance analysis . most striking are the peaks in figure [ ploteight]c that appear close to the orbital frequency at the isco , as well as @xmath188 and the first harmonic @xmath189 . the significance of these peaks increases with inclination ( compare with the smaller peaks at the same frequencies in fig . [ ploteight]b , where @xmath125 ) , as predicted by the hot spot model described in @xcite . this result is very intriguing , especially in light of a number of recent observations of black hole binaries that find multiple qpo peaks with a 2:3 frequency ratio @xcite . thus it is quite important to understand the robustness of such a result , if possible without requiring hundreds of hours more of supercomputer simulations . as shown below in section [ az_structure ] , the average density pertubation in the disk has a lifetime of less than a third of an orbital period . therefore two different light curves of the same disk viewed from @xmath190 should be uncorrelated when seen at high inclination , where the emission is dominated by special relativistic beaming from the half of the disk moving towards the observer . similarly , for an optically thick disk , an observer looking at the `` underside '' of the disk ( @xmath191 ) might also see an uncorrelated light curve . we calculated light curves and power spectra with @xmath192 and @xmath193 , and show the significance plots in figures [ ploteight]d - f . the light curves from @xmath194 are indeed uncorrelated . on the other hand , the top - bottom views appear to be strongly correlated . from this we infer that the largest hot spot perturbations extend vertically through the entire disk , and are thus visible from both sides simultaneously . as is clear from figures [ ploteight]d and [ ploteight]f , the 2:3 qpo peaks are not always present , while other peaks may appear at other frequencies . although the transience of this result is somewhat disappointing , it is actually consistent with the data from _ rxte _ , which detects the subtle hfqpos in only a small fraction of all observations @xcite . on the other hand , if the simulations most accurately represent the thermal state ( which is by no means certain , as evidenced by the large discrepancy in theoretical vs. observed power - law slope ) , one would not expect to see high - frequency qpos at all , because they are not generally observed in the state . yet another caveat is that around 230 hz , for @xmath171 msec , @xmath195 , so the distribution of @xmath183 is not quite normal . whereas a random variable with a gaussian distribution would have @xmath196 outliers only @xmath197 of the time , the actual distribution for @xmath198 ( roughly chi - squared with 10 degrees of freedom ) gives @xmath196 results with probability @xmath199 and @xmath200 outliers @xmath201 of the time . thus is it unlikely , but not impossible , to form a few large ( @xmath202 ) peaks from purely random fluctuations . in order to understand better the likelihood of these results , we performed a monte carlo calculation of @xmath203 simulated power - law power spectra . this is a similar approach to that taken by people searching for qpos in x - ray observations of sgr a@xmath204 and agn @xcite . for fixed values of @xmath205 and @xmath206 , we generate a random value for the power at each frequency according to the exponential distribution in equation ( [ fpnu ] ) . then we analyze this simulated power spectrum with the exact same methods described above , generating a plot of significance versus frequency like the ones shown in figure [ ploteight ] . repeating this procedure many times gives the probability of producing a random qpo signal at a given frequency and significance . for @xmath187 , figure [ plotnine ] plots this probability for @xmath207 3 , 4 , 5 , and 6 . the results of the monte carlo calculation appear to be very similar to the analytic estimates presented above , confirming the non - normal nature of the distribution . the `` steps '' or `` edges '' seen in figure [ plotnine ] are simply due to the finite number of frequency bins used in each qpo : for a fixed value of @xmath177 , @xmath155 increases with frequency , and every integer step corresponds to a small change in the significance , as can be seen from equation ( [ n_sigma ] ) . in the end , we are still not quite sure if the 2:3 ratio qpos are `` real , '' and if so , whether they represent the same qpo pairs seen in some observations . analogous to the observations , the only way really to improve our confidence in these results may be to run more simulations and see if the 2:3 peaks are reproducible and if so , with what range of qpo parameters . with more simulated data , we should also be better able to identify the exact mechanism that is causing the qpos . at this point , it seems somewhat unlikely that they are caused by a simple hot spot model as in @xcite , which requires much longer lifetimes for the quality factors seen here . to understand the robustness of these qpo pairs , in addition to longer simulation times , we must also explore the effects of black hole spin , one of the major ingredients in most hfqpo models . this effort , too , will require significant computational resources , but should be more productive than simply running the current schwarzschild simulation longer [ see eqn . ( [ a_significance2 ] ) ] . to gain a deeper understanding of the source of the light curve fluctuations , we must first characterize the behavior of the hydrodynamic fluctuations in the accretion disk . for all three emission models , and particularly the optically thin model , the variations in brightness are closely related to the surface density fluctuations . thus , to analyze the structure and evolution of the hot spots in the disk , we will primarily focus on the surface density @xmath208 defined as @xmath209 for each annulus of constant radius and time , we subtract out the mean and normalize the surface density such that [ normal_sigma ] @xmath210 and @xmath211 to estimate the distribution of hot spot lifetimes , we calculate the correlation between the surface density maps at two different times . the time - symmetric overlap function is defined as @xmath212^{1/2 } + \frac{1}{2}\left[\int d\phi \bar{\sigma}(r,\phi , t ) \bar{\sigma}(r^-,\phi^-,t-\delta t)\right]^{1/2}\ ] ] so that @xmath213 . here the advanced and retarded coordinates @xmath214 and @xmath215 are the spatial positions where a clump of matter at @xmath216 would be at times @xmath217 or @xmath218 . these coordinates are determined by integrating the time- and azimuth - averaged fluid velocities at each radius in the disk . for example , @xmath219 dt\ ] ] and @xmath220 dt,\ ] ] where @xmath221 is the average fluid transport velocity @xmath222 in the disk midplane . this more detailed approach is necessary for the inner disk where the radial velocities are not negligible and the orbital velocities change rapidly with @xmath17 . now the complete correlation function at each radius can be given by averaging over the duration of the simulation : @xmath223 the correlation is normalized so that if there is no evolution of the disk perturbations in the local fluid frame , then @xmath224 . of course , in practice there _ is _ significant evolution , and therefore we expect @xmath225 as the turbulent disk at late times is completely uncorrelated with the initial conditions . then the characteristic decay time for @xmath226 is a measure of the lifetime of a typical hot spot at that radius . to compare lifetimes at different radii , we rescale the time delay @xmath227 by the orbital period @xmath228 at each radius . these correlation functions are plotted in figure [ plotten ] for a range of @xmath17 , including the region inside the isco . remarkably , we find very similar behaviors for hot spots over a large range of radii @xmath229 , with nearly exponential decay at early times with a time constant of @xmath230 this behavior is quite similar to the simple hot spot model developed by @xcite where hot spots were created and destroyed around a single radius with random phases and exponentially - distributed lifetimes , naturally giving lorentzian peaks in the power spectrum at the geodesic frequencies corresponding to the chosen `` resonant '' radius . however , for the mhd simulations , there appears to be no special radius , and even for the potential qpos discussed above in section [ hfqpos ] , the geodesic hot spot model may not be appropriate . in addition to the typical lifetimes for the density perturbations , we have also analyzed the characteristic _ spatial _ dimensions of the hot spots in @xmath17 and @xmath15 . we would ideally like to follow each hot spot in its own locally non - rotating frame to remove the shearing effects of differential rotation in the disk . this would give a better estimate for the `` true '' size and shape of the local perturbations without including the global dynamics of the disk . in fact , there is a simple analytic transformation that can be applied to remove the shear effects from the entire disk simultaneously ( at least the region that behaves in a roughly keplerian way ; @xmath231 in our simulation ) . to construct this transformation , consider a sheared blob extending in @xmath17 from @xmath232 to @xmath233 . the orientation of the blob at its creation can be estimated by shifting each point backwards in time by @xmath234 . this amounts to a shift in azimuth of @xmath235 or @xmath236 where we have assumed a keplerian orbital velocity of @xmath237 ( recall this relation holds even for relativistic orbits in the schwarzschild metric ) . integrating equation ( [ ddphi_dr ] ) gives the azimuthal coordinate transformation @xmath238 , with @xmath239 figure [ ploteleven ] shows the effect of applying this transformation to a snapshot of the accretion disk surface density . in the `` before '' image ( fig . [ ploteleven]a ) , the differential rotation is clearly present , as material in the inner disk gets swept along faster , and shears the overdensities into diagonal arcs in the @xmath240 plane . after performing the transformation defined above , the blobs appear rather uniform in shape and orientation ( fig . [ ploteleven]b ) . finally , to compare the aspect ratios of blobs at different radii in the disk , we perform a simple coordinate transformation in the radial direction as well : @xmath241 as we will show , these coordinates highlight the invariance of @xmath242 for density perturbations throughout the disk . given a surface density map @xmath243 for each frame in time of the simulation , we normalize to get @xmath244 as in equation ( [ normal_sigma ] ) for all values of @xmath104 and @xmath33 . then , for each time step we create two partial fourier transforms , one in each spatial direction : @xmath245 and @xmath246 averaging ( in quadrature ) over the duration of the simulation gives @xmath247 and @xmath248 with this construction , @xmath249 is the time - averaged power spectrum of radial variations at a fixed azimuth @xmath250 . similarly , @xmath251 is the average azimuthal power for a fixed radius @xmath104 . figure [ plottwelve ] shows examples of these power spectra for representative values of @xmath104 and @xmath250 . it is clear from these plots that the variance @xmath252 can be well approximated by a broken power law for variations in both the radial and azimuthal directions . the `` break wavenumber '' gives the characteristic maximum size of a density perturbation . since we find the peak amplitude of the perturbations increases with their spatial extent , it is reasonable that the variance peaks at this break wavenumber . for each value of @xmath250 we fit the power spectrum @xmath249 to a broken power - law in @xmath253 and find an average value for @xmath254 . applying a similar method to @xmath251 gives an average azimuthal break wavenumber of @xmath255 . these values correspond to wavelengths of @xmath256 ( giving @xmath257 ) and @xmath258 , or an aspect ratio for the hot spots of @xmath259 . as we saw above in section [ hfqpos ] , the light curves from the top- and bottom - views of the optically thick disk are highly correlated . this suggests that the vertical extent of the density perturbations is limited by the disk thickness , which has a roughly constant value of @xmath260 ( see fig . [ plotone]d ) . thus the density perturbations throughout the disk are self - similar ellipsoids with principle axes in the ratio @xmath261 . to get a feeling for the stability of this wavenumber analysis , we plot in figure [ plotthirteen ] the best - fit parameters for the broken power - law at each slice in @xmath17 or @xmath250 . because of the rotational symmetry of the accretion system , it is expected that the fit parameters should be roughly constant in @xmath250 , so the curves in figures [ plotthirteen]d , e , f really give an estimate of the typical errors in our analysis . somewhat more surprising is the near constancy of the fit parameters as a function of @xmath262 , emphasizing the fact that density perturbations at all radii have similar extent in azimuth . lastly , the fact that the curves in figure [ plotthirteen ] do not appear as purely random fluctuations around the mean highlights the level of correlation between neighboring points in radius / azimuth . we have developed a generalized post - processor analysis tool that couples ray - tracing in the kerr metric with global grmhd simulations to produce light curves and power spectra of accreting black holes . using a variety of emission models , we probe different regions of the accretion disk and are better able to understand the underlying causes of time variability in the observed flux . the optically thick blackbody emission / absorption model is particularly useful for understanding the behavior of stellar - mass black holes in the thermal dominant state , which is characterized by a broad peak in the photon energy spectrum and very low levels of variability . by fixing the black hole mass and accretion rate , we can convert from dimensionless code units to physical units for local fluid density . then , assuming a radiation - pressure dominated gas in the inner disk , we can derive a physical temperature , and thus emissivity and absorption coefficients . the temperature and scale height of the disk compare well with the novikov - thorne predictions for a schwarzschild black hole accreting at 50% @xmath263 with a small torque at the inner boundary . despite the very different assumptions used in the analytic model and the computer simulations , the agreement is close enough to provide reasonable confidence in our conversion factors for density and temperature , and thus the conclusion that the simulations can be used to understand the thermal dominant state . we have also developed new methods for analyzing the azimuthal structure of the accretion flow , determining the characteristic sizes and lifetimes of density perturbations . over a large range of radii , we found the perturbations have a nearly exponential distribution of lifetimes , with @xmath264 . also , the characteristic shapes of the hot spots appears to be self - similar throughout the keplerian regions of the disk , with @xmath265 and @xmath266 . from these short coherence times , it seems clear that the hot spots formed by mhd turbulence can not survive long enough to produce the qpos with quality factors of @xmath267 described in @xcite . thus , if the transient high frequency qpos with 2:3 frequency ratios identified in section [ hfqpos ] are in fact robust signatures of grmhd disk dynamics , they are most likely _ not _ produced by geodesic hot spots . this conclusion is supported by the fact that the qpos are seen only from a single azimuthal viewing angle at any one time , perhaps suggesting some form of localized stationary wave in the disk . finally , as part of our search for qpos in the simulation data , we have also developed a general formalism for quantifying the significance of such features in simulations . these analytic results have been combined with monte carlo calculations to estimate our confidence limits for the qpos at @xmath268 . as argued throughout this paper , we believe the mhd simulations most closely resemble the thermal state of black hole x - ray binaries . this high - luminosity state is dominated by a broad thermal peak in the energy spectrum around @xmath269 kev and a relatively small contribution from a steep power - law tail at higher energies @xcite . the timing properties are characterized by a featureless power spectrum with @xmath270 , with typical power - law index @xmath271 . the total power in this state is also small , with ( rms / mean)@xmath272 hz@xmath273 above 1 hz . our simulated power spectra predict @xmath274 , which is rather greater than the typical slope . on the other hand , just this sort of steep power spectrum has been seen in the hard state observations of @xcite . one robust prediction of the mhd / ray - tracing simulations is that , independent of the emission mechanism , the integrated rms power increases with disk inclination . in principle , this should be an observable prediction that could be tested with _ data from black hole binaries in the thermal state . of course , there are complications when comparing two different binary systems , including different relative contributions to the thermal peak or power - law tail , different total luminosities , and different black hole masses . however , with a sufficiently large number of observations , it should be possible to account for these variables and extract a correlation between integrated spectral power and inclination . other observational trends that might be explored with future mhd simulations include the observed linear relation between the x - ray flux and rms , which appears to extend to agn as well @xcite . the timing properties of black hole binaries also vary greatly between their different spectral states , a general feature that is still not understood . to explore either of these relationships , we will require more sophisticated emission models and , quite likely , more simulation data with a broader range of disk parameters . however , even with improved simulations and light curve models , there still exist some inherent difficulties in comparing theory with observation . _ rxte _ typically requires thousands of seconds of observation to detect high - frequency qpos at rms amplitudes of a few percent ( limited largely by photon counting statistics and `` dead time '' corrections ) . to achieve a similar sensitivity , our mhd simulations require hundreds of hours corresponds to generate only 0.2 seconds of real time for a @xmath96 black hole ( we have perfect photon statistics , but are limited by the inherent variance in the power spectra of short time series ) . this is more closely analogous to the time scales of agn observations . if we were to scale the simulation to an agn of mass @xmath275 , it would correspond to a sampling rate of every 40 minutes for roughly a month , assuming _ perfect _ photon counting statistics ! thus it should be no surprise that qpos are still so difficult to detect with high significance from supermassive black holes @xcite . 99 abramowicz , m. a. , & kluzniak , w. 2001 , a&a , 374 , l19 agol , e. , & krolik , j. h. 2000 , apj , 528 , 161 armitage , p. j. , & reynolds , c. s. 2003 , mnras , 341,1041 balbus , s. a. , & hawley , j. f. 1998 , rev . phys . , 70 , 1 belanger , g. , et al . 2006 , apj , submitted ; pre - print [ astro - ph/0604337 ] belloni , t. , et al . 2006 , mnras , in press ; pre - print [ astro - ph/0603210 ] benlloch , s. , et al . 2001 , apj , 562 , l121 de villiers , j .- , & hawley , j. f. 2003 , apj , 592 , 1060 de villiers , j .- p . , hawley , j. f. , & krolik , j. h. 2003 , apj , 599 , 1238 done , c. , madejski , g. m. , & zycki , p. t. 2000 , apj , 536 , 213 evans , c. r. , & hawley , j. f. 1988 , apj , 332 , 659 fabian , a. c. , et al . 1995 , mnras , 277 , l11 fuerst , s. v. , & wu , k. 2004 , a&a , 424 , 733 gammie , c. f. , mckinney , j. c. , & toth , g. 2003 , apj , 589 , 444 hawley , j. f. , & krolik , j. h. 2001 , apj , 548 , 348 hawley , j. f. , & krolik , j. h. 2002 , apj , 566 , 164 homan , j. , et al . 2005 , apj , 623 , 383 king , r. e. , et al . 2001 , phys . rev . a , 64 , 053810 laor , a. 1991 , apj , 376 , 90 machida , m. , & matsumoto , r. 2003 , apj , 585 , 429 markowitz , a. , & edelson , r. 2004 , apj , 617 , 939 matthews , a. p. , & scheuer , p. a. g. 1990 , mnras , 242 , 623 mcclintock , j. e. , & remillard , r. a. 2005 , in _ compact x - ray sources _ , ed . w. h. g. lewin & m. van del klis , preprint [ astro - ph/0306213 ] mckinney , j. c. , & gammie , c. f. 2004 , apj , 611 , 977 miller , j. m. , et al . 2001 , apj , 563 , 928 miller , j. m. , et al . 2002 , apj , 578 , 348 petri , j. 2005 , a&a , 439 , 443 pollak , g. d. , delameter , n. d. , nash , j. k. , & hammel , b. a. 1994 , jqsrt , 51 , 303 rebusco , p. 2004 , pasj , 56 , 553 remillard , r. a. , et al . 2002 , apj , 580 , 1030 reynolds , c. s. , & begelman , m. c. 1997 , apj , 488 , 109 rezzolla , l. , yoshida , si . , maccarone , t. j. , & zanotti , o. 2003 , msras , 344 , l37 rybicki , g. b. , & lightman , a. p. 1979 , _ radiative processes in astrophysics _ ( new york : wiley - interscience ) schnittman , j. d. 2005 , apj , 621 , 940 schnittman , j. d. , & bertschinger , e. 2004 , apj , 606 , 1098 schnittman , j. d. , & craxton , r. s. 2000 , phys . plasmas 7 , 2964 schnittman , j. d. , & rezzolla , l. 2006 , apj , 637 , l113 smith , m. d. , norman , m. l. , winkler , k .- h . a. , & smarr , l. 1985 , mnras , 214 , 67 strohmayer , t. e. 2001a , apj , 552 , l49 strohmayer , t. e. 2001b , apj , 552 , l169 tregillis , i. l. , jones , t. w. , & ryu , d. 2001 , apj , 577 , 475 uttley , p. 2004 , mnras , 347 , l61 uttley , p. , mchardy , i. m. , & vaughan , s. 2005 , mnras , 359 , 345 vaughan , s. , & uttley , p. 2005 , mnras , 362 , 235 vanden berk , d. e. , et al . 2001 , aj , 122 , 549 van der klis , m. 2005 , in _ compact x - ray sources _ , ed . w. h. g. lewin & m. van del klis , preprint [ astro - ph/0410551 ]
we investigate the afterglow properties and large - scale environments of several short - duration gamma - ray bursts ( grbs ) with sub - arcsecond optical afterglow positions but no bright coincident host galaxies . the purpose of this joint study is to robustly assess the possibility of significant offsets , a hallmark of the compact object binary merger model . five such events exist in the current sample of 20 short bursts with optical afterglows , and we find that their optical , x - ray , and-ray emission are systematically fainter . these differences may be due to lower circumburst densities ( by about an order of magnitude ) , to higher redshifts ( by ) , or to lower energies ( by about a factor of 3 ) , although in the standard grb model the smaller-ray fluences can not be explained by lower densities . to study the large - scale environments we use deep optical observations to place limits on underlying hosts and to determine probabilities of chance coincidence for galaxies near each burst . in 4 of the 5 cases the lowest probabilities of chance coincidence ( ) are associated with bright galaxies at separations of , while somewhat higher probabilities of chance coincidence are associated with faint galaxies at separations of . by measuring redshifts for the brighter galaxies in three cases ( ) we find physical offsets of kpc , while for the faint hosts the assumption of leads to offsets of kpc . alternatively , the limits at the burst positions ( mag ) can be explained by typical short grb host galaxies ( l ) at . thus , two possibilities exist : ( i ) of short grbs explode kpc or kpc from the centers of or galaxies , respectively , and have fainter afterglows due to the resulting lower densities ; or ( ii ) of short grbs occur at and have fainter afterglows due to their higher redshifts . the high redshift scenario leads to a bimodal redshift distribution , with peaks at and . the large offset scenario leads to an offset distribution that is well - matched by theoretical predictions of ns - ns / ns - bh binary kicks , or by a hybrid population with globular cluster ns - ns binaries at large offsets and primordial binaries at offsets of kpc ( indicative of negligible kicks ) . deeper constraints on any coincident galaxies to mag ( using the _ hubble space telescope _ ) will allow us to better exclude the high - redshift scenario .
1007.0003
the bimodality of gamma - ray burst ( grb ) durations @xcite is indicative of separate progenitor populations for long- and short - duration grbs . while direct observational support exists for the massive star origin of long grbs ( e.g. , @xcite ) , the most popular progenitor model for short grbs is the coalescence of compact object binaries with neutron star and/or black hole constituents ( ns - ns / ns - bh ; @xcite ) . one of the key predictions of this model is that some systems will experience large velocity kicks at birth , leading to eventual mergers well outside of the host galaxies , in galactic halos and the intergalactic medium @xcite . these models predict that @xmath20 of all mergers will occur at offsets of @xmath21 kpc for milky way mass galaxies . in such environments the resulting afterglow emission is expected to be fainter than for bursts occurring in coincidence with their host galaxies due to the low ambient density @xcite . a subset of ns - ns binaries ( @xmath22 ) may be formed dynamically in globular clusters @xcite , leading to possible large offsets and low ambient densities despite an absence of kicks . the distribution of offsets for such binary systems can in principle be calculated from the spatial distribution of globular clusters , and initial predictions are that @xmath23 of these systems will have offsets of @xmath21 kpc ( depending on the mass of the galaxy ; @xcite ) . an additional expectation is that dynamically - formed binaries will be heavily skewed to lower redshifts due to the additional time delay between the formation and core - collapse of the globular clusters @xcite . other progenitor systems for short grbs have also been proposed , including young magnetars @xcite , accretion - induced collapse ( aic ) of neutron stars @xcite , and delayed magnetar formation through binary white dwarf mergers or white dwarf aic @xcite . these models are partially motivated by observations that can not be easily accommodated in the standard ns - ns merger model , such as extended soft @xmath0-ray emission on timescales of @xmath24 s ( e.g. , @xcite ) , or by phenomena such as short - duration giant flares from soft @xmath0-ray repeaters ( e.g. , @xcite ) . the general expectation is that these alternative progenitors will not experience kicks , and will therefore lead to bursts in coincidence with host galaxies . the detection of short grb afterglows starting in mid-2005 provided an opportunity to investigate the various progenitor models through a range of observational tests : the redshift distribution @xcite , the host galaxy demographics @xcite , the afterglow properties @xcite , and perhaps most importantly , their locations relative to the host galaxies @xcite . as of mid-2010 , x - ray and optical afterglows have been detected from 40 and 20 short grbs , respectively , with the latter sample providing accurate sub - arcsecond positions . of these 20 events , 15 directly coincide with host galaxies with a wide distribution of apparent magnitudes , and redshifts of @xmath25 or beyond ( e.g. , @xcite ) . however , the remaining five events do not appear to coincide with galaxies , and therefore provide an opportunity to assess the possibility of large progenitor offsets , and to test the validity of the ns - ns merger models . significant offsets have been claimed previously , in particular for grbs 050509b and 060502b with projected offsets of @xmath26 and @xmath27 kpc , respectively @xcite . however , in both cases only x - ray positions are available ( @xmath28 and @xmath29 radius , respectively ) , and the error circles contain several galaxies consistent with a negligible offset @xcite . moreover , in the case of grb050509b the x - ray error circle intersected the outer regions of the host , raising the possibility that the progenitor system was formed in , rather than kicked to , the outskirts of the host . this possibility raises a crucial point , namely that a substantial physical offset from the center of the host does not necessarily point to a progenitor kick if the burst still closely coincides with the host light distribution @xcite . an illustrative example of this point is grb071227 whose optical afterglow position coincides with the outskirts of edge - on disk galaxies , with an offsets of about 15 kpc from the host center @xcite . large offsets have also been speculated in a few cases with precise optical afterglow positions ( grbs 061201 , 070809 , and 080503 ; @xcite ) . however , these claims have not been investigated systematically , mainly because they were treated on a case - by - case basis , with probabilistic arguments that prevented conclusive associations . these cases , combined with the ambiguity inherent to x - ray positions , demonstrate that bursts with optical afterglows are essential for reaching any robust conclusions about progenitor offsets ( due to kicks and/or a globular cluster origin ) . here we present the first systematic study of short grbs with optical afterglows and no coincident hosts , which combines their afterglow properties with the large - scale environments . the purpose of this study is to statistically assess the possibility of offsets and to compare this with alternative explanations ( e.g. , a high redshift origin ) . to achieve this goal we set our study in the broader context of short grbs that have optical afterglows _ and _ coincident hosts , as well as short grbs with only x - ray positions . as we demonstrate throughout the paper , such a combined study is essential since offsets or high redshifts are expected to jointly affect both the afterglow properties and the large - scale environments . our study also provides a robust statistical assessment of _ a posteriori _ chance coincidence probabilities , and the expected number of spurious associations , which can not be properly assessed for individual bursts . the paper is organized as follows . in [ sec : obs ] we present deep optical observations of the environments of grbs 061201 , 070809 , 080503 , 090305 , and 090515 , as well as spectroscopic observations of bright galaxies near the burst positions for grbs 061201 , 070809 , and 090515 . we study trends in the afterglow and prompt emission properties of bursts with and without coincident hosts in [ sec : ag ] , and determine _ a posteriori _ probabilities of chance coincidence as a function of projected angular offset for galaxies near the position of each burst in [ sec : hosts ] . we also determine projected physical and host - normalized offsets , and use these in conjunction with the afterglow and prompt emission properties to address two scenarios for the short bursts with optical emission and no coincident bright hosts : ( i ) an origin in faint galaxies at @xmath12 ( [ sec : highz ] ) , or ( ii ) substantial offsets from galaxies at @xmath30 or @xmath7 ( [ sec : kick ] ) . finally , in [ sec : disc ] we present the offset distribution of all short grbs with optical afterglows ( in the context of scenario ii ) , and compare this distribution with predictions for ns - ns kicks and dynamically - formed ns - ns binaries . we further investigate whether the circumburst densities that are required for the measured optical magnitudes can be accommodated with a halo or igm origin . we draw conclusions about the progenitors of short grbs from our systematic study in [ sec : conc ] . we include in this investigation all 20 short grbs with optical afterglow detections as of june 2010 . this is the full subset of events for which sub - arcsecond positions are available . we stress that this sample represents only about 1/3 of all short grbs discovered to date , and about 1/2 of the sample with x - ray afterglow detections . thus , it is not a complete sample of short grbs , but it is the only subset for which we can investigate the possibility of large offsets with meaningful statistical significance . as we demonstrate in [ sec : disc ] , we do not expect this sample to be strongly biased with respect to circumburst density , at least for @xmath31 @xmath32 . the properties of the 20 short grbs , as well as events with deep optical limits , are summarized in table [ tab : shbs ] . as can be inferred from the table , some of the 20 events with only x - ray detections do not have optical follow - up observations , suggesting that the sample with optical afterglows may be largely representative . for the purpose of our investigation we define three sub - samples that will be used throughout the paper : ( i ) _ sample 1 _ : short grbs with detected afterglows and coincident host galaxies ( 15 bursts ) ; ( ii ) _ sample 2 _ : the 5 short bursts with detected optical afterglows and no bright coincident hosts ; and ( iii ) _ sample 3 _ : short grbs with detected x - ray afterglows ( from the _ swift _ x - ray telescope ) but no optical detections despite rapid follow - up observations ( 11 bursts ) . for the bursts in _ sample 2 _ we use deep space- and ground - based optical observations to place limits on the brightness of underlying galaxies , and to assess the probability of chance coincidence for nearby galaxies . for grb061201 we use _ hubble space telescope _ ( hst ) observations obtained with the advanced camera for surveys ( acs ) in the f814w filter , with a total exposure time of 2244 s @xcite . for grb070809 we use @xmath33-band observations obtained on 2008 january 14 ut with the low dispersion survey spectrograph ( ldss3 ) mounted on the magellan / clay 6.5-m telescope , with a total exposure time of 1500 s. for grb080503 we use hst wide - field planetary camera 2 ( wfpc2 ) observations in the f606w filter from 2009 january 30 ut , with a total exposure time of 4000 s @xcite . we also use the limits on a coincident host inferred from a deeper stack of hst observations by @xcite , with @xmath34 mag . for grb090305 we use ldss3 @xmath33-band observations obtained on 2010 may 8 ut with a total exposure time of 2400 s. finally , for grb090515 we use @xmath33-band observations obtained with the gemini multi - object spectrograph ( gmos ) mounted on the gemini - north 8-m telescope from 2009 may 15 ut with a total exposure time of 1800 s. the ground - based observations were reduced and analyzed using standard routines in iraf . the analysis of the hst observations is detailed in @xcite . the limiting magnitudes for all five observations are listed in table [ tab : nohost ] , and images of the five fields are shown in figures [ fig:061201][fig:090515 ] . in addition to the imaging observations , we obtained spectroscopic observations of galaxies near the positions of grbs 061201 , 070809 , and 090515 . our observations of a galaxy located @xmath35 from the afterglow position of grb061201 ( marked `` s4 '' in figure [ fig:061201 ] ) revealed a star forming galaxy at @xmath36 @xcite ; see figure [ fig:061201spec ] . for grb070809 we obtained spectra of two galaxies located @xmath37 and @xmath38 from the optical afterglow position ( marked `` s2 '' and `` s3 '' , respectively in figure [ fig:070809 ] ) using ldss3 on 2008 january 14 ut . the galaxy at a separation of @xmath37 was previously identified as a star forming galaxy at @xmath39 by @xcite . here we find that the object at a separation of @xmath38 is an early - type galaxy at @xmath40 , with no evidence for on - going star formation activity ( figure [ fig:070809spec ] ) . finally , for grb090515 we obtained multi - object spectroscopic observations with ldss3 for nearly 100 galaxies within a @xmath41 field centered on the grb position . these observations provide redshifts for several galaxies near the host position , including a star forming galaxy at @xmath42 ( @xmath43 offset ; `` s1 '' in figure [ fig:090515 ] ) , an early - type galaxy at @xmath44 ( @xmath45 offset ; s5 ) , and a star forming galaxy at @xmath46 ( @xmath47 offset ; `` s6 '' ) ; see figure [ fig:090515spec ] . in an upcoming paper we will demonstrate that the galaxy at @xmath44 is a member of a cluster . we begin our investigation by assessing the distribution of optical afterglow magnitudes for the three samples defined in [ sec : obs ] . the observed magnitudes and limits as a function of time after the burst are shown in figure [ fig : optag1 ] . the median observation time for the sample is about 8.5 hr after the burst . the distribution of detected optical afterglow magnitudes is broad , ranging from @xmath48 to @xmath49 mag . the limits range from @xmath50 to @xmath51 mag ( with the exception of grb080702 which only has a shallow limit of about 21 ab mag : @xcite ) . the mean brightness and standard deviation for _ sample 1 _ are @xmath52 mag . this is substantially fainter , by about an order of magnitude , than the afterglows of long grbs on a comparable timescale ( e.g. , @xcite ) . it is also remarkably similar to the prediction of @xmath53 mag at @xmath54 hr by @xcite . since the available limits are at least @xmath55 mag , we conclude that the bursts lacking optical detections are drawn from a population with fainter afterglows . for the 5 bursts in _ sample 2 _ we find a median and standard deviation of @xmath56 mag , about 1.4 mag fainter than the bursts with coincident hosts . the cumulative distributions for the three samples normalized to the fiducial time of @xmath57 hr after the burst are shown in figure [ fig : optag2 ] . a kolmogorov - smirnov ( k - s ) test indicates that there is only an @xmath58 probability that _ sample 1 _ and _ sample 2 _ are drawn from the same underlying distribution of optical afterglow brightnesses . similarly , the probability that _ sample 1 _ and _ sample 3 _ are drawn from the same underlying distribution is @xmath59 ( an upper limit since the bursts in _ sample 3 _ are not detected in the optical ) . on the other hand , the probability that _ sample 2 _ and _ sample 3 _ are drawn from the same underlying distribution is high , @xmath60 . the overall faintness of the optical afterglows in _ sample 2 _ and _ sample 3 _ can be explained in two primary ways . first , they could result from systematically lower circumburst densities . in the standard afterglow model with @xmath61 , the afterglow flux scales as @xmath62 for a uniform medium @xcite . thus , a difference of about @xmath63 mag can be explained with a density that is about an order of magnitude lower than for the bursts in _ sample 1_. alternatively , the fainter fluxes may be due to higher redshifts for _ sample 2 _ and _ sample 3 _ compared to _ sample 1 _ since the optical flux also scales as @xmath64 , where @xmath65 is the luminosity distance . the @xmath63 mag difference corresponds to @xmath66 ( @xmath67 ) for a _ sample 1 _ mean redshift of @xmath68 ( @xmath69 ) . similarly , the flux also depends on the total energy , with @xmath70 , and therefore a difference of @xmath63 mag can be explained with an energy release lower by about a factor of 3 . the various scenarios ( lower density , lower energy , or higher redshift ) can be further explored through a comparison of the prompt @xmath0-ray emission , and the relation between the optical and x - ray afterglow brightness . in the framework of the standard grb model we do not expect lower densities to impact the prompt emission since it is expected to be produced by internal processes ( shocks or magnetic dissipation ) that do not depend on the external medium . on the other hand , lower energies or higher redshifts will tend to systematically affect the prompt and afterglow emission . in figure [ fig : fg_fx_t90 ] we plot the distributions of @xmath0-ray fluence ( @xmath71 ) , afterglow x - ray flux at the fiducial time of 8 hr ( @xmath72 ) , and duration ( @xmath73 ) for the three samples . these distributions allow us to explore the underlying reason for the difference in optical afterglow brightness between the three samples . first , we find that the distribution of @xmath71 values for _ sample 1 _ has a mean value that is about a factor of 5 times larger than for _ sample 2 _ and _ sample 3_. this is indicative of higher redshifts for the latter two samples if the isotropic - equivalent energies of all short grbs are similar , or alternatively a lower energy scale if the redshifts are similar . in the same vein , we find that the distributions of @xmath72 values for _ sample 2 _ and _ sample 3 _ have lower means than for _ sample 1_. this result is indicative of overall fainter afterglow emission for the former two samples , and this can again be explained in the context of lower energies or higher redshifts . unlike in the case of the @xmath0-ray emission , a lower circumburst density would also account for the fainter x - ray fluxes if @xmath74 . finally , we find that the durations of the bursts in _ sample 2 _ and _ sample 3 _ are shorter by about a factor of 2 compared to the events with optical afterglows and coincident hosts , although there is substantial scatter in all three samples . the shorter durations are not trivially explained in the context of lower energies , higher redshifts , or lower densities . to explore this result in more detail we plot the observed fluence as a function of duration for the events in all three samples ( figure [ fig : t90_fg ] ) . we find that there is a mild positive correlation between the two quantities , but that the events in _ sample 1 _ appear to have larger fluences at a given duration compared to the events in _ sample 2 _ and _ sample 3_. this is indicative of lower @xmath0-ray fluxes for the latter two samples , possibly as a result of higher redshifts . higher redshifts will also shift the intrinsic durations of the bursts in _ sample 2 _ and _ sample 3 _ into better agreement with the bursts in _ sample 1_. we therefore conclude that the differences in prompt emission and optical / x - ray afterglow properties are consistent with a higher redshift origin for the bursts in _ sample 2 _ and _ sample 3_. the fainter afterglow emission is also consistent with lower density environments for these two samples , although this does not clearly explain the differences in prompt emission ( at least in the framework of the standard grb model ) . we return to the discussion of low density versus a high redshift origin in [ sec : hosts ] when we investigate the host galaxy properties . in addition to the overall faintness of the optical and x - ray afterglows , a substantial difference in density may also be imprinted on the _ ratio _ of optical to x - ray brightness . this is because the synchrotron cooling frequency depends on density as @xmath75 , and is therefore expected to transition across the x - ray band as the density decreases . for @xmath76 the x - ray and optical bands occupy the same portion of the synchrotron spectrum , with a resulting spectral index of @xmath77 to @xmath78 while for @xmath79 ( i.e. , high density ) , the spectrum between the two bands will be steeper , reaching a maximum value of @xmath80 when @xmath81 . in figure [ fig : opt_xray ] we plot the x - ray flux versus optical magnitude for all three samples . for each burst the fluxes in the optical and x - rays are taken at similar times after the burst , or extrapolated to a common time . the correction factors due these extrapolations are generally @xmath82 ( table [ tab : shbs ] ) . for the combined _ sample 1 _ and _ sample 2 _ we find a clear correlation between the fluxes in the two bands , leading to a mean optical to x - ray spectral index of @xmath83 . this is essentially indistinguishable from the ratio for long grbs , @xmath84 @xcite . the median values are consistent with @xmath85 , but exhibit dispersion that is likely due to scatter in the values of @xmath86 and/or the location of @xmath87 relative to the x - ray band . the similarity of @xmath88 for long and short grbs does not necessarily indicate that the densities are similar for the two samples . in particular , if @xmath87 is located close to the x - ray band for long grbs , while for short grbs @xmath89 ( due to a lower density ) , the effect on @xmath88 will be marginal , particularly within the overall observed scatter . for example , with @xmath90 the difference in @xmath88 between a model with @xmath87 exactly intermediate between the optical and x - ray bands , and a model with @xmath91 , is @xmath92 . on the other hand , the scatter resulting from a range of @xmath93 is of the same order , @xmath94 . similarly , the nearly equivalent median @xmath88 values may indicate that for both grb populations @xmath91 . in this case , the resulting lower limits on @xmath87 therefore prevent the use of @xmath88 as an indicator of density . comparing _ sample 1 _ and _ sample 2 _ , we find no clear difference in @xmath88 ( figure [ fig : opt_xray ] ) . the same is true for the bursts in _ sample 3 _ , which are all consistent with the same relation given the optical upper limits and a mix of x - ray detections and upper limits . thus , the ratio of optical to x - ray flux does not allow us to distinguish redshift / energy and density effects between the three samples . to summarize , the optical afterglows of short grb without coincident hosts ( or with only optical limits ) are systematically fainter than those of short grbs with coincident hosts . the same is true for their x - ray fluxes and @xmath0-ray fluences . the fainter afterglows may reflect lower densities ( by an order of magnitude ) , but this does not naturally explain the lower @xmath0-ray fluences . alternatively , the fainter afterglows and @xmath0-ray fluences can be explained as a result of higher redshifts ( @xmath1 ) or lower energies ( by about a factor of 3 ) . we next turn to an analysis of the large - scale environments of the bursts in _ sample 2 _ , partly in comparison to the hosts of bursts in _ sample 1_. as indicated in [ sec : obs ] , we place limits of @xmath95 mag on the brightness of any galaxy underlying the five short grb positions ( table [ tab : nohost ] ) . these limits are several magnitudes fainter than the measured brightnesses of short grb hosts at @xmath96 ( _ sample 1 _ ) . to assess the potential that galaxies near each of the 5 bursts are the hosts , we calculate their probability of chance coincidence . we follow the methodology of @xcite , namely , we determine the expected number density of galaxies brighter than a measured magnitude , @xmath97 , using the results of deep optical galaxy surveys @xcite : @xmath98 the probability for a given separation , @xmath99 , is then given by @xmath100 where we use the fact that for offsets substantially larger than the galaxy size , @xmath101 is the appropriate radius in equation [ eqn : prob ] @xcite . the resulting distributions for each field are shown in figure [ fig : prob1 ] . we include all galaxies that have probabilities of @xmath102 . we find that for 4 of the 5 bursts , faint galaxies ( @xmath103 mag ) can be identified within @xmath104 of the afterglow positions , with associated chance coincidence probabilities of @xmath105 ; in the case of grb090515 we do not detect any such faint galaxies within @xmath106 of the afterglow position . for grb080503 we also include the galaxy at an offset of @xmath107 and @xmath108 mag identified by @xcite based on their deeper stack of hst observations . on the other hand , for 4 of the 5 bursts we find that the galaxies with the lowest probability of chance coincidence , @xmath109 , are brighter objects with offsets of about @xmath110 from the burst positions ; only in the case of grb080503 the lowest chance coincidence is associated with the nearest galaxy ( see @xcite ) . for comparison we repeat the same analysis for short grbs from _ sample 1 _ which have faint coincident hosts ( grb060121 : 26.0 ab mag ; grb060313 : 26.6 ab mag ; and grb070707 : 27.3 ab mag ) . the results of the probability analysis are shown in figure [ fig : prob2 ] . we find that in all three cases , the coincident hosts exhibit the lowest probability of chance coincidence , @xmath111 . only in the case of grb070707 do we find galaxies with @xmath112 arcsec that have @xmath113 . thus , these three bursts are consistent with negligible offsets from faint galaxies , presumably at @xmath7 . the use of _ a posteriori _ probabilities to assign _ unique _ galaxy associations is fraught with difficulties . first , for a given apparent brightness , galaxies located further away from the grb position , potentially due to larger kicks and/or longer merger timescales in the ns - ns merger framework , have higher probabilities of chance coincidence . since we have no _ a priori _ model - independent knowledge of the range of possible kicks and merger timescales , we can not rule out galaxies at very large offsets for which @xmath114 . indeed , a reasonable constraint of @xmath115 km s@xmath116 and @xmath117 gyr leads to only a weak constraint on the offset of @xmath18 mpc . at @xmath118 ( @xmath69 ) this corresponds to about @xmath119 ( @xmath120 ) , a projected distance at which nearly all galaxies will have a chance coincidence probability of order unity . a second difficulty is that we are using angular offsets , which ignore the potential wide range of redshifts ( and by extension also luminosities ) of the various galaxies . for example , if the faint galaxies with small offsets are located at @xmath7 , the corresponding physical offsets are @xmath8 kpc , while if the galaxies at @xmath121 offsets are located at @xmath15 ( see [ sec : obs ] ) , the offsets are only somewhat larger , @xmath122 kpc . a galaxy at an even lower redshift , @xmath123 , with an offset of 50 kpc will be located about @xmath124 from the grb position and incur a large penalty in terms of chance coincidence probability . it is important to note , however , that galaxies at lower redshift will generally have brighter apparent magnitudes , partially compensating for the larger angular separations ( equations [ eqn : gal ] and [ eqn : prob ] ) . in only a single case ( grb070809 ) we find a galaxy with @xmath113 at @xmath125 ( which at @xmath126 for this galaxy corresponds to a physical offset of about 100 kpc ) . a final complication , which is not unique to this subset of events , is that we can only measure projected offsets , @xmath127 . the measured offsets can be used as lower limits on the actual offsets , while for the overall distribution we can apply an average correction factor of @xmath128 , based on the expectation value for the projection factor , @xmath129 . despite these caveats we can address the probability that _ all _ of the associations are spurious . this joint probability is simply the product of the individual probabilities @xcite . for the faint galaxies at small angular separations the probability that all are spurious associations is @xmath130 , while for the galaxies with the lowest probability of chance coincidence the joint probability is nearly 30 times lower , @xmath131 . conversely , the probabilities that _ none _ of the associations are spurious are @xmath132 and @xmath133 , respectively . these values indicate the some spurious coincidences may be present for _ sample 2_. indeed , the probabilities that 1 , 2 , or 3 associations are spurious are @xmath134 $ ] and @xmath135 $ ] , respectively . these results indicate that for the faint galaxies it is not unlikely that @xmath136 associations ( out of 5 ) are spurious , while for the brighter galaxies @xmath137 associations may be spurious . this analysis clearly demonstrates why a joint statistical study is superior to case - by - case attempts to associate short grbs with galaxies at substantial offsets . we therefore conclude that despite the weaknesses inherent to _ a posteriori _ probabilities , we conclude that there is stronger statistical support for an association of at least some of the 5 bursts in _ sample 2 _ with bright galaxies at separations of @xmath138 , than for an association with the faint galaxies at separations of @xmath139 . clearly , we can not rule out the possibility that in reality the hosts are a mix of faint and bright galaxies with a range of angular offsets of @xmath140 . we note that if deeper observations eventually lead to the detection of underlying galaxies ( @xmath141 ) at the level of @xmath142 mag , the associated chance coincidence probabilities will be @xmath143 per object , and the joint probabilities will be only slightly higher than for the bright galaxies with @xmath121 offsets . on the other hand , if we can achieve magnitude limits of @xmath19 mag on any coincident hosts , the resulting probabilities of chance coincidence will be larger than for the offset bright galaxies . thus , eliminating the possibility of underlying hosts at the level of @xmath19 mag is of the utmost importance . so far , only grb080503 has been observed to such a depth with no detected coincident host @xcite , but observations of the full sample are required for a robust statistical comparison . as we discuss in [ sec : highz ] below , such deep limits will also reduce the probability of underlying hosts based on redshift arguments . beyond the use of projected angular offsets , we note that the faint galaxies near the burst positions are likely to have projected physical offsets of about @xmath144 kpc . to assess the projected physical offsets for the galaxies with the lowest probability of chance coincidence we measured spectroscopic redshifts in three cases ( grbs 061201 , 070809 , and 090515 ; [ sec : obs ] ) . in the case of grb061201 , the galaxy redshift of @xmath36 leads to a projected physical offset of @xmath145 kpc @xcite . in the case of grb070809 , the lowest probability of chance coincidence is not associated with the star forming galaxy at @xmath39 identified by @xcite , but instead belongs to the early - type galaxy identified here , which has a redshift of @xmath40 , and hence an offset of @xmath146 kpc . finally , for grb090515 , the lowest probability of chance coincidence is associated with a cluster early - type galaxy at @xmath44 , leading to a physical offset of @xmath147 kpc . thus , two scenarios emerge from our investigation of the large - scale environments : ( i ) the bursts are spatially coincident with currently - undetected galaxies ( with @xmath148 mag ) ; or ( ii ) the bursts have substantial offsets of at least @xmath149 kpc depending on whether they are associated with faint galaxies at small angular separations , or brighter galaxies at @xmath30 ; for the ensemble of 5 events the larger offsets are statistically more likely than the @xmath8 kpc offsets . in the context of large offsets , even larger values may be possible if the bursts originated in galaxies with larger separations and @xmath150 . these two scenarios echo the possibilities that emerged from the analysis of the afterglow and prompt emission properties ( [ sec : ag ] ) . distinguishing between these possibilities is clearly of fundamental importance to our understanding of short grbs : the former scenario will point to a population of very faint hosts ( likely at high redshifts ) , while the latter scenario will provide evidence for large offsets ( due to kicks or a globular cluster origin ) and hence ns - ns / ns - bh progenitors for at least some short grbs . we can place upper limits on the redshifts of the grbs in _ sample 2 _ based on their detections in the optical ( i.e. , the lack of complete suppression by the ly@xmath151 forest ) . the afterglow of grb061201 was detected in the ultraviolet by the _ swift_/uvot and it is therefore located at @xmath152 @xcite . the remaining four bursts were detected in the optical @xmath153- or @xmath33-band , and can therefore be placed at @xmath154 ( table [ tab : shbs ] ) . we place additional constraints on the redshifts of any underlying hosts using the existing sample short grb host galaxies . in figure [ fig : mags ] we plot the @xmath33-band magnitudes as a function of redshift for the all available short grb hosts from _ sample 1_. for the faint hosts without known redshifts ( grbs 060121 , 060313 , and 070707 ) , we place upper limits on the redshift using optical detections of the afterglows ( table [ tab : shbs ] ) . a wide range of host magnitudes , @xmath155 mag , is apparent . we also plot the @xmath33-band magnitudes of long grb hosts @xcite , as well as the @xmath156 phase space that is traced by galaxies with luminosities of @xmath157 l@xmath11 . we use the appropriate value of l@xmath11 as a function of redshift , taking into account the evolving galaxy luminosity function @xcite . we find excellent correspondence between the hosts of long and short grbs , and the phase - space traced by @xmath158 l@xmath11 galaxies , at least to @xmath159 . in the context of these distributions , the available limits for the short grbs in _ sample 2 _ translate to redshifts of @xmath160 if they are 0.1 l@xmath11 galaxies , or @xmath161 if they are l@xmath11 galaxies . the latter lower limits are comparable to the redshift upper limits inferred from the afterglow detections . we note that for grb080503 , the limit of @xmath162 mag @xcite places even more stringent limits on the redshift of an underlying @xmath158 l@xmath11 galaxy . the redshifts of @xmath163 for putative 0.1 l@xmath11 hosts are consistent with the faintness of the optical afterglows , from which we inferred @xmath1 compared to _ sample 1 _ ( [ sec : ag ] ) . we note , however , that the one known short grb at @xmath12 ( grb090426 ; @xcite ) has a host galaxy luminosity of @xmath164 l@xmath11 , which may suggest that the appropriate redshift lower limits are @xmath161 . the possibility that the five bursts originated at @xmath161 leads to a bimodal redshift distribution ( figure [ fig : mags ] ) . nearly all of the bursts in _ sample 1 _ with a known redshift ( 9/10 ) have @xmath25 , with a median of @xmath165 ; the sole exception is grb090426 at @xmath166 . the three bursts with faint coincident hosts have upper limits of @xmath154 from afterglow detections , while lower limit of @xmath167 can be placed on these hosts if they have @xmath168 l@xmath11 . adding the _ sample 2 _ bursts with the assumption that they have @xmath161 will furthermore result in a population of short grbs with a median of @xmath169 , and leave a substantial gap at @xmath170 ( figure [ fig : mags ] ) . if the 5 bursts are instead hosted by @xmath171 l@xmath11 galaxies , the inferred lower limits on the redshifts ( @xmath160 ) lead to a potentially more uniform redshift distribution . it is difficult to explain a bimodal redshift distribution with a single progenitor population such as ns - ns binaries , without appealing to , for example , a bimodal distribution of merger timescales . another possibility is two distinct progenitor populations , producing bursts of similar observed properties but with distinct redshift ranges . while these possibilities are difficult to exclude , they do not provide a natural explanation for the short grb population . a final alternative explanation is that any underlying hosts reside at similar redshifts to the known hosts in _ sample 1 _ ( @xmath16 ) , but have significantly lower luminosities of @xmath172 l@xmath11 . this scenario would not naturally explain why the bursts in _ sample 2 _ have fainter optical and x - ray afterglows , as well as lower @xmath0-ray fluences . we therefore do not consider this possibility to be the likely explanation . while higher redshifts may explain the lack of detected hosts , the fainter afterglows , and the weaker @xmath0-ray fluences of the bursts in _ sample 2 _ , this scenario suffers from several difficulties outlined above . the alternative explanation is that the bursts occurred at significant offsets relative to their hosts , and hence in lower density environments that would explain the faint afterglow emission ( though possibly not the lower @xmath0-ray fluences ) . as we demonstrated in [ sec : prob ] , the offsets may be @xmath139 ( @xmath8 kpc ) if the bursts originated in the faint galaxies at the smallest angular separations , or @xmath121 ( @xmath173 kpc ) if they originated in the brighter galaxies with the lowest probability of chance coincidence . below we address the implications of these two possible offset groups through a comparison to the offsets measured for the bursts in _ sample 1 _ ( e.g. , @xcite ) . we plot the distributions of projected angular offsets for the short grbs with and without coincident hosts in figure [ fig : offset1 ] . the offsets for _ sample 1 _ have a mean and standard deviation of about @xmath174 and a range of about @xmath175 . modeled with a log - normal distribution , the resulting mean and width in units of arcseconds are @xmath176 and @xmath177 . if we associate the short bursts in _ sample 2 _ with the galaxies that have the lowest probability of chance coincidence , the resulting distribution has @xmath178 arcsec . this is clearly distinct from the distribution for _ sample 1_. the effect is less pronounced if we associate the bursts with the galaxies located at the smallest angular offsets ( figure [ fig : offset1 ] ) . even for these separations the mean and standard deviation are @xmath179 arcsec . as noted in [ sec : prob ] , the projected angular distances may not be the most robust quantity for measuring the offset distribution . an alternative quantity is the offset normalized by each host s effective radius @xcite . this quantity takes into account the varying sizes of the hosts due to both intrinsic size variations and redshift effects . it also gives a better indication of whether the burst coincides with the host light or is significantly offset . as shown in figure [ fig : offset2 ] , the host - normalized offsets of _ sample 1 _ have a mean and standard deviation of about @xmath180 @xmath181 , and a range of about @xmath182 @xmath181 . a log - normal fit results in a mean of @xmath183 and a width of @xmath184 . the bursts in _ sample 2 _ have much larger host - normalized offsets , with @xmath185 if they originated in the galaxies with the lowest chance coincidence probability . even if we associate the bursts with the nearest faint hosts , the distribution has a mean of about @xmath186 @xmath181 , reflecting the fact that the effective radii of the faint galaxies are smaller than those of the brighter ones . finally , we plot the projected physical offsets in figure [ fig : offset3 ] . the mean and standard deviation for _ sample 1 _ are @xmath187 kpc , and a log - normal fit results in a mean of @xmath188 and a width of @xmath189 . on the other hand , the bursts in _ sample 2 _ have a mean offset of about 19 kpc if they arise in the faint galaxies with small angular separation , or about 40 kpc if they arise in the brighter galaxies , again pointing to distinct distributions . the distributions of angular , physical , and host - normalized offsets exhibit a clear bimodality if we associate the bursts in _ sample 2 _ with the galaxies at @xmath30 . this is particularly apparent in the more meaningful quantities , namely physical and host - normalized offsets ( figures [ fig : offset2 ] and [ fig : offset3 ] ) . the effect is still apparent , though less pronounced in the case of association with the faint galaxies at @xmath7 . thus , if the offset scenario is correct , the resulting distributions point to a possible bimodality rather than a single continuous distribution of offsets . the cumulative distributions of physical offsets for _ sample 1 _ alone , and in conjunction with the two possible offset groups for _ sample 2 _ are shown in figure [ fig : offset4 ] . the combined distributions have a median of about 4 kpc , driven by the bursts with coincident hosts . however , there is a clear extension to larger physical offsets in the case of association with the brighter galaxies , with about @xmath190 of all objects having @xmath191 kpc . the cumulative distributions are particularly useful for comparison with ns - ns merger models since predictions exist for both the kick scenario and the globular cluster origin model . we turn to this discussion below . we have shown that the lack of host galaxy detections in coincidence with the bursts in _ sample 2 _ indicates either a high redshift ( @xmath12 ) origin , or substantial projected offsets of @xmath192 kpc ( @xmath193 @xmath181 ) . both scenarios can account for the lack of underlying galaxy detections and the fainter optical / x - ray afterglows ( either due to redshift or density effect ) ; the possibility of large offsets from galaxies at @xmath30 , however , does not naturally explain the lower @xmath0-ray fluences and somewhat shorter durations of the bursts in _ sample 2_. indeed , the lower fluences can be more easily accommodated in the case of undetected high redshift coincident hosts , or in the case of faint hosts with offsets of @xmath8 kpc . below we test the results of the offset scenario against predictions for ns - ns merger models and dynamically - formed ns - ns binaries in globular clusters . we also investigate whether the densities expected at offsets of @xmath194 kpc ( i.e. , galaxy halos or the intergalactic medium ) can accommodate the observed optical magnitudes . finally , we discuss the implications of the various scenarios for short grb energetics . the sample of short grbs with optical afterglows represents about @xmath195 of all short bursts , and may thus not be fully representative . one often - discussed bias is that the bursts with optical afterglows require a high circumburst density , and therefore have negligible offsets . however , from our analysis in this paper it is clear that one explanation for the lack of coincident hosts for the bursts in _ sample 2 _ is indeed large offsets , _ despite their detection in the optical band_. the bursts with only x - ray afterglow detections ( _ swift_/xrt ) have typical positional uncertainties of @xmath196 and therefore lead to a deeper ambiguity about the identity of the hosts . we have previously argued that most of these bursts are associated with galaxies consistent with a negligible offset ( or as large as tens of kpc ; @xcite ) . in some cases larger offsets have been advocated based on _ a posteriori _ chance coincidence probabilities ( e.g. , grb050509b : @xcite ; grb060502b : @xcite ) . however , from the analysis in [ sec : prob ] it is clear that while the inferred probabilities are only weakly dependent on the positional accuracy when @xmath197 ( i.e. , galaxies at large offsets ) , a large penalty is incurred for faint galaxies located _ within _ the x - ray error circles since in that case the appropriate value in equation [ eqn : prob ] is @xmath198 . as a result , the faint galaxies will have chance coincidence probabilities of @xmath150 . to avoid this complication in our comparison to ns - ns model predictions we restrict the analysis to the sample with optical afterglow positions . in recent work by @xcite we have shown that the offset distribution for short grbs with optical and/or x - ray positions , and accounting for incompleteness due to bursts with only @xmath0-ray positions , is broadly consistent with the predictions of ns - ns merger models @xcite . in particular , we concluded that @xmath199 of all short bursts have @xmath200 kpc , while @xmath201 have offsets of @xmath21 kpc . we repeat the analysis here using the offsets inferred in [ sec : kick ] for both the faint galaxies at small angular separations and the brighter galaxies at larger separations . as shown in figure [ fig : offset4 ] , the model predictions have a median of about 6 kpc , compared to about 4 kpc for the observed sample . on the other hand , the models predict @xmath20 of offsets to be @xmath202 kpc , in good agreement with the observed distribution in both the @xmath8 kpc and @xmath203 kpc scenarios . we note that the overall smaller offsets measured from the data may be due to projections effects . indeed , the mean correction factor of @xmath128 nicely reconciles the theoretical and observed distributions . in the previous section we noted a bimodality in the physical and host - normalized offsets for _ sample 1 _ and _ sample 2 _ ( figures [ fig : offset2 ] and [ fig : offset3 ] ) . in the framework of ns - ns binary kicks this bimodality may indicate that the binaries generally remain bound to their host galaxies , thereby spending most of their time at the maximal distance defined by @xmath204 ( i.e. , with their kinetic energy stored as potential energy ; @xcite ) . this would require typical kick velocities of less than a few hundred km s@xmath116 . we further compare the observed offset distribution to predictions for dynamically - formed ns - ns binaries in globular clusters , with a range of host galaxy virial masses of @xmath205 m@xmath206 @xcite . these models predict a range of only @xmath207 of all ns - ns mergers to occur within 10 kpc of the host center , in contrast to the observed distribution with about @xmath208 with @xmath209 kpc . we stress that this result is independent of what offsets we assign to the bursts in _ sample 2 _ since they account for only 1/4 of the bursts with optical afterglows . on the other hand , the globular cluster origin may account for the bimodality in the physical and host - normalized offsets ( figures [ fig : offset2 ] and [ fig : offset3 ] ) , with the objects in _ sample 2 _ arising in globular clusters and the objects with coincident hosts arising from primordial ns - ns binaries . this possibility also agrees with the predicted fraction of dynamically - formed ns - ns binaries of @xmath210 @xcite . the cumulative offset distributions for _ sample 2 _ alone ( assuming the hosts are the galaxies with the lowest probability of chance coincidence ) is well - matched by the range of predictions for dynamically - formed ns - ns binaries in globular clusters ( figure [ fig : offset4 ] ) . in this scenario , however , the implication is that short grbs outside of globular clusters do not experience kicks as expected for ns - ns binaries since the largest measured offset is only 15 kpc . unless the populations of short grbs with only x - ray or @xmath0-ray positions have fundamentally different offset distributions , we conclude that the measured offsets of short grbs and the predicted offsets for ns - ns kicks are in good agreement , _ if we treat all short grbs with optical afterglows as a single population_. alternatively , it is possible that the bimodal distributions of physical and host - normalized offsets point to a progenitor bimodality , with the bursts in _ sample 2 _ originating in globular clusters . large offsets are expected to result in low circumburst densities , and we now investigate whether the observed optical magnitudes for _ sample 2 _ can be used to place meaningful constraints on the offsets . the optical afterglow magnitudes do not provide a direct measure of the circumburst density since they depend on a complex combination of density , energy , and fractions of the burst energy imparted to the radiating electrons ( @xmath211 ) and magnetic fields ( @xmath212 ) . from our comparison of the x - ray and optical afterglows ( [ sec : ag ] ) we were unable to clearly locate the cooling frequency , which in principle can provide additional constraints on the density . however , we can still gain some insight into the required circumburst densities using a few basic assumptions . we assume that the @xmath0-ray energy ( @xmath213 ) is a reasonable proxy for the total energy , that @xmath214 , and that @xmath215 . we further use a fiducial redshift of @xmath68 ( figure [ fig : mags ] ) and a fiducial observation time of 8 hr ( [ sec : ag ] ) . to determine @xmath213 we use the @xmath216 kev fluences measured by _ ( table [ tab : shbs ] ) , and determine a correction factor to account for the incomplete energy coverage using bursts that have also been detected by satellites with a broader energy range ( figure [ fig : t90_fg ] ) . we find a typical correction factor of @xmath217 for the @xmath218 kev range . for the fiducial redshift of @xmath68 , the mean energy for short grbs with optical afterglows is @xmath219 erg , when we include the correction factor ( figure [ fig : z_egiso ] ) . with these assumptions we find the following relation between the optical brightness and the circumburst density @xcite : @xmath220 given the typical observed fluxes of about 2 @xmath221jy for _ sample 1 _ , and about 0.6 @xmath221jy for _ sample 2 _ , we infer typical minimum densities of @xmath222 and @xmath223 @xmath32 , respectively . thus , the observed optical brightnesses can be produced even at very low densities that are typical of the igm . this indicates that we can not rule out the large offset scenario based on density arguments . indeed , even if we allow for more typical values of @xmath224 , the resulting densities are @xmath225 and @xmath226 @xmath32 , respectively . these results compare favorably with predictions for ns - ns kicks , which suggest that most mergers will occur at densities of @xmath227 @xmath32 @xcite . the distribution of densities in globular clusters is not well known , preventing a meaningful comparison to our inferred minimum densities @xcite . finally , we investigate the energetics of the bursts in _ sample 2 _ in the context of a high - redshift origin ( @xmath161 ) , an origin in the faint galaxies at @xmath228 separations ( @xmath229 ) , and an origin in the galaxies with the lowest probabilities of chance coincidence ( @xmath230 ) . the resulting values of @xmath213 are shown in figure [ fig : z_egiso ] . for the lowest redshift origin , the inferred values are @xmath231 erg , for a @xmath229 origin they are @xmath232 erg , and for a @xmath161 origin they are @xmath233 erg . for comparison , the mean value for the bursts in _ sample 1 _ is @xmath234 erg , with a range of @xmath235 erg . however , there is a clear redshift dependence for the measured @xmath213 values ( figure [ fig : z_egiso ] ) , with @xmath236 erg at @xmath237 and @xmath238 erg at @xmath239 . thus , the bursts in _ sample 2 _ generally fit within the known range of isotropic energies regardless of their actual redshift . indeed , at @xmath169 they exhibit similar values of @xmath240 to that of grb090426 . as a result , we can not use the inferred energy release as a clear discriminant of the redshift range for _ sample 2_. if the bursts in _ sample 2 _ indeed originated at high redshifts , the resulting isotropic - equivalent energies suggest that there is either a large spread in the energy release of short grbs ( at least 2 orders of magnitude ) or a large variations in the ejecta geometry . if the typical energy for short grbs is about @xmath241 erg , as indicated by the nearest events , then the inferred values of @xmath242 at @xmath169 indicate opening angles of @xmath243 . this is similar to the opening angle of about @xmath244 that was inferred for grb051221 @xcite . we undertook the first systematic study of short grbs with detected optical afterglows ( and hence sub - arcsecond position ) but no coincident host galaxies to limits of @xmath148 mag . we find that the optical afterglows of these bursts are fainter by about 1.4 mag compared to the optical afterglows of short grbs with coincident hosts . they are similarly fainter in x - rays , and have somewhat lower @xmath0-ray fluences and slightly shorter durations . both samples have similar ratios of x - ray to optical flux , which are moreover similar to the ratios measured for long grbs . the fainter afterglows of the bursts lacking coincident hosts may be due to lower densities , lower energies , or higher redshifts . however , we note that only the scenarios with lower energies or higher redshifts naturally explain the faintness of the prompt emission _ in the context of the standard grb model_. this is because in the context of internal processes ( shocks or magnetic dissipation ) the external density should not play a role . we further use deep optical imaging to determine the probability of chance coincidence for galaxies in the field around each burst , and to place redshift limits for underlying hosts under the assumption that they are drawn from the same distribution of the detected short grb hosts ( @xmath10 l@xmath11 ) . this analysis leads to the following possible scenarios : ( i ) the underlying hosts are fainter than @xmath245 mag , indicative of redshifts of @xmath163 ( if @xmath246 l@xmath11 ) or @xmath161 ( if @xmath247l@xmath11 ) ; or ( ii ) the hosts are galaxies with substantial offsets either faint galaxies at separations of @xmath228 ( @xmath248 kpc for @xmath7 ) or brighter galaxies , which we find to be located at @xmath30 , with offsets of @xmath249 ( @xmath250 kpc ) . in the former scenario , unless the galaxies have @xmath246 l@xmath11 , the resulting redshift distribution is bimodal with peaks at @xmath16 and @xmath251 . such a scenario most likely requires a bimodal age distribution in the context of ns - ns mergers , or two distinct progenitor systems dominating at low and high redshifts . while this can not be ruled out by present data , the lack of overlap at @xmath252 is difficult to explain . we also conclude that it is unlikely that any faint underlying hosts are low - luminosity galaxies ( @xmath253 l@xmath11 ) at similar redshifts to the detected hosts , since this does not naturally explain the difference in afterglow and prompt emission properties . in the context of the large offsets scenario , the probability of chance coincidence ( both individually and for the sample ) is lower for the brighter galaxies at offsets of @xmath249 than for the faint galaxies at offsets of @xmath228 . however , it is not unlikely that the true associations are a mix of both populations , since in each case there is a non - negligible probability that @xmath137 of the associations are spurious . we note that optical positions are crucial for a uniform comparison of the chance coincidence probabilities for faint galaxies with small offsets and bright galaxies with larger offsets ; in the case of only x - ray positions there is an inherent bias ( in the sense of _ a posteriori _ ) against associations with faint galaxies inside the error circle , despite the fact that they are consistent with no offset . from spectroscopic observations for 3 of the 5 bursts we find that the galaxies with the lowest probability of chance coincidence are a star forming galaxy at @xmath36 ( grb061201 ) , an early - type galaxy at @xmath40 ( grb070809 ) , and an early - type cluster member galaxy at @xmath44 ( grb090515 ) . if these associations are indeed correct , they only slightly alter the host demographics , which are dominated by star forming galaxies @xcite . however , this does suggest that our present understanding of the relative ratios of star forming and elliptical hosts may be incomplete . the resulting distributions of angular , physical , and host - normalized offsets for the bursts with and without coincident hosts appear to be distinct , rather than continuous . however , the joint distribution of projected physical offsets is in good agreement with theoretical predictions for ns - ns binary mergers . on the other hand , the predicted distribution for dynamically - formed ns - ns binaries in globular clusters provides a much poorer fit to the entire data set , unless they account for only the bursts with large offsets ( _ sample 2 _ ) . in the case of a hybrid population of primordial and dynamically - formed binaries , with the latter accounting for only the large offsets , the fact that all the remaining offsets ( _ sample 1 _ ) are @xmath18 kpc , is indicative of no significant kicks . the large physical offsets also naturally explain the fainter afterglow emission as a result of lower circumburst densities . the resulting isotropic @xmath0-ray energies match the observed distribution for short grbs with coincident hosts , either at @xmath15 or at @xmath7 . our conclusion that large offsets of @xmath254 kpc ( corresponding to @xmath193 galactic effective radii ) are a likely explanation for the bursts with optical afterglows and no coincident hosts is of fundamental importance . the only progenitor model that naturally explains this result in the merger of ns - ns / ns - bh binaries , most likely due to kicks , or possibly with a minor contribution from a globular cluster population ( accounting specifically for the events in _ sample 2 _ ) . while a conclusive demonstration of a large offset requires an absorption redshift measurement that matches an offset galaxy redshift , the distribution of optical afterglow magnitudes indicates that this will be difficult to achieve . indeed , an absorption redshift is available for only one likely short grb @xcite . we end with several important observations . first , rapid optical follow - up of short grbs with 8-meter class telescopes is essential since observations to a depth of about 25 mag within a few hours after the burst may recover nearly all optical afterglows , regardless of circumburst density ( figure [ fig : optag1 ] ) . second , short grbs with only x - ray positions are unlikely to provide strong support for either negligible or large offsets due to the appreciable size of the error circles and the fact that the sample with optical afterglows exhibits wide host galaxy diversity , i.e. , events with coincident bright hosts at @xmath255 , events with coincident faint hosts likely at @xmath7 , and events with no coincident hosts likely due to offsets . third , a meaningful study of short grb offsets requires a statistical approach to mitigate the shortcomings of _ a posteriori _ chance coincidence probabilities , as well as to incorporate the relevant information from afterglow and prompt emission observations . finally , we note that from a wide range of observations of both the afterglows and host galaxies it appears that the case for ns - ns mergers as the progenitors of short grbs is gaining observational support . our main result here is that short grbs with optical afterglows and no detected host galaxies are somewhat less likely to be explained with high redshifts or with dwarf galaxy hosts at low redshift . instead , these bursts likely exploded @xmath8 kpc from galaxies at @xmath229 or tens of kpc from galaxies at @xmath15 . with larger samples , these possibilities will allow us to improve our understanding of the short grb redshift distribution , the host galaxy demographics , and predictions for gravitational wave detections . if high redshifts indeed turn out to be prevalent , this will have a significant effect on the possibility of multiple progenitor populations . we acknowledge helpful discussions with alicia soderberg , ryan chornock , and josh grindlay . this work was partially supported by swift ao5 grant number 5080010 , and made use of data supplied by the uk swift science data centre at the university of leicester . lcccccccl 0.08 in + 050709 & 0.07 & 0.161 & @xmath256 & 60.5 & @xmath257 & 34 & 2.3 & 13 + 050724 & 3.0 & 0.257 & @xmath258 & 11.8 & @xmath259 & 12 & 8.4 & 46 + 051221a & 1.4 & 0.546 & @xmath260 & 3.1 & @xmath261 & 3.1 & 5.8 & 78 + 060121 & 2.0 & @xmath262 & @xmath263 & 6.1 & @xmath261 & 7.4 & 8.8 & 911 + 060313 & 0.70 & @xmath264 & @xmath265 & 2.8 & @xmath266 & 2.8 & 10.8 & 1112 + 061006 & 0.42 & 0.438 & @xmath267 & 19.5 & @xmath268 & 14.9 & 2.9 & 11,13 + 061201@xmath269 & 0.80 & @xmath270,0.111 ? & @xmath271 & 8.6 & @xmath272 & 8.6 & 2.9 & 1415 + 070707 & 1.1 & @xmath262 & @xmath267 & 12.5 & @xmath273 & 11 & 1.9 & 16 + 070714b & 3.0 & 0.923 & @xmath274 & 24.4 & @xmath275 & 23.6 & 0.7 & 17 + 070724 & 0.40 & 0.457 & @xmath276 & 2.35 & @xmath277 & 2.3 & 5.0 & 1819 + 070809@xmath269 & 1.3 & @xmath264,0.473 ? & @xmath278 & 11.0 & @xmath279 & 11 & 0.8 & 2021 + 071227 & 1.8 & 0.381 & @xmath280 & 7.0 & @xmath281 & 7.0 & 1.6 & 13 + 080503@xmath269 & 0.32 & @xmath264 , & @xmath282 & 66.0 & @xmath283 & 25.9 & 0.3 & 22 + 080905 & 1.0 & 0.122 & @xmath284 & 18.4 & @xmath285 & 8.5 & 0.8 & 23 + 090305@xmath269 & 0.40 & @xmath262 , & @xmath286 & 0.45 & @xmath287 & 0.45 & 1.5 & 2425 + 090426 & 1.28 & 2.609 & @xmath288 & 2.0 & @xmath289 & 2.6 & 20.0 & 2627 + 090510 & 0.30 & 0.903 & @xmath290 & 6.4 & @xmath291 & 9.0 & 2.3 & 28 + 090515@xmath269 & 0.04 & @xmath262,0.403 ? & @xmath292 & 4.9 & @xmath293 & 1.9 & 0.1 & 29 + 091109b & 0.30 & & @xmath294 & 3.8 & @xmath295 & 5.7 & 0.5 & 30 + 100117 & 0.30 & 0.92 & @xmath296 & 8.4 & @xmath297 & 8.4 & 0.3 & 31 + + + 050509b & 0.04 & 0.225 & @xmath298 & 9.1 & @xmath297 & 2.1 & 0.7 & 3233 + 050813 & 0.60 & & @xmath299 & 94.9 & @xmath300 & 12.8 & 1.9 & 3436 + 051210 & 1.27 & & @xmath301 & 6.9 & @xmath302 & 19.2 & 1.6 & 11,37 + 060502b & 0.09 & & @xmath303 & 18.6 & @xmath304 & 16.8 & 0.7 & 11,38 + 060801 & 0.50 & 1.130 & @xmath301 & 10.9 & @xmath305 & 12.4 & 0.8 & 11 + 061210 & 0.19 & 0.409 & @xmath265 & 72.4 & @xmath306 & 2.1 & 1.4 & 11 + 061217 & 0.21 & 0.827 & @xmath307 & 38.6 & @xmath308 & 2.8 & 2.0 & 11 + 070429b & 0.50 & 0.902 & @xmath309 & 11.8 & @xmath310 & 4.8 & 0.6 & 39 + 080426 & 1.30 & & @xmath311 & 6.1 & @xmath312 & 7.5 & 2.6 & 40 + 080702 & 0.50 & & @xmath313 & 2.9 & @xmath314 & 12.1 & 12 . & 41 + 081226 & 0.40 & & @xmath315 & 3.2 & @xmath302 & 1.6 & 0.9 & 42 + 100206 & 0.12 & & @xmath284 & 6.9 & @xmath297 & 15.7 & 0.5 & 43 lcccc 0.1 in 061201 & hst / acs & f814w & 2224 & 26.0 + 070809 & magellan / ldss3 & @xmath33 & 1500 & 25.4 + 080503 & hst / wfpc2 & f606w & 4000 & 25.7 + 090305 & magellan / ldss3 & @xmath33 & 2400 & 25.6 + 090515 & gemini - n / gmos & @xmath33 & 1800 & 26.5 -band images of the location of grb070809 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . [ fig:070809],title="fig:",width=288 ] -band images of the location of grb070809 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . [ fig:070809],title="fig:",width=288 ] was found by @xcite based on a deeper stack of hst / wfpc2 observations . these authors also find that the galaxy marked `` s5 '' is located at @xmath316 , leading to a physical offset of 85 kpc . [ fig:080503],width=480 ] -band images of the location of grb090305 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . [ fig:090305],title="fig:",width=288 ] -band images of the location of grb090305 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . [ fig:090305],title="fig:",width=288 ] -band images of the location of grb090515 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . note that the object coincident with the cross - hairs is the optical afterglow . [ fig:090515],title="fig:",width=288 ] -band images of the location of grb090515 . galaxies near the position of the optical afterglow ( cross - hairs ) are marked . note that the object coincident with the cross - hairs is the optical afterglow . [ fig:090515],title="fig:",width=288 ] -ray fluence ( top ) , afterglow x - ray flux at 8 hr ( middle ) , and duration ( bottom ) for the three short grb samples discussed in this paper . the arrows mark the mean for each sample , indicating that the bursts in _ sample 2 _ and _ sample 3 _ have lower @xmath0-ray fluences , fainter x - ray fluxes , and shorter durations , than the bursts with detected afterglows and coincident hosts . [ fig : fg_fx_t90],width=480 ]
we review the recent work on the mechanics of fast moving strings in anti - de sitter space times a sphere and discuss the role of conserved charges . an interesting relation between the local conserved charges of rigid solutions was found in the earlier work . we propose a generalization of this relation for arbitrary solutions , not necessarily rigid . we conjecture that an infinite combination of local conserved charges is an action variable generating periodic trajectories in the classical string phase space . it corresponds to the length of the operator on the field theory side . calt-68 - 2518 + * + + notes on fast moving strings * + + +
hep-th0409040
the ads / cft correspondence is a strong - weak coupling duality . weakly coupled yang - mills is mapped to the string theory on the highly curved ads space . when ads space is highly curved , the string worldsheet theory becomes strongly coupled . therefore , the weakly coupled yang - mills maps to the strongly coupled string worldsheet theory . nevertheless , in some situations elements of the ym perturbation theory can be reproduced from the string theory side . one of the examples are the `` spinning strings '' . spinning strings are a class of solutions of the classical string worldsheet theory . they were first considered in the context of the ads / cft correspondence in @xcite . these are strings rotating in @xmath1 with a large angular momentum . it was noticed in @xcite that the energy of these solutions has an expansion in some small parameter which is similar in form to the perturbative expansion in the field theory on the boundary . then @xcite computed the anomalous dimensions of single trace operators with the generic large r - charge , making the actual comparison possible . in @xcite more general solutions were considered , having large compact charges both in @xmath1 and in @xmath2 . for all these solutions , computations in the classical worldsheet theory lead to the series in the small parameter which on the field theory side is identified with @xmath3 where @xmath4 is the thooft coupling constant and @xmath5 a large conserved charge . moreover it was shown in @xcite that the quantum corrections to the classical worldsheet theory are suppressed for the solutions with the large conserved charge ( see also the recent discussion in @xcite ) . this opened the possibility that the results of the calculations in the classical mechanics of spinning strings , which are valid a priori only in the large @xmath4 limit , can be in fact extended to weak coupling and therefore compared to the yang - mills perturbation theory . it was conjectured that the yang - mills perturbation theory in the corresponding sector is reproduced by the classical dynamics of the spinning strings . the following picture is emerging . single string states in @xmath6 correspond to single - trace operators in the @xmath7 supersymmetric yang - mills theory . ( we consider the large @xmath8 limit . ) the dynamics of the single - trace operators is described in the perturbation theory by an integrable spin chain . this spin chain has a classical continuous limit @xcite which describes a class of operators with the large r - charges . in this limit the spin chain becomes a classical continuous system . we have conjectured in @xcite that this classical system is equivalent to the worldsheet theory of the classical string in @xmath6 . the yang - mills perturbative expansion corresponds to considering the worldsheet of the fast moving string as a perturbation of the null - surface @xcite . the null - surface perturbation theory was previously considered in a closely related context in @xcite . in this paper we will try to make the statement of equivalence more precise . we will argue that the string worldsheet theory has a `` hidden '' @xmath9 symmetry which is defined unambiguously by its characteristic properties which we describe . this @xmath9 commutes with the group of geometrical symmetries of the target space . it corresponds to the _ length of the spin chain _ on the field theory side . we conjecture that the phase space of the classical continuous spin chain is equivalent to the hamiltonian reduction of the phase space of the classical string by the action of this @xmath9 . the equivalence commutes with the action of geometrical symmetries . we should stress that the hidden @xmath9 symmetry which we discuss in this paper was constructed already in @xcite , but the explicit calculation was carried out only at the first nontrivial order of the null - surface perturbation theory . the main new result of our paper is that we discuss this hidden symmetry from the point of view of the integrability . we conjecture the relation between the @xmath9 symmetry and the local conserved charges which if true gives a uniform description of this symmetry at all orders of the perturbation theory . the classical string on @xmath6 is an integrable system ( see @xcite and references there ) , and our @xmath9 corresponds to an action variable . the existence of the action variables for integrable systems with a finite - dimensional phase space is a consequence of the liouville theorem @xcite . the classical string has an infinite - dimensional phase space . we are not aware of the existence of a general theorem which would guarantee that the action variables can be constructed in the infinite - dimensional case . but we will give two arguments for the existence of one action variable for the string in @xmath6 , at least in the perturbation theory around the null - surfaces . the first argument gives an explicit procedure to construct the action variable order by order in the perturbation theory ( sections 3 , 4.4 and 4.6 ) . the second argument uses the existence of the local conserved charges @xcite ( known as higher pohlmeyer charges ) and the results of the evaluation of these charges on the so - called `` rigid solutions '' performed in @xcite . the arguments in section 4 of our paper together with the results of @xcite suggest that the action variable is an infinite linear combination of the pohlmeyer charges and allow in principle to find the coefficients of this linear combination . _ the plan of the paper . _ in section 2 we will review the classification of the null - surfaces following mostly @xcite and stress that the moduli space of the null - surfaces is a @xmath9-bundle over a loop space . therefore it has a canonically defined action of @xmath9 . in section 3 we will explain how to extend the action of @xmath9 from the null - surfaces to the nearly - degenerate extremal surfaces using the perturbation theory . a large part of section 3 is a review of @xcite . in section 4 we discuss the geometrical meaning of this @xmath9 as an action variable and argue that it is an infinite sum of the local conserved charges . _ note added in the revised version . _ the coefficients of the expansion of the action variable in the local conserved charges were fixed to all orders in the first paper of @xcite . here we consider only the pohlmeyer charges for the @xmath1 part of the string sigma - model . the role of the pohlmeyer charges for @xmath2 was discussed in the second paper of @xcite . in the special case when the motion of the string is restricted to @xmath10 the action variable discussed here corresponds to the action variable of the sine - gordon model , see the third paper of @xcite . a two - dimensional surface in a space - time of lorentzian signature is called a null - surface if it has a degenerate metric and is ruled by the light rays . there is a connection between null - surfaces and extremal surfaces . an extremal surface is a two - dimensional surface with the induced metric of the signature @xmath11 which extremizes the area functional . extremal surfaces are solutions of the string worldsheet equation of motion in the purely geometrical background ( no @xmath12-field ) . when the string moves very fast , the metric on the worldsheet degenerates and the worldsheet becomes a null - surface . therefore a null - surface can be considered as a degenerate limit of an extremal surface . in @xmath6 there are two types of the light rays . the light rays of the first type project to points in @xmath1 . the light rays of the second type project to the timelike geodesics in @xmath2 and the equator of @xmath1 . the operators of the large r - charge correspond to the null - surfaces ruled by the light rays of the second type . ] . it is straightforward to explicitly describe all the null - surfaces of the second type in @xmath6 . we have to first describe the moduli space of the null - geodesics of the second type . an equator of @xmath1 is specified by a point in the coset space @xmath13 . similarly , a timelike geodesic in @xmath2 is specified by @xmath14 . given @xmath15 and @xmath16 , let @xmath17 and @xmath18 be the corresponding equator in @xmath1 and timelike geodesic in @xmath2 , respectively . to specify a light ray in @xmath6 we have to give also a map @xmath19 which pulls back the angular coordinate on @xmath20 to the length parameter on @xmath21 ( see fig.[fig : ng ] ) . such maps are parametrized by @xmath22 . we see that each light ray is defined by a triple @xmath23 . therefore , the moduli space of light - rays of the second type in @xmath6 is geometrically : @xmath24\widetilde{\times } s^1\ ] ] a null - surface is a one - parameter family of light rays . therefore it determines a contour in @xmath25\widetilde{\times } s^1 $ ] . but we have to also remember that an arbitrary collection of the light rays is not necessarily a null - surface . it is a null - surface only if the induced metric is degenerate . to understand what it means , let us choose a space - like curve belonging to our surface . this space - like curve is a collection of points , one point on each light ray . for the surface to be null , the tangent vector to this curve at each point of the curve should be orthogonal to the light ray to which the point belongs . ( this condition does not depend on how we choose a space - like curve . ) what kind of a constraint does it impose on the contour ? the space @xmath25\widetilde{\times } s^1 $ ] is a @xmath9 bundle over @xmath26 . the condition of the degeneracy of the metric defines a _ connection _ on this bundle . the definition of this connection is : the curve in the total space is considered horizontal , precisely if the corresponding collection of light rays is a degenerate surface . what is the curvature of this connection ? both @xmath27 and @xmath28 are kahler manifolds ( if we forgive that the metric on the first coset is not positive - definite ) . let us denote the kahler forms @xmath29 and @xmath30 . the curvature of our @xmath9-bundle is @xmath31 . a curve in the base space @xmath32 can be lifted to the horizontal curve in the total space if and only if a two - dimensional film ending on this curve has an integer kahler area ( integral of @xmath31 over this film should be an integer ) . moreover , it is lifted as a horizontal curve almost unambiguously , except that there is a `` global '' action of @xmath9 shifting @xmath33 on every light ray by the same constant . therefore , the moduli space of null - surfaces is the @xmath9 bundle over the space of contours in @xmath32 subject to the integrality condition which we described . to summarize , the moduli space of the null - surfaces of the second type is : @xmath34 here @xmath35 means the space of maps from the circle to @xmath36 ; for @xmath36 a kahler manifold @xmath37 means the space of maps satisfying the integrality condition . at this point we consider the null - surfaces without a parametrization ; therefore we divide by the group @xmath38 of the diffeomorphisms of the circle . turning on the fermionic degrees of freedom on the worldsheet we get the moduli space of supersymmetric null - surfaces @xcite : @xmath39 here @xmath40 is the phase space of the continuous spin chain @xcite . therefore the moduli space of null - surfaces is `` almost '' equivalent to the phase space of the continuous spin chain , except for the fiber @xmath22 and the reparametrizations @xmath38 . we have to explain what happens to the fiber and why the null - surface actually comes with the parametrization . also , we have to explain how the symplectic structure is defined on the moduli space of null surfaces . let us start with the parametrization . the phase space of the classical string has a boundary which consists of strings `` moving with the speed of light '' . a string moving very fast can be approximated by a null - surface . but one null - surface can approximate many different fast moving strings . the null - surface as we defined it so far `` remembers '' only the _ direction _ of the velocity at each point of the approximated string , but it misses the information about the ratios of the relativistic factors @xmath41 at different points of the string . although @xmath42 in the null - surface limit , the ratio @xmath43 for two different points on the worldsheet remains finite . therefore , if we want to think of the moduli space of the null - surfaces as the boundary of the phase space , we have to equip the null - surfaces with an additional structure . this additional structure is the _ parametrization_. a null - surface is a one - parameter family of the light rays . the parameterization is a particular choice of the parameter . in other words , it is a monotonic function @xmath44 from the family of light rays forming the null - surface to the circle , defined modulo @xmath45 . one can also think of it as a density @xmath46 on the set of light rays forming the null - surface . this density is roughly speaking proportional to the density of energy on the worldsheet of the fast - moving string , in the limit when it becomes the null - surface . we will now give the definition of @xmath44 . consider the family of string worldsheets @xmath47 converging to the null - surface @xmath48 . we will introduce a parametrization @xmath46 of @xmath49 in the following way . consider a killing vector field @xmath50 on @xmath1 , corresponding to some rotation of the sphere : @xmath51 here @xmath52 parametrizes the @xmath1 : @xmath53 . when @xmath54 is large , @xmath47 is close to @xmath49 , the string moves very fast and the conserved charge corresponding to @xmath50 is very large . we can approximate this charge by an integral over a spacial contour on the null - surface @xmath49 of @xmath55 times some density @xmath46 : @xmath56 } d\sigma\ ; u^{ij } \ ; x_{0,s}^i\partial_{\tau}x_{0,s}^j+(\mbox{terms vanishing at } l\to \infty)\ ] ] here @xmath57 is the @xmath1-part of the null - surface ; we choose the @xmath58 coordinate on the null - surface to be the affine parameter on the light ray normalized by the condition @xmath59 . ( [ ncd ] ) with the condition @xmath60 is the definition of @xmath46 , and also the precise definition of the large parameter @xmath54 , modulo @xmath61 . we choose @xmath44 as the parametrization . we can now say that _ the moduli space of parametrized null - surfaces is the boundary of the phase space of a classical string . _ we say that a family @xmath47 of extremal surfaces has a _ parametrized _ null - surface @xmath49 as a limit when @xmath62 if and only if * @xmath47 has @xmath49 as a limit when @xmath62 , as a continuous family of smooth two - dimensional surfaces in a smooth two - dimensional manifold , and * the density of @xmath63 approaches eq . ( [ ncd ] ) in the limit @xmath62 . this definition of the parametrization does not depend on which particular geometrical symmetry @xmath50 we use . an alternative way to define the same parametrization is to use a special choice of the worldsheet coordinates on @xmath64 . let us choose the worldsheet coordinates @xmath65 so that @xmath66 where @xmath67 is the projection of the string worldsheet to @xmath2 and @xmath68 is the projection to @xmath1 . then we define @xmath69 . in the null - surface limit @xmath46 defines the parametrization of the null - surface . the moduli space of parametrized null - surfaces as a manifold depends only on the conformal structure of the target space . but we can introduce additional structures on this moduli space which use the metric on @xmath6 . an important additional structure is the closed 2-form which originates from the symplectic form of the classical string . strictly speaking a differential form in the bulk of the manifold does not automatically determine a differential form on the boundary . indeed , suppose that we have a differential form , for example a 2-form @xmath70 in the bulk . we can try to define the `` boundary value '' @xmath71 of @xmath70 on the boundary in the following way . given two vector fields @xmath72 on the boundary , we find two vector fields @xmath73 in the bulk such that @xmath74 and @xmath75 . then we define @xmath76 . but the problem is that this definition will depend on the choice of @xmath77 and @xmath78 . intuitively , if @xmath79 is some other choice of a pair of vector fields inducing @xmath80 on the boundary , and the `` vertical component '' of @xmath81 is not small enough near the boundary , then @xmath82 . given this difficulty , how do we define the symplectic form on the space of null - surfaces given the symplectic form on the string phase space ? when we lift the vector field @xmath83 on the boundary to the vector field @xmath84 in the bulk , let us require that @xmath85 goes to zero when @xmath62 . we define @xmath54 by eq . ( [ ncd ] ) ; it is only an approximate definition at @xmath62 , but this is good enough for the purpose of our definition : @xmath86 where @xmath77 and @xmath78 are such that @xmath87 . one can see that @xmath71 has a kernel , which is precisely the tangent space to the fiber @xmath22 in the numerator of eq . ( [ supersymmetric ] ) . the moduli space has a symmetry @xmath9 rotating this fiber ; we will discuss this symmetry in the next section ; we will call it @xmath88 . therefore @xmath71 is the symplectic form on the moduli space of null - surfaces modulo @xmath88 . ( [ supersymmetric ] ) implies that the moduli space of _ parametrized _ null - surfaces _ modulo _ @xmath88 is the space of parametrized contours in the grassmanian : @xmath89 one can see that @xmath71 is equal to the integral of the symplectic form on the super - grassmanian pointwise on the contour , with the measure @xmath46 . the symplectic area of the film filling the contour is the generating function of the shift of the origin of the circle . therefore the integrality condition guarantees that the symplectic form does not depend on the choice of the origin on @xmath22 ; the symplectic form is horizontal and invariant with respect to the shifts of the origin of @xmath22 . our definition of the symplectic form on the space of null - surfaces used the target space metric ( just the conformal structure would not be enough ) and also the fact that the target space is a product of two manifolds . our discussion in this and the next section will be limited to the classical bosonic string . the moduli space ( [ supersymmetric ] ) of null - surfaces is a @xmath9-bundle . the @xmath9 symmetry shifting in the fiber @xmath22 plays an important role in the formalism . we will call it @xmath88 . on fig . [ fig : nss ] we have shown schematically how @xmath88 acts on the null - surfaces . we conjecture that @xmath88 corresponds to the length of the spin chain . generally speaking , the length of the spin chain is not conserved in the yang - mills perturbation theory @xcite , but it is probably conserved in the continuous limit ( this should be related to the discussion of the `` closed sectors '' in @xcite ) . it should be conserved modulo the corrections vanishing in the continuous limit . we therefore conjecture that there is a continuation of @xmath88 from the space of null - surfaces to the phase space of the classical string , at least to the region of the phase space corresponding to fast moving strings . we conjecture that this continuation is uniquely defined by the following properties : 1 . the action of @xmath88 preserves the symplectic structure . 2 . the action of @xmath88 does not change the projection of the worldsheet to @xmath2 . moreover , it preserves the projection to @xmath2 of the null - directions on the worldsheet . we require that the orbits of @xmath88 are closed ( otherwise , we would not have called it @xmath9 ) . the restriction of @xmath88 to the null - surfaces acts as we described ( see fig . [ fig : nss ] ) . the second property reflects the fact that @xmath88 corresponds to the length of the operator rather than its engineering dimension . let @xmath90 denote the hamiltonian of @xmath88 . let @xmath36 denote the phase space of the classical string , and @xmath91 denote the hamiltonian reduction of the phase space on the level set of @xmath92 . the basic conjecture is : _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ there is a one - to - one map from the phase space of the spin chain of the length @xmath93 to the reduced phase space of the classical string @xmath91 preserving the symplectic structure and commuting with the action of @xmath94 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the reduction by @xmath88 was discussed in @xcite but only in a sector @xcite in which @xmath88 acts as some element of @xmath95 . the perturbation theory in this sector was discussed in @xcite ( see also section 2 of @xcite ) . in this subsection we will explain how to continue the action of @xmath88 from the boundary of the phase space . most of this section is a partial review . the next step is considering the action of the killing vector field @xmath96 where @xmath97 is the global time in @xmath2 on the invariants of @xmath88 and bringing the result to the form suitable for the comparison with the field theory computation . here we are discussing only the first step . ] of section 3 of @xcite . consider the phase space of a particle moving on @xmath1 , and restrict to the domain where the velocity of the particle is nonzero . this domain is naturally a bundle over the moduli space of equators of @xmath1 ; let @xmath98 denote the projection map in this bundle . a point of the phase space , corresponding to the position @xmath99 and the velocity @xmath100 , projects by @xmath98 to the equator going through @xmath101 and tangent to @xmath83 . see the discussion in @xcite . the symplectic form on the phase space is expressed in terms of the symplectic form on the base and the connection form @xmath102 : @xmath103 where @xmath104 , @xmath105 ( @xmath106 is the momentum of the particle ) and @xmath70 is the symplectic form on the moduli space of equators . the moduli space of equators @xmath107 is a kahler manifold , the symplectic form @xmath70 is the kahler form . now it is easy to construct the action of @xmath9 . one takes @xmath108 this is a vertical vector field , it does not act on the base . the coordinate @xmath109 is essentially the angle along the equator on which the particle is moving . more explicitly : @xmath110 it is easy to see that the trajectories of the vector field @xmath111 on the phase space of a particle on @xmath1 are periodic with the period @xmath112 . one has to remember that this vector field is defined only on the open subset of the phase space , where the velocity of the particle is nonzero . but we consider fast moving strings , and the region of the phase space where the velocity is nearly zero is not important for us . in some sense , a string is a continuous collection of particles . therefore , it is natural to apply a similar construction to the string . treating the string as a continuous collection of particles requires the choice of the coordinates on the worldsheet . we will therefore introduce the conformal gauge : @xmath113 ( \partial_{\tau}x,\partial_{\sigma}x)=0 \end{array}\ ] ] in this gauge the symplectic form is : @xmath114 in the hamiltonian formalism , we introduce @xmath115 the @xmath2-component of the momentum , and @xmath116 the @xmath1-component of the momentum . now we will interpret the string as a collection of particles parametrized by @xmath44 . we are tempted to interpret the vector field ( [ particleuone]),([explicitdpsi ] ) acting pointwise in @xmath44 as the required @xmath88 symmetry . the generator of this symmetry would be @xmath117 . but this would be wrong . this field preserves the symplectic structure , does have periodic trajectories and acts correctly on the null surfaces . but unfortunately it does not preserve the gauge ( [ conformalgauge ] ) . it only commutes with the second constraint , @xmath118 . but it does not commute with the first one , @xmath119 . indeed , it commutes with @xmath120 but not with @xmath121 . therefore we should modify this vector field so that it still has periodic trajectories , but also commutes with the constraint . there is a systematic procedure to do this , order by order in @xmath122 , developed in @xcite . let us summarize this procedure , or perhaps a variation of it . to make sure that the modified vector field is hamiltonian ( preserves the symplectic structure ) we construct it as a conjugation of @xmath123 with some canonical transformation , which we denote @xmath124 : @xmath125\right]\ ] ] or schematically @xmath126 . since @xmath124 is a canonical transformation , @xmath127 is automatically a hamiltonian vector field . since @xmath124 is single - valued , @xmath128 generates periodic trajectories . it remains to construct @xmath124 such that @xmath128 commutes with the constraint @xmath129 . but to require that @xmath130 commutes with @xmath129 is the same as to require that @xmath111 commutes with @xmath131 $ ] the pullback of @xmath132 by @xmath124 . therefore we have to find such a canonical transformation @xmath124 that the pullback of @xmath132 with @xmath124 is annihilated by the vector field @xmath111 . in other words , we have to find a canonical transformation which removes @xmath109 from @xmath132 ; after this canonical transformation @xmath133 becomes @xmath134 where all the @xmath135 for @xmath136 are in involution with @xmath137 and @xmath135 is of the order @xmath138 . this was done in section 3 of @xcite . the canonical transformation can be expanded in @xmath139 ; the corresponding generating function is expanded in the odd powers of @xmath140 . the authors of @xcite gave the explicit expression for @xmath124 to the first order in @xmath140 , but they also give a straightforward algorithm for constructing the higher orders . ( we will reconsider the higher orders from a slightly different point of view in section [ sec : alternative ] , perhaps making this algorithm more precise . ) at the first order we need to find @xmath141 such that the canonically transformed constraint , which is a function of @xmath44 : @xmath142\}\ ] ] has zero poisson bracket with @xmath143 up to the terms of the order @xmath144 , for every @xmath44 . and @xmath141 should be of the order @xmath140 . in other words , we should have : @xmath145 one can see that @xmath146 works . notice also that this @xmath141 is reparametrization invariant ( where @xmath147 transforms as a density of weight one ) . therefore it commutes also with the second constraint @xmath148 . therefore , to the first order in @xmath140 the canonical transformation we are looking for is generated by this @xmath141 . then the generator of the @xmath88 is , up to the terms of the order @xmath144 : @xmath149 & & = \int d\sigma \left[|p_s|+ { 1\over 4|p_s| } \left[(\partial_{\sigma}x_s)^2- \left(d_{\sigma}{p_s\over |p_s|},d_{\sigma}{p_s\over |p_s| } \right)- { ( p_s,\partial_{\sigma}x_s)^2\over ( p_s , p_s)}\right ] + \ldots\right ] \nonumber\end{aligned}\ ] ] one can see immediately that the trajectories of this charge are closed up to the terms of the order subleading to @xmath140 . indeed , we have explained in section [ caseofparticle ] why the leading term gives periodic trajectories . and the second term ( which as we have seen is needed to make the charge commuting with the virasoro constraints ) averages to zero on the periodic trajectories of the first term . therefore ( see for example section 3 of @xcite ) the trajectories of @xmath90 do not drift at this order . we will discuss the higher orders in section [ sec : alternative ] . we have seen that the null - surface perturbation theory has a `` hidden '' symmetry @xmath88 . the existence of @xmath9 symmetries acting on the phase space is typical for integrable systems , at least for those which have a finite - dimensional phase space . corresponding conserved charges are called action variables @xcite . classical string in @xmath6 is an integrable system . therefore , we should not be surprised to find such an action variable in an arbitrary ( polynomial ) potential . when the particle moves very fast , it does not feel the potential . all the trajectories are periodic in the limit of an infinite velocity . therefore on the boundary of the phase space , when the velocity is infinite , we have an action variable @xmath150 the absolute value of the momentum . it is well known that the perturbation theory in @xmath151 allows us to extend this action variable from the boundary inside the phase space , but _ only in the perturbation theory_. for an arbitrary potential , the perturbation series must diverge , because in fact there is no additional conserved quantity besides the energy . therefore the @xmath9 will be actually broken by effects which are not visible in the perturbation theory , unless if the potential is such that the system is integrable . we want to thank v. kaloshin and a. starinets for discussions of this subject . ] . the local conserved charges in involution for the classical string in @xmath6 are explicitly known . therefore , instead of constructing @xmath88 in the perturbation theory , we can try to build it as some linear combination of the already known conserved charges . in this section , we will argue that the coefficients of this linear combination are actually fixed by the calculation of @xcite . consider a string in the target space which is a product of two manifolds @xmath152 and @xmath153 . we assume that the metric on @xmath152 has the lorentzian signature , and the metric on @xmath153 has the euclidean signature . we will need @xmath154 and @xmath155 , but let us first consider the general @xmath156 . the string worldsheet will be denoted @xmath64 . the classical trajectory of the string is an embedding @xmath157 . we are going to use the fact that the target space is a direct product . a point of @xmath156 is obviously a pair @xmath158 where @xmath67 is a point of @xmath152 and @xmath68 is a point of @xmath153 . therefore for each point @xmath159 we have @xmath160 , where @xmath161 and @xmath162 . consider the 1-forms @xmath163 and @xmath164 on the string worldvolume , @xmath163 taking values in @xmath165 and @xmath164 in @xmath166 . in other words , @xmath167 $ ] is a differential of @xmath101 . the metric on @xmath156 has the lorentzian signature , and we consider the string worldsheets which have the induced metric with the lorentzian signature . pick two vector fields @xmath168 and @xmath169 on @xmath64 , which are both lightlike but have a nonzero scalar product : @xmath170 these vector fields have a simple geometrical meaning . since the worldsheet is two - dimensional , at each point we have two different lightlike directions . the vector @xmath168 points along one lightlike direction , and @xmath169 points along another . pick a spacial contour @xmath171 on @xmath64 , and a 1-form @xmath172 on @xmath64 such that @xmath173 and @xmath174 . consider the following functional : @xmath175}[x]= \oint_c \nu { \sqrt{(dx_s(\xi_+))^2}\over \nu(\xi_+)}\ ] ] we will prove that this functional does not depend on a particular choice of @xmath168 , @xmath169 , @xmath172 and @xmath171 . this is therefore a correctly defined functional on the phase space of the string . indeed , the only ambiguity in the choice of @xmath168 is @xmath176 where @xmath177 is some function on the worldsheet . but this function cancels in ( [ chargexinu ] ) . the ambiguity in the choice of @xmath169 and @xmath172 is also in rescaling which does not change ( [ chargexinu ] ) . it remains to prove that ( [ chargexinu ] ) does not depend on the choice of the integration contour @xmath171 . to prove that ( [ chargexinu ] ) is independent of @xmath171 , let us choose coordinates @xmath178 on the worldsheet in such a way that the induced metric is @xmath179 . then @xmath168 is proportional to @xmath180 and @xmath169 is proportional to @xmath181 . in these coordinates @xmath182}=\oint_c d\tau^+ \sqrt{(\partial_{+}x_s)^2}\ ] ] the variation of @xmath183}$ ] under the variation of the contour is measured by the differential of the form : @xmath184 but on the equations of motion @xmath185 . therefore the integral does not depend on the choice of the contour . let us explain why on the equations of motion we have @xmath185 . let @xmath8 be the second quadratic form of the surface , @xmath186 ( here @xmath187 is the normal bundle to @xmath64 in @xmath156 ) . the second quadratic form is defined in the following way : suppose that the particle moves on @xmath64 with the velocity @xmath83 , then the acceleration of the particle is @xmath188 modulo a vector parallel to @xmath189 . for the surface to be extremal , the trace of @xmath8 should be zero . the trace of @xmath8 is the contraction of @xmath8 with the induced metric on @xmath64 ; it is a section of @xmath190 . the trace of @xmath8 is proportional to @xmath191 , therefore we should have : @xmath192 but notice that @xmath193 therefore @xmath194 . another conserved charge is : @xmath195}= \oint_c d\tau^- \sqrt{(\partial_{-}x_s)^2}\ ] ] are there charges containing higher derivatives of @xmath68 ? let us consider the following expression : @xmath196}(\tau_+,\tau_-)={1\over |\partial_+x_s| } \left(d_+{\partial_+ x_s\over |\partial_+x_s| } , d_+{\partial_+ x_s\over |\partial_+x_s|}\right)\ ] ] even though @xmath185 it is not true that @xmath197}$ ] is zero . the covariant derivatives @xmath198 and @xmath199 do not commute , therefore @xmath200 we have @xmath201w = r(\partial_+x,\partial_-x).w\ ] ] where @xmath202 is the riemann tensor of @xmath156 . now we have to start using that @xmath153 is a sphere . for @xmath155 , the riemann tensor is constructed from the metric tensor , and @xmath203w=\partial_+x ( \partial_-x , w)- \partial_-x ( \partial_+x , w)\ ] ] now consider the following differential form : @xmath204 using ( [ specialriemann ] ) we can show that @xmath205 , therefore @xmath206 is a local conservation law . we use the formula @xmath207 which is special for @xmath1 . we will denote this charge @xmath208}$ ] . there is also a charge @xmath209 which is obtained from ( [ chargeuv ] ) by replacing @xmath210 with @xmath211 and @xmath212 or @xmath198 with @xmath213 or @xmath199 . these charges are just the first examples of an infinite family of charges , which are all in involution . this infinite family was constructed in @xcite . a particularly important linear combination is @xmath214}-\tilde{q}^{[1]})\ ] ] the construction of this charge requires only that the target space is a direct product of two manifolds . consider a local conserved charge @xmath215 acting trivially on the @xmath216 part of the worldsheet . in the conformal gauge , this means that @xmath215 is constructed as a contour integral of some combination of @xmath68 and @xmath217 . let us decompose @xmath215 in the inverse powers of @xmath147 : @xmath218 where @xmath219 is a non - negative integer , the `` order '' of the charge ; @xmath220 is of the order @xmath221 , @xmath222 is of the order @xmath223 etc . we have to require that @xmath215 is in involution with the virasoro constraints . in particular , it should be in involution with @xmath224 for an arbitrary @xmath44 . ( here we used that @xmath215 is trivial in ads - part . ) let us now apply the canonical transformation @xmath124 which we described in section [ stringonasphere ] . after this canonical transformation @xmath133 becomes @xmath134 where all the @xmath135 for @xmath136 are in involution with @xmath137 and @xmath135 is of the order @xmath138 . and @xmath225 becomes @xmath226 , where @xmath227 is the canonically transformed @xmath215 . we should have : @xmath228 for an arbitrary @xmath44 . at the leading order in @xmath147 this implies that @xmath229 is in involution with @xmath230 . at the next order , it follows that for all values of @xmath44 the expression @xmath231 is in involution with @xmath232 . this implies that : @xmath233 since the vector field generated by @xmath229 is periodic , this equation implies that @xmath229 is in involution with @xmath234 . an analogous argument at higher orders shows that all the @xmath235 commute with @xmath229 . therefore @xmath227 is in involution with the expression @xmath229 which is the generator of @xmath88 . the conserved charges of @xcite do have an expansion of the form ( [ qexpansion ] ) therefore they should commute with @xmath88 . this reinforces our conjecture that @xmath88 should be a combination of the local conserved charges . we can try to make more transparent the geometrical meaning of @xmath88 by drawing an analogy with the liouville theorem for finite - dimensional integrable systems . a mechanical system with @xmath236-dimensional phase space is integrable if there are @xmath237 functions @xmath238 in involution with each other , and the hamiltonian is a function of @xmath238 . then , there are @xmath237 action variables @xmath239 , each of them being some combination of @xmath238 : @xmath240 such that each @xmath241 generates @xmath9 ( has periodic orbits ) . in this paper we are dealing with an infinite - dimensional system , a classical string in @xmath6 . we can take the first pohlmeyer charge @xmath183}-\tilde{q}^{[1]}$ ] as a hamiltonian . this hamiltonian is presumably integrable , because there is an infinite family of higher charges commuting with it . on the other hand , it does not have any special periodicity properties ( we do not see any reason why it would ) . this means that the closure of the orbit of @xmath183}-\tilde{q}^{[1]}$ ] is an invariant torus . our @xmath88 commutes with @xmath183}-\tilde{q}^{[1]}$ ] ( this fact is seen immediately , because @xmath183}$ ] can be rewritten as @xmath242 and by definition @xmath88 does not act on the ads - part of the worldsheet . ) therefore @xmath88 should be a shift of one of the angles parametrizing the invariant torus of @xmath183}-\tilde{q}^{[1]}$ ] . the angles parametrizing the invariant torus are in correspondence with its one - dimensional cycles . which cycle corresponds to @xmath88 ? every invariant torus can be connected by a one - parameter family of invariant tori to a torus on the boundary of the phase space ( or the one very close to the boundary ) . this means that every 1-cycle is connected to some 1-cycle on a torus on the boundary the space of null - surfaces . we should take that 1-cycle which is connected to the orbit of @xmath88 on the null - surfaces , described in section [ chargedefinition ] . the corresponding action variable is @xmath90 the generator of @xmath88 . these arguments show the uniqueness of @xmath88 . the first pohlmeyer charge @xmath183}-\tilde{q}^{[1]}$ ] has a special property : it actually generates @xmath88 on the boundary . therefore the difference between @xmath183}-\tilde{q}^{[1]}$ ] and @xmath90 should be a combination of charges vanishing at the boundary . we expect that this is an infinite and _ linear _ combination . indeed , the construction of @xcite tells us that the charge we are looking for is local at each order in @xmath140 . a nonlinear combination of the charges would be non - local ( a product of integrals ) . in section 3 we constructed @xmath88 as @xmath243 where @xmath111 is generated by @xmath229 and @xmath124 is the canonical transformation such that @xmath244 commutes with @xmath245 . this canonical transformation is constructed in the perturbation theory , order by order in @xmath246 . a disadvantage of this procedure is that at each order we have to require that our @xmath88 commutes with @xmath245 _ for any _ @xmath44 . since there are infinitely many values of @xmath44 we have to impose infinitely many conditions on @xmath124 at each order . at the first order , we have seen in section [ stringonasphere ] that these conditions are not really independent ; one generating function @xmath141 takes care of all of them see eq . ( [ foranysigma ] ) . at the higher orders , this is not immediately obvious . therefore , we would like to propose a slightly different way of constructing @xmath124 . let us forget for a moment about the virasoro constraint ; instead of the phase space of the string consider the space of harmonic maps @xmath247 . instead of requiring that @xmath88 commutes with @xmath245 , let us require that @xmath88 commutes with @xmath183}=\int d\sigma |\partial_+ x_s(\sigma)|$ ] . we will see that the requirement that @xmath88 commutes with @xmath183}$ ] already fixes @xmath88 in the perturbation theory , and the resulting @xmath88 will automatically commute with the virasoro constraints . as in section 3 , we look for the generator of @xmath88 as a pullback by a canonical transformation of @xmath137 . in other words , let us look for such a canonical transformation @xmath124 that @xmath229 commutes with @xmath248}$ ] ( the pullback of @xmath183}$ ] by @xmath124 ) . we can construct such a canonical transformation order by order in the perturbation theory . let us denote @xmath249 . we have : @xmath250}=k+q_1+q_2+\ldots\ ] ] under the rescaling @xmath251 : @xmath252 , @xmath253 , @xmath254 , @xmath255 . the symplectic structure is of the degree 1 : @xmath256 , therefore the poisson brackets are of the degree @xmath257 : @xmath258 . we can construct @xmath124 order by order in this grading . we have : @xmath259})=k+q'_1+q'_2+\ldots+q'_m+\ldots\ ] ] suppose that we have already found @xmath124 such that @xmath260 commute with @xmath261 . at the order @xmath219 , we want to modify @xmath124 by the canonical transformation with the generating function @xmath262 of the order @xmath263 so that @xmath264 commutes with @xmath261 . since @xmath261 is periodic , we can decompose @xmath265 where @xmath266 . then we should take @xmath267 repeating this procedure at higher orders , we end up with the function @xmath124 such that @xmath268})\}=0 $ ] . the reparametrization invariance is manifestly preserved at each order , therefore the resulting charge @xmath243 will commute with @xmath269 for any @xmath44 . also , the fact that @xmath183}$ ] is reparametrization - invariant and the arguments analogous to the discussion at the end of section [ sec : commutes ] show that @xmath243 will automatically commute with @xmath245 , as well as with the higher pohlmeyer charges . indeed , we know that @xmath248}=k+q_1'+q_2'+\ldots$ ] commutes with @xmath270 & & \{k , \phi_1\ } + \{q_1',\phi_0\}= \{|p_s|^2(\sigma),q_2'\}\rightarrow \{k , \{k , \phi_1\}\}=0 \rightarrow \{k , \phi_1\}=0\nonumber\\[5pt ] & & etc.\nonumber\end{aligned}\ ] ] we used the periodicity of the trajectories of @xmath261 when we claimed that @xmath271 implies @xmath272 . indeed , for any functional @xmath273 on the phase space , if @xmath271 then @xmath274 is constant on the trajectories of @xmath261 . but if this constant were nonzero , then the change of @xmath273 along the trajectory of @xmath261 would accumulate over the period of @xmath261 , which would contradict the single - valuedness of @xmath273 on the phase space . expanding ( [ chargeuv ] ) in the conformal gauge in the powers of @xmath275 we get : @xmath276}-\widetilde{q}^{[2]}\right ) = \int d\sigma \left [ -2|p_s|+{3\over |p_s|}(\partial_{\sigma}x_s)^2 -\right.\\[5pt]&&\left . -{3\over |p_s|^3}(p_s,\partial_{\sigma}x_s)^2 + { 4\over |p_s|}\left(d_{\sigma}{p_s\over |p_s| } , d_{\sigma}{p_s\over |p_s|}\right)+\ldots\right ] \nonumber\end{aligned}\ ] ] and for @xmath183}$ ] we get : @xmath277}-\widetilde{q}^{[1]}= \int d\sigma \left [ 2|p_s|+{1\over |p_s|}(\partial_{\sigma}x_s)^2 -{1\over |p_s|^3}(p_s,\partial_{\sigma}x_s)^2 + \ldots\right]\ ] ] we have : @xmath278}-\widetilde{q}^{[1]})- { 1\over 2}(q^{[2]}-\widetilde{q}^{[2]})\right]= \nonumber\\[5pt]= & & \int d\sigma \left[|p_s|+{1\over 4|p_s| } \left [ ( \partial_{\sigma}x_s)^2- \left(d_{\sigma}{p_s\over |p_s| } , d_{\sigma}{p_s\over |p_s|}\right ) -{(p_s,\partial_{\sigma}x_s)^2\over ( p_s , p_s ) } \right]+\ldots \right ] \nonumber\end{aligned}\ ] ] this coincides with the result ( [ periodicsecondorder ] ) for @xmath90 which we know from the perturbation theory . we see that up to the terms of the order @xmath279 the hamiltonian of @xmath88 can be represented as a sum of the first two commuting local charges . we conjecture that @xmath88 is in fact an infinite combination of the local conserved charges . the perturbation theory construction suggests that it should be a worldsheet parity - invariant combination . the coefficients of this linear combination can be found from considering the conserved charges of particular solutions . there is a special class of fast moving strings , the so - called `` rigid '' strings . for these `` rigid '' strings , the corresponding field theory operators are known a priori . these operators provide local extrema of the anomalous dimension in the sector with the given charges . these `` rigid '' solutions were classified in @xcite . they are related to the solutions of the neumann integrable system . the local conserved charges of some rigid strings were computed in @xcite . in @xcite the local conserved charges are denoted @xmath280 . ( this agrees with our notation @xmath90 for the hamiltonian of @xmath88 . ) the precise definition of @xmath280 is given in section 3 of @xcite . the relation to our notations is : @xmath183}-\tilde{q}^{[1]}=2{\cal e}_2 $ ] , @xmath208}+2q^{[1]}-\tilde{q}^{[2]}-2\tilde{q}^{[1]}=-4{\cal e}_4 $ ] . the conserved charges have the following structure : @xmath281 where @xmath282 , and @xmath5 is a particular combination of the @xmath95 momenta . the coefficients @xmath283 depend on what kind of a rigid string is considered ( the ratio of spins ) . but the authors of @xcite noticed that the coefficients @xmath283 for different values of @xmath237 are not independent . for all the solutions they considered , they find that : @xmath284 this means that up to the terms of the order @xmath285 we should have : @xmath286 at first this formula looks rather strange , because it seems to imply that a certain combination of pohlmeyer charges ( which all commute with @xmath95 ) is equal to some component of the angular momentum ( which transforms in the adjoint of @xmath95 ) . we propose the following resolution of this puzzle . the right hand side of ( [ jfromcale ] ) is actually the action variable , which for a particular class of the solutions considered in @xcite happens to be equal to the @xmath95 charge @xmath5 ( because these particular solutions correspond to the chiral operators on the field theory side ; see section 2 of @xcite ) . in other words , for this particular class of solutions the angular momentum @xmath287 should be equal to our action variable @xmath90 . the general formula should have on the left hand side @xmath90 , the generator of @xmath88 , instead of @xmath287 : @xmath288 this gives the expansion of the generating function of @xmath88 to the order @xmath289 . it would be interesting to check explicitly , beyond the order @xmath140 , that this hamiltonian generates periodic trajectories . here we want to present a slightly different and perhaps simpler way of thinking about the continuation of @xmath88 in the perturbation theory . consider the hamiltonian vector field @xmath290 corresponding to the first pohlmeyer charge @xmath291 . consider the canonical transformation @xmath292 this canonical transformation is the hamiltonian flow generated by @xmath291 by the time @xmath112 . the trajectories of @xmath291 are almost periodic in the null - surface limit , therefore we can write @xmath293 where @xmath294 is a vector field of the order @xmath295 . this vector field can be constructed in the following way . let us choose the conformal gauge on the worldsheet . we know from ( [ ecal2 ] ) that @xmath296 where @xmath177 is of the order @xmath140 . taking into account that @xmath297 we get @xmath298+\ldots \right\ } \nonumber\end{aligned}\ ] ] this defines @xmath299+\ldots\ ] ] in the perturbation theory . ( notice that @xmath177 can be decomposed in the fourier series @xmath300 so that @xmath301 and then the leading term of @xmath83 is the zero mode @xmath302 ; this is the `` averaging '' procedure of @xcite . ) the vector field @xmath294 defines the vector field on the moduli space of null - surfaces as a limit @xmath303 ( where @xmath54 was defined in section [ sec : parametrization ] ) . this vector field determines the slow evolution of the null - surface ; it is the hamiltonian vector field of the landau - lifshitz model on the moduli space of the null - surfaces modulo @xmath88 @xcite . as in @xcite we can consider the improved currents @xmath304 . by definition @xmath304 is a linear combination of @xmath305 such that @xmath306 . the hamiltonian of the landau - lifshitz model is the null - surface limit of @xmath307 , more precisely @xmath308 . given that @xmath291 and @xmath307 are in involution , this implies that for some @xmath309 we have @xmath310 where @xmath311 is of the order @xmath312 . again , @xmath313 determines a vector field on the moduli space of null - surfaces . ( it can be also defined as @xmath314 . ) this vector field commutes with the time evolution of the landau - lifshitz model . we conjecture that this vector field is generated by the second conservation law of the landau - lifshitz model , which is proportional to the null - surface limit is invariant under the @xmath88 symmetry of the null - surfaces , because the @xmath88 symmetry of the null - surfaces is generated by the conserved quantity @xmath291 which commutes with @xmath304 . ] of @xmath315 , more precisely @xmath316 or @xmath317 . repeating this procedure we get @xmath318 in the perturbation theory . these arguments lead us to the following conclusion . first , we see once again that there is a linear combination @xmath319 generating periodic trajectories . second , the moduli space of null - surfaces in @xmath6 modulo @xmath88 is naturally equipped with the infinite tower of hamiltonians in involution which are the null - surface limit of the pohlmeyer charges . this is the generalized landau - lifshitz model . given a manifold with the metric of the lorentzian signature it is possible to construct the extremal surfaces in this manifold as perturbations of the null - surfaces . in the special case when the manifold is @xmath6 the ads / cft correspondence predicts that the extremal surfaces ( which are the same as classical string worldsheets ) correspond to the states of the large r - charge in the @xmath7 supersymmetric yang - mills theory . from this point of view considering the extremal surface as a perturbation around the null - surface corresponds to considering the state of the interacting yang - mills theory as a perturbation of the state of the free yang - mills theory . this correspondence has the following important features : 1 . locality . in the planar limit ( the limit of infinitely many colors ) the yang - mills perturbation theory is local in the following sense : the feynman diagramms involve only intractions of those elementary field operators which stand next to each other in the product under the trace . we expect that the correspondence between the parton chains and the string worldsheets is local in each order of the perturbation theory , and therefore the locality of the planar yang - mills perturbation theory should correspond to the locality of the string worldsheet theory . 2 . integrability . the classical string worldsheet theory in @xmath6 is an integrable system . because of the integrability , there is an infinite family of local conserved charges in involution . in this paper we have argued that an infinite linear combination of these local charges generates periodic trajectories on the string phase space . this statement can be verified order by order in the null - surface perturbation theory , and it is local at each order . this means that the `` slow evolution '' of nearly - degenerate extremal surfaces @xcite is essentially controlled by the pohlmeyer charges ( we will further discuss the slow evolution and how it is related to the pohlmeyer charges in the second paper of @xcite ) . it would be interesting to further study the null - surface perturbation theory from the point of view of the integrability . it would be especially interesting to study those manifestations of the integrability which are local . the bcklund transformations @xcite is one example . they allow us to construct a new extremal surface from a given extremal surface , and in the null - surface perturbation theory these transformations are well - defined and local at each order . the bcklund tranformations are closely related to the local conserved charges , and in fact the hidden symmetry @xmath88 can be considered as a consequence of the special properties of these transformations . we will discuss the relation between @xmath88 and the bcklund transformations in the third paper of @xcite . the general problem is , to study those aspects of the integrability which are local in the null - surface perturbation theory . ( without a reference to the null - surface perturbation theory , we would define the locality as some sort of an independence of the choice of the boundary conditions . ) this problem arises also on the field theory side . the feynman diagramms in the planar limit are local , but we usually compute the anomalous dimension of the single - trace operators which requires summing over the whole parton chain . the spectrum of single - trace operators at large @xmath8 is certainly an invariant of the theory , but it is non - local . if it is true that the planar @xmath7 yang - mills theory is integrable , it would be important to understand the integrability as much as possible in terms of the local properties of the parton chain ( perhaps on the level of the individual feynman diagramms ) . we have defined the @xmath88 strictly speaking in the perturbation theory , but it should be actually well - defined in the domain of the string phase space where the velocity of the string is large enough . in other words , the series defining the @xmath88 in fact converges if the string moves fast enough . it would be interesting to study the global properties of @xmath88 . an important question is what happens to @xmath88 after the quantization . to answer this question we should first include fermions . important steps in this direction were made recently in @xcite . it would be interesting to understand better why the `` length '' is conserved on the field theory side . ( why is there a quantum number @xmath54 with a well - defined classical limit ? ) to which extent the conservation of @xmath54 is related to the integrability of the planar yang - mills theory ? what happens to @xmath54 when we turn on the fermions ? null - surfaces are obviously an important ingredient in our construction . the correspondence between the null - surfaces and the `` engineering '' operators in the free field theory is rather straightforward ; the null - surfaces in @xmath6 appear very naturally in the description of the coherent states of the free theory @xcite . is there any way to see directly on the field theory side , that turning on the yang - mills interaction corresponds to the deformation of the null - surface into the extremal surface ? i would like to thank s. frolov , a. gorsky , v. kaloshin , a. kapustin , a. marshakov , s. minwalla , m. van raamsdonk , a. starinets , k. zarembo and especially a. tseytlin for discussions , and g. arutyunov for a correspondence on the local conserved charges . i want to thank a. tseytlin for comments on the text . i want to thank the organizers of the string field theory camp at the banff international research station for their hospitality while this work was in progress . this research was supported by the sherman fairchild fellowship and in part by the rfbr grant no . 03 - 02 - 17373 and in part by the russian grant for the support of the scientific schools nsh-1999.2003.2 .
we study which geometric structure can be constructed from the vierbein ( frame / coframe ) variables and which field models can be related to this geometry . the coframe field models , alternative to gr , are known as viable models for gravity , since they have the schwarzschild solution . since the local lorentz invariance is violated , a physical interpretation of additional six degrees of freedom is required . the geometry of such models is usually given by two different connections the levi - civita symmetric and metric - compatible connection and the weitzenbck flat connection . we construct a general family of linear connections of the same type , which includes two connections above as special limiting cases . we show that for dynamical propagation of six additional degrees of freedom it is necessary for the gauge field of infinitesimal transformations ( antisymmetric tensor ) to satisfy the system of two first order differential equations . this system is similar to the vacuum maxwell system and even coincides with it on a flat manifold . the corresponding `` maxwell - compatible connections '' are derived . alternatively , we derive the same maxwell - type system as a symmetry conditions of the viable models lagrangian . consequently we derive a nontrivial decomposition of the coframe field to the pure metric field plus a dynamical field of infinitesimal lorentz rotations . exact spherical symmetric solution for our dynamical field is derived . it is bounded near the schwarzschild radius . further off , the solution is close to the coulomb field .
gr-qc0510075
gr is a classical field theory for 10 independent variables the components of metric tensor @xmath0 . it is well known , however , that some problems inside and beyond einstein s gravity require a richer set of 16 independent variables the components of the coframe ( aka reper , vierbein , ... ) . in the following issues of gravity , the coframe is not only a useful tool but often can not even be replaced by the standard metric variable : * hamiltonian formulation @xcite,@xcite ; * positive energy proofs @xcite ; * fermions on a curved manifold @xcite,@xcite ; * supergravity @xcite ; * loop quantum gravity @xcite , @xcite . absolute ( teleparallel ) frame / coframe variables were introduced in physics by einstein in 1928 with an aim of a unification of gravitational and electromagnetic fields ( for classical references , see @xcite ) . in gr description , the additional six degrees of freedom do not have a physical sense and treated as a type of a gauge symmetry of the metric tensor . it was already noticed by einstein , that 16 reper components can not be completely equivalent to 10 components of the metric tensor . indeed , the supergravity and the loop quantum gravity models apparently can be formulated only in term of vierbein . so , it is natural to study which geometric structure can be constructed from the vierbein ( frame / coframe ) variables and which field models can be related to this geometry . this is a subject of the current paper . the frame field @xmath1 and its dual , the coframe field @xmath2 , have a well defined geometrical sense . in particular , even been considered as independent physical fields , they provide a special ( absolute ) reference basis . for the bases , @xmath3 , fixed at a point , the construction gives an invariant meaning to the components of a tensor , thus it emerges in violation of the rotational and lorentz invariance @xcite . however , when the _ global _ ( rigid ) lorentz transformations of the absolute basis fields are acceptable , the frame components of a tensor are transformed merely by the lorentz transformation law . thus , some interrelation between the lorentz invariant field theories and the diffeomorphism invariant gravity emerges . when the absolute basis field is restricted to a point , the lorentz invariance requires to consider it not alone but as a member of a class of equivalent bases . the equivalence relation for this class is provided by a group @xmath4 of transformations @xmath5 , which has to include local physics symmetries , i.e. , the rotations and the boosts . when we are dealing with the absolute basis fields on a manifold , the relation between the frames at different points is not governed by the constrains of local physics . consequently , three principally different possibilities are open : \(i ) _ riemannian geometry . _ the frames at distinct points are not related to one another at all , i.e. the dependence of the elements of the group @xmath4 on a point is absolutely arbitrary . it is possible only if the corresponding geometry , i.e the metric and the connection , respect the arbitrary transformations of the frame field . the unique connections with this property is the _ levi - civita connection _ of riemannian geometry . this case is realized in the standard gr when it is formulated in an arbitrary non - holonomic frame . the frame / coframe fields are not physical in this case and only play a role of a useful reference tool . they can be replaced even by holonomic bases generated by local coordinates . \(ii ) _ teleparallel geometry . _ another limiting case emerges when the frames in distinct points are strongly connected to one another . it means that when a frame in an arbitrary point is rotated in a certain angle , the frames at distant points immediately rotated in the same angle . the unique connection with this property is the flat _ weitzenbck connection . _ although the scalar curvature is zero for such a connection , the action functional can be constructed from the torsion tensor . it is natural to require this action functional to be invariant only under global transformations of the basis field . the analysis of this model shows that it has not spherical symmetric solutions with newtonian behavior at infinity . as a result , this absolute teleparallel gravity model is not physical . \(iii ) _ gauge geometry . _ besides the two limiting cases given above , there is a family of geometries with the field of frame rotations satisfying some system of differential equations . the basic geometrical quantities , metric and connection , have to be invariant under local transformations of the frame field satisfied these equations . in this construction , the frame field does not emerge explicitly , so we will refer to the corresponding geometrical structure as a _ gauge coframe geometry_. in alternative gravity models ( see @xcite @xcite and the reference given therein ) , the frame variable appears as an independent physical field . in the first order approximation @xcite , the coframe field is separated to a sum of two independent fields - a metric field and an antisymmetric field of the rang two . the same separation emerges in the lagrangian , in the field equation and also in the energy - momentum tensor . in the current paper we show that a geometrical and physical interpretation of the antisymmetric part can be prolongated in the higher order approximations . our main result is : in the viable coframe models , alternative to gr , the additional degrees of freedom can be interpreted as a new dynamical field which behavior in the first order approximation is described by the vacuum maxwell - type system . the aim of this paper is to derive the mentioned coframe gauge geometrical structure and its possible applications to classical fields . the paper is organized as followed . in section 2 , we construct a most general class of connections which are linear in the first order derivatives of the coframe field . this six - parametric family involves the levi - civita and the weitzenbck connections as special limiting cases . we also identify the sub - families of torsion - free and metric compatible connections . in section 3 , the behavior of the coframe connections under local @xmath6 transformations of the coframe is considered . besides the levi - civita and the weitzenbck connections , we identify a sub - family of gauge invariant connections . the corresponding constraints compose a system of 8 first order partial differential equation for six independent entries of a @xmath6 matrix . this situation is very similar to the standard maxwell system of eight field equations for six independent components of the electromagnetic field . in section 4 , we derive the same system of constrains from a different ( physical ) point of view . we require a most general quadratic coframe lagrangian to be invariant under @xmath6 transformations of the coframe . the gauge invariant lagrangians turn out to be in a correspondence with the known viable coframe models having the schwarzschild solution . in section 5 , we study the first order approximation to the system of the constrains . on a flat manifold , this system coincides with the vacuum maxwell equations . on a curved manifold , it turns out to be a system of covariant maxwell - type equations when the covariant derivatives are considered relative to the weitzenbck connections . in section 6 , we derive an exact spherical symmetric solution to our system of constrains . we also compare our model with the standard description of interaction between gravitational and electromagnetic fields . in contrast to the standard einstein - maxwell system , our model predicts mass dependence in the coulomb - type law . we derive the exact expression of the corresponding correction term . in section 7 , we give an outlook of the proposed alternative model and discuss how ( and if ) the additional degrees of freedom can be related to the ordinary electromagnetic field . we use the greek indices @xmath7 to identify the specific vector fields of the frame @xmath1 and the specific 1-forms of the coframe @xmath2 . the roman indices @xmath8 refer to local coordinates . summation from 0 to 3 is understood over repeated indices of both types ( einstein s summation convention ) . two types of indices and basically different . in particular , they can not be summed ( contracted ) in the expressions @xmath9 or @xmath10 . the lorentz metric is used with the sign agreement @xmath11 . the spatial indices are denoted as @xmath12 . we denote the coefficients of connection as @xmath13 and @xmath14 . such notation is useful for the exterior form representation . in order to go back to the ordinary tensorial notations @xmath15 , it is enough to move the last index to the first position . the symmetrization and antisymmetrization operators are used in the normalized form , i.e. , @xmath16 and @xmath17}= ( 1/{p!})(a_{i_1\cdots i_p}- \cdots)$ ] let a @xmath18-manifold be endowed with a frame field @xmath1 and a coframe field @xmath2 . all the structures are assumed to be smooth almost everywhere . in local coordinates @xmath19 , the fields are expressed respectively as @xmath20 i.e. , by two @xmath21 matrices @xmath22 and @xmath23 of smooth functions . these two basis fields are assumed to be reciprocal to each other : @xmath24so , in fact , only one field , @xmath1 or @xmath2 , is an independent variable . we prefer to use the coframe field @xmath2 as a basis variable since it is suitable for a compact exterior form representation . for a rigidly fixed coframe field , the components of an arbitrary tensor obtain an invariant sense when referred to the special bases @xmath25 . it emerges in hard violation of lorentz invariance . thus , in order to have a lorentz invariant field model , the coframe field has to be defined only up to global lorentz transformations . the gauge paradigm requires to localize the global transformations , so we define a _ coframe structure _ as a triplet @xmath26 where @xmath27 is a smooth manifold , @xmath2 is a smooth coframe field on @xmath27 , and @xmath28 is a local ( pointwise ) group of transformations @xmath29 . although the coframe variable itself is a pure geometrical object , it is not clear what geometry is generated by the structure ( [ str1 ] ) for different groups @xmath4 . we accept the cartan viewpoint that treats a geometrical structure as a pair of two independent objects : a metric tensor @xmath30 and an asymmetric connection field @xmath31 . for the specific coframe structure ( [ str1 ] ) , we require both fields to be explicitly constructed from the coframe components . in other words , we specify ( [ str1 ] ) to @xmath32 in correspondence to local physics , the metric tensor has to be defined as @xmath33 or , in components , @xmath34here @xmath35 is the lorentzian metric in the tangential vector space . indeed , in a small neighborhood of a point , the coframe field is approximately holonomic @xmath36 , so ( [ metr1 ] ) is locally approximated by lorentzian metric , see @xcite . due to the index content , ( [ metr ] ) is a unique construction up to a constant scalar factor . this factor can be neglected by rescaling the coframe field ( recall that the global @xmath37-transformations of the structure are admissible ) . hence ( [ metr ] ) is a unique construction of the metric tensor from the coframe components when the conformation with the local lorentz metric is required . this definition restricts the freedom of the coframe transformation to local pseudo - rotations , i.e. , at every point @xmath38 . the entries of this @xmath6-matrix remain arbitrary functions of a point . let a manifold be endowed with a field of asymmetric cartan connections . relative to a local coordinate chart @xmath19 , every connection is represented by a set of @xmath39 independent functions @xmath40 the coefficients of the connection . the only condition the functions @xmath40 have to satisfy is to transform , under a change of coordinates @xmath41 , by an inhomogeneous linear rule : @xmath42 where the derivatives are denoted as @xmath43 and @xmath44 . for the coframe connections , we require @xmath13 and @xmath45 to be explicitly constructed from the coframe components and their derivatives . moreover , we require @xmath13 to be linear in @xmath46 linear connection . this requirement is in parallel to the standard construction of riemannian geometry . indeed , the levi - civita connection can be treated as a unique linear connection which can be constructed from the first order derivatives of the metric components @xmath47 . two examples of the coframe connection satisfied the requirement above are well known : \(i ) the weitzenbck connection is defined as @xmath48 it is straightforward to check the transformation rule ( [ cc1 ] ) for this expression . the riemann curvature of this connection is identically zero , so the connection is flat . we denote the antisymmetric combination ( torsion ) of ( [ weiz2 ] ) and its trace as @xmath49k}\,,\qquad c_i = c_{mi}{}^m\,.\ ] ] when the transformation ( [ cc1 ] ) is applied to these quantities , the inhomogeneous parts are canceled . consequently , ( [ c - def ] ) change as tensors under the transformations of coordinates . their behavior under local transformations of the coframe is an independent property , which we will examine in the consequence . \(ii ) the levi - civita connection of riemannian geometry , @xmath50 can be rewritten as a linear combination of the first order derivatives of the coframe components . indeed , we substitute ( [ metr ] ) into ( [ lc - con ] ) to have @xmath51}+ \vt^\a{}_i\vt^\b{}_{[k , j]}\big)\nonumber\\ & = & \og_{(ij)k}+\og_{[ki]j}+\og_{[kj]i}\nonumber\\&= & \og_{ijk}-c_{ijk}+c_{kij}+c_{kji}\ , . \end{aligned}\]]the transformation law ( [ cc1 ] ) for this connection is a well known fact . the invariance of ( [ lc1 ] ) under arbitrary local @xmath6-transformations of the coframe field is clear from ( [ lc - con ] ) . we are looking now for a most general connection @xmath52 which is linear in the first order derivatives of the coframe , i.e. , for a generalization of ( [ weiz2 ] ) and ( [ lc1 ] ) . similarly to these special cases , the coefficients in the general linear combination of the first order derivatives have to be linear in the coframe components . in order to construct the most general coframe connection , we apply the well known fact : the difference of two arbitrary connections is a tensor . thus a general coframe connection can be represented as the weitzenbck connection plus a tensor , @xmath53 or , alternatively , as the levi - civita connection plus a tensor , @xmath54the tensor @xmath55 of ( [ con ] ) has to be itself linear in the first order derivatives of the coframe field . consequently it can be written as @xmath56}\big)= \chi_{ijk}{}^{mnl}\,c_{mnl}\,.\ ] ] the `` constitutive tensor '' @xmath57 can involve only the components of the metric tensor and the kronecker symbols . in view of the symmetry relation @xmath58l}$ ] , it is enough to restrict the general expression to @xmath59substituting into ( [ con ] ) we have @xmath60or , equivalently , @xmath61the @xmath62-terms in ( [ g - con1 ] ) are defined already on a linear connection manifold without a metric , the @xmath63-terms require a metric structure on a manifold . we give an equivalent form of the family ( [ g - con ] ) via the levi - civita connection , @xmath64 hence , in contrast to riemannian geometry with a unique levi - civita connection , we have , on the coframe manifold , an infinity 6-parametric family of connections constructed from the coframe components only . on a linear connection manifold , an asymmetric connection is characterized by its _ torsion tensor _ which is defined as @xmath65}{}^k\,.\ ] ] this skew - symmetric tensor ( @xmath66 ) does not depend on the metric structure . for the exterior form representation , see @xcite@xcite . for the weitzenbck connection , the torsion is given by @xmath67 . due to ( [ weiz2 ] ) , this tensor is zero only for closed coframe 1-forms which satisfy @xmath68 . the levi - civita connection is symmetric in the down indices , thus its torsion is zero . the torsion tensor of the general connection ( [ g - con ] ) is given by @xmath69 consequently , the relations @xmath70 extract a 3-parametric sub - family of identically symmetric ( torsion - free ) connections . geometrically , the non - zero torsion means a non - trivial parallel transport even on a flat manifold . the field models on a manifold with a connection of a non - zero torsion can include additional terms for interaction with torsion . the physical effects of a non - zero torsion was studied intensively , see the reports @xcite@xcite and the references given therein . probably the most important result of these numerous studies is the fact that non - zero torsion does not contradict the standard physical paradigm . another algebraic characteristic of an asymmetric connection can be given on a manifold endowed with a metric . the _ non - metricity tensor _ , @xmath71 , is defined by the covariant derivative of the metric tensor , @xmath72where @xmath73 . for the weitzenbck and the levi - civita connections , the non - metricity tensor is zero . for the general connection ( [ g - con ] ) , it is given by @xmath74hence , the connection ( [ con ] ) is metric - compatible ( has an identically zero tensor of non - metricity ) if @xmath75 two tensors , the torsion an the non - metricity , characterize the affine connection uniquely @xcite . in fact , we can decompose the general affine connection ( [ g - con ] ) as @xmath76consequently , a manifold endowed with an asymmetric connection can be treated as a riemannian geometry with two additional tensors of torsion and non - metricity . geometrically , a non - trivial non - metricity tensor means change of the lengths and angles when a parallel transport of vectors over a closed curve is applied . such violation of rotational and lorentz invariance bring us rather far from the standard physical paradigm . so , from the physical point of view , the metric - compatible constrains ( [ m - comp ] ) are better motivated then the torsion - free constrains ( [ tor - free ] ) . when the requirements ( [ tor - free],[m - comp ] ) are considered together , we come to a unique torsion - free and metric - compatible coframe connection given by the parameters @xmath77 certainly , it is no more than the ordinary levi - civita connection ( [ lc - con ] ) . although the riemannian curvature tensor is a classical subject of differential geometry , in the case of an asymmetric connection , slightly different notations are in use . we accept the agreements used in metric - affine gravity @xcite . the riemannian curvature tensor of a general asymmetric connection is defined as @xmath78 we are interested in the riemannian curvature of the coframe connection ( [ g - con ] ) . due to ( [ cl6 ] ) , @xmath79 involves the parameters @xmath80 in linear and quadratic combinations . it vanishes for zero values of the parameters , i.e. , for the weitzenbck connection . moreover , the weitzenbck connection @xmath81 is a unique coframe connection of an identically zero curvature , for a proof , see @xcite . for classical fields applications , we are interested in the scalar curvature , i.e. , in the hilbert - einstein lagrangian , @xmath82 when ( [ g - con ] ) is substituted here , we obtain @xcite @xmath83 where @xmath84}+\rho_3\big(c^{ijk}-2\,c^{im}{}_mg^{jk}\big)\,.\ ] ] the dimensionless coefficients @xmath85 are quadratic polynomials in the constants @xmath80 . the quadratic part of the ( [ he - tel ] ) is well known from the gravity fields models based on the coframe fields @xcite , @xcite , @xcite . up to a change of parameters , it is the most general scalar expression quadratic in the first order derivatives of the coframe field . usually , this expression is treated as an arbitrary linear combination of the squares of the weitzenbck torsion . note certain outputs of our coframe connections approach : * the quadratic coframe lagrangian is regarded as a standard einstein - hilbert lagrangian calculated on a general asymmetric coframe connection . * the six - parametric connection ( [ g - con ] ) is involved in the lagrangian ( [ he - tel ] ) only via three independent combinations . thus three additional requirements ( for instance , the compatibility to the metric tensor ) can be applied without changing the physical content of the model . let us examine now the behavior of the geometrical structure ( [ str2 ] ) under local transformations of the coframe field @xmath86the metric tensor itself is invariant under local lorentz transformations , thus , for every @xmath87 , the matrix @xmath88 . thus we are looking for a subclass of the connections ( [ g - con1 ] ) that invariant under pseudo - orthonormal transformations ( [ l - tr0 ] ) . since the coframe field appears in ( [ str2 ] ) only implicitly , ( [ l - tr0 ] ) is a type of gauge transformation . certainly , the levi - civita connection is in the desired subclass . consequently , our invariance condition does not bring us too far from the standard riemannian geometry . moreover , with this requirement , we can have only small additions also on the physical field level , see sect 5 . in a coordinate basis , the coframe components change under ( [ l - tr0 ] ) as @xmath89 in this paper , we will deal only with the infinitesimal version of the transformations ( [ l - tr0 ] ) . consequently , we restrict to @xmath90correspondingly , the matrix @xmath91 is antisymmetric . under ( [ inf - l ] ) , the components of the basis fields change as @xmath92it should be noted that our analyses is restricted to the linear approximation of the transformation matrix @xmath93 . in particular , we neglect with the second order term in the transformation of the metric tensor and assume it invariant under ( [ l - tr2 ] ) . we are examining now , under what conditions , the connections @xmath94 are invariant under the transformations ( [ l - tr2 ] ) . the weitzenbck connection changes as @xmath95 this connection is invariant , @xmath96 , only if @xmath97 , i.e. , only for global ( rigid ) transformations . recall , that the torsion of this connection is a building block of our construction . denoting its variation by @xmath98 , we have from ( [ w - change ] ) @xmath99correspondingly , @xmath100 for variation of the general connection , we use the form ( [ g - con2 ] ) , in which the invariant part ( the levi - civita connection ) is extracted explicitly . thus the invariance condition , @xmath101 , is expressed as @xmath102 let us examine now how this equation can be satisfied : \(i ) _ local transformations . _ certainly , ( [ g - con2x ] ) holds when all its coefficients are zero , i.e. , the connection is of levi - civita . the quantity @xmath98 is arbitrary in this case , thus also the transformation matrix @xmath103 is non - restricted . thus arbitrary @xmath6 transformations of the coframe are acceptable as it has to be in riemannian geometry . \(ii ) _ rigid transformations . _ another type of solution to ( [ g - con2x ] ) emerges by requiring @xmath104 . in this case we have only rigid transformations of the coframe , which are certainly admissible for all coframe connections . \(iii ) _ gauge transformations . _ we are looking now for a nontrivial solution to ( [ g - con2x ] ) , such that the the elements of the matrix @xmath103 are dynamical , i.e. , satisfy a well posed system of partial differential equations . a first consequence of ( [ g - con2x ] ) is its trace , i.e. , the contraction in two indices . since the left hand side of the equation is asymmetric , we have here three different traces all of the same form : @xmath105these conditions are necessary for invariance of the connection . here , the coefficients @xmath106 are equal to three different linear combinations of the parameters @xmath107 . in particular , all @xmath106 are zero for the levi - civita connection . we will consider , however , an alternative solution with at least one nonzero parameter @xmath106 . so we derive that the variations of the coframe field have to satisfy the equation @xmath108 substituting it into ( [ g - con2x ] ) we obtain @xmath109the completely antisymmetric combination of this equation yields @xmath110}=0\,.\]]again , for the levi - civita connection , the coefficient of this equation is zero . in the alternative case of a nonzero coefficient , the variations of the coframe field has to satisfy the equation @xmath111}=0\,.\ ] ] certainly this result holds only for @xmath112 , otherwise we do not have the condition ( [ trace3 ] ) at all . we neglect with this special case because it gives an undetermined system of conditions on @xmath103 . when ( [ trace1 ] ) and ( [ trace4 ] ) are substituted into ( [ g - con2x ] ) we remain with @xmath113we have to restrict now the coefficients , otherwise we obtain @xmath104 , i.e. , only the rigid transformations . hence , @xmath114 or , equivalently , @xmath115consequently , we have derived a family of coframe connections @xmath116}+ \nonumber\\ & & \qquad \a_2g_{ik}c_k+\a_3g_{jk}c_i+\b_1g_{ij}c_k\ , , \end{aligned}\]]which are invariant under local restricted variations of the coframe field . in this expression , the coefficients @xmath117 can be taken almost arbitrary . certainly , some exceptional values , for instance @xmath118 , are forbidden . the torsion of the connection ( [ g - con2 g ] ) is @xmath119}-(\a_2-\a_3)c_{[i}g_{j]k}\,,\end{aligned}\]]while the non - metricity tensor is @xmath120 starting with a coframe field on a manifold @xmath27 we derived a most general six - parametric family of connections which are linear in the first order derivatives of the coframe components . in this family , we identified , three subclasses of torsion - free , metric - compatible , and gauge invariant connections . let us compare the conditions determined these subclasses . since the gauge invariant condition ( [ par1 ] ) contradicts to ( [ tor - free ] ) , the gauge invariant connection can not be torsion - free . as for the conditions of metric - compatibility ( [ m - comp ] ) , they are correlated ( even partially overlap ) with ( [ par1 ] ) . consequently , we can introduce a subclass of _ gauge invariant metric compatible connections _ with the parameters @xmath121 two parameters @xmath122 and @xmath123 remain free . the corresponded connection is @xmath124}+ \a_3(g_{jk}c_i - g_{ij}c_k)\ , . \end{aligned}\]]or , equivalently , @xmath125}+ \a_3(g_{jk}c_i - g_{ij}c_k)\ , . \end{aligned}\ ] ] although two parameters @xmath122 and @xmath123 remain free , the torsion of ( [ join - con1 ] ) @xmath126}+\a_3c_{[i}g_{j]k}\,.\end{aligned}\ ] ] can not be zero . recall that some values of the parameters , in particular , @xmath127 , are forbidden . we summarize the relations between different coframe connections in fig = 2.25 in the standard levi - civita connection is invariant under arbitrary non - restricted variations of the coframe . this connection is a basis construction of the riemannian geometry and consequently of the einstein gravity theory . we have derived an alternative family of connections which are invariant under restricted local variations of the coframe field . the variations themselves have to satisfy the equations @xmath128}=0\ , , \qquad k^m{}_{am}=0\,.\end{aligned}\]]due to ( [ c - change ] ) this is a system of eight first order partial differential equations for six enters of the antisymmetric matrix @xmath103 . this is very similar to the standard maxwell system for the electromagnetic field strength . we will examine the system ( [ premax ] ) in section 5 . but first we will give a gravity model corresponding to these gauge connections . the lagrangian of a gauge invariant gravity model has to respect the gauge transformations ( [ l - tr2 ] ) restricted by ( [ premax ] ) . starting with invariant fields of metric and connection , such lagrangian can be constructed straightforwardly . namely , we can take the standard einstein - hilbert lagrangian calculated on our gauge invariant connections instead of the levi - civita connections @xmath129 moreover , since our connection is asymmetric , we can also consider the terms quadratic in torsion and non - metricity as acceptable additions to the lagrangians . all such terms are of the same physical dimensions so they have to be involved with free dimensionless coefficients . this way we come to a rather complicated lagrangian of mag @xcite . in our case , however , all the geometric quantities are constructed only from the coframe components and their first order derivatives . this fact simplifies very much the most general lagrangian we are seeking for . observe that the einstein - hilbert lagrangian ( [ sl1 ] ) can be rewritten as a linear combination of the first order derivatives of the coframe components plus a total derivative term . also the torsion and non - metricity tensors are linear in the first order derivatives of the coframe . consequently , the most general lagrangian can be given by a linear combination of terms which are quadratic in the first order derivatives of the coframe . a most general coframe lagrangian can be straightforwardly constructed from the exterior derivative of the coframe components @xmath130}$ ] , @xmath131}dx^i\wedge dx^j\,.\]]equivalently , we can use the components @xmath130}$ ] directly to deal with the tensorial expression @xmath132}\,,\qquad c_{ijk}=\eta_{\a\b}\vt^\a{}_k\vt^\b{}_{[i , j]}\,.\]]now it is enough to consider the possible quadratic combinations of @xmath133 or , equivalently , of @xmath134 . in this way , a general quadratic coframe action takes the known form @xcite , @xcite @xmath135where @xmath136 is a coupling constant of dimension @xmath137 , @xmath85 are free dimensionless parameters , while the `` elementary '' lagrangian densities are defined as [ cof - lag1 ] l_1 & = & d^d _ , + [ cof - lag2 ] l_2 & = & ( d_^ ) ( d_^ ) , + [ cof - lag3 ] l_3 & = & ( d_^ ) ( d_^ ) . the action ( [ cof - lag0 ] ) can be rewritten in a compact form @xmath138 where @xmath139}+\rho_3\big(c^{ijk}-2\,c^{im}{}_mg^{jk}\big)\,.\ ] ] the corresponding 1-form is @xmath140 to make physics on the coframe background we have to involve a lagrangian of a matter field @xmath141 . we do not specify the tensorial or spinorial content of this field . consequently , @xmath142 where the integrand is a 4-form density . the covariant derivative in ( [ mat - lag ] ) has to be taken with respect to the same connection @xmath143 which is involved in the coframe lagrangian . consequently , also @xmath144 involves the coframe components . apply now the variation of the total lagrangian @xmath145 with respect to the coframe field . we come to the coframe field equation which has a compact form in exterior calculus notation @xcite @xmath146where the right hand side involves the energy - momentum currents ( 3-form ) of the coframe and the matter fields . these quantities are related to the energy - momentum tensors @xmath147 in a regular way @xmath148 the standard einstein - hilbert action of gr is not completely distinct from the coframe action ( [ cof - lag0 ] ) . indeed , the term @xmath149 can be rewritten as a total derivative plus squares of first order derivatives of the metric . let us substitute @xmath150 and neglect with the total derivative . we remain with a scalar expression which is quadratic in the coframe components . since all such type expressions are encoded in ( [ cof - lag0 ] ) , there is a coframe lagrangian with a special set of parameters is equivalent ( up to a total derivative ) to the standard hilbert - einstein lagrangian . on the level of the field equations , we have here no more than the standard gr reformulated in the coframe variables . the explicit calculations @xcite give the special ( einstein ) choice of parameters @xmath151consequently , up to a total derivative , the hilbert - einstein action takes the form @xmath152 } + c^{ijk}- % 2\,c^{im}{}_mg^{jk } \big)\sqrt{-g}\nonumber\\ % & & + \partial(\cdots)\,,%\,{\rm total \ , derivative}\,.\]]note that although the energy - momentum tensor of the coframe field is defined also in this case , it can not be identified as an energy - momentum tensor of gravity . such identification quite often appears in literature , because one does not take into account a hidden symmetry of the coframe action . for ( [ e - choice1 ] ) , the action ( [ cof - lag0 ] ) is local lorentz invariant . this symmetry , however , is not preserved for the coframe energy - momentum tensor expression . let us examine now the behavior of the general coframe action ( [ cof - lag0 ] ) under the local lorentz transformations ( [ l - tr2 ] ) . using ( [ e - choice1 ] ) we rewrite ( up to a total derivative ) @xmath153+\rho_3\int r\sqrt{-g}\,{\rm vol}\,.\]]with respect to the transformations ( [ l - tr2 ] ) , this expression changes as @xmath154\nonumber\\ & = & \int c_{ijk}\big[2\rho_1 k^{ijk}+(2\rho_2+\rho_3)k^{[ijk]}\big ] \sqrt{-g}\,{\rm vol}\ , . \end{aligned}\ ] ] since the variations @xmath155 are independent , the action is invariant if and only if the following invariance condition holds @xmath156}=0\,.\]]this algebraic equation has to be satisfied identically ( for arbitrary coframe fields ) , thus we have three solutions to ( [ requ0 ] ) of basically different types : _ ( i ) `` the teleparallel equivalent of gr '' _ a first solution to ( [ requ0 ] ) can be taken as @xmath157 since the tensor @xmath155 is arbitrary now , the coframe lagrangian with ( [ requ2 ] ) is invariant under arbitrary lorentz transformations . consequently , six degrees of freedom of the coframe field are unphysical while the rest ten degrees are equivalent to the metric tensor . the underlining geometry for such coframe model is precisely riemannian and not teleparallel as it is claimed sometimes . the fact that the coframe lagrangian is expressed by the torsion of the weitzenbck connection does not mean a lot , specially in view of the formula ( [ lc - con ] ) , which express the levi - civita connection by the weitzenbck one . the coframe reformulation of gr can serve as a useful technical tool , for instance for search of new solutions in gr @xcite or for its hamiltonian formulation @xcite . it can not change however the main features of gr . in particular , the so - called teleparallel energy - momentum tensors of gravitational field are no more than the pseudo - tensors of the standard gr . _ ( ii ) the teleparallel gravity model . _ another solution to ( [ requ0 ] ) is @xmath158substituting into ( [ c - change ] ) we obtain @xmath159 , thus only global ( rigid ) lorentz transformations of the coframe field are admissible . consequently , if a coframe is given at a point , it is also given on the whole manifold . such teleparallel geometry is described by the weitzenbck connection . consequently , precisely the model ( [ requ1 ] ) ( and only this model ) has to be refereed to as a teleparallel gravity model . indeed only the lagrangians with @xmath160 do not have any local non - trivial group of transformations . this is in a complete correspondence to the flat teleparallel geometry . with a non - zero parameter @xmath161 , the coframe field equation ( [ cof - feq ] ) has spherical symmetric solutions of the type @xcite @xmath162 where @xmath163 is a function of the parameters @xmath85 . the corresponding metric does not have the newtonian behavior at infinity . consequently , the teleparallel model can not serve for description of @xmath18 gravitational field . \(iii ) _ a model alternative to gr _ the last solution to ( [ requ0 ] ) is @xmath164}=0\qquad \rho_1=0\,,\qquad 2\rho_2-\rho_3\ne 0\,,\]]the first equation here is a first order partial differential equation for the matrix @xmath103 . thus , in contrast to the previous cases , we are dealing now with _ dynamical local lorentz transformations_. for the coframe models with a zero parameter @xmath161 , the field equation ( [ cof - feq ] ) has a unique static spherical - symmetric solution of a `` diagonal form '' @xcite : @xmath165where @xmath166 . this coframe corresponds to the schwarzschild metric in the isotropic coordinates . another justification for the condition @xmath167 comes from consideration of the first order approximation to the coframe field model @xcite . in this case , the coframe variable is reduced to a sum of symmetric and antisymmetric matrices . it means that , in linear approximation , we can treat the coframe field as a system of two independent fields . it is natural to require all the field - theoretic constructions , i.e. the action , the field equation and the energy - momentum tensor , to accept the same separation to two independent expression . a remarkable fact that such separation ( free field limit ) appears if and only if @xmath167 . consequently we have derived the equation @xmath168}=0\,,\ ] ] as an invariance condition for a viable gravity field model . this is in an addition to the pure geometrical consideration given in the previous section . we turn now to the second constrain @xmath169 . distinctly , it comes from the symmetry a generic matter lagrangian . when the matter field @xmath141 is minimally coupled to gravity , its lagrangian involves the covariant derivatives taken with respect to some connection @xmath170since the matter fields themselves are invariant under coframe transformations , also the connection @xmath143 has to be invariant . in other words , @xmath171 in the case of a variety of connections the following problem emerges @xcite : to what connection the matter field is really coupled ? a natural answer can be proposed : the proper connection is this one which is already involved in the gravity sector of the model . in other words : the symmetries of the gravity and the matter sectors have to be conformed . we continue now with our alternative model ( [ requ3 ] ) . since the gravity sector is already restricted with the requirement @xmath172}=0 $ ] , it is natural to require the matter sector to respect this condition . this way we come to the family of connections ( [ g - con2 g ] ) which are invariant under the local lorentz transformations satisfied @xmath173 consequently , the equations ( [ premax ] ) emerge as the invariance conditions for the total action of a matter field coupling minimally to gravity . let us examine now what physical meaning can be given to the invariance conditions @xmath174}=0\ , , \qquad k^m{}_{im}=0\,.\end{aligned}\]]recall that the tensor @xmath175 depends on the derivatives of the lorentz parameters @xmath103 and on the components of the coframe field @xmath176 thus , in fact , we have in ( [ premax - x ] ) , two first order partial differential equations for the entries of an antisymmetric matrix @xmath103 . let us construct from this matrix an antisymmetric tensor @xmath177 @xmath178substituting into ( [ kk - def ] ) , we derive @xmath179}-\frac 12 \ , x_{\a\b}\big[(\vt^\a{}_k\vt^\b{}_i)_{,j}- ( \vt^\a{}_k\vt^\b{}_j)_{,i}\big]\nonumber\\ & = & f_{k[i , j]}-f_{km}c_{ij}{}^m-\frac 12 ( f_{mi}\og_{kj}{}^m- f_{mj}\og_{ki}{}^m)\ , . \end{aligned}\]]consequently , the first equation from ( [ premax - x ] ) takes the form @xmath180}=\frac 23 ( c_{ij}{}^mf_{km}+c_{jk}{}^mf_{im}+c_{ki}{}^mf_{jm})\,,\end{aligned}\]]while the second equation from ( [ premax - x ] ) is rewritten as @xmath181observe first a significant approximation to ( [ premax1-x][premax2-x ] ) . if the right hand sides in both equations are neglected , the equations take the form of the ordinary maxwell equations for the electromagnetic field in vacuum @xmath182}=0\ , , \qquad f^i{}_{j , i}=0\,.\]]in the coframe models , the gravity is modeled by a variable coframe field , i.e. , by nonzero values of the quantities @xmath81 . consequently , the right hand sides of ( [ premax1-x][premax2-x ] ) can be viewed as curved space additions , i.e. , as the gravitational corrections to the electromagnetic field equations . in the flat spacetime , when a suitable coordinate system is chosen , these corrections are identically equal to zero . consequently , in the flat spacetime , the invariance conditions ( [ premax - x ] ) take the form of the vacuum maxwell system . on a curved manifold , the standard maxwell equations are formulated in a covariant form . let us show that our system ( [ premax1-x][premax2-x ] ) is already covariant . we rewrite ( [ kk1-def ] ) as @xmath183consequently , @xmath184}\ , , \end{aligned}\]]where the covariant derivative ( denoted by the semicolon ) is taken relative to the weitzenbck connection . consequently , the system ( [ premax1-x][premax2-x ] ) takes the covariant form @xmath185}=0\ , , \qquad f^i{}_{j;i}=0\,.\ ] ] these equations are literally the same as the electromagnetic sector field equations of the maxwell - einstein system . the crucial difference is encoded in the type of the covariant derivative . in the maxwell - einstein system , the covariant derivative is taken relative to the levi - civita connection , while , in our case , the corresponding connection is of weitzenbck . observe that , due to our approach , the weitzenbck connection is rather natural in ( [ max - rie ] ) . indeed , since the electromagnetic - type field describes the local change of the coframe field , it should itself be referred only to the global changes of the coframe . as we have shown , such global transformations correspond precisely to the teleparallel geometry with the weitzenbck connections . the coupling between the electromagnetic and the gravitational field is an age - old problem . it is already related to the first observable prediction of gr about the bending of light rays of stars in the gravitational field of the sun . the electromagnetic and gravitational effects are of rather different orders of magnitude . however , the increasing precision of modern experimental techniques gives rise to the hope that the appropriate form of the coupling can soon be determined . in particular , we have in this context two independent but closely related problems : * does the gravitational field of a charged massive source depend on the electric charge ? * does the electromagnetic field of a charged massive source depend on the mass ? from a certain philosophical point of view ( `` everything has an influence on everything else '' ) the answer on both questions has to be positive . in classical ( non - relativistic ) physics , the both answers are negative : the newton force of attraction is independent on the charges , also the coulomb force is not sensitive to the masses . in the framework of gr , the coupling between the electromagnetic field and gravity is managed by the electromagnetic action itself @xmath186 where @xmath187 is a coupling constant . when the actions of the gravitational and the matter fields are added to ( [ em - act ] ) , the variation with respect to the metric yields the einstein field equation ( without cosmological constant ) @xmath188 here the electromagnetic energy - momentum tensor @xmath189 and the matter energy - momentum tensor @xmath190 are the sources of the gravitational field . the action ( [ em - act ] ) yields the electromagnetic field equation of the form @xmath191}=0\ , , \qquad f^i_{j;i}=j_j\,,\ ] ] where the semicolon is used for the covariant derivative taken relative to the levi - civita connection . the dependence of the metric is encoded here in the index raising procedure and in the covariant derivatives . the static spherical - symmetric solution to the einstein - maxwell system ( [ einst - eq1 ] , [ max - eq1 ] ) is given by the reissner - nordstrm metric @xmath192dt^2- \big[1-\lambda(r)\big]^{-1}dr^2- r^2d\omega^2,\end{aligned}\ ] ] with @xmath193 and the coulomb potential @xmath194 here @xmath195 and @xmath196 describes a mass @xmath197 with an electric charge @xmath198 . consequently , in the einstein - maxwell system , the gravitational field depends on the charge of the source . the electromagnetic field of a pointwise source remains the same as in the flat spacetime , i.e. , it is independent on the mass of the charge . let us look for a static spherically symmetric solution to our electromagnetic - type equations ( [ max - rie ] ) . we start with a spherically symmetric solution in a vacuum coframe model ( [ cof - lag0 ] ) with the parameter @xmath167 . we will use the isotropic coordinates @xmath199 with the isotropic radius @xmath200 recall that we identify the gravity variable with the coframe field defined up to an infinitesimal lorentz transformation . it is equivalent to the metric field . since the field equation ( [ cof - feq ] ) involves free parameters @xmath201 and @xmath202 , it is alternative to gr . although , the spherically symmetric gravity solution turns out to be the same . in particular , such solution can be taken in the form of a `` diagonal '' coframe @xcite @xmath203where @xmath204the non - zero components of the corresponding metric tensor are @xmath205 while the line element is @xmath206 recall that ( [ line ] ) with ( [ gr - f - anz ] ) substituted is no more than the standard schwarzschild line element in isotropic coordinates . the nonzero components of the weitzenbck connection ( [ weiz2 ] ) corresponding to the coframe ( [ gr - anz ] ) are @xmath207consequently , the independent nonzero components of the weitzenbck torsion tensor are @xmath208 the suitable ansatz for electromagnetic - type field of a pointwise charge can be taken as @xmath209 the first maxwell - type equation @xmath210}=0 $ ] is satisfied identically when ( [ c - anz ] ) and ( [ el - anz ] ) are substituted . indeed , @xmath211}&=&\frac 13(f_{0\hj,\hk}-f_{0\hk,\hj})\nonumber\\ & = & \d_{\hj\hm}(x^\hm f\psi^2)_{,\hk}-\d_{\hk\hm}(x^\hm f\psi^2)_{,\hj } = 0\,,\end{aligned}\ ] ] while @xmath212 the other components of the equation are zero because of ( anti)symmetry and independence of time . as for the second maxwell - type equation @xmath213 , we observe that for @xmath214 both sides vanish . for @xmath215 , we substitute ( [ c - anz ] ) and ( [ el - anz ] ) into ( [ premax2-x ] ) to obtain @xmath216this equation is straightforward integrated to @xmath217 where @xmath198 is a constant of integration . consequently , the non - zero field components are @xmath218 the coulomb - type force acted on a test charge @xmath219 ( of a small mass ) takes the form @xmath220 the ordinary cartesian radius @xmath221 is related to the isotropic radius @xmath222 as @xmath223 hence @xmath224 observe that the isotropic coordinates are defined only for @xmath225 . consequently , the modernized coulomb force takes the form @xmath226 finally , in the cartesian coordinates , the mass - correction terms take the form @xmath227 observe the complicated form of the solution in the standard polar coordinates . in fact , the choice of the isotropic coordinates can be viewed as a method to solve the differential equation @xmath213 exactly . unfortunately , this method is restricted to @xmath225 . [ cols="^,^,^,^,^ " , ] we depict the dependence of the force @xmath228 on the distances @xmath222 and @xmath221 on fig . the graphs start from @xmath229 and @xmath230 which are the minimal possible value . the deviation from the coulomb values appears only for small distances @xmath231 . the maximal value of the force between two charged particles predicts as @xmath232 for two electrons , it gives @xmath233 . we construct a complete class of connections linear in the first order derivatives of the coframe field . it involves the standard levi - civita connection and the flat weitzenbck connection . for special choices of parameters , the torsion free and the metric compatible sub - families of connections emerge . our main output is the identification of a sub - family of connections which are invariant under restricted local lorentz transformations . the corresponding conditions are a set of eight first order equations . in the first order approximations the rotations are described by an antisymmetric matrix . in this case , the set of invariance conditions turns out to be the standard vacuum maxwell system . we consider a lagrangian of the matter - coframe system @xmath234 in the standard consideration , the arbitrary variation of the coframe field is assumed . the corresponding coframe field equation is @xcite @xmath235 with the energy - momentum 3-form of the coframe - matter system @xmath236 in its right hand side . consequently ( [ feq ] ) represents a system of 16 independent equations . it can be covariantly decomposed to ten symmetric and six antisymmetric equations . @xmath237}&=&\vt^{[\b}\wedge \t^{\a]}\,,\end{aligned}\ ] ] consider the pure coframe system , and restrict to the viable case @xmath167 . the explicit calculations , see @xcite , show that both sides of ( [ feq2 ] ) involve the leading coefficient @xmath238 . consequently , for @xmath239 , only ten independent field equations ( [ feq1 ] ) remain . certainly , this symmetric system is equivalent to einstein equation . in this case , the coframe field is defined only up to arbitrary lorentz transformations . for the alternative models with @xmath240 , ( [ feq1]-[feq2 ] ) is a well posed system of 16 hyperbolic pde for 16 independent variables . the coframe field is defined uniquely ( up to global transformations ) . in the approach proposed in the current paper , only the symmetric variations of the coframe field are independent . the antisymmetric variations are constrained by the system ( [ max - rie ] ) . since the variation derivative of the lagrangian is covariantly decomposed to the symmetric and antisymmetric parts , only one field equation , namely the symmetric one , ( [ feq1 ] ) , is derived from the symmetric variation of the action . the second field equation comes from the invariance of the lagrangian . consequently our system of equations includes 16 independent equations for 16 independent variables @xmath241}=0\ , , \qquad f^i{}_{j;i}=0\,.\end{aligned}\ ] ] the first ( `` gravitational '' ) equation has to give a class of orthonormal coframes , i.e. , a solution that valid up to arbitrary transformations of the coframe . . the second ( `` electromagnetic '' ) equation has to define uniquely ( up to global transformations ) the field of local rotations . although , the well - posedness of the system ( [ feq3]-[feq4 ] ) requires a special consideration , it is rather reasonable . indeed : \(i ) in the first order approximation @xcite , the coframe field is decomposed to a sum of symmetric and antisymmetric fields . for @xmath167 , also the system of field equations is separated ( [ feq3 ] ) involves only the symmetric part , while ( [ feq2 ] ) describes the dynamics of the antisymmetric part . moreover , in this approximation , ( [ feq2 ] ) takes the form @xmath242 , which is a consequence of ( [ feq4 ] ) . \(ii ) for a spherical - symmetric ansatz , ( [ feq3 ] ) is equivalent to the vacuum einstein equation , i.e. , determines uniquely a class of orthonormal coframes . when the solution is substituted in ( [ feq4 ] ) also the field @xmath177 is determined uniquely ( see the previous section ) . \(iii ) the equation ( [ feq4 ] ) involves the torsion tensor @xmath243 only linearly . thus , it can be viewed as a linear combination of the derivatives @xmath244 . when these combinations are substituted into ( [ feq3 ] ) , the corresponding equation is a system of second order quasi - linear hyperbolic pde for @xmath177 . let the local lorentz rotations @xmath103 ( and , consequently , their approximations @xmath177 ) are assumed to be independent on a point . in this case , the field equations ( [ feq4 ] ) are satisfied . the remain 10 equations ( [ feq3 ] ) is an under - defined system for 16 independent components of the coframe field . the metric tensor , however , is defined uniquely . in particular , with a mentioned restriction @xmath167 , the system has a unique spherical - symmetric solution which corresponds to the schwarzschild metric . moreover , the coframe field model for gravity has even some advantage relative to the standard metric gr . it is well known that an energy - momentum tensor can not be constructed covariantly from the metric tensor and its first order derivatives . consequently , the energy of a gravity field can not be defined in gr . alternatively , for a coframe field , the corresponding tensor is defined @xcite . this tensor is not local lorentz in varian and so can not be prolongated into the standard metric gr . however , in the model proposed here , the arbitrary local lorentz transformations are not acceptable . moreover , these transformations are governed by their own field equations . let the coframe field be defined up to a local pseudo - rotation . thus the metric is defined uniquely . moreover , let a representative coframe be chosen . the set of invariance conditions now is a system of eight first order equations for six independent variables @xmath245 . in fact , we have here a linear field model for @xmath245 , which turns to a non - linear field model for @xmath103 , where @xmath246 a similar non - linear extension of electrodynamics based on orthonormal tensor was recently discussed @xcite . it can give an alternative to the born - infeld electrodynamics , which turns recently to a popular model in strings theory . recall , that in the current paper we consider only the first order approximation of the pseudo - orthonormal tensor @xmath245 . the corresponding invariance conditions are @xmath247}=0\ , , \qquad k^m{}_{im}=0\,,\end{aligned}\]]or , after a redefinition of the variables , @xmath248}=0\ , , \qquad f^m{}_{i;m}=0\,,\end{aligned}\ ] ] where the covariant derivative is taken relative to the flat weitzenbck connection . for a holonomic coframe field , these equations take the form of the standard maxwell system @xmath249}=0\ , , \qquad f^m{}_{i , m}=0\,.\end{aligned}\]]observe the main properties of the derived field equations : * the ( approximated ) maxwell - type system is not introduced by hand , but emerges in a natural way as a set of invariance condition for a geometry and for a viable lagrangian . * the complete field equation is nonlinear , what is in a correspondence with the string model consequences . * the spherical symmetric solution is bounded near the schwarzschild radius . further off , it is close to the coulomb field . * it is well known that in the models with an asymmetric connection , the torsion contribution is an obstacle for the standard definition of the potentials , @xmath250 . for torsion of a restricted type , the potentials can be defined , however , as @xmath251 and the modified gauge transformations are acceptable @xcite . * in the standard maxwell theory a charge is not connected to a mass . thus , unphysical massless charges are not forbidden . in our construction , the field of pseudo - rotations can be degenerate only in the singularities of the coframe field ( metric ) . with this requirement , the charges are necessary massive . certainly , the field of rotations can be regular in the whole space even the coframes are singular . in this case , we have uncharged masses . * a nonzero constant field @xmath177 is another unphysical solution of the maxwell system , which is usually removed by the boundary conditions . the modified field equation ( [ premax - xx ] ) does not have such a solution . * in the first order approximation , the coframe can be covariantly decomposed in a sum of its symmetric and antisymmetric parts . in this case , the difference between coordinate and coframe indices can be neglected . consequently , we can write @xmath252}=g_{ij}+\frac 1\beta f_{ij}\,.\ ] ] the free parameter @xmath253 has a dimension of a field strength . note that the structure of ( [ fin1 ] ) is similar to a combination appearing in the born - infeld electrodynamics , where @xmath253 is a maximal electromagnetic field . from our spherical symmetric solution @xmath254 . at this stage , it is only a speculation because @xmath253 has sense only in higher order approximations . * in this paper , we have considered only the vacuum case . a complete model has to include a lagrangian for fermions . due to fock and dirac , the fermionic lagrangian on a curved space is constructed with the spin connections . a remarkable fact that this dirac connection is proportional to the flat connection of weitzenbck . consequently , the covariant derivative is the same as in ( [ premax - xx ] ) . * qualitatively , the coframe structure in our model can be described as a `` paramagnetic substance '' . the gravitational sector establishes the coframe at every point , i.e. , the lengths of four covectors and the angles between them . the coframes do not interact with one another and are randomly oriented . the metric on the manifold is defined uniquely . the electromagnetic - type sector orders the coframes attached at different points in a smooth coframe field . different type of ordering correspond to different connections from the six parametric family described above . on a manifold endowed with a coframe ( vierbein , tetrad ) field , we constructed a most general linear asymmetric connection . besides the torsion - free and metric - compatible connections , we identified a subset of maxwell - compatible connections . the vacuum maxwell - type equations emerges as the conditions for invariance of the connection under lorentz transformations of the coframe . we derive the same equations as invariant conditions for a viable coframe lagrangian which has the schwarzschild solution even being alternative to gr . in our approach , the curved space maxwell - type equations emerges formally the same as in the standard maxwell - einstein system . the covariant derivatives is different , however , in proposed model they are taken relative to the flat connection . the result is a different behavior of the electromagnetic field near the schwarzschild radius . moreover , the model predicts an upper bound for the electromagnetic - type field of a source with given charge and mass . note that although the gravitational field of a charged source remains independent on the charge , it is only a result of the approximation used . already in the second approximation , the metric tensor is not invariant under the first order approximate lorentz transformations and a quadratic correction , @xmath255 , emerges in the metric tensor . in the current paper , we have treated only the vacuum case . the problem of the sources for the proposed maxwell - type field requires additional investigations . in particular , they can appear from a viable modification of dirac field on a coframe background . a part of this paper was written during my visit to mit . i am grateful to roman jackiw for warm hospitality and fruitful discussions . i thank shmuel kaniel , friedrich hehl , jacob bekenstein and david kazhdan for valuable remarks . many thanks are due to the referee for most constructive suggestions . 99 a. ashtekar , phys . d * 36 * , 1587 ( 1987 ) . p. g. bergmann , v. de sabbata , g. t. gillies and p. i. pronin , _ international school of cosmology and gravitation : 15th course : spin in gravity : is it possible to give an experimental basis to torsion ? _ , erice , italy , 13 - 20 may 1997 . k. hayashi and t. shirafuji , phys . d * 19 * , 3524 ( 1979 ) j. nitsch and f. w. hehl , phys . lett . b * 90 * , 98 ( 1980 ) ; f. mueller - hoissen and j. nitsch , phys . d * 28 * , 718 ( 1983 ) ; e. w. mielke , annals phys . * 219 * , 78 ( 1992 ) ; u. muench , f. gronwald and f. w. hehl , gen . rel . grav . * 30 * , 933 ( 1998 ) ; r. s. tung and j. m. nester , phys . d * 60 * , 021501 ( 1999 ) ; m. blagojevic and m. vasilic , class . * 17 * , 3785 ( 2000 ) ; y. itin and s. kaniel , j. math . phys . * 41 * , 6318 ( 2000 ) y. itin , int . j. mod . d * 10 * , 547 ( 2001 ) y. n. obukhov and j. g. pereira , phys . rev . d * 67 * , 044016 ( 2003 ) y. itin , j. math . phys . * 46 * 12501 ( 2005 ) ,
serre obtained a sharp bound on how often two irreducible degree complex characters of a finite group can agree , which has applications to a refined local - global principle for galois representations . we consider the more delicate question of finding a sharp bound when the characters are primitive . we answer this question for , the families and for general , and conclude a local - global application to primitive galois ( artin ) representations . this provides some insight on refined strong multiplicity one phenomena for automorphic representations of .
1607.07586
let @xmath5 be a finite group , and @xmath6 be two complex representations of @xmath5 with characters @xmath7 . we will study the quantities @xmath8 specifically , let @xmath9 be the minimum of @xmath10 as @xmath6 range over pairs of inequivalent irreducible @xmath0-dimensional representations of @xmath5 , with the convention that @xmath11 if there are no such pairs @xmath6 . note that @xmath9 tells us what fraction of elements of @xmath5 we must check to distinguish irreducible degree @xmath0 characters . put @xmath12 . an elementary consequence of orthogonality relations is [ prop1 ] we have @xmath13 . buzzard , edixhoven and taylor constructed examples to show this bound is sharp when @xmath0 is a power of 2 , which serre generalized this to arbitrary @xmath0 ( see @xcite ) . [ thm : serre ] for any @xmath0 , there exists @xmath5 such that @xmath14 , so @xmath15 . in particular , the infimum is a minimum . we will recall the proof of proposition [ prop1 ] and serre s construction in section [ sec : serre ] . for now , the main points to note are that serre s examples must be solvable and the representations are induced . in this paper , we consider two kinds of refinements of determining @xmath16 , and give an application to a strong local - global phenomenon for galois representations . the first refinement is about restricting to primitive representations and the second is about restricting to certain families of groups . define @xmath17 to be the infimum of @xmath10 where @xmath6 range over pairs of inequivalent irreducible primitive @xmath0-dimensional complex representations of @xmath5 . let @xmath18 . using the classical fact that the number of finite primitive subgroups of @xmath19 is bounded for a given @xmath0 , it is not hard to show that this infimum is also a minimum . from serre s theorem , we get a trivial bound @xmath20 . here is our first main result ( see theorems [ thm : deg2 ] and [ thm : deg3 ] ) . [ thm1 ] we have @xmath21 , @xmath22 and @xmath23 . furthermore @xmath24 if and only if @xmath5 is an extension of @xmath25 $ ] or @xmath26 $ ] , and @xmath27 if and only if @xmath5 is an extension of @xmath28 . by @xmath5 being an extension of @xmath29 ( by a normal subgroup @xmath30 ) , we mean @xmath31 . the groups @xmath25 $ ] and @xmath26 $ ] are the two groups of order 48 which are extensions of @xmath32 by the cyclic group @xmath33 and contain @xmath34 . the @xmath35 case is already contained in proposition [ prop1 ] as @xmath36 . for @xmath37 , these bounds are much better than the trivial bounds @xmath38 and @xmath39 . for @xmath40 , some related results were previously obtained by the second author in @xcite and will be discussed below . note that while @xmath16 is a strictly decreasing sequence for @xmath41 , our result says this is not the case for @xmath42 . in a slightly different direction , one can look for stronger lower bounds than @xmath43 for certain families of groups . we do not begin a serious investigation of this here , but just treat two basic families of finite groups of lie type which are related to the calculations for @xmath44 and @xmath45 . [ thm2 ] we compute @xmath9 and @xmath46 where @xmath47 and @xmath48 ; for @xmath0 not listed explicitly , @xmath49 . for @xmath50 with @xmath51 arbitrary or for @xmath47 with @xmath51 even , @xmath52 and @xmath53 whereas @xmath54 . for @xmath47 and @xmath51 odd , @xmath55 and @xmath56 whereas @xmath54 we remark that we completely determine @xmath57 for @xmath48 and @xmath2 in section [ sec : sl2q ] , but the exact formulas are a bit complicated and depend on divisibility conditions of @xmath58 . in particular , @xmath59 if and only if @xmath60 , and @xmath61 if and only if @xmath62 . the values for @xmath3 immediately give the following bounds . @xmath63 for @xmath51 any odd prime power greater than 3 . note the upper bound in the corollary for @xmath64 is the exact value of @xmath65 . even though theorem [ thm1 ] implies @xmath42 is not a decreasing sequence for @xmath41 , this corollary at least suggests the possibility of @xmath66 as @xmath67 . the proof of theorem [ thm1 ] relies on consideration of various cases according to the possible finite primitive subgroups of @xmath68 and @xmath69 , and about half of these are of the form @xmath2 or @xmath3 ( or @xmath70 ) for @xmath71 . thus theorem [ thm2 ] is a generalization of one of the ingredients for theorem [ thm1 ] . however , the main work in proving theorem [ thm1 ] is dealing with extensions of finite subgroups of @xmath68 and @xmath69 . one reason for interest in these questions is an application to density questions for galois and automorphic representations . namely , let @xmath72 be a number field and @xmath6 be irreducible @xmath0-dimensional artin representations for @xmath72 , i.e. , irreducible @xmath0-dimensional continuous complex representations of the absolute galois group @xmath73 of @xmath72 . for almost all places @xmath74 of @xmath72 , we can associate a well - defined frobenius conjugacy class @xmath75 of @xmath73 . the local invariant @xmath76 is of interest in number theory . the bound @xmath77 implies that if @xmath78 for a set of primes of density strictly greater than @xmath79 , then @xmath80 . on the other hand , there also exist inequivalent @xmath81 such that @xmath78 on a set of primes of density @xmath82 , so this bound is sharp . an analogous result for @xmath83-adic galois representations is due to rajan @xcite . ramakrishnan @xcite conjectured an analogous `` refined strong multiplicity one '' statement for cuspidal automorphic representations of @xmath84 , again with density bounds @xmath82 that should be sharp , and proved this for @xmath40 in @xcite . recently , in @xcite the second author showed that when @xmath40 one can in fact obtain stronger bounds under various assumptions , e.g. , the bound @xmath85 may be replaced by @xmath86 if one restricts to non - dihedral representations or by @xmath87 if the representations are not twist - equivalent . ( these bounds are also sharp . ) here non - dihedral is the analogue of primitive . all of these refined strong multiplicity one statements can be interpreted as stating , given an @xmath88-function @xmath89 associated to some object @xmath90 , how many euler factors @xmath91 one needs to know to determine @xmath92 . this project arose as an attempt to gain insight into what a conjectural analogue of @xcite ( specifically , the @xmath93 bound for non - dihedral representations ) might be for @xmath94 . since the problem for automorphic representations seems hard for @xmath95 , here we just settle for some results about artin representations as a test case . let @xmath81 , @xmath96 be inequivalent irreducible primitive @xmath0-dimensional artin representations for @xmath72 . suppose @xmath78 for a set of primes @xmath74 of @xmath72 of density @xmath97 . 1 . if @xmath40 and @xmath98 , then @xmath80 . if @xmath99 and @xmath100 , then @xmath80 . this follows immediately from the chebotarev density theorem which says that if @xmath101 is a finite normal extension with galois group @xmath5 and @xmath102 is a conjugacy class in @xmath5 , then the image of @xmath75 in @xmath103 lies in @xmath102 for a set of primes @xmath74 of density @xmath104and theorem [ thm1 ] . in general , chebotarev s theorem implies that if @xmath81 and @xmath96 are irreducible @xmath0-dimensional artin representations for @xmath72 which both factor through a finite galois extension @xmath101 with galois group @xmath5 , then @xmath105 implies @xmath80 ( with @xmath97 as in the corollary ) . when @xmath40 , if @xmath81 and @xmath96 are automorphic , i.e. , correspond to cuspidal automorphic representations of @xmath106 , then the above result already follows by @xcite . the strong artin conjecture asserts that all artin representations are automorphic . when @xmath40 , the strong artin conjecture for @xmath81 is known in many cases for instance , if @xmath81 has solvable image by langlands @xcite and tunnell @xcite , or if @xmath107 and @xmath81 is `` odd '' via serre s conjecture by khare - wintenberger @xcite . we remark that the methods of @xcite are quite different than ours here . the above corollary then suggests the following statement may be true : if @xmath108 are cuspidal automorphic representations of @xmath109 which are not induced from characters and @xmath110 for a set of primes @xmath74 of density @xmath111 , then @xmath112 . of course , not all cuspidal @xmath108 come from galois representations , so the @xmath113 bound may not be sufficient for general @xmath108 . however , it seems reasonable to think that coincidences of a large fraction of euler factors only happen for essentially algebraic reasons , so the density bounds are likely to be the same in both the galois and automorphic cases . the first author was partially supported by a simons collaboration grant . the second author was supported by forschungskredit grant k-71116 - 01 - 01 of the university of zrich and partially supported by grant snf pp00p2 - 138906 of the swiss national foundation . first we explain the proof of proposition [ prop1 ] and describe serre s construction . actually , we will prove a variant of proposition [ prop1 ] which will be of use later . suppose @xmath29 is a subgroup of @xmath5 and @xmath114 are irreducible @xmath0-dimensional representations of @xmath5 with characters @xmath115 . denote by @xmath116 the corresponding ratio @xmath117 for the restrictions of @xmath118 and @xmath119 to @xmath29 . also , let @xmath120 be the inner product of @xmath121 with @xmath122 . [ prop2 ] let @xmath115 be distinct irreducible degree @xmath0 characters of a finite group @xmath5 , and @xmath29 a subgroup of @xmath5 . then @xmath123 in particular , if @xmath115 are irreducible when restricted to @xmath29 , we have @xmath124 let @xmath125 be the set of elements of @xmath126 such that @xmath127 . then we have @xmath128 using the bound @xmath129 , we see @xmath130 by orthogonality relations , we have @xmath131 while the statement of this proposition does not appear to strictly contain that of proposition [ prop1 ] because of the averaging with @xmath132 , substituting @xmath133 into in fact yields proposition [ prop1 ] . we now recall serre s construction proving theorem [ thm : serre ] , which is briefly described in @xcite using observations from ( * ? ? ? * sec 6.5 ) . let @xmath29 be an irreducible subgroup of @xmath134 , containing @xmath135 for each @xmath0-th root of unity @xmath136 , such that its projective image @xmath137 in @xmath19 has order @xmath138 . this means that @xmath29 is of `` central type '' with cyclic center . such @xmath29 exist for all @xmath0 . for instance , one can take @xmath139 , where @xmath140 is an abelian group of order @xmath0 . some nonabelian examples of such @xmath141 are given by iwahori and matsumoto ( * ? ? ? * sec 5 ) . iwahori and matsumoto conjectured that groups of central type are necessarily solvable and this was proved using the classification of finite simple groups by howlett and isaacs @xcite . since @xmath142 and @xmath143 , the identity @xmath144 implies @xmath145 for each @xmath146 , i.e. , the set of @xmath147 such that @xmath145 has cardinality @xmath148 . let @xmath149 and take @xmath150 and @xmath151 to be representations of @xmath5 , where @xmath152 is the standard representation of @xmath29 and @xmath153 is the nontrivial character of @xmath154 . then @xmath155 for @xmath156 elements of @xmath5 . on the remaining @xmath157 elements of @xmath158 , @xmath159 and @xmath160 must differ on precisely @xmath0 elements , giving @xmath5 with @xmath161 as desired . we claim that @xmath81 and @xmath96 are induced . it suffices to show @xmath152 is induced . since @xmath141 is solvable , there is a subgroup of prime index @xmath162 , so there exists a subgroup @xmath163 of @xmath29 of index @xmath162 which contains @xmath164 . put @xmath165 . now @xmath166 . on the other hand @xmath167 , where @xmath168 is the decomposition of @xmath169 into irreducible characters of @xmath163 . thus @xmath170 and we must have equality , which means @xmath152 is induced from a @xmath171 . next we recall some definitions of finite linear groups . let @xmath5 be a finite subgroup of @xmath134 , so one has the standard representation of @xmath5 on @xmath172 . we say @xmath5 is reducible if there exists a nonzero proper subspace @xmath173 which is fixed by @xmath5 . we say @xmath5 is intransitive if there is a nontrivial decomposition @xmath174 such that @xmath175 and @xmath176 are invariant under the action of @xmath5 . by maschke s theorem , @xmath5 is intransitive if it is reducible , which means the standard representation is completely reducible . suppose @xmath5 is irreducible . if there exists a nontrivial decomposition @xmath177 such that @xmath5 acts transitively on the @xmath178 , then we say @xmath5 is imprimitive . in this case , each @xmath178 has the same dimension , and the standard representation is induced from a representation on @xmath175 . otherwise , call @xmath5 primitive . let @xmath179 denote the quotient map from @xmath134 to @xmath19 . similarly , if @xmath180 , let @xmath181 be the image of @xmath5 under this map . since @xmath182 is algebraically closed , @xmath183 . recall for groups @xmath5 and @xmath29 with @xmath184 a common subgroup of the centers @xmath158 and @xmath185 , the central product @xmath186 is the direct product @xmath187 modulo @xmath184 embedded diagonally . [ red - lem ] let @xmath5 be an irreducible finite subgroup of @xmath134 . then there exists a subgroup @xmath188 such that @xmath189 is the central product of @xmath190 and the cyclic group @xmath191 of order @xmath0 where @xmath192 . this is long known , e.g. , section 12 of @xcite . ( see the beginning of chapter 4 of @xcite for a more modern reference . ) the idea is to consider the group @xmath193 generated by @xmath5 and scalar matrices @xmath135 where @xmath136 ranges over the @xmath0-th roots of unity and take @xmath194 . thus classifying the irreducible finite subgroups of @xmath195 is essentially equivalent to classifying the irreducible finite subgroups of @xmath196 , which is essentially equivalent to classifying the irreducible finite subgroups of @xmath197 . if @xmath198 is an irreducible faithful representation , this says @xmath81 can be identified with a pair @xmath199 where @xmath200 is an irreducible faithful @xmath0-dimensional representation of @xmath190 and @xmath201 is an irreducible character of @xmath191 . let @xmath114 be irreducible @xmath0-dimensional representations of @xmath5 . we present a couple of basic ideas for bounding @xmath10 . let @xmath202 , for @xmath203 . then we may view @xmath114 as representations of @xmath204 and @xmath205 . so for our purposes , we may assume @xmath206 , i.e. , @xmath207 . first suppose one kernel is contained in the other , i.e. , @xmath208 , say , is trivial . then one can obtain a bound for @xmath209 from knowing bounds for finite primitive subgroups of @xmath134 . namely , @xmath119 is faithful , so we may view @xmath5 as a subgroup of @xmath134 , and thus @xmath210 . the bulk of the analysis for theorem [ thm1 ] is treating the case where one kernel does not contain the other , i.e. , @xmath211 and @xmath208 are nontrivial . in general , @xmath211 and @xmath208 are normal subgroups of the group @xmath212 , which means that @xmath213 . let @xmath214 . in particular , the image of @xmath215 in the quotients @xmath216 and @xmath217 are normal subgroups of @xmath216 and @xmath217 which are isomorphic to @xmath208 and @xmath211 , respectively . when @xmath218 and @xmath219 , @xmath220 and we can use the following obvious fact . [ prod - lem ] suppose @xmath221 . for @xmath222 , consider representations @xmath223 which we view as representations of @xmath5 by extending trivially on the other component . let @xmath224 be the union of the images of @xmath225 and @xmath226 . say @xmath227 takes on value @xmath228 for @xmath229 elements of @xmath230 . then @xmath231 we will also need the following generalization . [ prod - lem2 ] let @xmath5 be a group containing a subgroup of the form @xmath232 . regard @xmath211 and @xmath208 as subgroups of @xmath233 in the obvious way , and suppose @xmath211 , @xmath208 and @xmath232 are all normal in @xmath5 . put @xmath234 for @xmath203 and suppose @xmath235 is a faithful @xmath0-dimensional representation of @xmath230 which we also view as a representation of @xmath5 via pullback . let @xmath236 , so @xmath237 and @xmath238 . let @xmath239 be the possible values of @xmath225 and @xmath226 . fix @xmath240 , and let @xmath241 denote its image in @xmath230 . say @xmath242 is the number of times @xmath227 takes on value @xmath228 on the coset @xmath243 in @xmath230 . then @xmath225 and @xmath226 agree on exactly @xmath244 elements of the coset @xmath245 in @xmath5 . this follows by the same reasoning as for lemma [ prod - lem ] , after using the fact that @xmath246 ( traces on @xmath5 ) simply means @xmath247 ( traces on @xmath216 and @xmath217 ) . in this section we will prove the @xmath40 case of theorem [ thm1 ] . we used the computer package gap 4 @xcite for explicit group and character calculations in this section and the next . for instance , one can calculate all extensions of @xmath5 by @xmath30 in gap if @xmath248 and @xmath249 . we use the notation @xmath250 $ ] for the @xmath251-th group of order @xmath0 in the small groups library , which is accessible by the command ` smallgroup(n , m ) ` in gap . recall the classification of finite subgroups of @xmath253 . any finite subgroup of @xmath254 is of one of the following types : a. cyclic b. dihedral c. tetrahedral ( @xmath255 ) d. octahedral ( @xmath32 ) e. icosahedral ( @xmath256 ) now suppose @xmath5 is a subgroup of @xmath68 with projective image @xmath181 in @xmath257 . if @xmath181 is cyclic , @xmath5 is reducible . if @xmath181 is dihedral , then @xmath5 is not primitive . assume @xmath5 is a primitive subgroup of @xmath68 . then we have the following possibilities . 1 . suppose @xmath258 . as @xmath259 has no irreducible 2-dimensional representations , @xmath5 must be an extension of @xmath259 by @xmath33 . note @xmath255 has a unique nontrivial extension by @xmath33 , @xmath260 . note @xmath261 has three irreducible 2-dimensionals , which are all twist equivalent . they are necessarily primitive as @xmath5 has no subgroups of index @xmath262 . their characters differ on @xmath263 elements . + each such degree 2 character has absolute value @xmath262 on @xmath264 elements , absolute value @xmath265 on @xmath266 elements , and absolute value @xmath267 on @xmath268 elements . 2 . next suppose @xmath269 . here @xmath32 has one irreducible 2-dimensional representation , but it is neither faithful nor primitive . there are 3 nontrivial extensions @xmath270 $ ] , @xmath26 $ ] and @xmath271 $ ] ( the 3 groups of order 48 with frattini factor @xmath32 ) . the latter group does not contain @xmath34 or have faithful 2-dimensional representations . for @xmath270 $ ] or @xmath272 $ ] , @xmath5 contains @xmath34 and has 2 faithful irreducible representations , which are primitive and twists of each other by a quadratic character . their restrictions to @xmath34 are irreducible . these primitive characters differ on @xmath273 elements . + each of the primitive degree 2 characters has absolute value @xmath262 on @xmath274 elements , absolute value @xmath275 on @xmath273 elements , absolute value @xmath265 on @xmath276 elements , and absolute value @xmath267 on @xmath277 elements . last , suppose @xmath278 . as @xmath279 has no irreducible 2-dimensional representations , @xmath5 must be an extension of @xmath279 by @xmath33 . there is a unique such nontrivial extension , @xmath280 . now @xmath281 has two twist - inequivalent irreducible 2-dimensional representations , which are primitive . their characters differ on @xmath282 elements . + each such character has absolute value @xmath262 on @xmath283 elements , absolute value @xmath284 on @xmath285 elements , absolute value @xmath286 on @xmath285 elements , absolute value @xmath265 on @xmath287 elements , and absolute value @xmath267 on @xmath288 elements . consequently , we see that , up to isomorphism , there are 4 primitive subgroups of @xmath68 : @xmath34 , @xmath25 $ ] , @xmath26 $ ] and @xmath281 . since each of these are irreducible with center @xmath33 , lemma [ red - lem ] implies that these are also all possibilities for the primitive subgroups of @xmath252 . + + l|ccccc |g & 2 & 1 & 0 & 2 & 2 + a_4 & 112 & 23 & 14 & & + s_4 & 124 & 13 & 38 & & 14 + a_5 & 160 & 13 & 14 & 25 & [ lem : diffimage - sl2 ] let @xmath114 be two primitive 2-dimensional representations of a finite group @xmath5 with projective images @xmath289 , @xmath290 . if @xmath291 , then @xmath292 . let @xmath293 , so we can view @xmath235 as a faithful representation of @xmath230 , for @xmath294 . by above , each @xmath230 is isomorphic to one of @xmath34 , @xmath25 $ ] , @xmath26 $ ] and @xmath281 . we use table [ tab : sl2 ] , which lists , for each possibility for @xmath295 , the fraction of elements of @xmath5 with @xmath296 having a given value . if @xmath297 , then the fraction of @xmath240 such that @xmath298 is at most @xmath299 , i.e. , @xmath300 . if @xmath301 is either @xmath302 or @xmath303 , then we see @xmath304 . [ thm : deg2 ] for any finite group @xmath5 , @xmath305 , with equality being achieved if and only if @xmath5 is an extension of @xmath25 $ ] or @xmath26 $ ] . by lemma [ lem : diffimage - sl2 ] , we may assume the projective images of @xmath118 and @xmath119 are isomorphic , which we denote @xmath181 . for @xmath222 , put @xmath202 . as mentioned above , we may assume @xmath206 . put @xmath234 , so we may view each @xmath230 as a primitive subgroup of @xmath252 , and @xmath306 . as @xmath295 determines @xmath307 , we have @xmath308 . if @xmath309 , then we may view @xmath5 as a primitive subgroup of @xmath252 , with the four possibilities given above . in this case , we may use the above calculations for the primitive subgroups of @xmath68 to see @xmath310 , with equality only achieved in the octahedral case . so assume @xmath311 . suppose @xmath258 . then @xmath312 . as remarked above , @xmath211 is isomorphic to a normal subgroup of @xmath34 . the nontrivial normal subgroups of @xmath34 are @xmath33 , @xmath313 and @xmath34 . there are no nontrivial extensions of @xmath34 by @xmath33 , so if @xmath314 , then @xmath315 and one can check @xmath316 . suppose next @xmath317 . ( then there are two possibilities for @xmath5 , @xmath318 or @xmath319 $ ] . ) put @xmath320 . the primitive degree 2 characters @xmath321 of @xmath34 take on the following values on each conjugacy class l|ccccccc & 1 & 1 & 6 & 4 & 4 & 4 & 4 + _ 1 & 2 & -2 & 0 & & ^2 & - & -^2 + _ 2 & 2 & -2 & 0 & ^2 & & -^2 & - + for some ordering of conjugacy classes , where @xmath322 , and the sizes of the conjugacy classes are given in the top row . the normal @xmath313 subgroup of @xmath34 consists of the first 3 conjugacy classes in the above table . hence @xmath225 and @xmath226 agree on exactly @xmath323 elements of @xmath29 by lemma [ prod - lem ] . we also see from the character table of @xmath34 that on either nontrivial coset in @xmath324 , @xmath171 must take on at least 2 distinct values which occur for at most 4 elements each . hence by lemma [ prod - lem2 ] , @xmath225 and @xmath226 agree on at most @xmath325 elements of each nontrivial coset in @xmath326 . thus @xmath327 . ( equality actually occurs for @xmath328 $ ] . ) lastly , if @xmath5 is an extension of @xmath34 by @xmath34 which contains two normal subgroups isomorphic to @xmath34 with trivial intersection , then @xmath329 and @xmath330 by lemma [ prod - lem ] . suppose @xmath269 , so @xmath216 and @xmath217 are each isomorphic to either @xmath25 $ ] or @xmath26 $ ] . for either @xmath331 $ ] or @xmath26 $ ] , @xmath230 has 4 nontrivial normal subgroups : @xmath33 , @xmath313 , @xmath34 and @xmath230 . let @xmath29 be the preimage of @xmath332 under @xmath118 . since @xmath333 is normal in @xmath216 , if it is a proper subgroup then it lies inside the @xmath34 subgroup and @xmath334 . hence @xmath335 as @xmath29 has index 2 in @xmath5 . if @xmath336 or @xmath313 or @xmath34 , then @xmath337 by case i. then proposition [ prop2 ] implies @xmath338 . otherwise @xmath339 , so @xmath220 and @xmath330 by lemma [ prod - lem ] . suppose @xmath340 , so @xmath341 . the only nontrivial normal subgroups of @xmath281 are @xmath33 and @xmath281 . however , @xmath281 has no nontrivial extensions by @xmath33 and @xmath342 . otherwise @xmath343 and @xmath344 so @xmath330 by lemma [ prod - lem ] . we remark that rather than trying to get the best bound we can in each case , we aimed to get the desired bound using as little as possible . we will employ the same philosophy for @xmath99 . here we prove the @xmath99 case of theorem [ thm1 ] . first we review the classification of finite subgroups @xmath345 . the classification can be found in blichfeldt @xcite or miller blichfeldt dickson @xcite . we follow the classification system therein . the description involves 3 groups , @xmath346 , @xmath347 and @xmath348 of respective orders 36 , 72 and 216 , which will be specified later . any finite subgroup @xmath5 of @xmath69 with projective image @xmath181 is one of the following types , up to conjugacy : a. abelian b. a nonabelian subgroup of @xmath349 c. a group generated by a diagonal subgroup and @xmath350}$ ] d. a group generated by a diagonal subgroup , @xmath350}$ ] and a matrix of the form @xmath351}$ ] e. @xmath352 f. @xmath353 g. @xmath354 h. @xmath355 i. @xmath356 j. @xmath357 of these types , ( a ) , ( b ) are intransitive , ( c ) , ( d ) are monomial , and the remaining types are primitive . the first 3 primitive groups have non - simple projective image , whereas the latter 3 have simple projective image . the groups @xmath181 in cases ( h ) , ( i ) , ( j ) are simple . the groups in cases ( e ) , ( f ) , ( g ) are not . the latter groups are @xmath358 $ ] , @xmath359 $ ] , and @xmath360 $ ] . explicit matrix presentations for preimages in @xmath69 are given in ( * ? ? ? * sec 8.1 ) . their normal subgroups are described below . a. @xmath346 has 2 nontrivial normal subgroups , one of order 18 , one of order 9 ( @xmath361 ) . b. @xmath347 has 5 nontrivial normal subgroups , 3 of order 36 each isomorphic to @xmath346 , 1 of order 18 , and one of order 9 ( @xmath361 ) . c. @xmath348 has 3 nontrivial normal subgroups , one of order 72 , 1 of order 18 and one of order 9 ( @xmath361 ) . now we describe the irreducible degree 3 characters for each of the types ( e)(j ) . 1 . @xmath358 $ ] has no irreducible 3-dimensional representations . there are 2 nontrivial extensions of @xmath346 by @xmath362 : @xmath363 $ ] and @xmath364 $ ] . the latter has no irreducible 3-dimensional representations . the group @xmath365 $ ] has 8 irreducible 3-dimensional representations , which are all all primitive and faithful , and fall into 2 twist classes . any two degree 3 characters differ on at least @xmath366 elements . + each of these degree 3 characters has absolute value @xmath367 on @xmath368 elements , @xmath265 on @xmath369 elements and @xmath267 on @xmath370 elements . 2 . @xmath359 $ ] has no irreducible 3-dimensional representations . there are 2 nontrivial extensions of @xmath347 by @xmath362 : @xmath371 $ ] and @xmath372 $ ] . again , the latter has no irreducible 3-dimensional representations . the group @xmath373 $ ] has 8 irreducible 3-dimensional representations , which are all all primitive and faithful , and fall into 2 twist classes . any two degree 3 characters agree on at most @xmath374 elements . + each of these degree 3 characters has absolute value @xmath367 on @xmath375 elements , @xmath265 on @xmath376 elements and @xmath267 on @xmath377 elements . more precisely , each character takes on value @xmath378 for @xmath379 element , for each cube root of unity @xmath136 . 3 . @xmath360 $ ] has 1 irreducible 3-dimensional representation , though it is not primitive or faithful . there are 4 nontrivial extensions of @xmath348 by @xmath362 : @xmath380 $ ] , @xmath381 $ ] , @xmath382 $ ] , and @xmath383 $ ] . the former 3 have 7 irreducible 3-dimensionals each , and the latter has 3 irreducible 3-dimensionals . the 3-dimensionals for the latter group are neither primitive nor faithful . + for @xmath384 $ ] , @xmath381 $ ] or @xmath382 $ ] , @xmath5 has 6 faithful primitive 3-dimensional representations , which lie in 2 twist classes . two degree 3 characters differ on at least @xmath385 elements . + each degree 3 character has absolute value @xmath367 on @xmath386 elements , absolute value @xmath387 on @xmath388 elements , absolute value @xmath265 on @xmath389 elements , and absolute value @xmath267 on @xmath390 elements . @xmath391 has two 3-dimensional representations , which are faithful and primitive . the characters differ on @xmath392 elements . note @xmath5 has no nontrivial extensions by @xmath362 . + each degree 3 character has absolute value @xmath367 on @xmath393 elements , absolute value @xmath265 on @xmath394 elements , absolute value @xmath395 on @xmath396 elements , absolute value @xmath397 on @xmath396 elements , and absolute value @xmath267 on @xmath398 elements . @xmath399 has no irreducible 3-dimensionals , and one nontrivial extension by @xmath362 , @xmath400 $ ] . then @xmath5 has 4 3-dimensional representations , none of which are twists ( @xmath5 has no nontrivial 1-dimensionals ) . any two of the degree 3 characters differ on at least @xmath401 elements . + each of these two characters has absolute value @xmath367 on @xmath402 elements , absolute value @xmath265 on @xmath403 elements , absolute value @xmath395 on @xmath404 elements , absolute value @xmath397 on @xmath404 elements , and absolute value @xmath267 on @xmath405 elements . @xmath406 has 2 irreducible 3-dimensionals , which are not twists ( @xmath5 has no nontrivial 1-dimensionals as @xmath5 is simple ) . they differ on @xmath407 elements . note @xmath5 has no nontrivial extensions by @xmath362 . + each of these two characters has absolute value @xmath367 on @xmath408 elements , absolute value @xmath275 on @xmath407 elements , absolute value @xmath265 on @xmath409 elements , absolute value @xmath267 on @xmath410 elements . l|cccccc |g & 3 & 1 & 0 & 2 & 2 & 3 + g_36 & & 34 & 29 & & & + g_72 & 172 & 78 & 19 & & & + g_216 & 1216 & 58 & 727 & & & 19 + a_5 & 160 & 14 & 13 & 25 & + a_6 & 1360 & 38 & 29 & 25 & + ( 2,7 ) & 1168 & 38 & 13 & & 27 & the only primitive subgroups of @xmath69 with trivial center are @xmath279 and @xmath28 . consequently lemma [ red - lem ] tells us the following is a complete list of all primitive finite subgroups of @xmath345 up to isomorphism : @xmath363 $ ] , @xmath411 $ ] , @xmath380 $ ] , @xmath381 $ ] , @xmath382 $ ] , @xmath279 , @xmath412 , @xmath413 $ ] , @xmath28 , and @xmath414 . [ prop : deg3 ] let @xmath114 be two primitive 3-dimensional representations of a group @xmath5 with projective images @xmath289 , @xmath290 . suppose at least one of @xmath289 , @xmath290 is simple . then @xmath415 , with equality only if @xmath416 . let @xmath202 and @xmath230 be the image of @xmath235 for @xmath222 . as before , we may assume @xmath206 . then @xmath5 contains a normal subgroup isomorphic to @xmath232 whose image in @xmath216 is @xmath208 and image in @xmath217 is @xmath211 . say @xmath289 is simple . then by above , @xmath216 is isomorphic to one of @xmath279 , @xmath412 , @xmath413 $ ] , @xmath28 , and @xmath414 . suppose @xmath291 . if @xmath417 then table [ tabby ] shows @xmath415 just from considering elements with absolute character value @xmath275 . in fact the inequality is strict . if @xmath418 or @xmath419 and @xmath290 is not isomorphic to @xmath279 or @xmath419 , then considering elements with absolute character value @xmath420 shows @xmath421 . next , suppose @xmath418 and @xmath422 . then @xmath423 or @xmath412 and @xmath424 $ ] . since @xmath425 divides @xmath426 , we have that @xmath211 is a normal subgroup of @xmath413 $ ] of order divisible by 6 . the only possibility is that @xmath218 , so @xmath220 . by lemma [ prod - lem ] , one sees @xmath330 . suppose @xmath427 . first suppose @xmath211 or @xmath208 is trivial , say @xmath211 . then @xmath428 , and we can view @xmath119 as a ( not necessarily faithful ) representation of @xmath216 . the above descriptions of degree 3 characters of the possible @xmath216 ( together with calculations for @xmath429 and @xmath430 ) show @xmath415 , with the bound achieved when @xmath431 . now assume @xmath211 and @xmath208 are nontrivial . if @xmath218 or @xmath219 , then @xmath220 , and @xmath330 by lemma [ prod - lem ] . so we may assume @xmath211 is a proper nontrivial normal subgroup of @xmath217 , and similarly for @xmath208 and @xmath216 . then neither @xmath216 nor @xmath217 can be @xmath279 or @xmath28 . if @xmath432 , then @xmath433 , and either @xmath434 or @xmath435 . a similar statement holds for @xmath436 . consequently in every case we have @xmath437 and @xmath438 . suppose @xmath439 and @xmath434 . then @xmath5 contains @xmath361 as a normal subgroup with quotient @xmath279 . there is only one nontrivial extension of @xmath412 by @xmath362 , @xmath440 $ ] , but it does not possess @xmath361 as a normal subgroup . thus @xmath5 must be the trivial extension @xmath441 and then @xmath442 . suppose @xmath439 and @xmath443 . then @xmath5 contains @xmath444 as a normal subgroup with quotient @xmath362 , i.e. , @xmath5 is an extension of @xmath362 by @xmath445 . note an irreducible degree 3 character of @xmath279 takes distinct values on each conjugacy class . by lemma [ prod - lem ] , @xmath225 and @xmath226 agree on @xmath446 elements of @xmath29 . on @xmath447 , an irreducible degree 3 character has absolute value @xmath367 on @xmath262 elements , value @xmath267 on @xmath448 elements , absolute value @xmath265 on @xmath449 elements , absolute value @xmath284 on @xmath450 elements , and absolute value @xmath451 on @xmath450 elements . thus by lemma [ prod - lem2 ] , @xmath225 and @xmath226 agree on at most @xmath452 elements of @xmath126 . this implies @xmath453 . suppose @xmath454 $ ] so @xmath455 . consider a maximal subgroup @xmath456 of @xmath216 of the form @xmath412 and let @xmath29 be the preimage in @xmath5 . then @xmath457 is still irreducible and @xmath29 is an extension of @xmath412 by @xmath362 . since @xmath217 contains no such subgroups , @xmath458 , and as in case iia , @xmath459 . one can check from the character table that the traces of @xmath118 and @xmath119 differ on at least @xmath460 elements of @xmath29 , and at least @xmath461 elements of @xmath29 which are of order @xmath462 . there are 6 conjugates of @xmath456 in @xmath216 , and the intersection of any two is isomorphic to @xmath463 , and thus contains no elements of order @xmath464 . in particular , the intersection of any two of the distinct corresponding 6 conjugates of @xmath29 in @xmath5 contains no elements of order @xmath462 . thus @xmath225 and @xmath226 differ on at least @xmath465 elements of the union of the conjugates of @xmath29 . hence @xmath466 . suppose @xmath467 and @xmath434 . as in case iia , there is only one nontrivial extension of @xmath216 by @xmath362 , @xmath468 $ ] , but is does not contain @xmath361 as a normal subgroup so @xmath469 and @xmath470 . suppose @xmath467 and @xmath471 . as in case iib , let @xmath320 and lemma [ prod - lem ] tells us @xmath225 and @xmath226 agree on @xmath472 elements of @xmath29 . on @xmath126 , @xmath227 takes on absolute value 3 on @xmath262 elements , absolute value @xmath265 on @xmath473 elements , absolute value @xmath267 on @xmath474 elements and absolute value @xmath275 on @xmath475 elements . so by lemma [ prod - lem2 ] , @xmath225 and @xmath226 agree on at most @xmath476 elements . hence @xmath477 . to complete the determination of @xmath478 , we need the following facts about normal subgroups and restrictions of characters for the possible preimages of @xmath346 , @xmath347 and @xmath348 in @xmath345 . a. if @xmath479 then @xmath480 $ ] . this @xmath5 has 3 nontrivial proper normal subgroups : @xmath362 , @xmath481 $ ] ( the nonabelian group of order 27 with exponent 3 ) and @xmath482 $ ] ( the nontrivial extension of @xmath33 by @xmath481 $ ] containing 4 normal @xmath361 subgroups ) . if @xmath201 is an irreducible degree 3 character , then its restriction to @xmath482 $ ] or @xmath481 $ ] is also irreducible . on @xmath483 $ ] , such a character takes on value @xmath378 on exactly 1 element for each cube root of unity @xmath136 and takes on value @xmath267 on all other @xmath450 elements . on @xmath482 \setminus [ 27,3]$ ] , @xmath201 takes on 3 different values which are roots of unity , each of which occur for @xmath484 elements . on @xmath363 \setminus [ 54,8]$ ] , @xmath201 takes on at least 3 different values , which are roots of unity , each of which occurs for at most @xmath485 elements . b. if @xmath486 then @xmath487 $ ] . this @xmath5 has 6 nontrivial proper normal subgroups : 3 of index 2 each isomorphic to @xmath363 $ ] , one @xmath482 $ ] , one @xmath481 $ ] and one @xmath362 . if @xmath201 is an irreducible degree 3 character , then its restriction to any @xmath363 $ ] , @xmath482 $ ] or @xmath481 $ ] is still irreducible . on @xmath411 \setminus [ 108,15]$ ] , @xmath201 take on between 3 to 6 distinct roots of unity , each of which occurs for exactly @xmath485 or @xmath488 elements . c. if @xmath489 , then @xmath5 is @xmath490 $ ] , @xmath491 $ ] or @xmath492 $ ] . in each case there are 4 proper nontrivial normal subgroups of @xmath5 : @xmath371 $ ] , @xmath493 $ ] , @xmath483 $ ] and @xmath362 . these are nested subgroups . in each case , the 6 primitive 3-dimensional representations are necessarily irreducible when restricted to the normal @xmath411 $ ] subgroup , and thus also irreducible when restricted to the @xmath493 $ ] and @xmath483 $ ] normal subgroups . + on @xmath494 $ ] , an irreducible degree 3 character takes value @xmath267 on @xmath461 elements , takes on 6 different values of modulus @xmath387 , each exactly @xmath495 times , and either 6 or 3 different values of modulus @xmath265 each exactly @xmath488 or @xmath496 times , respectively . ( the former possibility happens whenever @xmath497 $ ] or @xmath498 $ ] , and the latter happens whenever @xmath499 $ ] . ) [ thm : deg3 ] let @xmath5 be a finite group . then @xmath500 , with equality if and only if @xmath5 is an extension of @xmath28 . let @xmath114 be inequivalent irreducible primitive 3-dimensional representations of @xmath5 . for @xmath294 , say @xmath235 has kernel @xmath501 , image @xmath230 and projective image @xmath295 . by proposition [ prop : deg3 ] , we may assume each @xmath502 . as before , we may assume @xmath503 . also assume @xmath504 . from section [ sec : sl3-subgps ] , if @xmath505 , i.e. , @xmath506 , then @xmath507 . so we may assume @xmath211 and @xmath208 are nontrivial . moreover , @xmath508 and @xmath333 are normal subgroups of @xmath217 and @xmath216 , respectively . as above , if these are not proper , then @xmath220 and lemma [ prod - lem ] implies @xmath330 ( looking at the absolute character values in table [ tabby ] is not enough for this , but it suffices to note from the character tables that any character value is obtained on less than half the elements ) . suppose @xmath509 , so @xmath510 $ ] . then @xmath438 , which is isomorphic to @xmath362 , @xmath483 $ ] or @xmath493 $ ] . suppose @xmath434 . then @xmath5 is an extension of @xmath216 by @xmath362 . there are only 2 such extensions , @xmath511 and @xmath512 $ ] . the latter only has one quotient group isomorphic to @xmath363 $ ] , thus @xmath513 . then @xmath514 . suppose @xmath515 $ ] . let @xmath456 be the normal @xmath482 $ ] subgroup and @xmath516 . now @xmath517 implies @xmath518 , so @xmath519 and @xmath520 is an index 2 subgroup in @xmath217 , so isomorphic to @xmath482 $ ] . consider @xmath118 and @xmath119 restricted to @xmath29 . from the character values listed above in ( i ) , lemma [ prod - lem2 ] implies @xmath225 and @xmath226 differ on at least @xmath521 elements of @xmath522 . also , by lemma [ prod - lem ] , @xmath225 and @xmath226 differ on @xmath523 elements of @xmath232 . now consider values on @xmath126 . on @xmath363 \setminus [ 54,8]$ ] , @xmath227 takes on at least 3 different values ( all roots of unity ) , each of which can occur at most on @xmath485 elements . thus , by lemma [ prod - lem2 ] and ( i ) , @xmath225 and @xmath226 differ on at least @xmath524 elements in each of the two cosets @xmath525 which are not contained in @xmath29 . this gives @xmath526 . for use in case ivb , we get a better bound . one can check that the 3 nontrivial cosets of @xmath527 $ ] by @xmath481 $ ] are each unions of 3 conjugacy classes of size 9 , and a degree 3 character of @xmath527 $ ] takes on distinct values on each of the 3 conjugacy classes within a given coset . hence @xmath225 and @xmath226 differ on at least @xmath528 elements of @xmath529 , whence @xmath530 . suppose @xmath531 $ ] . let @xmath532 . let @xmath533 and @xmath534 . an irreducible degree 3 character of @xmath482 $ ] attains 3 values of modulus 3 on 1 element each , 3 values of modulus 1 on 9 elements , and 0 on 24 elements . thus by lemma [ prod - lem ] , @xmath225 and @xmath226 differ on at least 2094 elements of @xmath29 . hence @xmath535 . suppose @xmath536 and @xmath537 , so @xmath538 $ ] and @xmath539 $ ] . then @xmath540 , @xmath208 is isomorphic to a normal subgroup of [ 216 , 88 ] and @xmath211 is isomorphic to a normal subgroup of @xmath363 $ ] . either @xmath541 $ ] and @xmath542 $ ] , or @xmath543 $ ] and @xmath544 $ ] suppose @xmath541 $ ] and @xmath542 $ ] , so @xmath232 has index 4 in @xmath5 . from ( i ) above and lemma [ prod - lem ] , we have that @xmath225 and @xmath226 differ on @xmath545 elements of @xmath232 . let @xmath546 $ ] be one of the 3 normal index 2 subgroups of @xmath216 and @xmath547 . since @xmath548 , @xmath334 , and thus @xmath549 $ ] . comparing character values on @xmath482 \setminus [ 27,3]$ ] and @xmath527 \setminus [ 54,8]$ ] in ( i ) and using lemma [ prod - lem2 ] , we see that the traces of @xmath118 and @xmath119 differ on at least @xmath550 elements of @xmath522 . hence @xmath551 so @xmath552 . suppose @xmath543 $ ] and @xmath544 $ ] , so @xmath553 has index @xmath262 in @xmath5 . by ( i ) and lemma [ prod - lem ] we see that the traces of @xmath118 and @xmath119 agree on at most @xmath554 elements . hence @xmath555 and @xmath556 . suppose @xmath557 and @xmath537 . then @xmath558 and @xmath559 so @xmath560 , and @xmath561 . however @xmath216 has no normal subgroups of order @xmath562 , @xmath563 or @xmath564 . this case does not occur . suppose @xmath565 . then @xmath566 $ ] , @xmath308 and @xmath211 , @xmath208 are both isomorphic to @xmath362 , @xmath483 $ ] , @xmath493 $ ] or @xmath363 $ ] . suppose @xmath434 . then @xmath5 is an extension of @xmath216 by @xmath362 . there are two such extensions : @xmath511 and @xmath567 $ ] . the latter has two normal subgroups isomorphic to @xmath362 but only one quotient isomorphic to @xmath216 . hence @xmath513 and @xmath568 . suppose @xmath515 $ ] . then @xmath232 has index @xmath569 in @xmath5 . let @xmath456 be a subgroup of @xmath216 of index @xmath262 , @xmath516 and @xmath570 , so @xmath29 has index @xmath262 in @xmath5 and @xmath571 $ ] . by case ib , @xmath572 . now one can check that each of the 4 cosets of @xmath411 $ ] modulo @xmath481 $ ] which are not contained in @xmath527 $ ] have 9 elements in 3 different conjugacy classes , and a degree 3 character takes on distinct values on each of these 3 conjugacy classes . hence by lemma [ prod - lem2 ] , on each of the 4 cosets of @xmath525 not in @xmath573 , @xmath225 and @xmath226 differ on at least @xmath574 elements . therefore @xmath575 . suppose @xmath531 $ ] so @xmath320 has index @xmath576 in @xmath5 . as in case ic , @xmath225 and @xmath226 differ on at least 2094 elements of @xmath29 . the 3 nontrivial double cosets of @xmath411/[54,8]$ ] each consist of 3 conjugacy classes of size 18 , and a degree 3 character takes on distinct values on these classes . so by lemma [ prod - lem2 ] , @xmath225 and @xmath226 differ on at least @xmath577 elements of each nontrivial coset of @xmath326 not contained in @xmath29 . hence @xmath578 . suppose @xmath579 $ ] so @xmath232 has index @xmath262 in @xmath5 . then by lemma [ prod - lem ] , @xmath225 and @xmath226 differ on at most @xmath580 elements of @xmath232 . hence @xmath581 . suppose @xmath557 and @xmath582 . then @xmath583 , and @xmath584 . any nontrivial proper normal subgroup of @xmath216 has order @xmath367 , @xmath585 , @xmath586 or @xmath587 . thus this case does not occur . suppose @xmath588 . so each of @xmath216 and @xmath217 are isomorphic to @xmath490 $ ] , @xmath491 $ ] or @xmath492 $ ] . then @xmath589 and @xmath590 are isomorphic to one of @xmath362 , @xmath483 $ ] , @xmath482 $ ] , and @xmath411 $ ] . suppose @xmath434 . then @xmath5 is an extension of @xmath216 by @xmath362 . there are two possible such extensions : @xmath511 and @xmath591 $ ] . in the former case , @xmath592 . suppose @xmath593 $ ] . then @xmath5 has four normal @xmath362 subgroups , with quotients isomorphic to @xmath380 $ ] , @xmath381 $ ] , @xmath382 $ ] , and @xmath383 $ ] . there are 18 primitive degree 3 characters , and any two differ on at least @xmath594 elements . suppose @xmath515 $ ] . let @xmath456 be the normal index 3 subgroup of @xmath216 , @xmath516 and @xmath595 , so @xmath596 $ ] . by case ivb , the traces of @xmath118 and @xmath119 differ on at least @xmath597 elements of @xmath29 . by ( iii ) and lemma [ prod - lem2 ] , @xmath225 and @xmath226 agree on at most @xmath598 elements of @xmath126 . hence @xmath599 . suppose @xmath531 $ ] . let @xmath29 , @xmath456 and @xmath600 be as in the previous case . from case ivc , the traces of @xmath118 and @xmath119 differ on at least @xmath601 elements of @xmath29 . by ( iii ) and lemma [ prod - lem2 ] , @xmath225 and @xmath226 differ on at least @xmath602 elements of @xmath126 . thus @xmath603 . suppose @xmath604 $ ] . then @xmath232 has index 3 in @xmath5 . from ( i ) and ( ii ) and lemma [ prod - lem ] , @xmath225 and @xmath226 differ on at least @xmath605 elements of @xmath232 . lemma [ prod - lem2 ] and ( iii ) imply that @xmath225 and @xmath226 differ on at least @xmath606 elements of @xmath607 . thus @xmath608 . we consider @xmath609 and @xmath610 , for even and odd prime powers @xmath51 . we separate these into three subsections : @xmath3 , @xmath51 odd ; @xmath611 , @xmath51 even ; and @xmath610 , @xmath51 odd . we refer to , and mostly follow the notation of , fulton harris @xcite for the representations of these groups . choose an element @xmath612 . we denote @xmath613 for the unique quadratic extension of @xmath614 . we can write the elements of @xmath615 as @xmath616 , where @xmath617 . the norm map @xmath618 is then defined as @xmath619 . we also denote @xmath620 to be the kernel of the norm map . a. @xmath622 . + b. @xmath623 . + c. conjugacy classes of the form @xmath624 $ ] , where @xmath625 where @xmath626 and @xmath627 . so there are four conjugacy classes , each of size @xmath628 . + d. conjugacy classes of the form @xmath629 $ ] , where @xmath630 with @xmath631 . since the conjugacy classes @xmath632 and @xmath633 are the same , we have @xmath634 different conjugacy classes , each of size @xmath635 . + e. conjugacy classes of the form @xmath636 $ ] , where @xmath637 where @xmath638 and @xmath639 . since @xmath640 we have @xmath641 conjugacy classes , each of size @xmath642 . + we give a brief description of the representations that appear in the character table . the first set of representations , denoted @xmath643 , are induced from the subgroup @xmath644 of upper triangular matrices . given a character @xmath645 , we can extend this to a character of @xmath644 , which we then induce to a @xmath646-dimensional representation @xmath643 of @xmath647 . if @xmath648 , then the induced representation is irreducible . if @xmath649 , then @xmath175 decomposes into its irreducible consituents : the trivial representation @xmath650 and the steinberg representation @xmath651 . if @xmath652 and @xmath653 , then it decomposes into two irreducible constituents denoted @xmath654 and @xmath655 . for the remaining irreducible representations , we consider characters @xmath656 and @xmath657 of the diagonal subgroup @xmath140 and the subgroup @xmath658 , respectively , where the characters agree when restricted to @xmath659 . then we construct a virtual character @xmath660 ( note that the virtual character will not depend on the specific choice of @xmath656 ) . when @xmath661 , @xmath662 decomposes into two distinct characters . in the case when @xmath657 is trivial , @xmath663 decomposes into the difference between the characters for the steinberg representation and the trivial representation . if @xmath657 is the unique ( non - trivial ) order 2 character of @xmath664 , then @xmath662 decomposes into two distinct irreducible characters of equal dimension ; we will label the corresponding representations @xmath665 and @xmath666 . if @xmath667 , then @xmath662 corresponds to an irreducible representation , which we denote as @xmath668 . two irreducibles @xmath668 and @xmath669 are equivalent if and only if @xmath670 or @xmath671 . we note that out of all the irreducible representations , the imprimitive representations are exactly all the @xmath643 ( for @xmath648 ) . + we define some notation that will appear in the character table for @xmath3 . let @xmath645 with @xmath672 , and @xmath657 a character of @xmath620 with @xmath673 . fix @xmath152 to be the non - trivial element of @xmath674 , and let @xmath675 lastly , we define @xmath676 to be the non - trivial element of @xmath677 . the character table is : assume @xmath684 . any two representations @xmath685 have trace characters that may differ only for the conjugacy classes @xmath686 $ ] . in the case of the conjugacy class @xmath687 $ ] , we note that @xmath688 has order 2 in @xmath689 . parametrize the nontrivial maps @xmath690 by @xmath691 as before . then @xmath692 if and only if @xmath693 have the same parity
let be a-diffeomorphism , , defined on a compact boundaryless-dimensional manifold , , and let be the homoclinic class associated to the hyperbolic periodic point . we prove that if there exists a neighborhood of such that for every the continuation of is entropy - expansive then there is a-invariant dominated splitting for of the form where is contracting , is expanding and all are one dimensional and not hyperbolic . [ multiblock footnote omitted ]
0903.2948
in this paper we study what are the consequences at the dynamical behavior of the tangent map @xmath13 of a diffeomorphism @xmath18 , assuming that @xmath10 is robustly entropy expansive . in this direction we obtain that the tangent bundle has a @xmath13-invariant dominated splitting of the form @xmath14 where @xmath15 is contracting , @xmath16 is expanding and all @xmath17 are one dimensional and not hyperbolic . let @xmath4 be a compact connected boundary - less riemannian @xmath3-dimensional manifold , @xmath5 , and @xmath19 a homeomorphism . let @xmath20 be a compact invariant subset of @xmath4 and @xmath21 a distance in @xmath4 compatible with its riemannian structure . for @xmath22 , @xmath23 and @xmath24 we say that @xmath15 @xmath25-spans @xmath26 with respect to @xmath10 if for each @xmath27 there is @xmath28 such that @xmath29 for all @xmath30 . let @xmath31 denote the minimum cardinality of a set that @xmath25-spans @xmath26 . since @xmath20 is compact @xmath32 . we define @xmath33 and the topological entropy of @xmath10 restricted to @xmath26 as @xmath34 the last limit exists since @xmath35 increases as @xmath36 decreases to zero . [ gammadeepsilon ] for @xmath37 let us denote @xmath38 we will simply write @xmath39 instead of @xmath40 when it is understood which @xmath10 we refer to . following bowen ( see @xcite ) we say that @xmath41 is _ entropy - expansive _ or _ @xmath42-expansive _ for short , if and only if there exists @xmath43 such that @xmath44 ( * ? ? ? * theorem 2.4 ) [ bowen ] for all homeomorphism @xmath10 defined on a compact invariant set @xmath20 it holds @xmath45 a similar notion to @xmath42-expansiveness , albeit weaker , is the notion of _ asymptotically @xmath42-expansiveness _ introduced by misiurewicz @xcite : let @xmath20 be a compact metric space and @xmath46 an homeomorphism . we say that @xmath10 is asymptotically @xmath42-expansive if and only if @xmath47 thus , we do not require that for a certain @xmath48 , @xmath49 but that @xmath50 when @xmath51 . it has been proved by buzzi , @xcite , that any @xmath52 diffeomorphism defined on a compact manifold is asymptotically @xmath42-expansive . the interessed reader can found examples of diffeomorphisms that are not entropy expansive neither asymptotically entropy expansive in @xcite . next we recall the notion of dominated splitting . [ domi ] we say that a compact @xmath10-invariant set @xmath53 admits a dominated splitting if the tangent bundle @xmath54 has a continuous @xmath13-invariant splitting @xmath55 and there exist @xmath56 , such that @xmath57 observe that if the topological entropy of a map @xmath18 vanishes , @xmath58 , then automatically @xmath10 is @xmath42-expansive . for instance morse - smale diffeomorphisms @xmath59 are @xmath42-expansive . we remark that morse - smale diffeomorphisms are @xmath8-stable under perturbations and so they constitute a class which is robustly @xmath42-expansive . here we are interested in diffeomorphisms that exhibit a chaotic behavior , i.e. : their topological entropy is positive . moreover , we restrict our study to homoclinic classes @xmath6 associated to saddle - type hyperbolic periodic points . recall that the homoclinic class @xmath6 of a saddle - type hyperbolic periodic point @xmath7 of @xmath60 is the closure of the intersections between the unstable manifold @xmath61 of @xmath7 and the stable manifold @xmath62 of @xmath7 . these classes persist under perturbations and we wish to establish the property of those classes under the assumption that @xmath42-expansiveness is robust . [ robuhagar ] let @xmath4 be a compact boundaryless @xmath52 manifold and @xmath18 be a @xmath1 diffeomorphism , @xmath2 . let @xmath6 be a @xmath10-homoclinic class associated to the @xmath10-hyperbolic periodic point @xmath7 . assume that there is a @xmath1 neighborhood @xmath63 of @xmath10 , such that for any @xmath11 it holds that the continuation @xmath12 of @xmath6 is @xmath42-expansive . then we say that @xmath64 is @xmath1-robustly @xmath42-expansive . in ( * ? ? ? * theorem b ) we obtain that if @xmath65 is isolated and the finest dominated splitting on @xmath65 is @xmath66 with @xmath15 contracting , @xmath16 expanding and all @xmath17 , @xmath67 , one dimensional and not hyperbolic , then @xmath68 is @xmath42-expansive . moreover , since the dominated splitting is preserved under @xmath8-perturbations this result holds for a @xmath8-neighborhood @xmath69 , i.e. : @xmath42-expansiveness is @xmath8-robust . roughly speaking , ( * ? ? * theorem b ) says that the domination property implies that small neighbourhoods in @xmath6 have an ` ordered dynamics ' and there can not appear ` arbitrarily small horseshoes ' , i.e : , horseshoes generated by homoclinic points in @xmath70 for @xmath71 arbitrarily small and @xmath72 periodic , as in the example given in @xcite[section 2 ] for a surface diffeomorphism . the presence of these arbitrarily small horseshoes would imply that @xmath73 for any @xmath74 . this paper is intended to continue @xcite in the reverse direction : we analyze the consequences of @xmath42-expansiveness to hold in a @xmath8-neighbourhood @xmath75 of @xmath10 . our main results are the following : [ prin1 ] let @xmath4 , @xmath18 and @xmath6 be as in definition [ robuhagar ] for @xmath76 . then @xmath6 has a dominated splitting @xmath55 . in fact ( * ? ? ? * example 2 ) shows that in dimension greater or equal to three the existence of a dominated splitting for @xmath6 is not enough tho guarantee @xmath42-expansiveness , so it is natural to search for a stronger property . let us recall the concept of _ finest dominated splitting _ introduced in @xcite . [ finest ] let @xmath53 be a compact @xmath10-invariant subset such that @xmath77 with @xmath78 @xmath13 invariant , @xmath79 . we say that @xmath80 is dominated if for all @xmath81 @xmath82 has a dominated splitting . we say that @xmath80 is the finest dominated splitting when for all @xmath83 there is no possible decomposition of @xmath78 as two invariant sub - bundles having domination . an improvement of theorem [ prin1 ] is the following . [ prin2 ] let @xmath4 , @xmath18 and @xmath6 be as in definition [ robuhagar ] for @xmath76 . then the finest dominated splitting in @xmath6 has the form @xmath14 where all @xmath17 are one dimensional and not hyperbolic . if @xmath6 _ is isolated _ then we may refine the previous result . before we announce precisely this result , let us recall the definitions of : chain recurrent set , isolated homoclinic class and heterodimensional cycles .. the chain recurrent set of a diffeomorphism @xmath10 , denoted by @xmath84 , is the set of points @xmath85 such that , for every @xmath48 , there is a closed @xmath86-pseudo orbit joining @xmath85 to itself : there is a finite sequence @xmath87 such that @xmath88 . we say that @xmath6 is isolated if there are neighborhoods @xmath9 of @xmath10 in @xmath89 and @xmath90 of the homoclinic class class @xmath6 in @xmath4 such that , for every @xmath91 , the continuation @xmath12 of @xmath6 coincides with the intersection of the chain recurrence set of @xmath92 , @xmath93 with the neighborhood @xmath90 . generically a recurrence class which contains a periodic point @xmath94 coincides with @xmath12 , @xcite . we say that @xmath95 is a cycle if @xmath96 , where @xmath97 are hyperbolic periodic points of @xmath10 and @xmath98 , for all @xmath99 . @xmath95 is called a heterodimensional cycle if , for some @xmath100 , @xmath101 . recall that the _ index _ of a hyperbolic periodic point @xmath7 is the dimension of its unstable manifold @xmath102 . [ prin3 ] let @xmath4 , @xmath18 and @xmath6 be as in definition [ robuhagar ] for @xmath76 . assume moreover that @xmath64 is isolated . then for @xmath92 in @xmath103 , @xmath12 has a dominated splitting of the form @xmath104 where @xmath15 is contracting , @xmath16 is expanding and all @xmath17 are not hyperbolic and @xmath105 . moreover , in case that the index of periodic points in @xmath12 are in a @xmath8 robust way equal to @xmath106 then for an open dense subset @xmath107 , @xmath12 is hyperbolic , i.e. : @xmath108 . on the other hand , if there are @xmath92 arbitrarily @xmath8-close to @xmath10 such that in @xmath12 there are periodic points of different index then @xmath6 is approximated by robust heterodimensional cycles , @xcite . if we do not assume that @xmath6 is isolated but we know that @xmath10 can not be approximated by @xmath92 exhibiting a heterodimensional cycle we have the following result : [ prin4 ] let @xmath109 , and @xmath6 be as in definition [ robuhagar ] for @xmath76 . assume that @xmath110 . then for @xmath92 in a residual subset @xmath111 , @xmath12 has a dominated splitting of the form @xmath112 where @xmath113 is not hyperbolic and @xmath114 , @xmath115 is contracting and @xmath116 is expanding . moreover , if @xmath117 then @xmath118 dominated . the proofs of theorems [ prin1 ] and [ prin2 ] go by contradiction : under the hypothesis that there is not a dominated splitting in @xmath119 , we profit from some ideas of @xcite and @xcite to create a flat tangency between @xmath62 and @xmath102 . we remark that in @xcite for the case that @xmath120 it was proved that if @xmath121 and @xmath92 has a homoclinic tangency then there are diffeomorphisms arbitrarily @xmath1-close to @xmath92 exhibiting persistent homoclinic tangencies ( thus generalizing results of @xcite , see also @xcite ) . in our case , since we can perform the perturbations in the @xmath8 topology , our arguments are simplier than theirs to obtain a @xmath122 diffeomorphism @xmath92 exhibiting a flat tangency , and afterward create an arc of tangencies between @xmath62 and @xmath102 . next we follow @xcite , to perform another @xmath8-perturbation with support in a small neighborhood of the arc of tangencies leading to the appearance of arbitrarily small horseshoes with positive entropy contradicting @xmath42-expansiveness . therefore @xmath123 admits a dominated spliting . moreover , either the finest dominated splitting ( see definition [ finest ] ) has the form @xmath124 where all @xmath17 are one dimensional and not hyperbolic or again we contradict robustness of @xmath42-expansiveness using ( * ? ? ? * theorem 6.6.8 ) . for the proof of theorem [ prin3 ] we assume some specific generic properties described in section [ genericos ] and that @xmath6 is isolated . these allow to prove that the extremal sub - bundles @xmath15 and @xmath16 are respectively contracting and expanding . moreover if the index of periodic points of @xmath12 is robustly the index of @xmath7 then for an open dense subset of @xmath103 the dominated splitting defined on @xmath119 is hyperbolic . this proof is done in two steps : ( 1 ) first we prove in lemma [ extremos1 ] that the extremal sub - bundles are hyperbolic using the fact that @xmath6 is isolated , @xcite . ( 2 ) second we show in lemma [ indiceconstante ] that if in a @xmath8-robust way the index of periodic points in @xmath12 are the same for @xmath125 then for an open and dense subset @xmath126 of @xmath103 we have that @xmath12 is hyperbolic . finally in theorem [ prin4 ] , where we do not assume that @xmath6 is isolated , we see , under the generic assumptions described at section [ genericos ] , that for a residual subset @xmath127 we have a dominated splitting @xmath128 defined on @xmath119 such that @xmath115 is contracting , @xmath116 is expanding and @xmath113 is dominated and at most two dimensional . for this we assume further that @xmath129 which allows to use ( * ? ? ? * maintheorem ) . in this section we prove theorem [ prin2 ] assuming that @xmath64 is robustly @xmath42-expansive . let @xmath6 be a @xmath10-homoclinic class associated to the hyperbolic periodic point @xmath7 . assume that there is a @xmath8 neighborhood @xmath130 of @xmath10 such that for any @xmath11 it holds that there is a continuation @xmath12 of @xmath6 such that @xmath12 is @xmath42-expansive . we may assume that @xmath7 is a _ hyperbolic fixed point _ since @xmath64 is @xmath42-expansive if and only if @xmath131 is @xmath42-expansive . this follows from the fact that for any compact @xmath10-invariant set @xmath132 we have that @xmath133 which implies that @xmath134 . let @xmath135 be a transverse homoclinic point associated to the periodic point @xmath136 . we define @xmath137 and @xmath138 . since @xmath7 is hyperbolic we have that @xmath139 . moreover , @xmath140 and @xmath141 are @xmath13-invariant , i.e. : @xmath142 and @xmath143 . denote by @xmath144 the set of the transverse homoclinic points associated to @xmath7 . then , it can be proved that @xmath145 . here @xmath146 stands for the closure in @xmath4 of the subset @xmath147 . so if we prove that there is a dominated splitting for @xmath144 we are done since we can extend by continuity the splitting to the closure @xmath6 . moreover , since @xmath122-diffeomorphisms are dense in the @xmath8-neighbourhood @xmath9 we may assume that @xmath10 is of class @xmath122 taking into account that we are assuming that @xmath42-expansiveness is @xmath8-robust . we will use the following result proved in @xcite : ( * ? ? ? * lemma 1.1)[fr ] let @xmath4 be a closed @xmath148-manifold , @xmath18 a @xmath8 diffeomorphism , and @xmath103 a given neighbourhood of @xmath10 . then , there exist @xmath149 and @xmath24 such that if @xmath150 , @xmath151 is a finite set , and @xmath152 are linear maps , @xmath153 , satisfying @xmath154 then there is @xmath155 satisfying @xmath156 and @xmath157 moreover , if @xmath90 is any neighborhood of @xmath158 then we may chose @xmath159 so that @xmath160 for all @xmath161 . the statement given there is slightly different from that above , but the proof of our statement is contained in @xcite . under the hypothesis of theorem [ prin1 ] , let us assume that @xmath10 is of class @xmath1 , @xmath121 and prove that there is a dominated splitting for @xmath144 . the proof goes by contradiction and it is done in several steps : ( 1 ) at lemma [ le : dom ] we perform a pertubation @xmath92 of @xmath10 exhibting a homoclinic point @xmath162 with small angle between @xmath163 and @xmath164 , ( 2 ) at proposition [ buenangulo ] we perform another perturbation ( that we still denote by @xmath92 ) of @xmath10 to create a tangency between @xmath165 and @xmath166 , @xmath167 , ( 3 ) at proposition [ chato ] through another pertubation of @xmath10 we create an arc of flat tangencies @xmath168 , ( 4 ) finally in subsection [ herradurita ] we perform a sequence of perturbations of @xmath10 leading to @xmath16 near @xmath10 presenting a sequence of two by two disjoint small horseshoes @xmath169 , @xmath170 as @xmath171 . moreover , we can select the sequence @xmath172 in such a way that none of then are a constant of @xmath42-expansiveness of @xmath16 . since the entropy of each of these small horseshoes is positive , we arrive to a contradiction to @xmath42-expansiveness of @xmath10 . to start , let us assume , by contradiction , that @xmath144 has no dominated splitting . then , by ( * ? ? ? * proof of theorem f ) it holds 1 . for all @xmath173 there exists @xmath174 such that for all @xmath175 , @xmath176 [ le : dom ] assume that ( ad ) holds . then , given @xmath177 and @xmath43 there is @xmath178 and @xmath92 an @xmath86-@xmath8-perturbation of @xmath10 with a homoclinic point @xmath179 associated to @xmath180 such that the angle at @xmath179 between @xmath163 and @xmath164 is less than @xmath181 . arguing by contradiction let us assume that there is @xmath182 such that for all @xmath92 in @xmath183 the angle at @xmath179 between @xmath163 and @xmath164 is greater or equal than @xmath184 . by hypothesis there exist vectors @xmath185 and @xmath186 with @xmath187 such that @xmath188 take @xmath43 small such that any @xmath8-@xmath86-perturbation of @xmath10 gives a diffeomorphism @xmath189 where @xmath183 is the @xmath8-neighborhood of @xmath10 where we have @xmath42-expansiveness . let @xmath190 be such that any perturbation of the derivatives along a finite orbit of @xmath10 can be realized via lemma [ fr ] by a @xmath8-@xmath86-perturbation of @xmath10 . let us define @xmath191 a linear map such that @xmath192 and @xmath193 , @xmath194 . note that @xmath195 stretches @xmath196 and left @xmath197 unchanged . let @xmath198 be a linear map satisfying @xmath199 in @xmath200 and @xmath201 in @xmath202 where @xmath203 is a linear map such that @xmath204 and @xmath205 . finally define @xmath206 , and @xmath207 for @xmath208 . by lemma [ fr ] there exists a diffeomorphism @xmath209 such that @xmath92 is @xmath86-near @xmath10 , keeps the orbit of @xmath174 unchanged for @xmath210 , and such that @xmath211 . we may assume ( and do ) that the support of the perturbation does not cut a small neighborhood of @xmath7 . it follows that @xmath174 continues to be a homoclinic point of @xmath92 . moreover , we do not change @xmath212 , @xmath213 , and @xmath214 is changed only for @xmath215 . thus such bundles are the stable and unstable directions of a homoclinic point of a diffeomorphism @xmath216 . we obtain that @xmath217 and after @xmath218 iterates we have @xmath219 . given @xmath220 we may find @xmath178 such that @xmath221 where @xmath182 is , by hypothesis of absurd , such that @xmath222 for all @xmath223 , @xmath216 , where @xmath224 stands for the angle between @xmath140 and @xmath141 . with this choice of @xmath218 , by ( * ? ? ? * lemma ii.10 ) we have @xmath225 @xmath226 dividing the inequality @xmath227 by @xmath228 and taking into account that by hypothesis @xmath229 we find @xmath230 arriving to a contradiction . hence @xmath231 , proving lemma [ le : dom ] . let us recall the following result which may be found in ( * ? ? ? * lemma 4.16 ) , see also ( * ? ? * lemma 3.8 ) . [ bdp ] let @xmath7 be a hyperbolic periodic point and @xmath6 its homoclinic class . assume that @xmath6 is not trivial . then there exists and arbitrarily small @xmath8-perturbation @xmath92 of @xmath10 and a hyperbolic periodic point @xmath232 of @xmath12 with period @xmath233 and homoclinically related with @xmath94 such that @xmath234 has only positive real eigenvalues of multiplicity one . observe that in the previous result , since @xmath235 , we have @xmath236 . so , to simplify notation , we may assume directly that @xmath237 and moreover that @xmath238 , and that @xmath7 is a fixed point . we order the eigenvalues of @xmath239 labeling them as @xmath240 so that the less contracting and the less expanding ones are respectively @xmath241 and @xmath242 . + by a small @xmath8-preturbations we may also assume that locally , in a neighborhood @xmath243 of @xmath7 , we have linearizing coordinates so that @xmath244 where we write @xmath245 for @xmath246 . the lines in @xmath247 corresponding to the eigenvalues @xmath248 may be extended to all of @xmath62 by backward iteration by @xmath10 giving us a foliation by lines of dimension @xmath249 . similarly for @xmath102 we have a @xmath250-foliation by lines obtained by forward iteration by @xmath10 . now , let us assume that @xmath92 is near @xmath10 , @xmath251 in a small neighborhood of @xmath7 and that there is a small angle between @xmath252 and @xmath253 where @xmath85 is a @xmath92-homoclinic point associated to @xmath7 . that is : there is @xmath181 small such that @xmath254 by theorem [ bdp ] , we may assume that all the eigenvalues of @xmath255 are positive with multiplicity one and that we have linearizing coordinates in a small neighborhood of @xmath7 . the next proposition stablishes that if the angle between @xmath252 and @xmath256 is small than we can create a tangency between @xmath257 and @xmath258 , for some @xmath159 near @xmath92 . [ buenangulo ] there is @xmath177 and @xmath259 so that for some @xmath260 there is a tangency between @xmath261 and @xmath262 if @xmath263 . moreover @xmath85 is a homoclinic point of @xmath264 , @xmath265 has dimension @xmath266 and there is @xmath267 so that if @xmath268 is the subspace common to @xmath261 and @xmath262 then @xmath269 is tangent to the line corresponding to the less contracting eigenvalue and @xmath270 is tangent to the line corresponding to the less expanding eigenvalue of @xmath271 . let @xmath103 , @xmath272 and @xmath36 be as in lemma [ fr ] . shrinking @xmath273 if it were necessary we may assume that @xmath274 . hence we may assume without loss of generality that there is some @xmath275 such that @xmath276 by hypothesis there is @xmath277 , @xmath278 and @xmath177 small so that @xmath279 taking @xmath280 , since @xmath263 , there exist @xmath281 and @xmath282 such that @xmath283 let @xmath284 be such that @xmath285 be defined by @xmath286 then we have @xmath287 take a neighborhood @xmath90 of @xmath288 such that @xmath289 . using lemma [ fr ] we find @xmath290 such that @xmath291 outside @xmath90 , and @xmath292 hence @xmath293 since its forward and backward orbits continue to converge to @xmath294 . moreover @xmath295 and so the intersection of @xmath296 and @xmath297 is not transverse at the point @xmath85 . since the eigenvalues of @xmath239 are all real positive and of multiplicity one and @xmath251 in a small neighborhood of @xmath7 , by @xmath298 forward iterations we have a vector @xmath299 almost tangent to the straight line @xmath300 corresponding to the less contracting eigenvalue at @xmath7 . again by lemma [ fr ] we can perturb @xmath264 outside a small neighborhood of @xmath7 to let the direction of @xmath301 coincide with @xmath300 . similarly we obtain @xmath302 tangent to the line corresponding to the less expanding eigenvector of @xmath303 . from proposition [ buenangulo ] we may assume for @xmath10 itself that there is a homoclinic point of tangency @xmath135 with properties analogous to those of @xmath264 . the next lemma asserts that under these hypothesis , we can obtain an arc @xmath304 of non - tranversal homoclinic points in @xmath305 . [ chato ] let @xmath7 be a hyperbolic fixed point for @xmath10 of index @xmath249 and @xmath135 such that the intersection at @xmath85 is not transversal . then by an arbitrarily small @xmath8-perturbation we may obtain a diffeomorphism @xmath92 with @xmath306 such that the intersection at @xmath85 is flat , there exists a small arc @xmath304 contained in the intersection of the stable and unstable manifolds of @xmath7 . moreover , there is @xmath267 such that @xmath307 is tangent to the eigenvector corresponding to the less contracting eigenvalue and analogously @xmath308 is tangent to the eigenvector corresponding to the less expanding eigenvalue . since @xmath7 is a hyperbolic saddle , @xmath62 is an euclidean @xmath249-dimensional hyperplane and @xmath102 an euclidean @xmath250-dimensional hyperplane both immersed in @xmath4 . if the intersection at @xmath85 of @xmath62 and @xmath102 is not transversal we should have a vector @xmath309 in @xmath310 , i.e. : we have a tangency between @xmath62 and @xmath102 at the homoclinic point @xmath85 . using lemma [ fr ] we may assume that the subspace generated by @xmath311 is the unique in common between @xmath312 and @xmath313 , that is @xmath314 has dimension @xmath266 . moreover , we also may assume that @xmath315 ( otherwise we may take @xmath316 instead of @xmath10 ) and , again by lemma [ fr ] , that the tangent space @xmath317 intersects trivially @xmath318 the orthogonal complement of @xmath319 . under these assumptions the orthogonal projection of @xmath320 into @xmath321 is locally a diffeomorphism in a suitable neighborhood of @xmath85 . let us choose @xmath322 a small disk and @xmath267 such that @xmath323 , and let @xmath324 be a small disk in @xmath325 such that @xmath326 . @xmath324 projects onto @xmath327 diffeomorphicaly . via a local coordinate map we may identify @xmath328 with @xmath329 with @xmath85 identified with the origin @xmath330 and @xmath311 having the direction of @xmath331 which is tangent at 0 to @xmath332 too . @xmath324 may be viewed as the graph of a map @xmath333 with @xmath334 . to simplify notation we write @xmath335 and @xmath336 . hence if @xmath337 then @xmath338 , where , given @xmath332 , @xmath339 is a local coordinate map from a neighborhood of 0 in @xmath340 to @xmath328 . [ claim30 ] there exists a @xmath8 perturbation of @xmath10 that produces a diffeomorphism @xmath341 with a flat intersection at @xmath342 , with @xmath343 and @xmath344 . this flat intersection contains a small arc @xmath304 . define @xmath345 by @xmath346 here @xmath16 is a @xmath52-bump function , @xmath347 , that vanishes in the boundary of the ball @xmath348 , is equal to 1 in @xmath349 , and such that @xmath350 , where @xmath351 means the gradient . let us see that @xmath42 is a diffeomorphism @xmath352-@xmath8-close to the identity . 1 . @xmath42 is injective : indeed , @xmath353 implies that @xmath354 . hence @xmath355 therefore @xmath356 where we have used that @xmath357 for all @xmath358 . taking into account that @xmath359 we obtain that @xmath360 . therefore @xmath361 here @xmath362 is a point in the segment joining @xmath363 with @xmath364 . let us choose @xmath190 so small that @xmath365 . it follows that @xmath366 by induction we have that for all @xmath23 @xmath367 therefore @xmath368 and @xmath42 is injective . 2 . @xmath42 is a diffeomorphism : indeed , we have @xmath369 + here @xmath370 , analogously @xmath371 only depends on @xmath372 , and @xmath373 is the transpose of @xmath95 . as @xmath374 we have that @xmath375 is small if @xmath352 is sufficiently small and the same is true with respect to @xmath376 and @xmath377 , taking into account that @xmath378 and @xmath379 . thus @xmath380 is invertible . items ( a ) and ( b ) above prove that @xmath42 is a diffeomorphism as @xmath8-close to the identity map as we wish and @xmath381 off a small ball @xmath382 . now consider @xmath383 . then @xmath92 is a small pertubation of @xmath10 . [ arcobeta ] @xmath85 is a flat @xmath92-homoclinic point and there is an arc @xmath384 with @xmath385 . indeed , since @xmath386 we have that @xmath387 and so @xmath85 is neither forward recurrent nor backward recurrent . this implies that we may choose the support , @xmath382 , of the perturbation in such a way that for @xmath388 , @xmath389 . hence if @xmath390 then for @xmath43 small we obtain that @xmath391 . but @xmath42 sends and arc @xmath304 passing through @xmath85 in @xmath392 onto an arc @xmath181 included in @xmath393 and passing through @xmath85 too . therefore @xmath394 sends the arc @xmath395 into @xmath304 which iterated sucessively by @xmath316 converges to @xmath7 . hence @xmath304 is an arc contained in _ both _ the local stable and unstable manifold of @xmath85 which is contained in @xmath396 . thus @xmath304 is an arc of flat intersection between @xmath397 and @xmath398 . this finishes both the proofs of claim [ arcobeta ] and claim [ claim30 ] . it is not difficult to see that this perturbation @xmath92 may be done in such a way that for @xmath267 great enough @xmath399 is tangent to the eigenvector corresponding to the less contracting eigenvalue and analogously @xmath400 is tangent to the eigenvector corresponding to the less expanding eigenvalue . all together finishes the proof proposition [ chato ] . the previous result gives a diffeomorphism @xmath92 , @xmath8-near @xmath10 , such that the intersection between @xmath398 and @xmath397 , in a local chart around @xmath85 such that @xmath401 , contains a segment @xmath402 . moreover , @xmath403 is tangent to the line corresponding to the less contracting eigenvector of @xmath404 and @xmath405 is tangent to the line corresponding to the less expanding eigenvector of @xmath404 . next we shall do a perturbation of @xmath92 , which will give a diffeomorphism @xmath16 such that @xmath16 coincides with @xmath92 outside a small neighborhood of @xmath304 , similar to those of ( * ? ? ? * lemma 5.1 , lemma 6.3 ) in order to create a sequence of small horseshoes @xmath406 associated to @xmath407 and @xmath408 . these horseshoes will have positive topological entropy and will be built in such a way that neither @xmath43 , nor @xmath409 will be constants of @xmath42-expansiveness for @xmath410 . therefore the diffeomorphism @xmath16 is not @xmath42-expansive , contradicting our hypothesis . to do so we proceed as follows : first , since we are working in a @xmath8-neighborhood of @xmath10 and @xmath411 diffeomorphisms are dense in @xmath89 we may assume that @xmath92 , the diffeomorphism obtained at proposition [ chato ] , is of class @xmath412 . we split the proof into two cases , according to the index of @xmath7 . let us assume first that @xmath7 is of index @xmath266 , i.e. : @xmath414 . this will simplify the techniques involved . we may assume , taking a large positive iterate by @xmath92 and possibly reducing @xmath36 , that @xmath304 , the segment of tangency , is contained in the local stable manifold of @xmath7 in a local chart which is a linearizing neighborhood @xmath415 of @xmath7 . let @xmath416\to { \mbox{$i\!\!r$}}$ ] be a @xmath52 bump function satisfying : 1 . @xmath417 , for @xmath418 $ ] . this implies that @xmath419 for all @xmath420 . 2 . @xmath421 for @xmath422 . @xmath423 for all @xmath424 $ ] , this implies that @xmath425 for all @xmath420 . 4 . @xmath426 for @xmath427 . @xmath428 for all @xmath429 $ ] , this implies that @xmath430 for all @xmath420 . next , consider @xmath431\to{\mbox{$i\!\!r$}}$ ] such that @xmath432\ , , \ ] ] @xmath433\ , , \ ] ] @xmath434\ , , \ ] ] and in general @xmath435\ , .\ ] ] put also @xmath436\ , .\ ] ] it is easy to see that @xmath437 is @xmath52 at @xmath438 $ ] . we may assume that for @xmath439 $ ] , @xmath440 and @xmath441 , for some @xmath442 . + hence for @xmath443 $ ] we have @xmath444 @xmath445 and @xmath446 therefore @xmath447 and @xmath448 when @xmath449 . setting @xmath450 we have that @xmath451 and @xmath452 is of class @xmath122 on @xmath453 $ ] . let @xmath454 be the unit vector in @xmath455 tangent to the expanding eigenvector of @xmath456 . recall we are assuming that @xmath457 . then @xmath454 is not contained in @xmath458 since @xmath459 is tangent to @xmath460 . recall that @xmath461 are the coordinates of @xmath304 in a local chart and that the interval @xmath462,0)$ ] is totally contained in @xmath304 . in the plane given by the origin @xmath330 ( identified with @xmath85 ) and the vectors @xmath311 and @xmath454 we consider the graph of the function @xmath463\to{\mbox{$i\!\!r$}}$ ] given by @xmath464\ , .\ ] ] observe that for @xmath465 $ ] , @xmath466 vanishes at @xmath467 and @xmath468 and it has a maximum value equals to @xmath469 at @xmath470 . now we extend @xmath471 to @xmath453 $ ] in the following way : @xmath472\ , , \ ] ] @xmath473\ , , \ ] ] and in general for @xmath474 : @xmath475\ , .\ ] ] for @xmath476 $ ] , @xmath471 vanishes only at @xmath477 and @xmath478 and it has a maximum value @xmath479 at @xmath480 . we complete the definition of @xmath471 in @xmath453 $ ] setting @xmath481 elsewhere . finally , let @xmath482 for all @xmath483 $ ] . then @xmath484 is @xmath52 in @xmath438 $ ] and @xmath122 in @xmath453 $ ] . put coordinates in the local chart @xmath485 and denote by @xmath486 a small @xmath487-dimensional disk around @xmath85 contained in a fundamental domain of @xmath488 whose coordinates in the local chart are @xmath489 . analogously denote by @xmath490 a small @xmath491-dimensional disk contained in @xmath398 around @xmath85 whose coordinates in the local chart are @xmath461 . note that @xmath486 is characterized by @xmath492 ; and @xmath490 is the arc @xmath304 contained in @xmath486 , parameterized by @xmath493 $ ] . the point @xmath85 is identified with @xmath494 . now , pick another @xmath52 bump function @xmath495 such that @xmath495 vanishes outside a @xmath86 neighborhood of @xmath304 , @xmath496 , and is equal to 1 in the @xmath497 neighborhood of @xmath304 . let @xmath345 be given by @xmath498 and @xmath381 outside @xmath499 where @xmath86 is such that the @xmath86-neighborhood of @xmath304 does not intersect @xmath500 . now , letting @xmath501 we get , by construction , that @xmath16 is a small perturbation of @xmath92 , and , as in proposition [ chato ] , it is not difficult to see that @xmath502 and @xmath503 furthermore , it is straightforward to show that @xmath504 and @xmath505 intersect transversely at the points @xmath506 and the absolute value of the tangent of the angles at the points @xmath507 we denote by @xmath508 the graph of @xmath484 in the plane @xmath509 . if we choose @xmath86 , @xmath510 with @xmath511 and @xmath36 small , we may obtain the perturbation @xmath512 to be @xmath8 small ( see @xcite ) . moreover , we can also assume that : 1 . @xmath513 on @xmath514 , where we recall that @xmath515 is a linearizing neighborhood of @xmath7 . 2 . @xmath516 and @xmath517 . here @xmath518 states for a suitable small positive number , 3 . @xmath519 . in particular @xmath520 . @xmath521 for all @xmath522 and there is @xmath523 such that @xmath524 for all @xmath525 , 5 . @xmath526 . we point out that item ( 5 ) above follows from the fact that we may reduce the value of @xmath36 , if it were necessary , in order to ensure it . [ inesperado ] there exists a sequence @xmath511 such that @xmath16 is not @xmath42-expansive . recall that we are working in a linearizing neighborhood @xmath90 of @xmath7 with respect to @xmath92 . set @xmath527 let @xmath528 and denote by @xmath529 the coordinates of the end points of @xmath530 corresponding respectively to @xmath531 and @xmath532 . in the same way we label all points in @xmath530 corresponding to the _ transverse _ intersections of @xmath304 with @xmath508 : @xmath533 corresponds to @xmath534 and @xmath535 corresponds to @xmath536 , @xmath537 corresponds to @xmath538 and @xmath539 corresponds to @xmath494 , @xmath540 corresponds to @xmath541 and @xmath542 corresponds to @xmath543 , and so on , labeling the image by @xmath544 of all the points of transverse intersection between @xmath304 and @xmath508 . take small arcs @xmath545 and @xmath546 contained in @xmath547 tangent to the the direction of the eigenvector corresponding to the weakest contracting eigenvalue of @xmath548 at the points @xmath533 and @xmath535 . multiply them by a @xmath549-dimensional disk @xmath550 of diameter @xmath551 . analogously take small arcs @xmath552 and @xmath553 tangent to the direction corresponding to the eigenvector of the expanding eigenvalue of @xmath548 at the points @xmath534 and @xmath535 and contained in @xmath554 . by the @xmath555-lemma , @xcite[lemma 7.1 ] , the forward orbits of @xmath552 and @xmath553 contain arcs arbitrarily @xmath8 near @xmath505 and the backward orbits of @xmath556 and @xmath557 contain @xmath487-dimensional disks arbitrarily @xmath8 near @xmath504 . by the way we have chosen @xmath545 and @xmath546 and the assumption about the eigenvalues of @xmath558 ( all positive real ) , we have that there is @xmath559 such that for @xmath560 in @xmath90 we have @xmath561 and @xmath562 . moreover , we may choose @xmath563 small such that @xmath564 and @xmath565 cut @xmath508 but is contained in the @xmath566 neighborhood of @xmath304 and therefore @xmath567 there . in the local coordinates we have chosen , we pick a thin rectangle @xmath568 with top and bottom given by @xmath569 and @xmath570 and bounded in its sides by segments parallel to the @xmath454-axis which is transverse to @xmath571 . increasing @xmath572 and reducing @xmath551 , @xmath545 and @xmath546 , if it were necessary , we may assume that @xmath573 is contained in the @xmath551-neighborhood of the graph of @xmath508 restricted to @xmath574 $ ] . set @xmath575 and let @xmath576 and consider @xmath577 then @xmath578 contains a horseshoe @xmath579 ( see @xcite ) and therefore @xmath580 has positive topological entropy . since this horseshoe is arbitrarily small we may assume that there is a periodic point @xmath581 such that @xmath582 see definition [ gammadeepsilon ] , where @xmath583 . moreover , the periodic point @xmath584 is homoclinically related to @xmath7 since by the @xmath555-lemma we have that positive iterates by @xmath585 give thin subrectangles crossing all of @xmath568 and hence the stable manifold of @xmath584 cuts @xmath586 and analogously positive iterates by @xmath587 gives subrectangles close to @xmath508 in the hausdorff metric and therefore the unstable manifold of @xmath584 cuts @xmath588 . [ paraentender ] there is @xmath589 such that for every @xmath172 it is associated a horseshoe @xmath590 with @xmath591 and @xmath592 . let us choose @xmath593 and construct @xmath594 . for this , pick @xmath595 such that @xmath569 and @xmath570 are at a distance greater than @xmath596 from @xmath489 . since @xmath597 for all @xmath598 we have that no part of the graph of @xmath484 for @xmath599 $ ] cuts @xmath568 . we found a new rectangle @xmath600 disjoint from @xmath568 contained in @xmath601 with @xmath602 applying again the @xmath555-lemma . increasing @xmath603 and reducing the corresponding values of @xmath604 , @xmath605 and @xmath606 , if it were necessary , we may assume that @xmath607 is contained in the @xmath604-neighborhood of the graph of @xmath508 restricted to @xmath608 $ ] . by construction when we iterate by @xmath16 the images of @xmath568 and @xmath600 can not intersect since in @xmath609 there are only one iterate of @xmath568 and one iterate of @xmath600 ( namely @xmath568 and @xmath600 ) . we then have for @xmath16 two disjoint small horseshoes , @xmath610 both with periodic points @xmath611 homoclinically related to @xmath7 ( the proof that @xmath612 is homoclinically related to @xmath7 is the same than that to @xmath584 ) . hence both @xmath579 and @xmath613 are included in @xmath410 . next we choose @xmath614 so that @xmath615 and @xmath616 are at a distance greater than @xmath617 from @xmath489 . for such @xmath617 , there is a horseshoe @xmath618 disjoint from @xmath619 and @xmath594 but still contained in @xmath410 . this construction follows the same steps as before : first find a thin rectangle @xmath620 cutting the graph of @xmath484 only for @xmath621 $ ] , @xmath622 , @xmath623 . then find an appropriate positive real number @xmath624 such that @xmath625 is contained in the @xmath626-neighborhood of the graph of @xmath508 restricted to @xmath627 $ ] . in this way we may pick the sequence @xmath172 such that for every @xmath148 it is associated a horseshoe @xmath590 satisfying ( 1 ) @xmath628 , ( 2 ) @xmath629 and ( 3 ) @xmath630 for all @xmath631 . this proves claim [ paraentender ] . since the topological entropy of @xmath590 is positive for all @xmath148 , and @xmath591 , we conclude that @xmath632 is not @xmath42-expansive , violating robustness of @xmath42-expansiveness . the proof of lemma [ inesperado ] is complete . then , the final conclusion is that hypothesis ( ad ) described in the begining of this section can not hold . in another words , we conclude that there exists @xmath178 such that for all homoclinic point @xmath72 there is @xmath633 such that @xmath634 following ( * ? ? ? * theorem a ) , it can be built a dominated splitting for the homoclinic points of @xmath65 as required , and then extend it by continuity to the whole @xmath65 using that the closure of the homoclinic points coincide with @xmath65 . thus in the case of @xmath7 a periodic point of index @xmath266 the proof of theorem [ prin1 ] follows . let us point out that even though we can assume that @xmath92 , the diffeomorphism with a segment of homoclinic tangencies , is @xmath52 , the bump function @xmath484 , used to perturb it , is just @xmath122 . hence it seems that a similar construction can be used to prove the stronger result that @xmath635 is not asymptotically @xmath42-expansive . recall , @xcite , that @xmath52- diffeomorphisms are asymptotically @xmath42-expansive so that a @xmath52 perturbation of a @xmath52 diffeomorphism does not disprove asymptotically @xmath42-expansiveness . for the general case of @xmath636 the proof is similar , the perturbation @xmath42 of @xmath92 given @xmath512 has to be adapted as we sketch below . let @xmath454 be the unit vector in @xmath455 tangent to the less expanding eigenvector of @xmath456 . then @xmath454 is not contained in @xmath458 see propositions [ buenangulo ] and [ chato ] . in a local chart around @xmath85 , @xmath461 represent the coordinates of the arc @xmath304 but the coordinates @xmath637 are such that @xmath158 is a @xmath638-dimensional vector , and @xmath639 a @xmath250 dimensional vector that we split as @xmath640 with @xmath641 one - dimensional . as in the codimension one case we have that @xmath462,0,0)$ ] is totally contained in @xmath304 . in the plane given by the origin @xmath330 ( identified with @xmath85 ) and the vectors @xmath311 corresponding to @xmath642 and @xmath454 corresponding to @xmath643 we , as above , consider the graph of the function @xmath463\to{\mbox{$i\!\!r$}}$ ] given by @xmath464\ , .\ ] ] now we extend @xmath471 to @xmath453 $ ] and define the @xmath122-function @xmath484 as in the codimension one case . put coordinates in the local chart @xmath644 and denote by @xmath486 a small @xmath249-dimensional disk around @xmath85 contained in a fundamental domain of @xmath488 whose coordinates in the local chart are @xmath645 . analogously denote by @xmath490 a small @xmath646-dimensional disk contained in @xmath398 around @xmath85 whose coordinates in the local chart are @xmath647 . note that @xmath486 is characterized by @xmath648 ; and @xmath490 contains the arc @xmath304 contained in @xmath486 , parameterized by @xmath493 $ ] . the point @xmath85 is identified with @xmath649 . now , pick a @xmath52 bump function @xmath495 such that @xmath495 vanishes outside a @xmath86 neighborhood of @xmath304 , @xmath496 , and is equal to 1 in the @xmath497 neighborhood of @xmath304 . let @xmath345 be given by @xmath650 and @xmath381 outside @xmath499 where @xmath86 is such that the @xmath86-neighborhood of @xmath304 does not intersect @xmath500 . now , letting @xmath501 we get , by construction , that @xmath16 is a small perturbation of @xmath92 , and , as in proposition [ chato ] , it is not difficult to see that @xmath502 and @xmath651 the remaining of the proof of theorem [ prin1 ] follows in a similar way to that of the codimension one case . in this section we prove both theorems [ prin2 ] and [ prin3 ] . for this , let us first remark that after @xcite , @xmath8-generically the finest dominated splitting has a very special form . thus , before we continue , let us first put @xmath10 in that context . 1 . @xmath10 is kupka - smale , ( i.e. : all periodic points are hyperbolic and their stable and unstable manifolds intersect transversally ) 2 . for any pair of saddles @xmath653 , either @xmath654 or @xmath655 . 3 . for any saddle @xmath7 of @xmath10 , @xmath65 depends continuously on @xmath656 . 4 . the periodic points of @xmath10 are dense in @xmath657 . the chain recurrent classes of @xmath10 form a partition of the chain recurrent set of @xmath10 . every chain recurrent class containing a periodic point @xmath7 is the homoclinic class associated to that point . taking into account ( * ? ? ? * corollary , 6.6.2 , theorem 6.6.8 ) , that guarantees that the homoclinic tangency can be associated to a saddle inside the homoclinic class , the next result is proved in ( * ? ? ? * corollary 3 ) : 1 . for any neighborhood @xmath90 of @xmath65 and any @xmath8-neighborhood @xmath9 of @xmath10 there is a diffeomorphism @xmath661 with a homoclinic tangency associated to a saddle of the homoclinic class @xmath662 , where @xmath94 is the continuation of @xmath7 . or 2 . there is a dominated splitting @xmath663 with @xmath664 and @xmath665 for all @xmath42 and @xmath666 . moreover , the sub - bundles @xmath667 are not hyperbolic . let @xmath668 be a homoclinic class robustly entropy expansive , i.e. , there is a neighbourhood @xmath669 such that @xmath670 , there is a continuation @xmath12 of @xmath6 for all @xmath661 and @xmath12 is @xmath42-expansive . by theorem [ prin1 ] we have a dominated splitting defined on @xmath119 . moreover , by ( * ? ? ? * theorem 6.6.8 ) , we have that in @xmath12 there is a finest dominated splitting which has the form @xmath671 with @xmath15 , @xmath16 and @xmath667 @xmath13-invariant sub - bundles , @xmath672 , and all @xmath667 one - dimensional , and @xmath673 otherwise , by the theorem of @xcite cited above , we may create with an arbitrarily small @xmath8-perturbation a tangency _ inside _ the perturbed homoclinic class . after that we repeat the arguments of [ herradurita ] contradicting @xmath42-expansiveness . theorem [ prin2 ] is proved . * proof of theorem [ prin3]*. by @xcite there is residual subset @xmath674 of @xmath675 such that , for every @xmath676 , any pair of homoclinic classes of f are either disjoint or coincide . for @xmath676 , by @xcite , the number of different homoclinic classes of @xmath10 is locally constant in @xmath674 . we split the proof into two cases : ( 1 ) this number is finite ( and in this case @xmath10 is _ tame _ ) or ( 2 ) there are infinitly many distinct homoclinic classes ( and in this case @xmath10 is _ wild_. * @xmath10 is tame * in this case @xmath6 is isolated . before we continue , recall that if @xmath677 and @xmath678 is the maximal invariant set of @xmath10 in @xmath243 , i.e. : @xmath679 , then set @xmath678 is robustly transitive if there is a @xmath8-neighbourhood @xmath9 of @xmath10 such that @xmath680 and @xmath681 is transitive for all @xmath661 ( i.e. : @xmath681 has a dense orbit ) . since @xmath6 is isolated it is a robustly transitive set maximal invariant in a neighbourhood @xmath682 and hence , according to @xcite[theorem d ] , the extremal sub - bundles @xmath15 and @xmath16 are contracting and expanding respectively . under the same hypothesis of the previous lemma either we have that in a @xmath8-robust way the index of periodic points in @xmath12 , @xmath92 near @xmath10 , are the same and equal to @xmath106 or there are @xmath92 arbitrarily @xmath8-close to @xmath10 such that in @xmath12 there are periodic points of different index . in the first case we have we follow the lines of the proof at ( * ? ? ? * section 6 ) . since @xmath6 is isolated by ( * ? ? ? * corollaire 1.13 ) or ( * ? ? ? * theorem a ) it is robustly isolated . let @xmath15 and @xmath26 be sub bundles such that @xmath684 is @xmath218-dominated , for all @xmath685 , with @xmath686 . we need to prove that @xmath687 as @xmath688 and @xmath689 as @xmath688 for any @xmath167 in order to prove that @xmath12 is hyperbolic . let us show only that @xmath687 as @xmath688 , the other one being similar . for this , it is enough to show that for any @xmath167 there exists @xmath690 such that @xmath691 thus there is a perturbation @xmath42 of @xmath92 such that @xmath42 has a non hyperbolic periodic point in @xmath697 . after a new perturbation we obtain periodic points @xmath698 and @xmath699 contained in a small neighborhood @xmath90 of @xmath697 and with different indeces . since @xmath6 is @xmath8-robustly isolated @xmath700 contradicting our assumption that in a @xmath8-robust way the index of periodic points in @xmath6 are the same and equal to @xmath106 . this finishes the proof of theorem [ prin3 ] in this case . in the second case , that is , there are @xmath92 arbitrarily @xmath8-close to @xmath10 such that in @xmath12 there are periodic points of different indeces , by @xcite , @xmath8-generically the diffeomorphism @xmath92 , and hence @xmath10 , can be @xmath8 approximated by diffeomorphisms exhibiting a heterodimensional cycle . next we show that in this case the eigenvalues of periodic points are robustly in @xmath701 . [ extremos2 ] let us assume that there is a periodic point @xmath702 with expanding complex eigenvalues such that @xmath703 . then there is an arbitrarily @xmath8-small perturbation of @xmath10 creating a tangency inside the perturbed homoclinic class @xmath12 . @xmath8 generically we may assume that there is a robust heterodimensional cycle between @xmath7 and @xmath232 and that @xmath704 contains a compact arc @xmath705 homeomorphic to @xmath706 $ ] , ( see @xcite ) . let us consider a disk @xmath707 of the same dimension @xmath708 of @xmath62 and contained in @xmath62 such that @xmath707 is homeomorphic to @xmath706\times [ -1,1]^{s-1}$ ] by a homeomorphism @xmath42 such that @xmath709\times \{0\}^{s-1}=l$ ] . iterating by @xmath710 this arc @xmath705 spiralizes around @xmath232 while @xmath707 stretches approaching @xmath711 . since @xmath712 there is a @xmath8 small perturbation of @xmath10 creating a tangency between @xmath713 and @xmath714 . under the hypothesis of the lemma we may create tangencies inside @xmath6 and by another @xmath8- perturbation an arbitrarily small horseshoe in the intersection between @xmath715 and @xmath716 contradicting @xmath42-expansiveness . thus corollary [ acima ] implies that the eigenvalues of periodic points in @xmath6 are real numbers in a robust way . by @xcite for @xmath8-generic diffeomorphisms the set of indices of the ( hyperbolic ) periodic points in a homoclinic class form an interval in @xmath717 . thus by @xcite[theorem 2.1 ] there are diffeomorphisms arbitrarily @xmath8-close to @xmath10 with @xmath8-robust heterodimensional cycles . [ solita ] if @xmath64 is @xmath8 robustly @xmath42-expansive and @xmath6 is an isolated homoclinic class then for a dense open subset @xmath718 either @xmath64 is hyperbolic and we have @xmath719 or there is a robust heterodimensional cycle in @xmath12 for @xmath92 arbitrarily close to @xmath10 . if we have that in a @xmath8-robust way the index of periodic points in @xmath12 are the same and equal to @xmath720 by lemma [ indiceconstante ] there is an open dense subset @xmath721 of @xmath722 such that @xmath12 is hyperbolic for @xmath723 . hence we are done . otherwise we have an open subset @xmath722 in any neighborhood @xmath724 of any diffeomorphism @xmath725 exhibiting a heterodimensional cycle , @xcite . this finishes the proof theorem [ solita ] , which in its turn gives the proof of theorem [ prin3 ] . * @xmath10 is wild * now let us assume that @xmath6 is not isolated . either there is a small @xmath8-perturbation @xmath92 of @xmath10 such that @xmath12 is isolated or @xmath6 is persistently not isolated , i.e. : @xmath12 is not isolated for any @xmath92 close to @xmath10 . in the first case we are done by theorem [ solita ] . [ crovisier ] there exists a dense @xmath726 subset of @xmath727 such that each homoclinic class @xmath728 has a dominated splitting @xmath729 which is partially hyperbolic and such that each central bundle @xmath730 has dimension 0 or 1 . universidade federal do rio de janeiro , = facultad de ingenieria , m. j. pacifico , j. l. vieitez , + instituto de matematica , instituto de matematica , + universidade federal do rio de janeiro , facultad de ingenieria , + c. p. 68.530 , cep 21.945 - 970 , universidad de la republica , + rio de janeiro , r. j. , brazil . cc30 , cp 11300 , + montevideo , uruguay + [email protected]_
we quantified and calibrated the metallicity and temperature sensitivities of colors derived from nine wide field camera 3 ( wfc3 ) filters aboard the hubble space telescope ( hst ) using dartmouth isochrones and kurucz atmospheres models . the theoretical isochrone colors were tested and calibrated against observations of five well studied galactic clusters : m92 , ngc 6752 , ngc 104 , ngc 5927 , and ngc 6791 , all of which have spectroscopically determined metallicities spanning . we found empirical corrections to the dartmouth isochrone grid for each of the following color magnitude diagrams ( cmd ) ( f555w f814w , f814w ) , ( f336w f555w , f814w ) , ( f390m f555w , f814w ) and ( f390w f555w , f814w ) . using the empirical corrections we tested the accuracy and spread of the photometric metallicities assigned from cmds and color - color diagrams ( which are necessary to break the age - metallicity degeneracy ) . testing three color - color diagrams [ ( f336w f555w),(f390m f555w),(f390w f555w ) , vs ( f555w f814w ) ] , we found the colors ( f390m f555w ) and ( f390w f555w ) , to be the best suited to measure photometric metallicities . the color ( f390w f555w ) requires much less integration time , but generally produces wider metallicity distributions , and , at very - low metallicity , the mdf from ( f390w f555w ) is wider than that from ( f390m f555w ) . using the calibrated isochrones we recovered the overall cluster metallicity to within 0.1 dex in when using cmds ( i.e. when the distance , reddening and ages are approximately known ) . the measured metallicity distribution function ( mdf ) from color - color diagrams show this method measures metallicities of stellar clusters of unknown age and metallicity with an accuracy of 0.2 - 0.5 dex using f336w f555w ,.15 - 0.25 dex using f390m f555w , and.2 - 0.4 dex with f390w f555w , with the larger uncertainty pertaining to the lowest metallicity range .
1310.4553
metallicity , age and mass are fundamental characteristics of a stellar population . metallicity distributions , in conjunction with chemical evolution models , provide evolutionary information about both enrichment and gas inflow and outflow in the form of galactic winds . while in recent years measuring large numbers of spectroscopic metallicities has become more feasible with multiobject spectrographs @xcite , it is still challenging to observe samples large enough to find rare objects or substructure in metallicity distribution functions , or to observe faint enough to build up large samples in nearby galaxies . photometric metallicities , though not as accurate as spectra , provide measurements for every star in the field , including those with fainter magnitudes than can be reached spectroscopically . the general technique to assign photometric metallicities relates color to metallicity . for instance , in a color magnitude diagram ( cmd ) fiducial ridgelines from several clusters with similar age , and a range of known metallicities can be interpolated to estimate the metallicity of an unmeasured cluster based upon the location of its ridgeline , provided that the cluster is also of similar age ( e.g. , saviane et al . 2000 ; da costa et al . 2000,2002 ) . fiducial ridgelines have been used to derive empirical relations between color and metallicity at a given absolute magnitude @xcite . alternatively , one can use theoretical isochrones , however , the isochrones need to be empirically calibrated to match observed sequences ( e.g. , brown et al . 2005 ; lianou et al . 2011 ) . an issue with deriving metallicities from color arises because a giant can be redder either because it is older , or because it is more metal rich . one method to break the age - metallicity degeneracy uses a color - color diagram , where one color is constructed using a metallicity sensitive filter , and the other color is constructed from a pair of filters that provide a temperature estimate with minimal dependence on metallicity . there is a long history in astronomy of using specifically designed filters to isolate stellar characteristics such as metallicity . proposed photometric indices as a means of stellar classification . the @xmath4 index estimates the stellar metallicity of horizontal branch , red giant branch and main sequence stars when used in conjunction with a temperature - sensitive index such as @xmath5 . the washington system was developed to measure photometric metallicities , temperatures and the amount of cn line blanketing for giant g and k stars @xcite . another consideration with photometric metallicities is that medium and broad band colors lose nearly all sensitivity at low metallicity . the caby system , developed in the 1990 s , modified the strmgren system , replacing the v filter with a filter centered on the ca h @xmath6 k lines . the caby system measures stellar metallicities 3 times more accurately , especially at low metallicity where the m@xmath7 index loses sensitivity @xcite . several filters on the hst / wfc3 were designed to provide information on the metallicities of resolved populations . our wfc3 calibration program ( 11729 , pi holtzman ) was designed to collect images of clusters with well established spectroscopic metallicities in order to map wfc3 colors to metallicity . another program designed to study the galactic bulge ( 11664 , pi brown ) imaged the same clusters with different filters , also as a calibration . this paper utilizes observations from both programs to present a broad range of calibrating filters and their metallicity sensitivities . observations were obtained of m92 , ngc 6752 , ngc 104 , ngc 5927 , and ngc 6791 . the five calibration clusters are well studied spectroscopically , and span a wide range of metallicity : @xmath8 @xmath1 . individual stars can be observed throughout the local group , opening up possibilities for studying populations within the milky way , as well as local group galaxies . our primary interest is in using these calibrations to measure metallicity distribution functions in several local group dwarf galaxies @xcite where we only have sufficient accuracy to measure giants , but the calibrations presented here may have broader applicability . as will be shown , photometric indices that measure metallicity also have a sensitivity to surface gravity , but for objects outside the milky way , it is trivial to separate giants from dwarfs based on their observed brightness . we structure the paper as follows : in section 2 we use model atmospheres and isochrones to explore the capacity of various filter combinations to measure metallicity ; in section 3 we present the observations of the stellar clusters . section 4 describes how we calibrate a set of isochrones to the observed sequences , and demonstrate how well these can be used to recover metallicities . in section 5 we summarize our conclusions . hst wfc3 observations were obtained in the following filters : f336w , f390 m , f390w , f395n , f410 m , f467 m , f547 m , f555w , f814w , f110w and f160w . information on the filter widths and system throughputs are listed in table [ throughput ] . the system responses for the uvis filters are shown in figure [ bandpasses ] , along with kurucz model stellar spectra of typical giant branch stars with log g = 1.5 , t = 4000 k and three different metallicities , @xmath9 , @xmath10 and @xmath11 . llrcl[b ] f336w & u , strmgren & 51.1 & 0.20 & 5.04 + f390w & c , washington & 89.6 & 0.25 & 4.47 + f390 m & ca ii continuum & 20.4 & 0.22 & 4.63 + f395n & ca ii , 3933/3968 & 8.5 & 0.22 & 4.58 + f410 m & v , strmgren & 17.2 & 0.27 & 4.42 + f467 m & b , strmgren & 20.1 & 0.28 & 3.79 + f547 m & y , strmgren & 65.0 & 0.26 & 3.12 + f555w & wfpc2 v & 156.2 & 0.28 & 3.16 + f814w & wfpc2 wide i & 153.6 & 0.23 & 1.83 + f110w & wide yj & 443.0 & 0.56 & 1.02 + f160w & wfc3 h & 268.3 & 0.56 & 0.63 many of these filters are comparable to those from well established systems . for example , f336w , f410 m , f467 m and f547 m are analogous to strmgren u , v , b and y , respectively . while designed for measuring the balmer decrement , the f336w and f410 m filters both cover spectral regions with many absorption features from metals . the other two strmgren filters , f467 m and f547 m , sample regions mostly clear of spectral features ; historically f467m f547 m ( i.e. b y ) has been used as a temperature indicator . filters f390 m and f395n cover the ca h @xmath6 k spectral features . the f395n filter is narrow ( 85 ) while the f390 m filter is broader , leading to a higher throughput , but also including a cn feature at @xmath12 @xmath13 3885 . the f390w filter has a wide bandpass ( 896 ) , and is similar to the ground - based washington c filter . the washington c filter was designed to evaluate the total effects of line blanketing by cn ( bands at 3595 , 3883 and 4215 ) as well as the ch molecular transition at 4304 , commonly known as the g band @xcite . f555w and f814w are wide band filters designed to cover the same spectral regions as the wfpc2 and acs filters of the same name ; they are similar to , but broader than , johnson - cousins v and i. the f555w and f814w filters measure mostly continuum ; one notable exception is the mgh feature at @xmath2 5100 @xmath14 in the f555w bandpass . as part of the observation campaign images in two near ir filters , f110w and f160w , were obtained in an effort to explore reddening free indices @xcite . however , we found that the photometric uncertainties from a two color reddening free index seemed to be too large to significantly add to our present analysis . stellar colors are a function of gravity , metallicity and effective temperature . for a given star , increasing the metallicity lowers the effective temperature and enhances line blanketing effects at a given mass , both of which cause redder color . for populations of comparable age the color is directly related to the metallicity . this relation has been used extensively in older stellar populations , e.g. , globular clusters , to determine metallicities . for an old population , we calculate the metallicity sensitivity of cmd colors using dartmouth stellar isochrones @xcite . these cover the metallicity range @xmath15 @xmath16 . since we wish to understand sensitivity to metallicity at lower metallicities , we extend these down to @xmath17 by assuming that stars with @xmath18 have the same effective temperatures and luminosities as those with @xmath19 , while adopting the colors from a grid of model atmospheres @xcite that extends down to @xmath17 . in table [ cmdsens ] , we report metallicity sensitivity for rgb stars , in units of dex of per 0.01 mag of color change , for a range of different color choices , including most of the wideband uvis wfc3 filters . in these units , small numbers represent higher sensitivity to metallicity . these sensitivities were computed for stars at m@xmath20 ; for more luminous giants , the sensitivity is better , while for fainter ones , it is worse . sensitivities are reported for several different ranges in metallicity , demonstrating the smaller color sensitivity at lower metallicity . generally , sensitivity increases as the wavelength separation of the filters increases . however , each filter has different photometric precision for a fixed exposure time . the second to last column in table [ cmdsens ] gives the relative color errors ( normalized to @xmath21 ) , estimated using the exposure time calculator ( etc ) on the wfc3 online data handbook for a k5 iii giant and a fixed exposure time . the last column of table [ cmdsens ] lists the sensitivity , over the metallicity range @xmath22 @xmath23 , scaled by the relative photometric error . the implication is that ideal filter choices rely on both the color sensitivity to metallicity and the precision to which the color can be measured . the most metallicity sensitive colors will not be the optimum choice when observing time and errors are considered . however , if you use a color that only changes minimally with metallicity , no amount of increased photometric accuracy will improve the metallicity determination . the last column of table [ cmdsens ] suggests that f475w f814w is the optimal color for maximum metallicity sensitivity , and this has been adopted by many studies ( e.g. , gallart 2008 ) ; although this choice does depend to some extent on the color of the target stars . other commonly used metallicity sensitive colors are f390w f814w and f555w f814w . as pointed out in the last paragraph there is a trade off between metallicity sensitivity and photometric accuracy ( from a reasonable amount of integration time ) . the sensitivities in table [ cmdsens ] show that these wide band colors have simultaneously greater throughput and less metallicity sensitivity than the medium and narrow band filters listed in table [ cmdsens ] . additionaly , at very - low metallicity , these broad - band colors have little sensitivity ( @xmath24 and @xmath25 dex , respectively , per 0.01 mag of color change ) , and since it is challenging to reduce photometric errors below 0.01 mag , this leads to a fundamental limit on the accuracy of derived metallicities . for some colors , at low metallicity the color change from metallicity becomes smaller than the typical photometric accuracy of 0.01 mag . in table [ cmdsens ] we report this as @xmath26 1 dex of /0.01 mag color change . the narrower filters , such as f395n and f390 m , retain sufficient sensitivity ( @xmath27 dex per 0.01 mag of color change ) to provide useful metallicity estimates even at very - low metallicity . lccccccc f336w f814w & 0.270 & 0.100 & 0.021 & 0.007 & 0.005 & 44.2 & 0.31 + f390w f814w & 0.667 & 0.185 & 0.036 & 0.012 & 0.009 & 4.5 & 0.05 + f390m f814w & 0.454 & 0.133 & 0.028 & 0.008 & 0.007 & 31.3 & 0.26 + f395n f814w & 0.333 & 0.094 & 0.027 & 0.010 & 0.007 & 69.7 & 0.72 + f410m f814w & @xmath261 & 0.476 & 0.060 & 0.014 & 0.009 & 11.6 & 0.17 + f438w f814w & @xmath261 & 0.400 & 0.053 & 0.020 & 0.013 & 2.9 & 0.06 + f467w f814w & @xmath261 & @xmath261 & 0.149 & 0.038 & 0.017 & 3.6 & 0.13 + f475w f814w & @xmath261 & @xmath261 & 0.097 & 0.031 & 0.017 & 1.3 & 0.04 + f547m f814w & @xmath261 & @xmath261 & 0.198 & 0.050 & 0.023 & 1.4 & 0.07 + f555w f814w & @xmath261 & @xmath261 & 0.182 & 0.048 & 0.022 & 1.0 & 0.05 + f606w f814w & @xmath261 & @xmath261 & 0.284 & 0.075 & 0.027 & 0.79 & 0.06 + f625w f814w & @xmath261 & @xmath261 & 0.402 & 0.114 & 0.033 & 0.90 & 0.10 + f775w f814w & @xmath261 & @xmath261 & @xmath261 & 0.741 & 0.270 & 0.94 & 0.70 for a population of mixed age , the color - metallicity relation breaks down because younger giants , which are more massive , are hotter than older giants of the same metallicity . the color changes due to age and metallicity are demonstrated in figure [ agez ] , for two different ages ( 12.5 and 4 gyr , solid and dashed lines , respectively ) , and for a range of metallicities ( different colors ) . the effect is quantitatively shown in figure [ age_col ] , where color as a function of metallicity is plotted for several ages ( 2 , 7 and 12 gyr ) for a giant branch star with a @xmath28 = @xmath29 ( with comparable results along the entire giant branch ) . at higher metallicity this leads to an uncertainty in derived metallicity of a few tenths of a dex , but the uncertainty can be significantly larger at lower metallicity . the uncertainties in metallicity also become greater with younger populations ; as figure [ age_col ] shows , the color difference between 2 and 7 gyr is almost double that between 7 and 12 gyr . the key to breaking the age - metallicity degeneracy is separating the color effects of metallicity and temperature . two color indices have the potential to break the degeneracy because one color can measure a metallicity dependent feature in the spectrum , while the other can control for the temperature . the more the two color s sensitivities to metallicity and temperature differ , the more effective the color combination will be at measureing metallicity . we examine the temperature and metallicity sensitivity for all the filter combinations by using kurucz stellar atmosphere models of metallicity ranging from @xmath30 @xmath31 . we integrate the wfc3 transmission curve for each filter over the synthetic spectra with parameters of typical giant branch stars ( log g = 2.5 and t@xmath32 = 4500 k ) . we compute the color change with temperature ( @xmath33 color/@xmath33 t@xmath34 ) and metallicity ( @xmath33 color / @xmath33 ) , at a range of metallicities . the ratio of the two gives the relative sensitivity to temperature and metallicity at equal color difference , where small values of @xmath33 t@xmath34/@xmath33 indicate a smaller dependence on metallicity than temperature . the relative sensitivity of temperature to metallicity ( @xmath33 t@xmath34/@xmath33 ) for all colors is plotted in figure [ dtdz ] . as expected , the colors whose bandpasses contain fewer metal - features ( e.g f555w f814w and f467m f547 m ) are the ones that are least sensitive to metallicity . figure [ dtdz ] shows that the f467m f547 m and f555w f814w colors have comparably smaller sensitivities to metallicity . however , the relative color error for f467m f547 m is over 3.5 times larger than f555w f814w for a fixed exposure time . f814w stands out as the optimal temperature index , considering minimal metallicity sensitivity and photometric accuracy . while f555w f814w certainly changes with metallicity ( cf . its use as a metallicity indicator in clusters at fixed age as discussed in the previous section ) it has the largest relative sensitivity of temperature to metallicity of all the filters compared in this section ( see figure [ dtdz ] ) . figure [ agez2 ] demonstrates the age independence and metallicity separation of a color - color diagram . the plot includes isochrones of two different ages , 12.5 and 4 gyr , as solid and dashed lines respectively , for a range of metallicities in steps of 0.5 dex . the solid and dashed lines closely follow each other throughout the color - color diagram . the bifurcation of the isochrones at cooler temperatures ( especially seen at higher metallicity with the dashed and solid purple lines ) is due to dwarf and giant stars colors showing increased gravity sensitivity at higher metallicity . the gravity sensitivity causes the metallicity - sensitive color of dwarfs to be bluer than giants at the same value of the temperature - sensitive color , however , for targets at a common distance , apparent magnitude can be used to separate the evolutionary stage . based upon the color sensitivities shown in figure [ dtdz ] , we selected the more metallicity - sensitive colors to use with the temperature - sensitive color ( f555w f814w ) in color - color diagrams . figure [ sensitivity ] shows giant branch isochrones in color - color diagrams for five metallicity sensitive colors : f336w f555w , f390m f555w , f390w f555w , f395n f555w and f410m f555w . the effectiveness of these color combinations to determine metallicity are quantified in table [ spans ] by the color separation due to metallicity predicted from stellar isochrones in color - color diagrams , using units of dex of /0.01 mag of color change . while all the filter sets have similar sensitivity around solar metallicity ( @xmath35 @xmath36 ) this breaks down at lower metallicities . lcccccccc[h ] f336w f555w & 0.9 & 0.427 & 0.143 & 0.047 & 0.023 & 0.026 & 9.8 & 0.24 + & 1.0 & 0.290 & 0.107 & 0.045 & 0.023 & 0.024 & 9.8 & 0.23 + & 1.1 & 0.236 & 0.087 & 0.051 & 0.023 & 0.024 & 9.8 & 0.23 + f390w f555w & 0.9 & 0.943 & 0.287 & 0.098 & 0.041 & 0.049 & 1.0 & 0.04 + & 1.0 & 0.704 & 0.199 & 0.090 & 0.038 & 0.046 & 1.0 & 0.04 + & 1.1 & 0.541 & 0.149 & 0.084 & 0.038 & 0.043 & 1.0 & 0.04 + f390m f555w & 0.9 & 0.667 & 0.206 & 0.074 & 0.027 & 0.036 & 6.9 & 0.19 + & 1.0 & 0.493 & 0.146 & 0.062 & 0.024 & 0.035 & 6.9 & 0.16 + & 1.1 & 0.370 & 0.110 & 0.052 & 0.023 & 0.034 & 6.9 & 0.16 + f395n f555w & 0.9 & 0.515 & 0.151 & 0.066 & 0.036 & 0.049 & 15.4 & 0.56 + & 1.0 & 0.369 & 0.104 & 0.057 & 0.037 & 0.047 & 15.4 & 0.56 + & 1.1 & 0.277 & 0.079 & 0.052 & 0.038 & 0.047 & 15.4 & 0.60 + f410m f555w & 0.9 & @xmath26 1 & 0.820 & 0.254 & 0.090 & 0.063 & 2.6 & 0.23 + & 1.0 & @xmath26 1 & 0.510 & 0.288 & 0.070 & 0.051 & 2.6 & 0.16 + & 1.1 & @xmath26 1 & 0.364 & 0.350 & 0.063 & 0.044 & 2.6 & 0.16 weighting the metallicity sensitivity by the relative uncertainty ( last column of table [ spans ] ) shows that f390w f555w is the most sensitive for a fixed exposure time . below @xmath37 the color separation for f395n f555w , f390m f555w and f336w f555w are the most sensitive to metallicity . although the f395n f555w color retains sensitivity it also requires 2 to 3 times more exposure time to get comparable accuracy , making it prohibitive to use . at extremely low metallicity the most sensitive color is f395n f555w , with sensitivity decreasing respectively for the colors : f336w f555w , f390m f555w and f390w f555w . based upon the metallicity sensitivity and photometric efficiency we find that the most promising metallicity indicating filters are the f390w , f390 m and f336w filters . in the remainder of the paper we will focus our analysis on these filters . reddening adds uncertainty to any photometric metallicity derivation , especially when the uncertainty in reddening is large , or if there is differential reddening throughout the field . in a color - color plot uncorrected reddening will be confused with a change in metallicity if the reddening vector is in the direction of the metallicity separation . for the majority of the color - color diagrams the reddening vectors are predominantly in the same direction as the isochrones , mitigating the influence of reddening . however , for ( f336w f555w , f555w f814w ) the reddening vector is @xmath2 40@xmath38 shallower than the isochrones , which increases the uncertainty in derived metallicities when there is differential reddening or if the uncertainty in the reddening is large . reddening vectors are indicated in figure [ sensitivity ] for e(b v ) = 0.1 . the vector is defined by the reddening of the colors along the x and y axes . for a given color , the reddening e(filter@xmath7filter@xmath39 ) = @xmath40 , where a@xmath41 = r@xmath41 e(b v ) , and the r@xmath42 values are listed in table [ throughput ] . to get an estimate of the effect of reddening on metallicities we consider an uncertainty of @xmath43 = 0.02 , as might be appropriate in a low reddening target . at shorter wavelengths , where all of our metallicity sensitive filters are located , the uncertainty in reddening will have an increased effect due to the uv extinction law increasing as wavelength decreases . in this case the @xmath44 adds a systematic color uncertainty of @xmath20.1 mag to all of the metallicity sensitive colors . however , the color change from reddening will be along the reddening vector , and because the angles between the isochrones and reddening vectors vary ( e.g. figure [ sensitivity ] ) and the colors have different sensitivity to metallicity ( e.g@xmath45 table [ spans ] ) , this leads to metallicity uncertainties ranging from a tenth to a few tenths of a dex . the resulting systematic uncertainties in are reported in table [ error ] . as expected , larger uncertainties are found at lower metallicities . historically , reddening - free indices have been proposed as a means to correct colors for extinction . a reddening - free index , q , is defined as : @xmath46 where e@xmath47 is the color excess for the color @xmath47 @xcite . however , making our metallicity indicator reddening - free still leaves the problem that the temperature indicator is affected by reddening . to completely remove reddening effects from the color - color diagram , both indices need to be reddening - free . @xcite suggest using two reddening - free indices ( 5 total filters , including two in the ir ) to predict metallicity . we looked into using these additional filters and found that with our low reddening targets the additional filters do not significantly add to the metallicity resolution . lccc[hb ] f336w f555w & 0.30 & 0.14 & 0.15 + f390m f555w & 0.20 & 0.09 & 0.13 + f390w f555w & 0.23 & 0.11 & 0.12 + f395n f555w & 0.10 & 0.07 & 0.09 + f410m f555w & 0.33 & 0.12 & 0.08 while is commonly used as a proxy for the overall stellar metallicity , elements do not vary in lockstep from star to star . variations in abundance ratios can alter spectral features within a bandpass , changing the stellar color and causing the overall photometric metallicity to be misinterpreted if fixed abundance ratios are assumed . a common abundance variation is @xmath48 element enhancement ( e.g. ) , often present at low metallicity in globular clusters ( see and references there in ) . the cmd of an @xmath48-enhanced population resembles a cmd of higher with solar abundance ratios . figure [ cmdafeh ] shows how @xmath48 enhancements affect different colors at low and high metallicity . different combinations of and [ @xmath48/fe ] can give the same colors . using dartmouth isochrones , we compute the relative color change from @xmath48 abundances ( @xmath33 color/ @xmath33 [ @xmath48/fe ] ) and metallicity ( @xmath33 color/ @xmath33 ) , to find the relative sensitivity of various colors to and @xmath48 enhancement . the ratio of the two as a function of metallicity is presented in table [ da_dz ] ; unsurprisingly the colors with the most @xmath48 sensitivity are the the two colors ( f395n f555w and f390m f555w ) whose bandpasses are dominated by the h @xmath6 k lines . metallicities assigned assuming solar abundance ratios can be considered as a metallicity indicator that is a combination of the actual and [ @xmath48/fe ] , following the relation : @xmath49}_{phot}=\textrm{[fe / h]}+ \left ( \frac{\delta \textrm{[fe / h ] } } { \delta[\alpha/\textrm{fe } ] } \right ) [ \alpha/\textrm{fe}]\ ] ] where the amount of relative color change due to @xmath48 enhancements ( listed in table [ da_dz ] ) can be used to get the relation . the differing sensitivity of ( @xmath33/@xmath33[@xmath48/fe ] ) in different colors implies that @xmath48 abundance can be separated with photometry using similar techniques described in section 2.3 . however , additional colors with differing sensitivities would be necessary to isolate the @xmath48 abundance , such as f395n f555w and f410m f555w or f390m f555w and f390w f555w . we should note that the majority of globular clusters show large star - to - star variations in light elements ( li to si ) ( for a review see gratton et al@xmath45 2012 , and references therein ) . additionally there are cluster - to - cluster abundance variations , which also have the potential to affect the stellar colors . @xcite found that abundance enhancements of mg , si , and to some extent ca , will change the color and thus location of the giant branch , while variations in oxygen will affect the height of the subgiant branch and the luminosity of the ms turnoff . of particular interest in this study , c and n enhancements can affect the stellar colors measured with f390 m and f390w , since a few cn and ch bands fall in the same spectral region covered by these filter bandpasses ; see section 2.1 for details . since we can not explicitily measure any of these abundance variation in this study due to the limited resolution of photometric metallicities , any such variations could affect the stellar colors and our inferred metallicities . lccc[hb ] f336w f555w & 0.27 & 0.39 & 0.64 + f390w f555w & 0.22 & 0.19 & 0.09 + f390m f555w & 0.65 & 0.37 & 0.34 + f395n f555w & 0.88 & 0.84 & 0.79 + f410m f555w & 0.09 & 0.17 & -0.09 wfc3 images were obtained of five clusters during cycle 17 as part of the calibration program 11729 , ( pi holtzman ) . additional images of the five clusters come from the calibration portion of the program 11664 ( pi brown ) . the five clusters were chosen because they are well studied and span a range of ; 5 of the 6 clusters are discussed by @xcite for acs calibration . lccccc[h ] f336w & 850 , 30 & 1000 , 30 & 1160 , 30 & 950 , 30 & 800 , 30 + f390 m & 1400 , 50 & 1400 , 50 & 1400 , 50 & 1400 , 50 & 1400 , 50 + f390w & 2290 , 10 & 2460 , 10 & 2596 , 10 & 2376 , 10 & 2250 , 10 + f395n & 1930 , 90 & 2100 , 90 & 2240 , 90 & 1015 , 90 & 1850 , 90 + f410 m & 1530 , 40 & 1600 , 40 & 1600 , 40 & 1600 , 40 & 1490 , 40 + f467 m & 700 , 40 & 800 , 40 & 900 , 40 & 730 , 40 & 700 , 40 + f547 m & 445 & 445 & 445 & 445 & 445 + f555w & 1360.5 & 1360.5 & 1360.48 & 1381 & 1360.5 + f814w & 860.5 & 1020.5 & 1160.48 & 961 & 810.5 + f160w & 1245 , 8.3 & 1245 , 12.5 & 1245 , 16.7 & 1245 , 12.5 & 1245 , 4.2 + f110w & 1245 , 8.3 & 1245 , 12.5 & 1245 , 16.7 & 1245 , 12.5 & 1245 , 4.2 [ exposure ] we present data for four globular clusters , m92 ( ngc 6341 ) , ngc 6752 , ngc 104 ( 47 tuc ) , ngc 5927 , and one open cluster ngc 6791 . exposure times are listed in table [ exposure ] , and include short and long exposures to increase the dynamic range of the photometry . reduced images were obtained from the hst archive using its on - the - fly processing . a deep image was created by summing all of the exposures in f467 m , and stars were identified on this image . for each individual frame , a pixel area correction was applied to account for the modification in fluxes by flat fielding , and an astrometric solution was derived relative to the reference frame . aperture photometry was done on all of the stars using aperture radii of 0.12 , 0.25 , 0.375 , and 0.5 arcsec . using this initial aperture photometry and model psfs calculated using tinytim , photometry was redone on each individual star after subtracting off all neighbors ; this process was iterated twice . to adopt isochrones for the clusters we search the literature for well established cluster parameters . the distance modulus and the reddening are needed to get the absolute magnitudes of the cluster stars . additionally we use the age , metallicity , and @xmath48 enhancement to select isochrones for our study . we have compiled a sampling of the reported cluster parameters , and list the literature values adopted in this study in table [ obsdat ] . * m92 * is the only very metal - poor cluster in our sample . spectroscopic metallicity measurements include those by @xcite , @xmath50 , , @xmath51 , and @xcite,@xmath52 @xmath53 and @xmath52 @xmath54 .. we average the spectroscopic metallicities , and adopt @xmath55 . isochrone studies by @xcite and @xcite both found @xmath48 enhancements of [ @xmath48/fe ] @xmath56 , however our isochrone grid is in steps of 0.2 , so we adopt [ @xmath48/fe]@xmath57 . we adopt the age , and reddening found by @xcite , 13.5 gyr and e(b v ) @xmath58 . after adopting these isochrone parameters we adjust the distance modulus to be fully consistent with our isochrone models , by assuming ( m m)@xmath59 , which is consistent with the values used by @xcite . * ngc 6752*. measured spectra of seven stars for an overall metallicity @xmath60 , and [ @xmath48/fe ] @xmath61 @xmath620.01 . measured @xmath63 from high resolution spectra of 14 stars , they also measured multiple @xmath48 elments for an average [ @xmath48/fe ] @xmath64 . in a previous study , found @xmath65 . other spectroscopic metallicity measurements include those by @xcite , @xmath66 , , @xmath67 , @xcite , @xmath52 @xmath68 . we average the spectroscopic and adopt a @xmath69 , which is slightly lower than the average but is consistent with the larger [ @xmath48/fe ] adopted to fit our isochrone grid , [ @xmath48/fe]@xmath57 . we assume a reddening of e(b v ) @xmath70 as measured by from 118 stars . we adopt an age of 12.5 gyr , reported by @xcite , and also by @xcite in a study of comparative ages of globular clusters . from these adopted isochrone prameters we adjust our distance modulus to ( m m)@xmath71 , which is consistent with the distance to ngc 6752 measured by @xcite using the white dwarf cooling sequence ; ( m m)@xmath72= 13.17 . * 47 tuc*. spectroscopic measurement include those by @xcite , @xmath73 , who found @xmath74 , @xcite , @xmath52 @xmath75 , @xcite found @xmath76 based on seven rgb stars , and measured @xmath77 from the medium resolution spectra of 147 stars and @xmath78 from the high resolution spectra of 11 stars , and measured an [ @xmath48/fe]@xmath79 . we average the adopting values of @xmath80 and [ @xmath48/fe ] = 0.4 . we use the e(b v ) @xmath81 reported by . @xcite found an age of 12 gyr , @xcite derived an age of 12.9 gyr using diffusive models , which we average for an adopted age of 12.5 gyr . based upon the adopted isochrone parameters we adjust the distance modulus to ( m m)@xmath82 which is smaller than the values measured by @xcite and @xcite who both measured the distance modulus ( 13.36 and 13.27 , respectively ) to 47 tuc using white dwarf cooling models . * ngc 5927 * is more complicated to study photometrically due to the high differential reddening within the cluster ( @xmath33 e(v - i)=0.27 measured by and @xmath83(e(b v)=0.169 measured by bonatto et al . @xcite found @xmath84 ; @xcite measured @xmath85 , while @xcite found @xmath86 . we adopt a of @xmath87 and [ @xmath48/fe ] of + 0.2 . @xcite found the reddening to be e(b @xcite did a study on the relative ages of gc s and found the age to be @xmath13 10 gyr , while the relative gc age study by @xcite determined an age of @xmath2 11.3 gyr , we assume an age of 11.5 gyr . based upon the adopted isochrone parameters we adjust the distance modulus to ( m m)@xmath88 . * ngc 6791 * , unlike the rest of our calibrating clusters , is an open cluster ( although recent work by @xcite suggests that ngc 6791 might be the remains of a stripped globular cluster ) . determined a metallicity of @xmath89 . @xcite used infrared spectroscopy of 6 stars to measure a = @xmath900.02 and solar @xmath48 abundance ratio . @xcite found a @xmath91 . we average the spectroscopic metallicities and round to the nearest isochrone grid spacing and adopt @xmath92 and [ @xmath48/fe]=0.0 . we assume the reddening found by @xcite , e(b v ) = 0.10@xmath93 . we use the age found by both @xcite and @xcite of 8 gyr . found an age of @xmath28.3 gyr , and @xcite also reported a consistent age of 8 gyr between the white dwarf cooling sequence and the msto age . from this we find a distance modulus of ( m m)@xmath94 , which is consisten with who used an eclipsing binary to determine ( m m)@xmath72 = 13.46@xmath62 0.1 , and with @xcite , who found ( m m)@xmath95 . as previously stated the adopted and [ @xmath48/fe ] are not perfectly matched to the literature values but are rounded to the spacing within the isochrone grid ( in steps of 0.05 of and 0.2 of [ @xmath48/fe ] ) . when adjusting the metallicity to our grid we note that the adjustment for a change in [ @xmath48/fe ] varied for each color according to equation ( 2 ) and table [ da_dz ] ; therefore we made an average metallicity adjustment that best fit all colors simultaneously . llrlllrr[ht ] m 92 & 17 17 07.05 & + 43 07 58.2 & 0.023 & 14.60 & 2.50 , 2.14 ( 2.30 ) & + 0.30 ( + 0.4 ) & 13.5 + ngc 6752 & 19 10 54.86 & 59 59 11.2 & 0.046 & 13.20 & 1.54 , 1.43 ( 1.45 ) & + 0.27 ( + 0.4 ) & 12.5 + ngc 104 & 00 24 15.26 & 72 05 47.9 & 0.024 & 13.20 & 0.77 , 0.60 ( 0.70 ) & + 0.44 ( + 0.4 ) & 12.5 + ngc 5927 & 15 28 00.20 & 50 40 26.2 & 0.45 & 15.78 & 0.67 , 0.30 ( 0.40 ) & . . ( + 0.2 ) & 11.5 + ngc 6791 & 19 20 53.00 & + 37 46 30.0 & 0.10 & 13.45 & + 0.35 , 0.40 ( @xmath96 ) & 0.0 ( 0.0 ) & 8.0 from the reported reddening , and distance modulus we calculate the absolute magnitudes for each filter . all magnitudes reported are in the vegamag system with zeropoints taken from the wfc3 handbook . the absolute magnitudes for each filter were calculated using : m@xmath97 m@xmath98 - ( ( m m)@xmath99 - a@xmath99 ) - a@xmath100 . the distance modulus , ( m m)@xmath99 , as reported in table [ obsdat ] , was corrected for the v band extinction , @xmath101 = 3.1e(b v ) , and addtionally corrected for the extinction within the given filter a@xmath102 = r@xmath102e(b v ) . the ratio of total to selective extinction , r@xmath102 as listed in table [ throughput ] , was calculated by taking the integral over the stellar atmosphere sed convolved with the filter transmission curve , divided by the integral over the same atmosphere and filter convolved with the galactic extinction curve . we used an atmosphere with @xmath103 , temperature of 5000 , and log g = 2.0 . it is well known that theoretical isochrone models do not perfectly match observed colors due to a combination of uncertainties from stellar evolution models , model atmospheres , and instrumental systematics . when comparing the isochrone of the mean spectroscopic value to the cmds the models often did not match the shape of the cluster ridgelines . however , it should be noted that uncertainties in distance , age , reddening and composition complicate the fitting process . the cluster cmds are shown in figures [ cmds ] - [ medcmd ] , where dashed lines represent the literature valued isochrones adopted for each cluster as reported in table [ obsdat ] . in order to improve the metallicity determinations we derived an empirical correction to the isochrones using the five cluster cmds as benchmarks . for small steps in magnitude along the giant branch , we found the distance between the theoretical isochrone colors and the cluster ridgeline , then squared and summed the distances . using the canonical metallicities we found the empirical corrections with the smallest summed distance between the corrected isochrone and the cluster . these corrections lead to consistent cluster metallicities across all cmds . all isochrone corrections were based on color , gravity and metallicity adjustments . the various filter combinations require different functional forms of the generic equation : @xmath104}}){\rm(color)}_{\rm{obs } } \\ + \gamma(\text{\textrm{[fe / h]}}+\delta)^2 + \epsilon \nonumber \\ + ( \zeta + \eta \textrm{[fe / h]})(\theta - \text{log g } ) \nonumber \end{aligned}\ ] ] equation ( 3 ) was modifed by inspection of the various cmds . the ( f555w f814w , f814w ) cmd in figure [ cmds ] shows the uncorrected literature valued isochrones falling redward of the cluster ridgelines to varying degrees depending on the magnitude and metallicity of the isochrone . the most metal - poor cluster was especially discrepant , with larger deviations along the lower giant branch than the upper . for this cmd the first order color correction includes a dependence on metallicity , which allows for larger corrections at lower metallicity . the first order color dependent term shifts and changes the shape of the isochrone , while the second order metallicity dependent term adds the same shift to a given isochrone without changing the shape . the additional gravity dependent term corrects the bend of the giant branch ( where log g @xmath105 3.4 ) , with more bend required as metallicity decreases . in figure [ ucmd ] the ( f336w f555w , f814w ) cmd shows the uncorrected isochrones redward of the cluster ridgelines , except for m92 where the isochrone along the giant branch falls blueward . this cmds shows the largest difference between the clusters and the models . the bend of the isochrone along the giant branch ( where log g @xmath105 3.4 ) is corrected with a gravity and metallicity dependent term , with increasing tilt as metallicity decreases . in this cmd the corrected isochrone for 47 tuc ( ngc 104 ) falls @xmath20.03 mag redward of the cluster ridgeline , while all of the other clusters are within @xmath20.01 mag of their ridgelines . we believe this slight offset to be an artifact of rounding the average spectroscopic metallicity to fit the grid spacing . this color offset is smaller than the color difference between a isochrones of [ fe / h ] @xmath106 and@xmath107 , therefore we do not force and overcorrection which would potentially lead to distorted mdfs . in figure [ widecmd ] the uncorrected isochrones in the ( f390w f555w , f814w ) cmds are the least discrepant of the cmds we examined . however a few corrections are still needed to improve the fits , including a metallicity dependent gravity term , with smaller corrections required for lower metallicities . lllll[hb ] ( f336w f555w , f814w ) & m92 & -2.28 & 0.114 & 937 + . . . . . . . . . . . . . . . & n6752 & -1.45 & 0.073 & 481 + . . & n104 & -0.74 & 0.059 & 2096 + . . . & n5927 & -0.37 & 0.067 & 1518 + . . & n6791 & 0.39 & 0.079 & 320 + ( f390m f555w , f814w ) & m92 & -2.32 & 0.089 & 899 + . . & n6752 & -1.43 & 0.087 & 468 + . . & n104 & -0.68 & 0.066 & 2071 + . . & n5927 & -0.42 & 0.108 & 1504 + . . & n6791 & 0.40 & 0.098 & 311 + ( f390w f555w , f814w ) & m92 & -2.33 & 0.092 & 906 + . . & n6752 & -1.39 & 0.060 & 462 + . . & n104 & -0.66 & 0.060 & 2067 + . . & n5927 & -0.46 & 0.124 & 1443 + . . & n6791 & 0.39 & 0.082 & 297 the ( f390m f555w , f814w ) cmds presented in figure [ medcmd ] shows the uncorrected isochrones blueward of the clusters . additionally the isochrone curve along the giant branch is too steep to match the shape of the clusters . we applied a first order color term that has an additional metallicity dependence , making the color coefficient greater for lower metallicities . the isochrone corrections for all cmds are listed in table 8 . finding a single correction to align all five isochrones to the clusters gives us reasonable confidence to interpolate our corrections . smoothly changing corrections can be applied to the entire set of isochrones , then fit to an unknown population , although caution must be taken when extending beyond our calibrating clusters , i.e. m92 at @xmath108 , and ngc 6791 at @xmath109 . for the majority of cluster cmds the empirical corrections improve the isochrone - ridgeline alignment such that the average difference between the two is @xmath110 0.01 mag along the giant branch . in figures [ cmds ] - [ medcmd ] the dash - dot lines represent the uncorrected isochrones , while the solid lines are the empirically corrected isochrones . all the color corrections listed in table 8 are shown to @xmath111 . below this metallicity we replace all of the metallicity dependent terms with the value of the term when @xmath19 keeping the corrections constant where the color change is small and we are unable to verify the corrections . we tested the empirical isochrone corrections by re - deriving the cluster metallicities . we assign photometric metallicities to stars in the cmds by searching isochrone grids with spacing in of 0.05 dex ; each star was assigned the metallicity of the closest isochrone . we adopted [ @xmath48/fe ] @xmath112 since we can not separate the color effects due to @xmath48 and with only three filters . we selected giant branch stars with errors @xmath1050.03 mag in all filters . the photometric metallicities adopted from ( f336w f555w , f814w ) , ( f390m f555w , f814w ) and ( f390w f555w , f814w ) were used to create mdfs for every cluster , which are shown on the left side of figure [ plotmdf ] . for all the clusters the mdf peaks are within @xmath62 0.06 dex of the spectroscopically derived values . the recovered peaks , widths and the number of stars measured are listed in table 9 . the mdf dispersion within each cluster does not vary much across the metallicity range , although we do see slightly larger numbers for the differentially reddened ngc 5927 and for the lowest metallicity cluster , m92 , reflecting the diminished accuracy expected at low metallicity , as discussed in section 2.2 . additionally for ngc 104 and ngc 5927 the horizontal branch stars are clearly seen as small shoulders left of the mdf peak . the horizontal branch metallicities are systematically lower than the true metallicities because we only use the isochrone colors along the giant branch to assign metallicity , so horizontal branch stars were incorrectly matched to gb stars . overall , the average mdf dispersion for all clusters is @xmath20.10 dex and is consistent with what is expected from photometric errors . we simulated this by creating synthetic cmds based on the imf and photometric errors within our data set . at each magnitude along the gb ridgeline we used the measured photometric errors to randomly distribute the relative number of stars found at each magnitude . the synthetic cmd was then put through the same isochrone matching processes to get photometric metallicities and a mdf for each cluster . in order to determine the systematic effect of using uncorrected isochrones to assign photometric metallicities we compared the spectroscopically determined metallicities to those adopted using the uncorrected isochrone grid to assign metallicities . consistent cluster metallicities could not be found across the various cmds . for a given cmd the peak of the metallicity distribution varied by @xmath62 0.4 dex from the canonical values , with the larger discrepancies occurring at low metallicity . cllll[h ] -2.30 & 0.6208 & -0.0853 & 0.3473 & 0.2647 + -2.30 & 0.6212 & -0.0839 & 0.3476 & 0.2651 + -2.30 & 0.6217 & -0.0825 & 0.3480 & 0.2657 + . . + -1.45 & 0.6804 & -0.0160 & 0.4533 & 0.3075 + -1.45 & 0.6807 & -0.0149 & 0.4534 & 0.3077 + -1.45 & 0.6809 & -0.0137 & 0.4536 & 0.3080 + . . . + -0.70 & 0.7542 & 0.3142 & 0.8049 & 0.5268 + -0.70 & 0.7545 & 0.3152 & 0.8052 & 0.5270 + -0.70 & 0.7550 & 0.3165 & 0.8058 & 0.5276 + . . . + -0.40 & 0.7740 & 0.4647 & 0.9201 & 0.6125 + -0.40 & 0.7743 & 0.4656 & 0.9203 & 0.6129 + -0.40 & 0.7746 & 0.4668 & 0.9209 & 0.6133 + . . . + + 0.40 & 0.9087 & 1.0424 & 1.4213 & 0.9728 + + 0.40 & 0.9089 & 1.0435 & 1.4217 & 0.9731 + + 0.40 & 0.9094 & 1.0451 & 1.4223 & 0.9737 to test our metallicity recovery without using age as a known parameter , we apply the empirical corrections to isochrones in color - color diagrams ( figure [ cc ] ) then adopt metallicities . for each star we found the closest isochrone in the grid ( spaced by 0.05 dex in with the [ @xmath48/fe ] held constant at solar ) and assigned the corresponding metallicity . the corrected fiducial isochrone colors shown as black lines in the color - color diagram of figure [ cc ] are available in the online version in table 10 . the mdfs derived from color - color diagrams are shown on the right side of figure [ plotmdf ] . the recovered peaks , widths and the number of stars measured are listed in table 11 . we selected the same giant branch stars as before , i.e. stars with photometry better than 0.03 mag in all filters . we exclude poorly fit stars , e.g. stars more than 0.25 mag outside of the color range covered by the isochrone grid . this additional constraint accounts for the varying number of stars measured for different color - color diagrams in the last column of table 11 , though the variation is @xmath113 . for four of the five clusters , only the giant branch stars are used to adopt metallicities in the color - color plots . for the fifth cluster , ngc 6791 , there are not enough prominent giant branch stars , necessitating that we additionally use the sub - giant branch and main sequence stars . theoretically the different evolutionary states should not affect the metallicity determination because we selected the corresponding portion of the isochrone to match the less evolved stars . the mdfs from color - color diagrams , as compared to those from cmds , tend to have slightly larger offsets from the spectroscopically derived values , and systematically larger mdf dispersions . however , these metallicities were adopted without any age information , which is necessary for measuring metallicities for a population of mixed ages . the mdfs from the ( f390m f555w ) color - color diagrams have mean metallicities within @xmath62 0.07 dex for all clusters . the mdfs from the ( f390w f555w ) color - color diagrams produces mean metallicities within @xmath62 0.05 dex for m92 , ngc 104 ( 47tuc ) and ngc 6791 , and peaks within @xmath2 0.2 dex for ngc 6752 and the differentially reddened ngc 5927 . the mdfs from the ( f336w f555w ) color - color diagram finds mean metallicities of @xmath20.05 dex for most clusters , with the exception of ngc 104 which is discrepant by 0.1 dex . the larger mdf widths from the color - color diagrams as opposed to the cmd are a natural consequence of using color on the y - axis as opposed to magnitude . the change in color for the total range in metallicity is generally smaller than that for magnitude ( see section 2 ) , thus there is decreased resolution when using color - color diagrams ; additionally , the errors for colors are larger than for magnitudes . together , the decreased color span and the increased errors both add to the uncertainty in the photometric metallicities adopted from color - color diagrams widening the mdfs . the derived mdf widths are consistent with what is expected based upon the photometric errors . lllll[hb ] ( f336w f555w , & m92 & -2.22 & 0.471 & 879 + f555w f814w ) & n6752 & -1.73 & 0.484 & 454 + . . & n104 & -0.79 & 0.285 & 2067 + . . & n5927 & -0.38 & 0.365 & 1147 + . . & n6791 & 0.37 & 0.075 & 320 + ( f390m f555w , & m92 & -2.25 & 0.244 & 906 + f555w f814w ) & n6752 & -1.46 & 0.172 & 448 + . . . & n104 & -0.68 & 0.201 & 2110 + . . & n5927 & -0.47 & 0.264 & 1322 + . . & n6791 & 0.37 & 0.061 & 314 + ( f390w f555w , & m92 & -2.27 & 0.396 & 891 + f555w f814w ) & n6752 & -1.57 & 0.250 & 447 + . . & n104 & -0.70 & 0.232 & 2107 + . . . & n5927 & -0.58 & 0.296 & 1298 + . . & n6791 & 0.36 & 0.073 & 303 [ mdfcc ] of the three color - color diagrams tested to assign metallicity the ( f336w f555w , f555w f814w ) diagram does the worst . for ngc 104 the peak metallicity is off by 0.09 dex and the mdf dispersions are twice to three times the dispersions from ( f390m f555w , f555w f814w ) depending on the cluster . f336w , while predicted to remain sensitive at low metallicity , does not show this in practice , as seen from the wide dispersions for the clusters . the overestimate of the sensitivity could be due to the fact that the color separation , as seen from the clusters , is smaller than predicted from the isochrones , both of which are shown in figure [ ucmd ] . the ( f390m f555w , f555w f814w ) color - color diagram generally does the best job deriving metallicities of all the colors . the mdfs have peaks within @xmath62 0.07 dex of the spectroscopically determined metallicities for all clusters . the mdf dispersions are @xmath110 0.2 dex , except for the very - low metallicity m92 at 0.24 dex and the differentially reddened ngc 5927 at 0.26 dex . the mdfs derived using the ( f390m f555w ) color are better than those from ( f390w f555w ) and ( f336w f555w ) . in particular the mdf dispersion in the very - metal poor regime is the narrowest of all the mdfs from color - color diagrams by almost a factor of two . the greater accuracy in photometric metallicities below @xmath114 when using the f390 m makes this filter the most useful when working in the very - low metallicity regime . the ( f390w f555w , f555w f814w ) color - color diagram recovers peaks within @xmath62 0.2 dex of the spectroscopically determined metallicities . the mdf dispersions are @xmath2 0.25 dex , except for m92 , the very - low metallicity cluster , where the dispersion is 0.4 dex . however , f390w is very economical in terms of integration time , and accurate to @xmath20.25 dex above @xmath115 , therefore this filter is most useful when working in the more metal rich regimes . in order to determine the systematic effect of using uncorrected isochrones to assign photometric metallicities we compared the spectroscopically determined metallicities to those adopted using the uncorrected isochrone grid in color - color diagrams to assign metallicities . in the mdfs from the color - color diagrams of ( f336w f555w , f555w f814w ) and ( f390w f555w , f555w f814w ) the adopted cluster metallicities were within @xmath20.2 dex of the spectroscopically derived values for the clusters with metallicities between @xmath22 @xmath116 , while the very metal poor and metal rich cluster are 3 to 4 times more discrepant . in the color - color diagram of ( f390m f555w , f555w f814w ) the metallicities are all systematically higher by 0.4 to 0.8 dex , again , with more disparate values for very metal poor and metal rich clusters . we explored the metallicity and temperature sensitivities of colors created from nine wfc3/uvis filters aboard the hst using dartmouth isochrones and kurucz atmospheres models . the theoretical isochrone colors were tested and calibrated against observations of five well studied clusters . we found that ( f390w f555w ) , and ( f390m f555w ) , are the most promising colors in terms of metallicity sensitivity . for almost all of the clusters f390 m has slightly better metallicity sensitivity , and narrower mdf dispersions , although the f390w filter requires much less integration time . additionally , at low metallicity the photometric metallicities from ( f390m f555w ) are nearly twice as accurate as those from ( f390w f555w ) . using photometry of m92 , ngc 6752 , ngc 104 , ngc 5927 and ngc 6791 , all of which have spectroscopically determined metallicities spanning @xmath0 @xmath1 , we found empirical corrections to the dartmouth isochrone grid for each of the following cmds ( f555w f814w , f814w ) , ( f336w f555w , f814w ) , ( f390m f555w , f814w ) and ( f390w f555w , f814w ) . using the empirical corrections we tested the accuracy and spread of the photometric metallicities adopted from cmds and color - color diagrams . from the color - color diagrams we were able to recover the spectroscopic metallicities independent from any assumptions about cluster ages , which allows us to apply the color - color diagram method of determining metallicities to complex stellar populations with confidence that the method breaks the age - metallicity degeneracy . when using color - color diagrams to assign metallicity , we found the ( f390m f555w ) color to have the greatest accuracy and consistency across the entire metallicity range , with the main advantage being the increased sensitivity at low metallicity , while ( f336w f555w ) and ( f390w f555w ) both lose accuracy in this range . we showed that by using the calibrated isochrones we could assign the overall cluster metallicity to within @xmath2 0.1 dex in when using cmds ( i.e when the distance , reddening and ages are approximately known ) . the measured mdfs from color - color diagrams show this method measures metallicities of stellar clusters of unknown age and metallicity with an accuracy of @xmath2 0.2 - 0.58 dex using f336w , @xmath20.15 - 0.3 dex using f390 m , and @xmath20.2 - 0.52 dex with f390w , with the larger uncertainty pertaining to lowest metallicity range . support for program 11729 was provided by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . @xmath117 } } ) \rm{(f555w - f814w)}_{o } + 0.136 $ ] & log g @xmath26 3.4 , @xmath6 -2.5 @xmath118 @xmath105 -1.4 + @xmath117 } } ) \rm{(f555w - f814w)}_{o } + 0.136 - ( 0.015\rm{\textrm{[fe / h]}})(3.4-\log{g})$ ] & log g @xmath105 3.4 , @xmath6 -2.5 @xmath118 @xmath105 -1.4 + @xmath117 } } ) \rm{(f555w - f814w)}_{o } + 0.05(\rm{\textrm{[fe / h]}}+0.55)^2 + 0.10 $ ] & log g @xmath26 3.4 , @xmath6 -1.4 @xmath118 @xmath105 0.5 + @xmath117 } } ) \rm{(f555w - f814w)}_{o}+0.05(\rm{\textrm{[fe / h]}}+0.55)^2 + 0.10-(0.015\rm{\textrm{[fe / h]}})(3.4-\log{g})$ ] & log g @xmath105 3.4 , @xmath6 -1.4 @xmath118 @xmath105 0.5 + + @xmath119 & log g @xmath26 3.4 , @xmath6 -2.5 @xmath105 @xmath118 -1.65 + @xmath120}})(3.4-\log{g})$ ] & log g @xmath105 3.4 , @xmath6 -2.5 @xmath105 @xmath118 -1.65 + @xmath121}}+0.80)^2 - 0.02 $ ] & log g @xmath26 3.4 , @xmath6 -1.65@xmath105 @xmath118 0.05 + @xmath122}}+0.80)^2 -0.02 + ( 0.015 - 0.036\rm{\textrm{[fe / h]}}(3.4-\log{g } ) $ ] & log g @xmath105 3.4 , @xmath6 -1.65@xmath105 @xmath118 0.05 + @xmath123 & log g @xmath26 3.4 , @xmath6 0.05@xmath105 @xmath118 0.5 + @xmath124}})(3.4-\log{g } ) $ ] & log g @xmath105 3.4 , @xmath6 0.05@xmath105 @xmath118 0.5 + + @xmath125 } } ) \rm{(f390w - f555w)}_{o}+0.01(\rm{\textrm{[fe / h]}}+2.5)^2 + 0.05 $ ] & log g @xmath26 3.3 , @xmath6 -2.5 @xmath105 @xmath105 1.65 + @xmath125 } } ) \rm{(f390w - f555w)}_{o}+0.01(\rm{\textrm{[fe / h]}}+2.5)^2 + 0.05 + ( 0.07 + 0.006\rm{\textrm{[fe / h]}})(3.3-\log{g } ) $ ] & log g @xmath105 3.3 , @xmath6 -2.5 @xmath105 @xmath105 1.65 + @xmath125 } } ) \rm{(f390w - f555w)}_{o}+0.057 $ ] & log g @xmath26 3.3 , @xmath6 -1.65 @xmath105 @xmath105 + 0.5 + @xmath125 } } ) \rm{(f390w - f555w)}_{o}+0.057 + ( 0.07 + 0.006\rm{\textrm{[fe / h]}})(3.3-\log{g } ) $ ] & log g @xmath105 3.3 , @xmath6 -1.65 @xmath105 @xmath105 0.5 + + @xmath126 } } ) \rm{(f390m - f555w)}_{o } + 0.07(\rm{\textrm{[fe / h]}}+2.50)^2 + 0.005 $ ] & log g @xmath26 3.2 , @xmath6 -2.5 @xmath105 @xmath105 -1.65 + @xmath126 } } ) \rm{(f390m - f555w)}_{o } + 0.07(\rm{\textrm{[fe / h]}}+2.50)^2 + 0.005 + ( 0.018 - 0.011\rm{\textrm{[fe / h]}})(3.2-\log{g})$ ] & log g @xmath105 3.2 , @xmath6 -2.5 @xmath105 @xmath105 -1.65 + @xmath126 } } ) \rm{(f390m - f555w)}_{o}+0.056 $ ] & log g @xmath26 3.2 , @xmath6 -1.65 @xmath105 @xmath105 + 0.5 + @xmath126 } } ) \rm{(f390m - f555w)}_{o}+0.056 + ( 0.018 - 0.011 \rm{\textrm{[fe / h]}})(3.2-\log{g})$ ] & log g @xmath105 3.2 , @xmath6 -1.65 @xmath105 @xmath105 + 0.5
a procedure for calculation of rotation - vibration states of medium sized molecules is presented . it combines the advantages of variational calculations and perturbation theory . the vibrational problem is solved by diagonalizing a hamiltonian matrix , which is partitioned into two sub - blocks . the first , smaller sub - block includes matrix elements with the largest contribution to the energy levels targeted in the calculations . the second , larger sub - block comprises those basis states which have little effect on these energy levels . numerical perturbation theory , implemented as a jacobi rotation , is used to compute the contributions from the matrix elements of the second sub - block . only the first sub - block needs to be stored in memory and diagonalized . calculations of the vibrational - rotational energy levels also employ a partitioning of the hamiltonian matrix into sub - blocks , each of which corresponds either to a single vibrational state or a set of resonating vibrational states , with all associated rotational levels . physically , this partitioning is efficient when the coriolis coupling between different vibrational states is small . numerical perturbation theory is used to include the cross - contributions from different vibrational states . separate individual sub - blocks are then diagonalized , replacing the diagonalization of a large hamiltonian matrix with a number of small matrix diagonalizations . numerical examples show that the proposed hybrid variational - perturbation method greatly speeds up the variational procedure without significant loss of precision for both vibrational - rotational energy levels and transition intensities . the hybrid scheme can be used for accurate nuclear motion calculations on molecules with up to 15 atoms on currently available computers . molecular rotation ; vibration ; variational ; perturbation theory ; infra red spectra ; nuclear motion * acknowledgment * this work was supported by the erc under advanced investigator project 267219 .
1411.6098
approaches used to compute vibration - rotation energy levels and wave functions include methods based on perturbation theory , effective hamiltonians and on the use of the variational principle . each of these methods has its advantages and disadvantages when used for larger molecules such as those with more than ten atoms . historically , perturbation theory was the first method employed to treat the many - body anharmonic problem @xcite , see also the review by klein @xcite . second - order perturbation theory is the most common analytic treatment for estimating the vibrational energy levels and including contributions from terms in the hamiltonian beyond the harmonic approximation ( see , for example , refs . @xcite ) as well as infrared intensities @xcite . however , there are drawbacks inherent in this method that prevent extensive use of it in practice . in particular the results obtained depend on the specific form of the hamiltonian and the distribution of degenerate oscillators . polyatomic molecules often show quasi - degeneracies between vibrational energy levels and use of second - order perturbation theory can result in significant errors . furthermore , in practice anharmonic effects in polyatomic molecules are sufficiently large that they are often not converged with a second - order treatment . as a result perturbation theory calculations usually overestimate the anharmonic corrections , sometimes by a factor of two ; see the final columns of tables [ tab : en3 ] and [ tab : en4 ] given below . variational methods for the calculation of anharmonic energy levels were developed independently by a number of authors @xcite . early implementations in general computer codes focused on the use of basis functions for triatomic molecules @xcite . more recent developments have involved the increasing use of the discrete variable representation @xcite and the extension of the work to polyatomic molecules @xcite . the main advantage of this method is that it allows the almost exact calculations of the vibrational - rotational energy levels and wave functions for a given anharmonic potential @xcite . however , variational methods work much better for few - atom systems since the size of the basis set grows factorially with the number of vibrational degrees of freedom in the molecule ; hence the size of the hamiltonian matrices , which must be computed and diagonalized , also grows very rapidly . for this reason , variational calculations for polyatomic molecules are difficult even on modern computers and therefore are not routinely used for molecules with more than six atoms , when only small basis sets can be afforded . the computational errors associated with incomplete basis sets are normally significantly larger than the errors from perturbation approaches ; this point will be discussed further in the following section . the treatment of ro - vibration states adds extra complexity to the calculation . in this case a rotational basis set , usually described with analytical rigid - rotor functions , must be introduced . for high rotational quantum number , @xmath0 , this leads to large hamiltonian matrices even for small polyatomic molecules . however a two - step variational procedure can be used to mitigate the effects of this problem @xcite meaning that it has long been possible to compute rotational excitation up to dissociation for small molecules @xcite . there are a number of modifications of the variational method which facilitates ( ro-)vibrational calculations on larger molecules . for example , the use of vibrational self - consistent field theory by gerber , ratner and others @xcite . bowman , carter and handy @xcite used vibrational configuration interaction ( vci ) to reduce the size of the diagonalization for large molecules . if the molecule has separable degrees of freedom , a significant reduction in the time taken to diagonalize the hamiltonian matrices can be obtained . scribano and benoit @xcite used a hybrid approach where a modification of the vci method take into account the interactions of individual configurations using perturbation theory . similarly the general code multimode @xcite allows such treatments of larger systems @xcite with approximations involving the degree of coupling between vibrational modes . here we propose a new hybrid variational - perturbation theory method based on a physically reasonable division of the large , full , variational hamiltonian matrix into weakly interacting sub - blocks . second - order perturbation theory is used to include cross - interaction effects between these sub - blocks , which are then diagonalized separately . this allows one to replace a single diagonalization of a large hamiltonian matrix by a series of diagonalizations of much smaller sub - matrices . we show that our hybrid method can greatly accelerate the variational procedure by eliminating diagonalization of large matrices without significant loss of precision in the computed vibrational - rotational energy levels and the intensity of transitions between them . this hybrid scheme is able to perform calculations for large molecules containing up to 15 atoms on currently available computers . finally we note that a different version of the hybrid variational - perturbation approach has recently been proposed by fabri _ et al _ @xcite and implemented in the general eckart - watson hamiltonian code dewe @xcite . their hybrid scheme uses a traditional single - state ro - vibrational perturbation method based on the variationally computed vibrational ( @xmath1 ) eigenfunctions as a zero - order solution , where the vibrational problem is solved variationally and the ro - vibrational energies are derived using the second order perturbation theory expressions . this is different from our two - step hybrid scheme , where the perturbation method is an integral part of calculations . we believe that this is the key to extending variational calculations to larger molecules . the method we propose is not dependent on the precise form of the hamiltonian . instead the requirement of the hamiltonian is that it should result in a diagonally dominant matrix representation . for semi - rigid molecules this criterion is naturally satisfied the eckart - watson hamiltonian @xcite as well as number of related hamiltonians @xcite . however the method is unlikely to perform so well for hamiltonians whose emphasis is not on making producing diagonally dominant matrices such those expressed in polyspherical coordinates @xcite . in this work , all calculations are performed using angmol @xcite , a variational program for calculating ro - vibrational spectra of general polyatomic molecules . angmol uses a hamiltonian expressed in curvilinear internal coordinates and an eckart embedding which is discussed briefly in the appendix and extensively elsewhere @xcite . to start we consider the nature of the error arising from use of the variational method for calculating vibrational ( @xmath2 ) energy levels of polyatomic molecules . in the variational rayleigh - ritz procedure , the wave function , @xmath3 , is generally approximated by a finite expansion @xmath4 in terms of some appropriate basis functions @xmath5 mathematically , this procedure is equivalent to finding the eigenvalues and eigenvectors of the hamiltonian matrix whose elements are given by : @xmath6 eigenvalues of this matrix give values of the vibrational energy levels , @xmath7 , and the eigenvectors , @xmath8 , give the wave function , @xmath3 , when combined with the basis functions . for ease of use and better convergence of the basis functions , the @xmath5 are generally chosen to form a complete orthonormal set . exact values of the energy levels and eigenfunctions are only guaranteed with an infinite number of basis functions . the accuracy of the calculated energies depend on the number of basis functions used . as the basis set is extended , the calculated energy levels tend ( mostly smoothly and monotonically ) to their theoretical limit . the variational limit for a group of ( lowest ) energy levels is reached when the number of basis functions is sufficient for the chosen energy levels to differ from the exact value by less than some prescribed error . the physical meaning of the variational limit is that basis functions representing highly excited vibrational levels make only a small contribution to the eigenfunctions of the low - lying levels . the number of basis functions needed to reach the variational limit depends on the number of energy levels to be calculated . in general , the greater the number of energy levels calculated , the larger the number of basis functions required . however , a good choice of basis functions also speeds the convergence : the closer the basis functions are to desired eigenfunctions , the fewer of them are required to achieve the variational limit . mathematically , this means that the off - diagonal matrix elements ( @xmath9 ) become small , the eigenvalues become close to the corresponding diagonal element ( @xmath10 ) , and the corresponding eigenvectors tend towards being unit vectors . details of our method are give in the appendix . we start by defining the vibrational basis as a product of one - mode functions @xmath11 where @xmath12 is a either a morse or a harmonic oscillator function , @xmath13 is the excitation number of the @xmath14 oscillator and @xmath15 is the number of vibrational degrees of freedom . morse oscillator functions are commonly used for the x h stretches and other stretching motion whose vibrations are strongly anharmonic , such as doubly - bonded co. harmonic oscillator functions are used for skeletal bonds in molecules with small anharmonicity , for example the cc bonds in hydrocarbons , and for all deformation ( angular ) oscillations . this form of the basis set helps to reduce the off - diagonal elements in the hamiltonian matrix and thus has been found to give rapid convergence to the variational limit . in variational procedures it is often considered desirable to use a complete , orthonormal basis . the bound states of the morse oscillator form a variational basis of finite length basis and so are strictly speaking not longer complete . it is possible to develop a complete set based on the morse oscillators @xcite ; however , our practical computation shows that rapid convergence can be achieved with a finite number of morse oscillators allowing the variational limit to be reached . our implementation relies on the particular structure of the hamiltonian matrix ordered by increasing polyad ( total vibrational excitation ) number , @xmath16 @xmath17 where @xmath18 is some integer weighting which is often roughly proportional to the inverse of the frequency @xcite . for simplicity in this work we use @xmath19 for all @xmath20 . this gives the size of the basis set , @xmath21 in terms of the maximum polyad number , @xmath22 , @xmath23 which increases factorially with the number of vibrational degrees of freedom , @xmath15 . ) hamiltonian matrix for water ( @xmath24 , @xmath25 ) : region a contains large ( near-)diagonal elements ; region b contains only small off - diagonal elements . the blue continuous lines are used to separate the two regions . the brightness of the red points is proportional to the magnitude of the individual matrix elements.,scaledwidth=40.0% ] with this definition of the basis set , the hamiltonian matrix has a semi - banded structure , as shown in fig . [ fig:1 ] , and can be factorized into two regions : the region around the diagonal ( region a ) , containing large off - diagonal elements , and the outer region ( region b ) with ` small ' or zero off - diagonal elements . the width of the band , as discussed in the appendix , depends on the form of the hamiltonian and obviously on the precise definition of ` small ' . for example , with a potential function represented as a fourth - order polynomial in terms of the coordinate displacements and harmonic oscillator basis functions , the band only includes matrix elements with excitations whose basis functions @xmath12 are separated from the diagonal by @xmath26 . use of the morse oscillator basis functions gives a similar picture . clearly distinguishable in fig . [ fig:1 ] are eight off - diagonal bands corresponding to elements connected by @xmath27 , @xmath28 , @xmath29 , and @xmath30 . thus for a fourth - order , anharmonic potential function , all matrix elements separated by @xmath31 lie in region b which becomes increasingly large as either @xmath16 or @xmath15 increases . this rule has similarities with the well - known slater s rules for configuration interaction matrix elements in electronic structure calculations . as an example we use the program angmol to compute the vibrational band origins and corresponding band intensities for h@xmath32o and hno@xmath33 . in these calculations we employ semi - empirical potential energy functions represented as fourth - order polynomials which were obtained by fitting the corresponding potential parameters to experimentally derived energies @xcite . these empirical potential energy functions reproduce the experimental vibrational term values of the fundamental and first overtone states of hno@xmath33 and h@xmath32o with the root - means - squares ( rms ) errors less then 0.1 and 0.4 @xmath34 , respectively . initial values of the potential parameters were obtained using _ ab initio _ calculations at the ccsd(t)/aug - cc - pvqz level of theory . the corresponding dipole moments were represented in the form of a second - order polynomial fitted to the ccsd(t)/aug - cc - pvqz _ ab initio _ values . full details will be given elsewhere @xcite . all calculations were performed using curvilinear vibrational coordinates , see appendix . to represent the stretching coordinates we use morse coordinates @xmath35 , where @xmath36 in the bond length displacements of the @xmath37 bond and @xmath38 is the standard morse parameter . valence angles were represented by @xmath39 , where @xmath40 and @xmath41 are the @xmath42 inter - bond angle and its equilibrium value , respectively . basis functions were constructed as a direct product of the morse oscillator functions for the stretches and harmonic oscillator functions for the angles . the harmonic part of the hamiltonian is constructed to be diagonal for the bending part . this form of variational basis functions has the advantage that the vibrational hamiltonian matrix has relatively small off - diagonal elements , i.e. it is close to a diagonal form . this is due to ( i ) the property of the morse oscillators which give a good description of the stretching modes and ( ii ) small bending anharmonic terms . besides , in this basis the hamiltonian matrix elements all have simple analytical forms and thus can be efficiently evaluated . tables [ tab : en1 ] and [ tab : en2 ] present calculated values of the wavenumbers and intensities of the vibrational transitions for h@xmath32o and hno@xmath33 as a function of the number of basis functions used in the variational calculations . in the following , the parameter @xmath43 ( @xmath44 ) will be used to reference the polyad number for energy values targeted in the variational calculations and @xmath45 ( @xmath46 ) to reference the corresponding number of vibrational states . the results for the water molecule suggest that to compute all vibrational levels associated with the polyad number @xmath47 to better than 5 @xmath34 , the basis set must include functions with @xmath22 at least up to @xmath48 . for example , to obtain all energies up to the second overtones ( @xmath49 ) with this accuracy , @xmath22 must be at least 7 . this is in accord with our assumption that for the potential energy function given as a fourth - order polynomial , the large matrix elements which belong in region 1 are all associated with functions given by @xmath50 . calculating all @xmath51 vibrational term values to an accuracy better than 0.3 @xmath34 requires basis functions with @xmath52 . this means that the hamiltonian matrix must include all the basis functions for which the difference in @xmath13 is up to 8 . for water , using the lowest possible basis , _ i.e. _ @xmath53 , gives errors of 28 @xmath34 for the @xmath54 fundamental band , 84 @xmath34 for the first overtone @xmath55 and 154 @xmath34 for @xmath56 , see table [ tab : en1 ] . for hno@xmath33 the situation even worse , with @xmath57 , the band centres of the fundamentals are not reproduced to within 500 @xmath34 . it is important to note that the number of the target energy levels defined by @xmath43 grows rapidly with the number of degrees of freedom @xmath15 . for a triatomic molecule like water ( @xmath58 3 ) , to reach an accuracy better than 0.5 @xmath34 for the fundamental bands only ( @xmath59 , @xmath60 ) requires @xmath61 220 basis functions . for a pentatomic molecule , such as hno@xmath33 ( @xmath58 9 ) , this requires @xmath61 48 620 basis functions and for an 8-atom molecule such as c@xmath32h@xmath62 ( @xmath15 = 18 ) , @xmath63 is 4 686 825 . the vibrational energies of polyatomic molecules are often characterized by ( accidental ) resonances , such as fermi resonances , resulting from strong mixing between vibrationally excited states . one consequence of this is that for hno@xmath33 @xmath22 is increased to @xmath64 . comparison of columns 2 of tables [ tab : en2 ] and tables [ tab : en4 ] shows that even when for @xmath61 48 620 basis functions corresponding to @xmath60 , the convergence of the fundamental energy levels , _ i.e. _ @xmath65 , is only good to 4 @xmath34 . for nitric acid the resonances can also give rise to large errors in band intensities for vibrational states involved in these resonances , especially when small basis sets are used . however , the sum of the bands intensities for all resonant states is less sensitive to the size of the basis set . we note , however , that for water even individual vibrational band intensities show relatively small dependence on the basis set size , see table [ tab : en1 ] . tables [ tab : en1 ] and [ tab : en2 ] also show that the stretching energy levels converge faster , requiring less basis functions , than the bending modes . this is because the morse oscillators are more compact than the harmonic oscillator basis functions and also because the stretching quanta of hydrogen bonds is normally about two times larger than that of the bending modes , i.e , it takes fewer stretching quanta to reach the same energy . however , the harmonic oscillator functions used for the bending modes also represent a reasonable choice giving small coupling matrix elements , which satisfy the requirements of perturbation theory . variational calculations on many - atom systems rapidly become impractical . to address this problem we implement a mixed variational - perturbation theory approach . the idea is based on the observation that the calculated energies and wave functions of the lower - lying vibrational levels depend differently on different parts of the hamiltonian matrix . as described above , the general hamiltonian matrix can be factorised into two regions as illustrated in fig . [ fig:2 ] . the first region contains the zeroth - order contribution to the calculated vibrational energies as well as all other states strongly coupled to them . as noted above , these strongly - coupled levels are those for which the hamiltonian matrix contains large off - diagonal elements with the states of interest . for a potential function represented as a fourth - order polynomial , the first block should contain couplings to the targeted states with all contributions with @xmath26 . the second region of the matrix is generally much larger . this region includes all the remaining basis states that only have a small effect on the calculated energy levels in question . these basis states are ones for which the polyad numbers @xmath16 differ by more than 4 from those of the levels of interest , @xmath66 , because the corresponding off - diagonal elements are small . the essence of our method is that instead of diagonalizing the whole hamiltonian matrix , we diagonalize the smaller block 1 ( see fig . [ fig:2 ] ) with the matrix elements corrected by the second order perturbative contributions from region 2 as described in detail below . assuming the size @xmath21 of the overall hamiltonian matrix is defined by @xmath22 in connection with eq . ( [ e : n_bmax ] ) and the size of block 1 , @xmath67 , is given by @xmath68 where @xmath69 is the number of the largest polyad included in block 1 . to reach an accuracy of 1 @xmath34 or better , the following two conditions have to be met : @xmath70 and @xmath71 . in the first step of our approach all the matrix elements from blocks 1 are computed along with the diagonal matrix elements of block 3 . the second step involves computing the off - diagonal elements of block 2 and accounting for their effect on the matrix elements of the block 1 using one jacobi rotation @xcite . considering a contribution from the block 2 off - diagonal element @xmath72 , which couples the diagonal elements @xmath10 in block 1 and @xmath73 in block 3 , see fig . [ fig:2 ] , these two diagonal elements are perturbatively adjusted using the jacobi formula @xmath74 where @xmath75 and @xmath76 here @xmath10 is an initial ( unperturbed ) diagonal element and @xmath77 is an adjusted one . this formula automatically allows for cases where the unperturbed vibrational states are ( quasi-)degenerate , i.e. @xmath78 . in the absence of the degeneracy , @xmath79 , we have @xmath80 , which is equivalent to the energy correction given by second - order perturbation theory ( see e.g. @xcite ) . similarly , if there is degeneracy or quasi - degeneracy @xmath81 , we have @xmath82 and for degeneracy @xmath83 , we have @xmath84 . that is @xmath85 is always satisfied . strictly speaking , an off - diagonal element @xmath9 in block 2 should be included as perturbative contribution both to the relevant diagonal @xmath10 and off - diagonal @xmath86 elements from block 1 . however , the changes to the off - diagonal are of the second - order leading to small contributions and are thus omitted here without significant loss of accuracy , as will be shown below . in the final step of the algorithm , block 1 of the corrected hamiltonian matrix is diagonalized . the resulting eigenfunctions are represented as expansions in terms of the basis functions associated with block 1 only , while the eigenvalues include contributions from all three regions of the matrix . as in the examples above , we use h@xmath87o and hno@xmath88 to illustrate the method ( see also ref . @xcite , where vpt2 was used to study the ir spectroscopic properties of hno@xmath88 ) . tables [ tab : en3 ] and [ tab : en4 ] show similar results obtained using our hybrid method , where the vibrational term values calculated with the size of block 1 , @xmath67 , increasing . we note that when @xmath89 , our hybrid method automatically becomes a full variational calculation . the hybrid method provides accurate results for the vibrational energy levels even when the dimension of the final diagonalized block @xmath90 is much smaller than the dimension of the full hamiltonian matrix @xmath91 . for example , to calculate the vibrational energy levels of water with an average accuracy of 1 @xmath34 , it is sufficient to explicitly include the basis functions with @xmath92 . that is , block 1 includes all the larger off - diagonal elements ( at least @xmath93 ) and block 2 contains the smaller off - diagonal elements contributing to the target vibrational energies ( at least @xmath94 ) . this reduction is especially valuable for larger molecules . for example , the @xmath95 rule for the eight - atomic molecule c@xmath32h@xmath62 ( @xmath15 = 18 ) means that to calculate the fundamental band centres to within 1 @xmath34 , the dimension of block 1 is only @xmath96 ( @xmath97 5 ) , while the full size of the hamiltonian matrix is @xmath98 ( @xmath99 9 ) . for hno@xmath33 we have @xmath100 and @xmath101 ( @xmath99 9 ) . however in case of a strong fermi resonance between vibrational states the accuracy condition must be increased at least by 1 unit . for example , in order to obtain all vibrational term values hno@xmath33 defined by the polyad number @xmath102 to within 4.0 @xmath34 , the basis set should include at least @xmath103 contributions , see table [ tab : en4 ] . even with very small basis sets the hybrid method gives reasonable results . for example , with the minimum possible size defined by @xmath104 the error in the band centres of @xmath54 , @xmath55 , and @xmath56 of h@xmath32o is 2 @xmath34 , 6 @xmath34 and 15 @xmath34 , respectively , which can be compared to the error of 28 @xmath34 , 84 @xmath34 and 154 @xmath34 , respectively , obtained using the purely variational procedure with the same basis set size . this and other examples collected in tables [ tab : en3 ] and [ tab : en4 ] show that our hybrid method inherits the advantage of perturbation theory which performs reasonably well with small basis sets . in the limit of @xmath105 , our variation - perturbation method turns into the pure perturbation calculation with the perturbation treated numerically rather than analytically . for comparison tables [ tab : en3 ] and [ tab : en4 ] also give results obtained from perturbation theory using a single jacobi rotation for each off - diagonal element . tables [ tab : en3 ] and [ tab : en4 ] also illustrate the performance of the hybrid approach for the vibrational band intensities of these two molecules . these results suggest that for small basis sets the hybrid intensities are slightly better than intensities obtained using a direct variational treatment of the same size . this is due to the reasonable quality achieved , even with small basis sets , of the corresponding energy levels and eigenfunctions , see above . it should be noted that the hybrid method requires that the off - diagonal elements @xmath9 of the block 2 are small compared to the elements of block 1 . this requirement can be satisfied if basis functions are close to the true eigenfunctions . our tests show that the morse oscillators used for the stretching coordinates and harmonic oscillators used for the other degrees of freedom satisfy this criterion . finally we note that it is not necessary to compute off - diagonal elements in block 3 . this means that the total number of the off - diagonal elements calculated is @xmath106 , which is much less than the overall size of the hamiltonian matrix @xmath107 . inclusion of rotational motion into the problem adds an extra level of complexity , especially when highly excited rotational states are required . the dimension of the ro - vibrational problem @xmath108 is normally increased by the factor of @xmath109 , where @xmath0 is the total angular momentum quantum number . in the variational approach this leads to matrices which can be several order of magnitude larger than the corresponding pure vibrational ( @xmath1 ) hamiltonian matrices . in order to make the vibrational part of the basis set more compact , we construct the ro - vibrational basis set as a direct product of the rigid - rotor winston s @xmath110 and a set of selected vibrational eigenfunctions of the @xmath1 problem obtained as described above . this approach is sometimes referenced in the literature as the @xmath1 contraction @xcite . let us use @xmath111 as the number of vibrational eigenstates for this purpose . @xmath112 can be significantly reduced compared to the total number of vibrational states , @xmath67 , from the previous step as defined by the size of block 1 . this is because not all of the @xmath67 vibrational states are equally important for the target vibrational bands @xmath113 . indeed , because our expansion of the kinetic energy operator is truncated at the second order ( see appendix ) , many matrix elements with @xmath114 are either exactly zero ( for the harmonic oscillators ) or very small . as an example , consider the situation where we would like to compute a spectrum involving all ro - vibrational energies of hno@xmath33 up to @xmath115 @xmath34 . this range corresponds to @xmath116 = 4 or @xmath113 = 715 . from our experience of the hno@xmath33 molecule , the corresponding ro - vibrational basis set has to include excitations at least up to @xmath117 @xmath34 to reach the convergence of 1 @xmath34 , which means including about @xmath118 = 8000 vibrational levels . then the corresponding rotational contribution may include up to @xmath119 . this situation is common in the exomol project @xcite , which computes spectra of hot molecules . this leads to a set of ro - vibrational hamiltonian matrices with the dimensions @xmath120 ranging from 8000 to @xmath121 = 1 608 000 . in the following we show how to exploit the advantages of the special structure of the ro - vibration hamiltonian matrix in order to simplify this problem . for large polyatomic molecules , the ro - vibrational hamiltonian matrix has a pronounced quasi - block character built around vibrational states . this is due to the fact , see appendix , that the off - diagonal elements @xmath122 between two different vibrational states @xmath123 and @xmath124 always contribute substantially less to the calculated energy levels than the off - diagonal elements within a vibrational state , @xmath125 , where @xmath126 is the quantum number giving the projection of @xmath0 on the body - fixed z - axis ( @xmath127 ) and @xmath123 runs over the vibrational basis set ( @xmath128 ) . physically , this corresponds to small coriolis interactions between different vibrational states , which is achieved by the use of the eckart embedding @xcite . figure [ fig:3 ] illustrates the structure of the ro - vibration matrix for the first four vibrational states of molecules hno@xmath33 ( @xmath129 ) . we can see that most of the off - diagonal @xmath130 elements are zero when @xmath131 reflecting the quadratic differential form of the kinetic energy operator . it can also be seen that the non - zero off - diagonal elements gather near the main diagonal and the matrix has an almost tridiagonal form . the near - diagonal elements @xmath132 are mostly associated with changes in the effective geometry of the rotating molecule through interactions with the corresponding vibrational state @xmath123 . the off - diagonal elements @xmath130 are associated with coriolis coupling between vibrational states @xmath123 and @xmath124 . for the first four vibrational states , @xmath133 ( density of red is proportional to the magnitude of the individual matrix element ) ; b : the contributions from the off - diagonal elements to the final energy levels ( the density of red is proportional to the magnitude of the corresponding contribution ) . the squares along the diagonal depict the pure rotational sub - matrices for vibrational states @xmath123 = 1 , 2 , 3 , 4 respectively.,title="fig:",scaledwidth=80.0% ] + a for the first four vibrational states , @xmath133 ( density of red is proportional to the magnitude of the individual matrix element ) ; b : the contributions from the off - diagonal elements to the final energy levels ( the density of red is proportional to the magnitude of the corresponding contribution ) . the squares along the diagonal depict the pure rotational sub - matrices for vibrational states @xmath123 = 1 , 2 , 3 , 4 respectively.,title="fig:",scaledwidth=80.0% ] + b importantly , the contribution of the off - diagonal elements @xmath130 corresponding to the coupling between different vibrational states is much smaller than that of the off - diagonal elements @xmath134 belonging to the same vibrational state , because of the large energy separation between vibrational levels . this is illustrated in fig . [ fig:3 ] , where the actual contributions from the off - diagonal matrix elements to the final ro - vibrational energy levels for molecules hno@xmath33 are shown . typically contribution from off - diagonal elements corresponding to different vibrational states are very small , less than 0.001 @xmath34 . this feature of the vibration - rotation hamiltonian matrix is common for large molecules and underpins the ro - vibrational version of our hybrid approach , which is schematically represented in fig . [ fig:4 ] . again we use the second - order perturbation theory ( as defined by the jacobi rotation ) to transform the @xmath130 matrix to a block diagonal form @xmath135 built from rotational sub - matrices corresponding to different vibrational states @xmath123 . thus the dimension of each degenerate rotational sub - block is @xmath109 only and we only consider @xmath113 sub - matrices that correspond to @xmath113 vibrational states . in the case of strong resonances between different vibrational states , these states can be combined into one , enlarged sub - matrix and treated together , i.e. for an @xmath136-fold symmetry vibrational degeneracy , the dimension of the sub - block is @xmath137 . for details see the appendix . equation ( [ e : h - nondiag ] ) is a symmetrised version of the standard formula for the single jacobi rotation with respect to the indices @xmath144 and @xmath145 . it should be noted that in the limit @xmath146 , eq . ( [ e : h - nondiag ] ) is identical with the corresponding expression for second - order perturbation theory : @xmath147 the resulting block - diagonal form is then diagonalized for each @xmath123 sub - matrix separately , where @xmath123 indicates either a single vibrational state or a set of strongly interacting vibrational states . thus our algorithm replaces the diagonalization of a huge [ @xmath148 ro - vibrational matrix with a number of diagonalizations of much smaller [ @xmath109 or @xmath149 matrices , where @xmath136 is the number of resonance sub - states . it should be noted that in our algorithm , the ro - vibration wave functions , in contrast to those from an exact diagonalization , do not contain any contribution from other vibrational states . this leads to some errors in the subsequent calculation of the transition dipole moments and intensities of the ro - vibrational transitions @xcite . however , the relative error is rather small for the stronger and most important transitions because of the relatively small perturbation contribution from the off - diagonal ( @xmath150 ) matrix elements @xmath130 . for other states , where intensity stealing is important @xcite , the resonant coupled vibrational states procedure can be used @xcite . finally , we note that our procedure is both very quick and easily parallelized . ; the off - diagonal matrix elements from region 4 are generally smaller and treated using the perturbation theorynote indices @xmath151 and @xmath152 are denoted @xmath153 and @xmath154 , respectively , in the text which distinguished between vibrational and rotational basis functions.,scaledwidth=40.0% ] although the calculation of the ro - vibration energy levels and wave functions may be long , it is only an intermediate stage . the final step is the calculation of transition intensities which involves computing matrix elements of the dipole moment functions between the computed ro - vibration wave functions . for hot molecules , where a huge number of transition dipoles must be evaluated , this step becomes very time consuming and even prohibitively expensive @xcite . thus , even for hno@xmath33 , the problem of calculating the vibration - rotation spectrum in the range of 04500 @xmath34 using a standard variational method without approximations is extremely difficult using existing computers . for larger molecules such calculations are currently almost impossible . the separable form of our hybrid rotation - vibration wave functions , which are obtained in separate diagonalizations for each vibrational state , allows a speed - up , by several orders of magnitude , of the subsequent evaluation of transition intensities . this is because the wave functions are much more compact and represented by a number of independent parts . as we will show below , this approximation in case of hno@xmath88 and h@xmath87o , leads to relatively small errors in the overall shape and magnitude of the ro - vibrational spectra computed at room or higher temperatures . tables [ tab : en5 ] and [ tab : en6 ] show the pure rotational energy levels of h@xmath32o and hno@xmath33 for several @xmath0 values obtained using three different methods to find eigensolutions of the corresponding hamiltonian matrices : full diagionalisation ( f ) , the hybrid ro - vibrational approach described above ( h ) and a rigid rotor calculation represented by separate diagonalisations of the rotational sub - blocks with the off - diagonal couplings with other vibrational states neglected ( r ) . the results for water , table [ tab : en5 ] , are of interest because the molecule exhibits relatively large off - diagonal elements @xmath130 and thus large centrifugal effects . the difference @xmath155 illustrates the magnitude of the centrifugal and coriolis contributions , while @xmath156 shows the quality of our hybrid method for different @xmath0 . for example , for @xmath157 the maximum value of @xmath158 is only 0.03 @xmath34 and @xmath159 is within 0.005 @xmath34 and for @xmath160 , @xmath161 is within 0.6 @xmath34 and @xmath159 is within 0.05 @xmath34 . at higher @xmath0 these residuals grow significantly , up to 161 and 44 @xmath34 , respectively , for @xmath162 . for a given @xmath0 , both @xmath158 and @xmath163 increase with increasing rotational energy . for example , for @xmath164 for the lowest term value of 741.25 @xmath34 ( @xmath165 ) , @xmath166 @xmath34 and @xmath167 @xmath34 , while for the highest term value 1805.18 @xmath34 ( @xmath168 ) we obtain @xmath169 @xmath34 and @xmath170 @xmath34 . thus , we conclude that for a small molecule with large centrifugal effects , like water , the hybrid method does not provide accurate ro - vibrational energies . indeed , it is well known that the @xmath1 contraction performs poorly for water @xcite because of issues with linear geometry . however , for such systems purely variational calculations do not present a computational problem @xcite . for larger molecules it is generally not necessary to deal with issues associated with quasi - linearity . for larger molecules with smaller rotational constants , centrifugal distortion effects are expected to decrease significantly . therefore we also expect the hybrid method to perform better for such systems . tables [ tab : en6 ] illustrates this effect for hno@xmath33 . for example , for @xmath171 for the maximal residuals we obtain @xmath172 @xmath34 and @xmath173 @xmath34 , which also correspond to the highest term value 1556.46 @xmath34 for the @xmath174 manifold ( @xmath168 ) . for the rotational levels in the range @xmath175 , the average error of the hybrid method is within 0.02 @xmath34 , i.e. much better than that for water for the same energy range . the range @xmath176 is selected as it represents the states which are important for the room temperature ro - vibration spectrum of hno@xmath33 . it should be noted that for large molecules with small rotational constants , like hno@xmath88 , individual absorption lines are often not resolved because of the high density of the lines even at room temperature . for example , the spectrum of hno@xmath88contains @xmath177 lines in the region 0 4000 @xmath34 with intensities above 10@xmath178 @xmath179molecules at 600 k. this suggests that calculations with the accuracy within 0.02 @xmath34 should be sufficient for most purposes . as an example , fig . [ fig:5 ] shows the absorption cross - sections calculated in the region of the @xmath180 and @xmath181 bands of hno@xmath33 for @xmath182 298 k using the full variational ( ` full ' ) calculation as well as the difference with the hybrid calculation . here a voigt line profile was used with parameters @xmath183 @xmath34 ( a half width at half maximum ( hwhm ) of 0.153 @xmath34 ) , chosen to match spectra from the pnnl database @xcite . further details of these spectra will be given elsewhere @xcite . for the central part of the band ( @xmath184 ) , where the rotational lines are strong enough to be resolved , the two methods ( full and hybrid ) give almost identical results . at the edges of the bands ( @xmath185 ) slight differences in the frequencies and intensities of the individual ro - vibration absorption lines appear . however the rotational structure is poorly resolved due to superposition of a large number of lines . besides the intensity of the band diminishes rapidly with @xmath0 due to the decreasing population of the highly excited rotational levels , therefore the absolute difference is also negligible . calculated using the full diagonalization of the ro - vibration hamiltonian matrix ( red curve ) , and our hybrid method ( blue curve ) . the green curve gives the difference between these two methods . a : @xmath180 and @xmath181 bands ; b : @xmath186-branch of the @xmath180 band . , title="fig:",scaledwidth=80.0% ] + a calculated using the full diagonalization of the ro - vibration hamiltonian matrix ( red curve ) , and our hybrid method ( blue curve ) . the green curve gives the difference between these two methods . a : @xmath180 and @xmath181 bands ; b : @xmath186-branch of the @xmath180 band . , title="fig:",scaledwidth=80.0% ] + b in this paper two hybrid variation - perturbation methods for computing vibration and ro - vibrational energies for large molecules are proposed . these methods yield significant speed - ups for the computation of ro - vibrational spectra , with only a minor loss of accuracy , at least for semi - rigid polyatomic molecules . we plan to use method to compute extensive line lists and cross sections for large molecules , including nitric acid @xcite , as part of the exomol project @xcite . as an illustration of the efficiency of the hybrid method , calculations of the vibrational and ro - vibrational energies of a tri- and a penta - atomic molecule are presented . we can use these to project to even larger systems . assuming that it is currently feasible to diagonalize a matrix of about 1 000 000 by 1 000 000 and using eq . ( 5 ) , we can expect to achieve an accuracy of 1 @xmath34 for all ro - vibrational term values for polyads @xmath187 for a given potential energy surface for a molecule containing up to 15 atoms using our hybrid procedure . for molecules containing up to 25 atoms , the expected accuracy is about 5 @xmath34 . our hybrid methods have several advantages : 1 . they are relatively easy to implement as an extension to a pure variational computer program , which readily provides all required components . for example , all the calculations presented here were performed using angmol @xcite , a variational program for ro - vibrational spectra of general polyatomic molecules . it required changes to less than 1% of the entire program to implement the hybrid methods . they allow efficient parallelization and vectorization of the code . they are very fast compared to full diagonalizaion . not only is the diagonalization time greatly reduced , but also only a small proportion of the hamiltonian matrices needs to be computed . furthermore , the coupling elements can be evaluated on the fly and do not need to be stored in the memory . our method adopts several approximations arising from second - order perturbation theory and the rigid rotor model to full diagonalization approaching the variational limit . for the vibration - only problem , this approximation is easily controlled by altering the size of the block 1 . our method has scope for further improvement . for example : * for the purely vibrational hamiltonian matrix , one can change not only the diagonal but also the off - diagonal elements in block 1 , as we do for the ro - vibrational hamiltonian matrix . this will increase the accuracy at the expense of increased computer time . * one could use not only one but two or more consecutive jacobi rotations . again , this will increase the accuracy at the expense of increased computer time . * simple perturbation theory corrections generally overestimate the effect of the perturbation . one could reduce this effect by multiplying the computed perturbations by some empirical coefficient @xmath188 . this will increase the accuracy without increasing the computer time . the problem with this approach is that the magnitude of @xmath189 depends on the number of degrees - of - freedom involved , the structure and size of the hamiltonian matrix and value of perturbation adjustment . * finally , we note that in our present implementation no advantage is taken of the pronounced polyad structure shown by many molecules . this could be achieved by a suitable choice of the @xmath190 coefficients in eq . ( [ e : n_v ] ) . this work was supported by the erc under advanced investigator project 267219 . 93 [ 1]`#1 ` j.h . van vleck , phys . rev . * 33 * , 467 ( 1929 ) . jordahl , phys . rev . * 45 * , 87 ( 1934 ) . shaffer , h.h . nielsen and l.h . thomas , phys . rev . * 56 * , 895 ( 1939 ) . nielsen , rev . * 23 * , 90 ( 1951 ) . c. bloch , nucl * 6 * , 329 ( 1958 ) . g. amat , h. h. nielsen , _ rotation - vibration of polyatomic molecules _ ( dekker , new york , 1971 ) . jrgensen , f and t. pedersen , mol . * 27 * , 33 ( 1974 ) . s. califano , _ vibrational states _ ( john wilay and sons , new york , london , 1976 ) . tyuterev and v.i . perevalov , chem . phys . lett . * 74 * , 494 ( 1980 ) . klein , j. chem . phys . * 61 * , 786 ( 1974 ) . mills , j. mol . spectrosc . * 5 * , 334 ( 1961 ) . mills , j. mol . spectrosc . * 17 * , 164 ( 1965 ) . v. barone , j. chem . phys . * 122 * , 014108 ( 2005 ) . v. barone , j. bloino , c.a . guido and f. lipparini , chem . lett . * 496 * , 157 ( 2010 ) . gribov , opt . spectrosc * 31 * , 456 ( 1971 ) . i. suzuky , bull . japan * 44 * , 3277 ( 1971 ) . bucknell , n.c . handy and s.f . boys , mol . * 28 * , 759 ( 1974 ) . bucknell and n.c . handy , mol . * 28 * , 777 ( 1974 ) . r.j . whitehead and n.c . handy , j. mol . spectrosc . * 55 * , 356 ( 1975 ) . g.d . carney and c.w . kern , intern . j. quantum chem . * s9 * , 317 ( 1975 ) . r.j . whitehead and n.c . handy , j. mol . spectrosc . * 59 * , 459 ( 1976 ) . carney and r.n . porter , j. chem . phys . * 65 * , 3547 ( 1976 ) . carney , s.r . langhoff and l.a . curtiss , j. chem . phys . * 66 * , 3724 ( 1977 ) . carney and r.n . porter , chem . . lett . * 50 * , 327 ( 1977 ) . carney , s. giorgianni and k.n . rao , j. mol . spectrosc . * 80 * , 158 ( 1980 ) . carney and r.n . porter , phys . * 45 * , 537 ( 1980 ) . s. carter and n.c . handy , mol . phys . * 47 * , 1445 ( 1982 ) . j. tennyson and b.t . sutcliffe , j. chem . phys . * 77 * , 4061 ( 1982 ) . g. brocks and j. tennyson , j. mol . spectrosc . * 99 * , 263 ( 1983 ) . s. carter , n. handy and b.t . sutcliffe , mol . phys . * 49 * , 745 ( 1983 ) . j. tennyson and b.t . sutcliffe , j. chem . phys . * 79 * , 43 ( 1983 ) . d. cropek and g.d . carney , j. chem . phys . * 80 * , 4280 ( 1984 ) . j. tennyson and b.t . sutcliffe , mol . phys . * 51 * , 887 ( 1984 ) . j. tennyson , comput . commun . * 29 * , 307 ( 1983 ) . j. tennyson , comput . * 42 * , 257 ( 1986 ) . j. tennyson and s. miller , comput . commun . * 55 * , 149 ( 1989 ) . s. carter , p. rosmus , n.c . handy , s. miller , j. tennyson and b.t . sutcliffe , comput . . commun . * 55 * , 71 ( 1989 ) . light , i.p . hamilton and j.v . lill , j. chem . * 82 * , 1400 ( 1985 ) . z. bai and j.c . light , annu . chem . * 40 * , 469 ( 1989 ) . light and t. carrington jr , adv . phys . chem . * 114 * , 263 ( 2000 ) . kozin , m.m . law , j. tennyson and j.m . hutson , comput . commun . * 163 * , 117 ( 2004 ) . yurchenko , w. thiel and p. jensen , j. mol . spectrosc . * 245 * , 126 ( 2007 ) . p. botschwina , chem . phys . * 40 * , 33 ( 1979 ) . p. botschwina , h. haertner and w. sawodny , chem . lett . * 74 * , 156 ( 1980 ) . p. botschwina , chem . phys . * 68 * , 41 ( 1982 ) . p. botschwina , chem . phys . * 81 * , 73 ( 1983 ) . p. botschwina and p. sebald , j. mol . spectrosc . * 100 * , 1 ( 1983 ) . ponomarev and a.i . pavlyuchko , optika i spektroskopiya * 44 * , 1262 ( 1978 ) . yukhnevich , e.g. kokhanova , a.i . pavlyuchko and v.v . volkov , j. molec . ( theochem ) * 23 * , 1 ( 1985 ) . pavlyuchko and l.a . gribov , optika i spektroskopiya * 58 * , 1247 ( 1985 ) . pavlyuchko , optika i spektroskopiya * 67 * , 286 ( 1989 ) . pavlyuchko , j. struct . chem . * 36 * , 204 ( 1995 ) . pavlyuchko , j. struct . chem . * 36 * , 210 ( 1995 ) . pavlyuchko , j. struct . chem . * 38 * , 959 ( 1997 ) . vigasin , f. huisken , a.i . pavlyuchko , l. ramonat and e.g. tarakanova , j. mol . spectrosc . * 209 * , 81 ( 2001 ) . vigasin , a.i . pavlyuchko , y. jin and s. ikawa , j. mol . struct . * 742 * , 173 ( 2005 ) . pavlyuchko and l.a . gribov , j. struct . chem . * 51 * , 444 ( 2010 ) . m. pavanello , l. adamowicz , a. alijah , n.f . zobov , i.i . mizus , o.l . polyansky , j. tennyson , t. szidarovszky , a.g . csszr , m. berg , a. petrignani and a. wolf , phys . rev . lett . * 108 * , 023002 ( 2012 ) . j. tennyson and b.t . sutcliffe , mol . phys . * 58 * , 1067 ( 1986 ) . s. miller and j. tennyson , chem . * 145 * , 117 ( 1988 ) . h.y . mussa and j. tennyson , j. chem . * 109 * , 10885 ( 1998 ) . gerber and m.a . ratner , chem . . lett . * 68 * , 195 ( 1979 ) . bowman , acc . chem . res . * 19 * , 202 ( 1986 ) . gerber and m.a . ratner , adv . phys . * 70 * , 97 ( 1988 ) . s. carter , j.m . bowman and n.c . handy , theor . acc . * 100 * , 191 ( 1998 ) . handy and s. carter , mol . phys . * 102 * , 2201 ( 2004 ) . y. scribano and d.m . benoit , chem . . lett . * 458 * , 384 ( 2008 ) . y. scribano , d.m . lauvergnat and d.m . benoit , j. chem . phys . * 133 * , 094103 ( 2010 ) . bowman , s. carter and x.c . huang , intern . j. quantum chem . * 22 * , 533 ( 2003 ) . senent , p. palmieri , s. carter and n.c . handy , chem . lett . * 354 * , 1 ( 2002 ) . c. fabri , t. furtenbacher and a.g . csszr , mol . phys . * 112 * , 2462 ( 2014 ) . e. mtyus , g. czak , b.t . sutcliffe and a.g . csszr , j. chem . phys . * 127 * , 084102 ( 2007 ) . watson , mol . phys . * 15 * , 479 ( 1968 ) . gribov and a.i . pavlyuchko , _ variational methods for solving anharmonic problems in the theory of vibrational spectra of molecules _ ( nauka , moscow , 1998 ; ( in russian ) ) , ( in russian ) . srensen , in _ topics in current chemistry _ , edited by m. j. s. dewar et al , vol . 82 ( springer - verlag , dordrecht , 1979 ) , p. 99 . yurchenko , w. thiel and p. jensen , j. mol . spectrosc . * 245 * , 126 ( 2007 ) . f. gatti and c. iung , phys . * 484 * , 1 ( 2009 ) . j. tennyson , p.f . bernath , l.r . brown , a. campargue , m.r . carleer , a.g . csszr , l. daumont , r.r . gamache , j.t . hodges , o.v . naumenko , o.l . polyansky , l.s . rothmam , a.c . vandaele , n.f . zobov , a.r . al derzi , c. fbri , a.z . fazliev , t. furtenbacher , i.e. gordon , l. lodi and i.i . mizus , j. quant . . transf . * 117 * , 29 ( 2013 ) . rothman , i.e. gordon , y. babikov , a. barbe , d.c . benner , p.f . bernath , m. birk , l. bizzocchi , v. boudon , l.r . brown , a. campargue , k. chance , e.a . cohen , l.h . coudert , v.m . devi , b.j . drouin , a. fayt , j.m . flaud , r.r . gamache , j.j . harrison , j.m . hartmann , c. hill , j.t . hodges , d. jacquemart , a. jolly , j. lamouroux , r.j . le roy , g. li , d.a . long , o.m . lyulin , c.j . mackie , s.t . massie , s. mikhailenko , h.s.p . mller , o.v . naumenko , a.v . nikitin , j. orphal , v. perevalov , a. perrin , e.r . polovtseva , c. richard , m.a.h . smith , e. starikova , k. sung , s. tashkun , j. tennyson , g.c . toon , v.g . tyuterev and g. wagner , j. quant . . transf . * 130 * , 4 ( 2013 ) . pavlyuchko , s.n . yurchenko and j. tennyson , j. chem . ( 2014 ) , ( submitted ) . jacobi , crelle s j. * 30 * , 51 ( 1846 ) , ( in german ) . wilkinson , _ the algebraic eigenvalue problem _ ( oxford university press , oxford , uk , 1965 ) . k.j . feierabend , d.k . havey , m.e . varner , j.f . stanton and v. vaida , j. chem . phys . * 124 * , 124323 ( 2006 ) . yurchenko , r.j . barber , a. yachmenev , w. thiel , p. jensen and j. tennyson , j. phys . chem . a * 113 * , 11845 ( 2009 ) . j. tennyson and s.n . yurchenko , mon . not . r. astron . soc . * 425 * , 21 ( 2012 ) . c. eckart , phys * 47 * , 552 ( 1935 ) . nielsen , intern . j. quantum chem . * 1 * , 217 ( 1967 ) . s. miller , j. tennyson and b.t . sutcliffe , j. mol . spectrosc . * 141 * , 104 ( 1990 ) . s.n . yurchenko and j. tennyson , mon . not . r. astron . soc . * 440 * , 1649 ( 2014 ) . light , r.m . whitnell , t.j . pack and s.e . choi , in _ supercomputer algorithms for reactivity , dynamics and kinetics of small molecules _ , edited by a. lagan , _ nato asi series c _ , vol . 277 ( kluwer , dordrecht , 1989 ) , p. 187 barber , j. tennyson , g.j . harris and r.n . tolchenov , mon . not . r. astron . soc . * 368 * , 1087 ( 2006 ) . sharpe , t.j . johnson , r.l . sams , p.m. chu , g.c . rhoderick and p.a . johnson , appl . spectrosc . * 58 * , 1452 ( 2004 ) . pavlyuchko , a. yachmenev , j. tennyson and s.n . yurchenko 2014 , ( work in progress ) . .fully variational calculation of the vibrational band origins ( upper row , @xmath34 ) and vibrational band intensities ( lower row , km / mole ) for h@xmath32o as a function of total vibrational excitation number , @xmath191 , and number of basis functions , @xmath21 [ cols= " < , > , > , > , > , > , > , > , > , > , > , > , > " , ] our rotation - vibration hamiltonian written in curvilinear internal coordinates and an eckart embedding has the form @xcite @xmath192 where @xmath193 are the rotational coordinates and @xmath194 is the vibrational part of the hamiltonian @xmath195 @xmath196 here @xmath197 are internal , vibrational curvilinear coordinates given by changes in bond lengths or changes in the valence bond angles , dihedral angles , etc . ; @xmath198 , @xmath199 , @xmath200 are the euler angles between the axes of the equilibrium moment of inertia tensor and external cartesian coordinate axes ; @xmath201 are elements of the inverse of the moment of inertia tensor , @xmath202 ; @xmath203 is the vibrational kinetic energy operator and @xmath204 are elements of the kinetic energy coefficients matrix @xmath205 and @xmath206 $ ] . finally , @xmath207 is the molecular potential energy . after transformation , the vibrational kinetic operator can be written as @xmath208 where @xmath209\right\ } , \end{split}\ ] ] is the so - called pseudo - potential @xcite ; @xmath210 are elements of @xmath211 . @xmath205 is a complicated function of the internal coordinates , with the elements in general case given elsewhere @xcite . if the coordinates associated with both indices @xmath151 and @xmath152 represent changes in the bond lengths then @xmath212 if the coordinate @xmath151 represents a change in the bond length and @xmath152 is an angular coordinate then @xmath213 and if @xmath151 and @xmath152 both represent angular coordinates it becomes @xmath214 in these expressions @xmath215 is bondlength of the @xmath126-th bond and @xmath216 represents the angular coordinates . the computation of @xmath205 is achieved using a second - order taylor expansion in the angular coordinates @xmath217 @xmath218 ~~~,\ ] ] @xmath219 ~.\ ] ] these formulae are exact as @xmath220 , @xmath221 and @xmath222 are quadratic functions of angles when the angular coordinates . the @xmath216 are represented as cosine differences @xmath223 , where @xmath224 is the instantaneous equilibrium angle for bond angles , and sine differences for dihedral angles . to calculate the vibrational hamiltonian matrix elements in block 2 , which give the perturbative contribution to block 1 , the vibrational kinetic energy coefficients are expanded in the polynomial form and truncated after the second order @xmath225 this form is convenient because it allows faster ( by an order - of - magnitude or more ) computation of the coefficients without significant loss of accuracy . the potential energy function used by us is a fourth - order polynomial @xmath226 where a morse transformation , @xmath227 , is used to represent changes in terminal bonds such as x h and @xmath228 for angular coordinates and the coordinates of skeletal changes in the bond lengths . in variational calculations of vibrational energy levels , the hamiltonian matrix elements @xmath229 are computed using the product form @xmath230 of the basis functions , which are eigenfunctions of the morse or harmonic oscillators . morse oscillator functions , @xmath231 , are used for the stretching coordinates , @xmath232 , for which the potential is given using a morse transformation . harmonic basis functions , @xmath233 , are used for the other coordinates which are represented using the curvilinear normal coordinates @xmath234 expressed as a linear sum over the internal coordinates , @xmath197 , for which the potential function is defined as a taylor series . the coordinates @xmath235 are those which diagonalize the harmonic part of the hamiltonian given in the internal coordinates @xmath197 . with these definitions , all multidimensional integrals required to calculate the hamiltonian matrix elements are separated into products of one - dimensional integrals between either morse functions or harmonic oscillators . all these integrals have a simple analytic form which results in high - speed computation of the hamiltonian matrix elements . diagonalizing the hamiltonian matrix gives the anharmonic vibrational energy levels @xmath236 and the corresponding wave functions @xmath237 . for ro - vibrational energy levels it is necessary to calculate elements of the complex hermitian hamiltonian matrix @xmath238 using the variational basis functions @xmath239 where @xmath240 is a ( complex ) wigner function . in this case @xmath241 this expression is equivalent to @xmath242 where @xmath243 and @xmath244 to calculate @xmath245 , the matrix elements @xmath201 are expanded to the second - order as a taylor series @xmath246 where @xmath247 @xmath248 ~.\ ] ] in this case @xmath249 @xmath250 and all the integrals reduce to products of one - dimensional integrals over either morse or harmonic oscillators . when analyzing the off - diagonal elements of the vibration - rotation hamiltonian matrix @xmath251 it can be seen that they differ significantly in magnitude , depending on whether they are diagonal in the vibrations , @xmath252 , or couple different vibrational states , @xmath253 . for semi - rigid molecules with small values of the vibrational quantum numbers the following condition usually holds @xmath254 this results from the slight change in the effective geometry of the molecule upon vibrational excitation @xmath255 and @xmath256 when calculating the vibrational - rotational energy levels , the off - diagonal elements @xmath257 corresponding to different vibrational states @xmath253 give a much smaller contribution ( change in the diagonal elements in the block that will be diagonalized ) to the calculated energy levels than the non - diagonal elements @xmath258 within the vibrational state in question . these changes are given approximately by @xmath259 @xmath260 however , @xmath261 involves only a change in the rotational energy level , while @xmath262 involves also a change in the vibrational energy level .
we present a detailed comparison between far - uv / optical colour magnitude diagrams obtained with high - resolution _ hubble space telescope _ data and suitable theoretical models for three galactic globular clusters : m3 , m13 and m79 . + these systems represents a `` classical '' example of clusters in the intermediate metallicity regime that , even sharing similar metal content and age , show remarkably different horizontal branch morphologies . as a consequence , the observed differences in the colour distributions of horizontal branch stars can not be interpreted in terms of either first ( metallicity ) or a second parameter such as age . + we investigate here the possible role of variations of initial helium abundance ( ) . thanks to the use of a proper setup of far - uv filters , we are able to put strong constraints on the maximum y ( ) values compatible with the data . we find differences between the clusters with m13 showing the largest value ( ) and m3 the smallest ( ) . in general we observe that these values are correlated with the colour extensions of their horizontal branches and with the range of the observed na - o anti - correlations . [ firstpage ] globular clusters : individual ( m3 , m13 , m79 ) ; stars : evolution horizontal branch ; ultraviolet : stars
1212.4419
horizontal branch ( hb ) stars are the progeny of low - mass red giant branch stars ( rgb ) burning helium in their cores and hydrogen in a shell around it ( hoyle & schwarzschild 1955 ) . as first noticed by iben & rood ( 1970 ) , the different hb star colour distributions observed in old stellar systems , is the reflection of the amount of mass lost during the rgb phase . + the scientific community agrees from nearly fifty years about the fact that the principal parameter governing the shape of hbs in galactic globular clusters ( ggcs ) is metallicity . the general rule is that metal - rich systems have red hbs , while in the metal - poor ones stars are distributed on average at higher effective temperatures ( bluer colours ) . several exceptions have come out during the last decades ; remarkable cases the cases of ngc6388 and ngc6441 ( rich et al . 1997 ) , which despite their metallicity ( @xmath5\sim-0.6 $ ] ) show some of the bluest hbs known among ggcs ( busso et al . 2007 ; dalessandro et al . moreover several clusters , sharing similar metal content , reveal different hb morphologies , typical cases being the pairs ngc5927 - ngc6388 at high metallicities ( @xmath5\sim-0.4 $ ] ) , m3 - m13 at intermediate metallicity regime ( @xmath5\sim-1.5 $ ] ; ferraro et al . 1997 ) and m15 - m92 at low metallicities ( @xmath5\sim-2.3 $ ] ) . + these noticeable exceptions have required the introduction of a second ( freeman & norris 1981 ) and possibly a third parameter in order to explain the hb distributions in all ggcs . what we can call now the _ `` i - th parameter problem '' _ is still a hot topic , as stressed by several authors , we recall the reader to catelan 2009 for a nice review ( see also dotter et al . 2010 and gratton et al . 2010 ; hereafter d10 and g10 respectively ) . + an accurate knowledge of the physical parameters playing a role in shaping the hb is extremely important also for an appropriate interpretation of distant unresolved stellar populations . in fact it is well known that the hb morphology can have a strong impact on the integrated light of stellar populations , affecting colours and line indices ( lee et al . 2002 ; schiavon et al . 2004 ; percival & salaris 2011 ; dalessandro et al . 2012 ) . + despite the huge efforts made to address this problem , its solution is not obvious and still different scenarios are proposed . one of the reasons that complicates the identification of the mechanisms other than metallicity at work in shaping the observed luminosity and effective temperature distribution of stars along the hb is that there are many possible culprits ( mass - loss , age , helium abundance ... ; see rood 1973 for example ) and some of them are not well constrained from theory . + age has been identified as the natural global second parameter by many authors in the past years ( lee et al . 1987 , 1988 , 1990 ; lee , demarque & zinn 1994 ; sarajedini & king 1989 ) . according to this interpretation older clusters tend to have bluer hbs , while younger ones should have on average redder hb morphologies . this scenario appeared in agreement with the picture for the galaxy formation and its early evolution ( searle & zinn 1978 ; zinn 1985 ) . by means of high resolution hst data for a large sample of ggcs , d10 found that the existence of outer halo gcs with anomalously red hbs fits well the scenario in which age is the second parameter . in fact , the behaviour of the 4 - 5 relatively younger clusters in their sample could be reproduced in term of correlation between age and hb morphology , while the bulk of the analyzed targets is peaked around old ages ( see ages reported by salaris & weiss 2002 , g10 , d10 ) and does nt show any obvious correlation . also results by g10 agree on the fact that age is the second main parameter driving the hb morphology . + it is also worth noticing that most of these results are based on optical cmds and hb morphology parameters ( like the well known hbr from lee et al . 1994 ) , which tend to minimize the importance of blue tails . on the contrary using proper combinations of ultra - violet ( uv ) and optical filters has an important impact both in term of hb classification and comparison with theoretical models . still , age is not able to explain exhaustively the hb morphology . detailed cluster to cluster comparisons have shown that there are systems with similar iron content and age , but remarkably different hb morphologies . a clear example is given by the three clusters m3 - m13 - m80 , as shown by ferraro et al . ( 1997 , 1998 ) and at present there is hardly a scenario able to give a satisfactory explanation for their different morphologies . + as suggested by buonanno et al . ( 1985 ) and fusi pecci et al . ( 1993 ) , age might be one of many and probably the most important hb second - parameter , but not the only one . buonanno et al . ( 1995 ) argued that it is not possible to reproduce the complex hb zoology with a single `` global '' parameter , but more likely we can explain it by thinking of a `` global combination '' of `` non - global '' quantities and phenomena related to the formation , chemical and dynamical evolution of each cluster . + the necessity of at least a third parameter transpires also from d10 and g10 analyses , in the form of either the luminosity cluster density or stellar density ( @xmath6 ) as already suggested by fusi pecci et al . ( 1993 ) which might correlate with the hot extension of the hbs , or a variation of the initial helium abundance ( y ) , respectively . + indeed dantona et al . ( 2005 ) and dalessandro et al . ( 2011 hereafter paperi ) have shown that for ngc2808 the main parameter that determines the hb morphology is y. in particular in paperi we have been able to satisfactory reproduce the cluster complex hb morphology by assuming three different sub - populations with he abundances compatible with what inferred from the multimodal main sequence ( ms ; piotto et al . 2007 ) and spectroscopic analyses ( bragaglia et al . 2010 ; pasquini et al . 2011 ) . it is worth noticing however that only few clusters show evidences of multiple sub - populations with different he abundances as inferred from their ms or spectroscopic analyses ( ngc2808 , @xmath7 centauri bedin et al . 2004 , ngc6752 milone et al . 2010 , ngc6397 milone et al . 2012 ; ngc1851 gratton et al . 2012 ) + in order to further investigate the _ `` i - th parameter problem '' _ , we selected a triplet of ggcs with similar metallicity ( @xmath5\sim-1.50 $ ] ) and age , namely ngc5272 ( m3 ) , ngc6205 ( m13 ) and ngc1904 ( m79 ) as templates for a comparative analysis of their hbs . these targets are basically the same used by ferraro et al . ( 1997 , 1998 ) , with the only exception of m80 , which have been replaced by m79 since recent spectroscopic analyses ( carretta et al . 2009 ) show that it is slightly more metal - poor than the others . as shown by ferraro et al . ( 1998 ) , m13 has a very extended hb with two clear gaps ( named g1 and g2 after their analysis ) ; while , the hb of m3 covers a much narrower extension in effective temperature and shows only a mild indication for a gap at the same @xmath8 as g1 in m13 . m79 represents an intermediate case between the two . + as already done for ngc2808 ( paper i ) we will use a combination of optical and uv photometric images . details about data reduction are presented in section 2 . in section 3 we will focus on the long - standing debate about the age - differences between these clusters . comparisons with theoretical models and cluster - to - cluster differences will be discussed in section 4 and 5 . the data - set used in the present work consists of hubble space telescope ( hst ) wide field planetary camera 2 ( wfpc2 ) images . all targets have been observed in the @xmath9 , @xmath10 and @xmath11 bands , the only exception is m3 for which no f160bw data are available . ( see table 1 ) . images and photometry obtained in other bands are also available for some of the clusters , however as shown in rood et al . 2008 and paper i the combination of @xmath9 , @xmath10 and @xmath11 seems to be the best filters setup to study both rhbs and stars lying at the very hot end of the hb . the photometric catalogues of m3 and m13 have been already partially presented by ferraro et al . ( 1998 ) , while the photometry of m79 has been already used by lanzoni et al . . moreover far - uv photometry of m79 and m13 has been presented by rood et al . ( 2008 ) in order to describe in detail the advantages of using the combination of @xmath11 and @xmath9 magnitudes for comparison with hb models . + + in all cases the photometric analysis has been performed by using romafot ( buonanno et al . the reduction strategy has been optimized to be particularly sensitive to hot stars , as already described by ferraro et al . for the present analysis attention has been paid to the cross - identification of stars detected in the @xmath11 images . stars observed only in these bands have been force - fitted in the other bands ( see paper i ) . + the instrumental magnitudes have been calibrated to the vegamag photometric system by using zero - points and the gain settings reported in table 5.1 of the wfpc2 _ _ data handbook manual__. magnitudes thus obtained have been also corrected for charge transfer efficiency ( cte ) effect by means of the prescriptions by dolphin ( 2000 ) and updated equations listed in the dedicated page of dolphin s web site . + the resulting colour magnitude diagrams ( cmds ) for each cluster are shown in both the ( @xmath12 , @xmath13 ) and ( @xmath14 , @xmath15 ) planes in figs [ cmdopt ] and [ cmdfuv ] . star lists have been reported to the absolute coordinate system by mean of proper roto - traslations . in particular , for each cluster the instrumental coordinates have been reported to the gscii astrometric standard system by using a large number of primary and secondary standards in common , consisting typically of wide field ground - based data - sets , in the wfpc2 fields of view ( fovs ) ( see lanzoni et al . 2007 for details ) . at the end of the procedure , the typical error of the astrometric solution is @xmath16 . the identification of variable stars lying along the hb is a crucial step to perform a suitable comparison between observations and theoretical models . this is particularly true in case of m3 , that hosts tens of rr lyrae stars . + however , none of the data - sets used here is designed for variability search , both because of the short time baseline covered and number of images acquired . in our cmds , variables are basically sampled at random phase . thus we decided to use publicly available catalogues of variable stars in order to cross - identify them in our photometric catalogues , as already done in paper i. we retrieved position , magnitude , and classification lists from the `` catalogue of variable stars in galactic globular clusters ( 2011 ) '' ( clement et al . we refer to that paper for all details and references . + we focused only on rrlyrae stars , neglecting all the other different variable types . we first roto - traslated the catalogues of variables to our reference frames by using cataxcorr , a software developed at the bologna observatory ( montegriffo , private communication ) . we then looked for stars in common with our photometry . all the known rrlyrae stars in our fovs have been recovered in our catalogues . we found 59 rrlyrae in m3 , 7 in m13 and 5 in m79 ( see table 1 ) . their positions in the optical cmds are shown in figure [ cmdopt ] . an accurate determination of metallicities and ages is a crucial point for hb morphology studies and for the correct interpretation of comparative analyses . + the three ggcs in our sample have a similar iron content : according to the recent reassessment of the gc metallicity scale by carretta et al ( 2009 ) , metallicities are equal to [ fe / h]=@xmath17 , @xmath18 and @xmath19 for m3 , m13 and m79 , respectively . consistent results were also found by sneden at al . ( 2004 ) who obtained [ fe / h]=@xmath20 and @xmath21 for m3 and m13 ( for homogeneity we report only values obtained from the lick hamilton spectra ) . only in the analysis by cohen & melendez ( 2005 ) m3 and m13 are slightly more metal - rich , i.e. [ fe / h]=@xmath22 and [ fe / h]=@xmath23 . in addition in this case the observed metallicity difference ( @xmath24[fe / h]@xmath25 ) is larger than the uncertainties . + gyr and @xmath26 gyr . the right panel shows synthetic cmds obtained from the adopted isochrones and the appropriate photometric errors.,width=370 ] on the contrary the determination of their ages is a more debated argument . in literature there is a quite general consensus that m3 and m79 are old and coeval ggcs . there is instead more uncertainty about the age of m13 . ferraro et al . ( 1997 ) put constraints on the possible difference in age between m13 and m3 by comparing their ridge mean lines obtained with the optical cmds used in this work , and theoretical models by dorman et al . the age difference resulted to be smaller than @xmath27 gyr . the same conclusions have been reached also by johnson & bolte ( 1998 ) and rosenberg et al . according to the analysis by grundahl et al . ( 1999 ) m13 may be younger than m3 by @xmath28 gyr , while for rey et al . ( 2001 ) it is older by about @xmath29 gyr . + ages reported by other authors for the clusters of our analysis agree quite well within the error bars . salaris & weiss ( 2002 ) found a typical difference smaller than @xmath30 gyr independently of the metallicity scale used . ages by d10 obtained using the isochrone - fitting method are basically identical ; in fact t=@xmath31 gyr and t=@xmath32 gyr for m3 and m13 respectively , while m79 is not in their sample . the same results were obtained by marin - franch et al . for g10 instead , m13 is older than m3 and m79 of about @xmath33 1.75 gyr . + it is worth noticing that even moderate differences of the average y among these three clusters do not affect appreciably the results of the methods employed to determine their relative ages ( see , e.g. , marin - franch et al . 2010 ) . + we investigated any possible age difference by comparing the ( @xmath34 ) cmds of the three clusters of our analysis . but for m3 and m13.,width=370 ] we used m13 as reference and shifted m79 and m3 according to the difference of @xmath35 values and distance moduli ( see sections 3.2 , 3.3 and 3.4 for details about the values adopted ) . as apparent from the upper panels of figure [ m13m79 ] and [ m13m3 ] , once relative differences in reddening and distance are taken into account , the cmds match well at the to level and along the rgb , consistently with the fact that they have almost the same age and share a similar metallicity . + in the bottom panels we focus on their turn off ( to ) region . also from a close inspection , the cmds have been found to nicely overlap at the to and sgb region . this already rules out any significant age difference between m13 , m3 and m79 . however in order to perform a more accurate comparison , we also used two isochrones of proper metallicity and different age from the basti database ( pietrinferni et al . in particular we used as reference the isochrone that best fits the observed cmd ( @xmath36 gyr ) and an additional one that is @xmath37 gyr older . as apparent from the left bottom panels of figures [ m13m79 ] and [ m13m3 ] , the older isochrone does not match either the to or the sgb . + starting from these isochrones , and using the estimated photometric errors we build two synthetic populations of similar size to the observed ones for each comparison . with the typical uncertainties of our hst catalogues at the to magnitude level , an age difference of @xmath37 gyr would be clearly visible and detectable . we then conclude that m13 , m3 and m79 differ in age by less than @xmath38 gyr . this result is in agreement with estimates by several authors ( johnson & bolte 1998 ; rosenberg et al . 1999 ; salaris and weiss 2002 ; marin - franch et al . 2009 ; d10 ) , while it is incompatible with the age differences proposed by g10 . it is worth noticing that an age difference of about @xmath39 gyr would be compatible with the observed data only assuming extremely different he abundances ( @xmath40 ) between these clusters . we have shown that m3 , m13 and m79 have similar metallicities and are coeval within typical uncertainties . therefore in this case , the scenario proposed by g10 and d10 , identifying the _ second parameter _ with age , is not applicable . we therefore investigate here the possible role of variations of helium abundances in the framework of gcs having experienced multiple formation bursts ( dercole et al . 2008 ; conroy & spergel 2011 ; valcarce & catelan 2011 ) on a short timescale ( @xmath41 myr ) in environments enriched by material polluted by agb stars ( ventura et al . 2002 ) or fast rotating stars ( decressin et al . 2007 ) . as highlighted before , helium has been proposed also by g10 to be the possible hb _ third parameter_. + as in paper i , we have compared our photometric data with a suitable set of basti @xmath42-enhanced hb tracks ( pietrinferni et al . we have considered theoretical models for [ fe / h ] = @xmath431.62 , which is the closest value available in the basti database to the clusters spectroscopically determined [ fe / h ] . in our analysis we take advantage of all initial he - abundances available in the basti database , i.e. y=0.246 , 0.300 , 0.350 and 0.400 . bolometric corrections and extinction effects have been calculated as in paper i. recently , gratton et al . ( 2011 ) have found spectroscopical evidence of a moderate na - o anti - correlation in stars belonging to the hb of ngc2808 , at @xmath44 below @xmath4512000 , that marks the onset of radiative levitation effects ( see discussion below ) . a similar result has been found for m 4 by marino et al . ( 2011 ) , the general trend being that the bluer hb stars are on average more na - rich and o - poor . as a consequence of these empirical results , we verified by performing a number of tests ( see appendix a ) how adequate is the use of standard @xmath42-enhanced spectra to determine bolometric corrections in this temperature regime . we found that results obtained by using standard @xmath42-enhanced bolometric corrections , are unaffected by the presence of the observed cnona anti - correlations . regarding the effect of radiative levitation , spectroscopic investigations by behr ( 2003 ) , moehler et al . ( 2003 ) , fabbian et al . ( 2005 ) have disclosed that - similar to the case of ngc2808 - in these three clusters hb stars hotter than @xmath4512000 k display surface abundances of metal species like fe , cr and ti around solar or higher , due to radiative levitation in the atmosphere . also , the surface he abundances are depleted by a factor of ten or more as a consequence of gravitational settling . following the approach used in paper i , we mimic this effect using standard @xmath42-enhanced hb stellar models by applying bolometric corrections appropriate for [ fe / h]=0.0 ( and scaled - solar mixture ) when @xmath44 is above 12000 k. this is of course a crude approximation , which we are forced to adopt because of the lack of extended grids of both hb stellar evolution and atmosphere models with a large range of chemical compositions , that include consistently the effect of radiative levitation . the underlying assumptions are that _ ( i ) _ the hb model evolutionary lifetimes , _ ( ii ) _ the evolution of @xmath44 and _ ( iii ) _ bolometric luminosity , and _ ( iv ) _ the structure of the corresponding model atmospheres are not affected by the radiative levitation , and that the relevant bolometric corrections are determined mainly by the enhanced metals . with these assumptions the effect of radiative levitation on bolometric corrections makes @xmath12 fainter by about 0.15 and @xmath14 and @xmath46 brighter by about 0.1 and 0.05 respectively . paper i has shown how this approximation allowed us to recover the three different he - abundances inferred by the multimodal ms , along the blue hb of ngc2808 . we are therefore confident that a similar procedure will be adequate to constrain the he distribution along the hb of these three clusters , that we are going to discuss separately in the following section . the theoretical analysis performed in this paper is based on synthetic hb calculations , performed as described in paper i. the analysis of ngc2808 was simplified by the fact that the initial he content of the cluster subpopulations is somewhat quantized , as shown by the observed trimodal ms ( piotto et al . in case of m13 , m3 and m79 there is no apparent quantization of the ms , and we need to consider the possibility of a continuous distribution of y along the hb . + we stress here that in these cases the modeling of the y variations along the hb becomes more uncertain since calculations require more assumptions . therefore the main goal of this work is not to attempt an exact description of the y distribution , that would be affected by our assumptions , but to understand whether helium plays a role in shaping these very different hbs . uv filters are much better suited than optical ones to address this issue . our synthetic hb calculations require the specification of at least 4 parameters , plus the cluster age , that we assume to be equal to 12 gyr ( see section 3 ) . these four parameters are the minimum value of y ( @xmath47 ) , the range of he abundances ( @xmath48 ) , the mean value of the mass lost along the rgb @xmath49 that in first instance ( and for simplicity ) we assume to be the same for each y and the spread around this mean value ( @xmath50 ) . throughout our analysis we will assume that y varies according to either a uniform probability distribution , from @xmath47 to @xmath51 , or a gaussian one . in this latter case @xmath48 is the 1@xmath52 spread around a prescribed mean value . also for the rgb total mass loss we will assume a gaussian distribution around the mean value @xmath49 or a uniform distribution with minimum value equal to @xmath49 the idea behind this type of simulations ( see , e.g. , caloi & dantona 2005 ) is that the colour extension of the hb is driven mainly by the variation of y rather than mass loss efficiency . a higher initial y implies a lower mass at the tip of the rgb ( for a fixed cluster age ) hence , for a fixed value of @xmath49 , a smaller mass along the hb and a bluer colour ( the he - core mass is affected by the variation of y to a smaller extent ) . our synthetic hb code first draws randomly a value of y , and determines the initial mass of the star at the rgb tip ( @xmath53 ) from interpolation among the basti isochrones of the prescribed age . the mass of the corresponding object evolving along the hb ( @xmath54 ) is then given by @xmath55 plus a gaussian random perturbation @xmath50 ( or a uniform probability between @xmath49 and @xmath49+@xmath50 ) . the wfpc2 magnitudes of the synthetic star are then determined according to its position along the hb track with appropriate mass and y obtained by interpolation among the available set of hb tracks after an evolutionary time t has been randomly extracted . we determined t assuming that stars reach the zahb at a constant rate . we employed a flat probability distribution ranging from zero to @xmath56 , where @xmath56 denotes the time spent from the zahb to the he - burning shell ignition along the early agb . the value of @xmath56 is set by the mass with the longest lifetime ( the lowest masses for a given y and z ) . this implies that for some synthetic object the randomly selected value of t will be longer than its @xmath56 or , in other words , that they have already evolved to the next evolutionary stages . finally , the derived synthetic magnitudes are perturbed with a gaussian 1@xmath52 error determined from the data reduction procedure . we detail now the effects on these results of possible variations of age and chemical composition within the uncertainties described in section 3 . the mass distribution along the hb ( as derived in our analysis ) does not depend on the cluster age , but changing the cluster age changes the mass loss necessary to reproduce the hb mass distribution . an increase of age of @xmath57 gyr leads to a decrease of the total mass loss along the rgb by @xmath58 ( because of the decreased evolving mass ) . + one can simulate the effect of a variation of [ fe / h ] , [ @xmath42/fe ] and/or their sum by considering that at these metallicities scaled solar and @xmath42-enhanced models ( also for the hb ) with the same total global metallicity [ m / h ] are coincident ( salaris et al . 1993 ) . we have therefore considered a change @xmath59=+0.3 $ ] dex that corresponds to changes of [ fe / h ] or [ @xmath42/fe ] or [ fe / h]+[@xmath42/fe ] . we have two effects here . the first one is on the derived mass distribution ( at fixed y ) along the hb . for the redder part of the hb we use the ( @xmath14,@xmath15 ) colour . at fixed colour in the horizontal part of the ( @xmath60 ) cmd , an increase @xmath24[m / h]=0.3 decreases the derived value of @xmath54 by @xmath450.04 @xmath61 . for the blue tails we use the ( @xmath62 ) fixed ( @xmath63 ) an increase @xmath24[m / h]=0.3 decreases the derived value of @xmath54 by @xmath64 @xmath61 . this means that if a larger mass loss is needed to explain the blue hb stars ( compared to the mass loss needed to explain the redder population ) , an increase of [ m / h ] tends to reduce this mass loss difference . the second effect is on the rgb mass ( hence on the absolute value of the rgb mass loss ) . an increase @xmath24[m / h]=0.3 causes an increase of @xmath65 by @xmath66 @xmath61 . the distribution of y along the observed hbs is basically unaffected by the precise age and [ m / h ] , because it is based on the variation of magnitude ( in @xmath14 and @xmath12 , depending on the star colour ) with y , a differential property that is unaffected by the exact model metallicity and age . in this section we will discuss separately the theoretical analysis of our three selected clusters . we stress again that allowing y to vary within individual clusters , increases the number of free parameters needed to match the observed hb star distribution in the cmd . leaving aside for a moment the open problem of the rgb mass loss , the distribution of y among stars fed onto the hb will depend on the convolution of the unknown ( at least for these three clusters ) initial y distribution among ms stars , with the y - dependent evolutionary timescales along ms and rgb predicted by stellar models . as discussed before , in absence of constraints from the ms , we assumed here as a first order approximation either gaussian or uniform probability distribution for y among hb stars . as a consequence , the derived distribution of y and mass range ( at fixed y ) are most probably just a reasonable approximation to the real ones . a much more solid result is the total range of y covered by the stars in each individual cluster , because thanks to our uv cmds this is strongly constrained by the observed stellar magnitudes @xmath12 ( when available ) . ) cmd of m13 hb . zahb sequences for y=0.246 ( solid line ) , 0.300 ( dashed line ) and 0.350 ( dot - dashed line ) are also displayed . the rectangular box marks the rhb region ( see text for details).,width=317 ] spectroscopic analyses of this cluster reveal a well developed na - o anti - correlation ( sneden et al . 2004 , carretta et al . 2009 ) , comparable to the case of ngc2808 , but photometric analyses currently available do not show evidences for a multimodal ms . here we analyze our hb photometry with as stated before the main aim of constraining the range of initial he values of the cluster stars . the use of the f160bw filter not considered in the investigations by caloi & dantona ( 2005 ) and dantona & caloi ( 2008 ) will set strong constraints on the maximum value of y. the first step of our analysis consists in defining the highest possible y value necessary to fit the observed hb . to this purpose we calculated a synthetic hb with this preliminary set of parameters ( uniform distributions for y and gaussian for m ): @xmath47=0.246 , @xmath48=0.104 , @xmath49=0.21@xmath67 , @xmath50=0.025@xmath67 . even though this selection of parameters is not optimized for a perfect fit to the observed stellar distribution along the hb , it will suffice to put a first general constraint to the maximum possible range of initial y. the number of stars in this simulation is higher than the observed sample , to reduce the effect of statistical number fluctuations on the synthetic hb . the parameter @xmath48 has been arbitrarily fixed to reach the maximum he mass fraction y=0.35 . once @xmath47 is chosen , the mean rgb total mass loss @xmath49 is constrained by matching the red boundary of the observed hb ( the subpopulation with the lowest y has the largest hb mass at fixed age and @xmath49 ) . the value of @xmath50 then introduces a general dispersion in the relationship between hb colour and local he abundance . the smaller the value of this parameter , the narrower the range of y in a given colour bin along the hb . we notice that a very small value of @xmath50 ( @xmath50=0.005 for example ) , would produce a sloped hb ( brighter towards bluer colours because of a very tight colour - y relationship ) in the ( @xmath14 , @xmath15 ) cmd at colour ( @xmath15)@xmath68 0.5 , instead of the observed roughly horizontal structure ( see figure [ synthm13_1 ] ) . but for the ( @xmath62 ) cmd ( see text for details ) . in the right panel we also show two selected hb tracks corresponding to @xmath69 and @xmath70 for each zahb . , width=317 ] as done in paper i , we determined the cluster distance modulus by fitting the synthetic data to the observed red hb ( rhb ) ( defined here as the region of the hb contained in the square box plotted in figure [ synthm13_1 ] ) in the ( @xmath71 ) plane . notice how the rhb is roughly horizontal in this cmd , and allows us to estimate straightforwardly the cluster distance by comparing the synthetic data to the observed number distribution of objects in @xmath14 , assuming a reddening e(b@xmath43v)=0.02 ( see for example ferraro et al . a fit to both the lower envelope defined as the faintest magnitude bin where the star counts are at least 2@xmath52 above zero of the observed binned distribution in @xmath14 , and the mean value of the @xmath14 rhb magnitudes ( @xmath72=15.26 ) provides @[email protected] , in fair agreement within the uncertainties with ferraro et al . ( 1999 ; @xmath75 ) . the left and right panels of figure [ synthm13_1 ] show the observed and simulated ( @xmath71 ) cmd , respectively . we have also over - imposed three zahbs for y=0.246 ( solid line ) , 0.300 ( long dashed line ) and 0.350 ( dot - dashed line ) . figure [ synthm13_2 ] is equivalent to figure [ synthm13_1 ] , but for the ( @xmath76 ) cmd . whilst in the ( @xmath77 ) cmd zahb sequences and synthetic stars with differing initial y are largely degenerate along the bluest hb tail as a consequence of the huge increase of the bolometric correction in the optical filters , this same tail appears as a roughly horizontal sequence at ( @xmath63)@xmath78 in the ( @xmath62 ) cmd . theoretical zahb sequences display here different shapes at varying y ( see also pietrinferni et al . it is clear from a comparison of the observed and synthetic hb that y=0.35 for the bluest stars is too high . the lower envelope of the observed stellar distribution lies around the y=0.30 zahb . this implies that the maximum y ( @xmath79 ) is @xmath450.30 . as a conclusion , in the hypothesis that the colour extension of the hb is due to a range of initial y values , @xmath47=0.246 at the red end of the observed hb implies @xmath800.30 at the blue end , i.e. the initial he - abundance spans a range @xmath810.05 this constraint on @xmath79 is very general ; the critical discriminating factor is provided by the @xmath12 magnitude of the zahb at the blue end of the hb . but for one realization of the best fit synthetic hb model discussed in the text.,width=317 ] one can of course try a more accurate fit to the stellar distribution along the observed hb by comparing theoretical and observed histograms of star counts as a function of colour and magnitude . in this case a fine tuned calibration of the set of free parameters is necessary . in the assumption ( for simplicity ) of either uniform or gaussian probability distributions for the mass loss and initial y , one needs to split the simulation into three different sections , because no single continuous distribution of y and mass allows to match the observed cmds . as a general method - applied also to m3 and m79 - , for each choice of the free parameters we have produced 100 synthetic distributions , each of them with the same number of objects as in the observational sample . from this ensemble of simulations we have determined the mean value and associated error for each colour or magnitude bin used in their comparison with data . the star counts along the rhb on the red side of the gap at ( @xmath15)@xmath82 0.5 are best fit with @xmath47=0.246 , @xmath48=0.01 and uniform probability distribution for y , @xmath49=0.21@xmath67 , with a gaussian spread @xmath50=0.01@xmath67 . the resulting distance modulus is @[email protected] , quite similar to the value obtained from the simulation in figure [ synthm13_1 ] . the second section of the simulation covers the region from the blue boundary of the gap to a colour ( @xmath83)@xmath82 3.0 along the hb blue tail . in this case one needs a gaussian distribution of y , with mean value @xmath84=0.285 and 1@xmath52 spread @xmath85=0.012 , mass loss @xmath49=0.235 @xmath67 with a gaussian spread @xmath50=0.01@xmath67 . the last section of the simulation covers the extreme tail of the hb , that is best reproduced with @xmath84=0.300 with a gaussian 1@xmath52 spread @xmath85=0.003 , and @xmath49=0.266 @xmath67 with a gaussian spread @xmath50=0.002@xmath67 . figure [ synthm13_3 ] and figure [ synthm13_4 ] display a comparison between the observed hb in the ( @xmath60 ) and ( @xmath62 ) cmds , and one realization of the simulations with the parameter choice described before . from the parameters of this simulations , it is clear that a good match to the observed star counts can be obtained only by relaxing the assumption of constant @xmath49 for all initial he contents . in particular , it appears necessary to slightly increase on average @xmath49 when y increases , as already highlighted in paper i for the case of ngc2808 . however it is important to note that , the maximum y range is fully constrained independently of the assumptions on mass loss . note that the slight increase of @xmath49 with @xmath86 could be in principle due to an age variation at increasing @xmath86 . in this scenario the variation of @xmath49 corresponds to a variation of @xmath53 with y , keeping the total rgb mass loss unchanged . however , these mass differences imply an increase of age with increasing y ( lower mass evolving along the rgb ) of the order of @xmath87 that is hard to justify . in fact such a large age difference would be detectable from the cluster to and sub - giant branch regions , at odds with the empirical evidence ( figure [ m13m79 ] and figure [ m13m3 ] ) . the mean he abundance along the hb of this simulation is @xmath84=0.270 , with a 1@xmath52 spread @xmath88=0.02 . notice that salaris et al . ( 2004 ) determined for this cluster [email protected] from the r - parameter , consistent with the results of the simulation . but for one realization of the best fit synthetic hb model discussed in the text.,width=317 ] we finally address the question of what happens if @xmath47 in rhb stars is increased . as a test , we have first considered a constant y=0.280 for the rhb as proposed by caloi & dantona ( 2005 ) and dantona & caloi ( 2008 ) and produced the corresponding synthetic population by tuning appropriately the value of the mass range along the rhb . the distance modulus derived from the @xmath14 magnitudes of rhb stars is 0.14 mag larger than the case of our simulation with @xmath47=0.246 . given that the @xmath12 magnitude along the extreme blue tail of the observed hb increases with increasing y ( see figure [ synthm13_2 ] ) , this higher distance modulus would imply [email protected] along the extreme hb ( ( @xmath63 ) @xmath893.5 ) , i.e. a value lower than along the rhb , a result justifiable only assuming a strong anti - correlation between mass loss and y. + this test also implies that in principle , for a fine tuned @xmath47 intermediate between 0.246 and 0.280 , the whole hb of m13 can be also modeled at least from the point of view of the magnitude levels in @xmath14 and @xmath12 employing a single value of y. we find that when @xmath900.265 , the ( @xmath60 ) and ( @xmath62 ) observed cmds and star counts can be reproduced with a constant he abundance , for a total mass spread of @xmath450.14 @xmath67 . even if we can not discard this scenario on the basis of the hb analysis , we consider it very unlikely . in fact it would mean that m13 lost completely its first generation of standard he stars , while preserving most of the he - enriched population . this would be hardly compatible with results from n - body simulations ( see for example decressin , baumgardt & kroupa 2008 ) . ) cmd of m3 . the zahb sequences for y=0.246 and y=0.300 are also displayed . the rectangular box marks the region considered for the fit ( see text for details).,width=317 ] [ [ m3 ] ] m3 ~~ spectroscopic analyses of m3 rgb stars reveal a na - o anti - correlation that is reduced in extension compared to m13 ( see , e.g. , sneden et al . 2004 , johnson et al . 2005 and references therein ) . also in this case , the currently available photometric data do not show evidence of a multimodal ms . also , as well known , the hb of m3 displays a much shorter colour extension compared to m13 . here we discuss its properties only in the ( @xmath60 ) cmd , because no @xmath11 images are available for this cluster . given the way in which hb stars are distributed in all cmds involving @xmath91 , it does nt give any additional information respect to the optical plane . also given the relatively short extension of the hb of m3 , we decided not to use the shorter wavelength @xmath91 filter . + we considered in our analysis all objects within the rectangular box in figure [ synthm3_1 ] . our synthetic calculations show that all these stars have @xmath44 below the threshold for the onset of radiative levitation . there are three stars outside the box , that may be genuine hb objects , and appear spread along the bluest tail of the cluster hb , that is vertical also in this cmd . given their extremely small number compared to the total sample of hb stars , and the fact that their location does not provide any further constraint on the cluster he distribution , we do not consider them in the theoretical analysis . following the same philosophy as for the case of m13 , in the assumption of either uniform or gaussian distributions for the mass loss and initial y , observed star counts as a function of the @xmath14 magnitude and ( @xmath15 ) colour are best reproduced with @xmath47=0.246 , @xmath48=0.02 and uniform probability distribution , a minimum rgb total mass loss @xmath49=0.122@xmath67 with a range ( again uniform probability distribution ) of values @xmath50=9.0 @xmath92 ( y-0.246 ) @xmath67 , where y is the actual initial he mass fraction of the synthetic star . it is obvious that this dependence of @xmath50 on y not imposed a priori but constrained by the fit to star counts as a function of magnitude and colour implies an increase of the mass loss spread with increasing y , that is required by the observed shape of the cmd in the ( @xmath60 ) plane . ) colour along the hb of m3 ( grey crosses ) and m13 ( black dots).,width=317 ] in fact as shown in figure [ synthm3_1 ] the reddest part of the zahb describes a steeply sloped sequence , before turning horizontal at @xmath93 . to match the observed star counts we need to increase the mass loss spread with increasing y. this is necessary to reproduce both the steep increase of @xmath14 with decreasing colour when ( @xmath15 ) is larger than @xmath450.2 e.g. , the lower he subpopulations have to cover a relatively small mass range and the almost constant @xmath14 when the colour ( @xmath15 ) is bluer that this limit e.g. , the higher he subpopulations have to cover an increasingly larger mass range . the mean he abundance obtained from the synthetic hb is @xmath84=0.256 , with a 1@xmath52 spread @xmath88=0.006 , consistent with the value [email protected] obtained from the r - parameter by cassisi et al . ( 2003 ) . in the synthetic hb calculations we have disregarded all objects lying in the rrlyrae instability strip , whose red and blue edges have been fixed at , respectively , log@xmath44=3.80 and log@xmath44=3.88 ( see , e.g. , the discussion in dantona & caloi 2008 ) , which define a very narrow colour width of @xmath94 at fixed @xmath14 approximatively in the range @xmath95 . the distance modulus obtained as described in the case of m13 is @[email protected] with standard reddening value e(b@xmath43v)=0.01 which is fully in agreement with previous estimates ( ferraro et al . 1999 ; harris 1996 catalogue version 2010 ) . in this case the maximum value of y ( @xmath96 ) is essentially constrained by the brightest limit of the stellar distribution at ( @xmath970.2 . figure [ synthm3_3 ] displays stellar mass and y distributions as a function of the ( @xmath15 ) colour in one representative synthetic realization of the cluster hb , compared to the corresponding best - fit composite simulation of m13 . not only the range of he values is obviously much larger in m13 , but also the shape of the y trend with colour is completely different in these two clusters . the mass distribution forms a very smooth continuous sequence , with the maximum and minimum masses along the hb distribution in m3 being appreciably larger than the corresponding values in m13 . a detailed comparison with recent results by catelan et al ( 2009 ) and caloi & dantona ( 2008 ) is reported in appendix b. ) cmd of m79 hb . the box marks the region used for the estimate of the cluster distance ( see text for details).,width=317 ] spectroscopic analyses of m79 rgb stars disclose an extension of the na - o anti - correlation that is intermediate between that of m13 and m3 ( see sneden et al . 2004 , carretta et al . 2010 ) , and photometric analyses do not reveal evidences of a multimodal ms . also the colour extension of its hb is intermediate between m13 and m3 . following the same approach as for m13 and m3 , we find that observed star counts as a function of magnitudes and colours are best reproduced with @xmath47=0.246 , @xmath48=0.035 and uniform probability distribution , a rgb total mass loss ( at a given y ) ranging from @xmath49=0.11@xmath67 to @xmath49=0.27@xmath67 , with a uniform probability distribution . one realization of the best fit simulation is displayed in figures [ synthm79_1 ] and [ synthm79_2 ] . the distance modulus has been determined from the fit to the stars contained within the box displayed in figure [ synthm79_1 ] ( denoted here as rhb ) as described for m13 . assuming e(b@xmath43v)=0.01 ( harris 1996 ) we have obtained @[email protected] , values well in agreement with both ferraro et al . ( 1999 ) and harris ( 1996 ) . for the rhb stars we determine an average @xmath84=0.263 , with a 1@xmath52 spread @xmath98=0.010 . the whole hb is characterized by @xmath84=0.265 , with a 1@xmath52 spread @xmath99=0.010 . this means that the full range of y values is already present in the rhb sample . in the reasonable assumption that the spread of initial y scales with the extension of the na - o anti - correlation , we should expect to find among rhb stars the full range of na - o abundances observed along the rgb . a range of he abundances is necessary along the rhb to match the observed thickness in @xmath14 , and the large range of @xmath49 ( at fixed y ) is required to match the extended and almost horizontal structure of the rhb in the ( @xmath100 ) cmd . figure [ synthm79_5 ] displays the mass and y distribution as a function of the ( @xmath15 ) colour in one representative synthetic realization of the cluster hb , compared to the m13 counterpart . the range of y and the shape of the trend with colour are very different . the maximum mass of hb stars in m79 reaches higher values compared to m13 , and the minimum hb mass is also higher than in m13 . but for the ( @xmath62 ) cmd ( see text for details).,width=317 ] @|p1.1cmc||p0.7cmc||p1.0cmc||p0.6cmc||p0.6cmc||p0.5cmc| cluster & filter & @xmath101 & @xmath102 & @xmath79 & @xmath84 + m3 & f336w & 158 & 59 & 0.266 & 0.256 + & f555w & & & & + m13 & f160bw & 228 & 7 & 0.300 & 0.270 + & f336w & & & & + & f555w & & & & + m79 & f160bw & 130 & 5 & 0.281 & 0.265 + & f336w & & & & + & f555w & & & & + [ tab : tot ] there is a general consensus that the first parameter shaping the globular cluster hb morphology is metallicity and the second one has been suggested by several authors ( lee et al . 2002 ; g10 ; d10 ) to be the age . while in general this picture may be surely adequate , however it is able to explain only few young clusters in the galaxy , while it fails to reproduce a number of `` classical '' cases where an additional ingredient ( sometime called third parameter ) is required . the `` third parameter '' has been proposed to be the stellar density or luminosity density ( fusi pecci et al . 1993 ; d10 ) or the initial helium abundance ( g10 ) . + in this framework we considered three ggcs with similar age and metallicity , m3 , m13 and m79 , with the aim of studying their hbs . according to g10 , m3 is younger by @xmath45 2gyr than m13 , however we find , in agreement with a number of authors ( salaris & weiss 2002 ; marin - franch et al . 2009 ; d10 ) , that these clusters differ in age by less than @xmath103 . this triplet is well known to have extremely different hbs , with m13 displaying by far the bluest morphology , and showing also evidence of at least 2 gaps at different effective temperatures , while m3 has been proposed over the years as a prototype of a `` normal '' hb . + as done in paper i , we compared hst uv and optical cmds with theoretical hb models , with the aim of understanding the impact ( if any ) of cluster - to - cluster y variations on hb morphologies . at odds with the case of ngc2808 , the three gcs analyzed in this work , do not show ( so far ) evidences of quantized ms , so our analysis has been performed without any a priori knowledge of the y distribution . as explained in section 4 , this forced us to enlarge the parameters space : the quantities needed are the minimum value of y ( @xmath104 ) and its full range ( @xmath48 ) , the mean value of the mass lost along the rgb @xmath105 and the dispersion around the mean value . as a consequence , it is hard in these cases to determine uniquely a full and detailed representation of the y variations along the observed hbs . a more stringent and robust derivation involves instead the largest value of y ( @xmath79 ) that is clearly constrained by the distribution of stars in the @xmath106 cmds . for the case of m3 , given the shorter extension of its hb , the magnitudes @xmath14 give already a strong constraint on @xmath1 . + with these caveat in mind , we performed a detailed analysis by comparing observed cmds with a large set of synthetic cmds and zahbs from the basti database ( pietrinferni et al . the effects of levitation for stars with @xmath107 k and anti - correlations along the hbs ( marino et al . 2011 ; gratton et al . 2011 ) have been considered . + we find differences @xmath108 between these ggcs . in particular m13 displays the largest value ( @xmath109 ) , m3 ( @xmath110 ) the smallest one , and m79 is an intermediate case with @xmath111 . these values are summarized in table 1 . they seem to qualitatively correlate with the differences in the temperature ( colour ) extensions of the cluster hbs . it is also interesting to note that our estimates of @xmath79 for these three clusters nicely correlate with the observed range of light - element variations . in particular , m13 shows the most extreme na - o anti - correlation ( sneden et al . 2004 ) , with stars reaching [ o / fe]@xmath112 and [ na / fe]@xmath113 , while m3 the least extended one , [ o / fe]@xmath114 and [ na / fe]@xmath115 . this is in line with the strict correspondence between hb colours and na - o abundances observed in ngc2808 by gratton et al . + the comparison between m13 and m3 is particularly interesting , because of several previous analyses of their hb morphology and y distribution . for example catelan ( 2009 ) performed a detailed analysis comparing the mean masses of the hbs of these clusters with age differences proposed by different authors . by using several mass - loss recipes , he found that there is no way to reproduce the different hbs of m3 and m13 only in terms of age , but at least one additional ingredient should be invoked to account for the blue hb extension observed in m13 . our best - fit model for m13 requires to split the simulation in three steps , given that no single continuous distribution of y and mass loss allows to match the observed cmds . thus we divided the hb in three groups : 1 ) the rhb with a typical @xmath116 and a mean mass lost = @xmath117 , 2 ) stars with @xmath118 best reproduced with @xmath119 and @xmath120 and 3 ) stars with @xmath121 which show a @xmath122 and @xmath123 . a fit with a population of stars with uniform @xmath124 can not be ruled out on the basis of only the hb analysis , while a single population of stars with @xmath125 , as proposed by caloi & dantona ( 2005 ) and dantona & caloi ( 2008 ) , is incompatible with the distribution of stars in our uv cmd . + for m3 a single synthetic population with @xmath116 and distributed according to a uniform probability distribution with @xmath126 is required . a total mass loss of @xmath127 and a linear increase as a function of y , as constrained by the fit of star counts as a function of magnitude and colours , is needed . in the appendix b we will discuss the y distribution derived from our simulations in the context of the constraints posed by the analyses by catelan et al . ( 2009 ) and dantona & caloi ( 2008 ) ; it turns out that our estimate of @xmath96 is robust . + as highlighted by g10 , while m13 seems to behave as other relatively massive clusters , m3 appears to be peculiar and a more extended hb would have been expected in this case . the age difference proposed by g10 between these two clusters would justify at least in part the different hb morphology , in the framework in which age is the second parameter . our results would instead lead to think that m3 and m13 experienced a different amount of enrichment of light elements . this would be compatible with the scenario proposed by g10 ( see also carretta et al . 2009 ) that invokes a delayed cooling flow in the case of m3 . in particular the hb simulations and derived @xmath86 distributions would suggest that m13 is qualitatively similar to ngc2808 , and that it likely experienced a similar star formation , while m3 ( and m79 ) probably had a less complex formation history . + analyses based on a suitable combination of uv to optical photometry and when available observed rr lyrae period distributions and synthetic hb simulations can provide not only insights on the hb second parameter problem , but in general they can potentially give some clues on the first stages of gcs formation and chemical evolution . + we thank the referee aaron dotter for useful comments and suggestions that improved the presentation of this work . this research is part of the project cosmic - lab funded by the european research council ( under contract erc-2010-adg-267675 ) . sc acknowledges the financial support of the ministero della ricerca scientifica e delluniversita prin miur 2007 : , the italian theoretical virtual observatory project and prin inaf 2011 * appendix a * + + * the effect of anti - correlations on bolometric corrections * + we checked the impact of cnona anti - correlations on the @xmath42-enanched bolometric corrections by performing a number of tests . in particular , we have chosen three representative pairs of ( @xmath44 , @xmath128 ) values along our theoretical @xmath42-enhanced zahb with [ fe / h][email protected] , y=0.246 , namely ( 5600 k , 2.40 ) , ( 7000 k , 2.80 ) , ( 10000 k , 3.50 ) . for each of these pairs we have calculated a synthetic spectrum with the synthe code by r.l . kurucz ( sbordone et al . 2005 ) in the wavelength range between 1200 and 7500 @xmath129 , including all the atomic and molecular lines available in the latest version of the kurucz / castelli linelist . the spectra were computed employing the lte , plane - parallel model atmospheres calculated with the atlas12 code ( castelli 2005 ) . atlas12 adopts the `` opacity sampling '' method in the line opacity calculations , and allows to generate models for arbitrary chemical abundance mixtures . all the model atmospheres and synthetic integrated spectra were computed with [ fe / h][email protected] and y=0.246 and both the standard [ @xmath42/fe]=0.4 metal distribution used in the stellar evolution models , and the cnona2 mixture by sbordone et al . this latter mixture displays compared to our reference @xmath42-enhanced mixture enhancements of n and na by 1.44 dex and 0.8 dex by mass , respectively , together with depletions of c and o by 0.6 dex and 0.8 dex , respectively , and unchanged cno sum . this mixture is representative of extreme values of the cn and nao anti - correlations observed in globular cluster stars ( see sbordone et al . 2011 for details ) . it is also very important to notice that low - mass stellar evolution models calculated for this mixture turn out to be identical to models calculated with the standard @xmath42-enhanced metal distribution . from the synthetic spectra we have calculated bolometric corrections to the f160bw , f336w and f555w filters used in our analysis , for both metal mixtures . only at @xmath1305600 k there is a difference for the f336w filter ( we did not consider the f160bw at this low temperature , because our photometry does not register cold hb stars in this passband ) that amounts however to only 0.03 mag ( cnona2 mixture stars appearing fainter ) . given the small value of this difference and the fact that such extreme values of the anti - correlations appear not to be reached at such low temperatures along the hb , the outcome of our analysis of ngc2808 and the results of this paper , based on standard @xmath42-enhanced bolometric corrections , are unaffected by the presence of the observed cnona anti - correlations . we have explicitly determined also the effect of he variations ( at fixed metal content ) on the bolometric corrections . we considered the point with ( 7000 k , 2.80 ) i.e. around the blue edge of the rr lyrae instability strip and calculated spectra for our standard @xmath42-enhanced metal distribution , with y=0.35 . compared to the case with standard y=0.246 , both f336w and f555w bolometric corrections are changed by less than 0.01 mag . the f160bw filter is affected at this level by less than 0.02 mag . + + * appendix b * + + * comparison with previous studies on the hb morphology of m3 . * + we address here two issues that arise from a comparison with recent results from the literature about the hb morphology of m3 . first of all , the range of y derived from our simulations is about twice the value determined by catelan et al . ( 2009 ) from a comparison of theoretical zahb magnitudes and gravities with both strmgren photometry , and spectroscopically measured surface gravities and @xmath131 . the size of the sample with spectroscopic gravities and @xmath131 estimates is relatively small , and within the corresponding error bars models with @xmath132 can not be easily ruled out . more compelling seems to be the constraint from the strmgren cmds . a fit of theoretical zahb models to the magnitude ( @xmath133 , @xmath134 and @xmath135 ) of the lower envelope of the observed hb at both the red side of the instability strip and at the knee of the stellar distribution beyond the blue edge of the strip , before the sparsely populated and not very extended blue vertical tail where zahbs of different y tend to become degenerate discloses a range of y at most equal to @[email protected] , with y increasing at the blue side of the strip . this disagreement with our results is however only apparent , for it can be very easily explained by recalling that to any given @xmath0 value corresponds it is associated a dispersion in colour along the hb ( see section 4.1 ) . figure [ synthm3_4 ] illustrates clearly this point by comparing a theoretical zahb for y=0.256 and @xmath44 larger than the blue edge of the instability strip shifted by our derived distance modulus and assumed reddening to the observed data . notice how for ( @xmath136 ) , that corresponds to the region of the knee in the strmgren cmds , this zahb fits nicely the observed lower envelope of the cmd , even though the average y in that colour range inferred by the best - fit synthetic model , is 0.260 . given the existence of a range of hb masses at fixed y , this colour interval is also populated by stars with y as low as @xmath450.255 , that are on average the faintest objects and hence determine the observed lower envelope of the magnitude distribution . the second issue is related to the cluster rich rrlyrae population . the period distribution of fundamental rrlyrae is strongly peaked , and in the recent literature attempts have been made to reproduce both the peaked period distribution and the hb colour extension . castellani , castellani , & cassisi ( 2005 ) have shown that when keeping y constant , a suitable bimodal mass distribution with two different mass dispersions is able to reproduce the observed features . on the other hand , dantona & caloi ( 2008 ) assumed a range of y and a very small mass dispersion for the mass loss along the rgb , and reproduced the observed hb colour extension and rrlyrae periods by calibrating the y distribution ( instead of the hb mass distribution ) . it is interesting to notice that dantona & caloi ( 2008 ) find from their simulations that 90% of hb stars are expected to have [email protected] , in agreement with our results , and only 10% of objects in their simulations reach y values above our estimated maximum y. here we have considered the synthetic objects in the best fit simulations that fall within the boundary of the instability strip ( using the same @xmath44 boundaries employed by dantona & caloi 2008 ) and were excluded from the comparison with the observed cmd . we have calculated their pulsational periods using the relationships by di criscienzo et al . ( 2004 ) and compared with the observed distribution ( corwin & carney 2001 ) . our predicted period distribution does not reproduce the observed ones , which is probably not surprising given our simple assumptions about the distribution of y among stars fed onto the hb . ) cmd of m3 using the y distribution proposed by dantona & caloi ( 2008 ) . open symbols denote synthetic objects with [email protected] . the zahb sequences for y=0.246 and y=0.300 are also displayed . the rectangular box marks the region considered in the comparison ( see text for details).,width=317 ] as a test , we have calculated additional synthetic hb models for m3 , employing this time the @xmath139 y distribution proposed by dantona & caloi ( 2008 see their figure 4 ) to reproduce both the observed period distribution in the instability strip and the @xmath140 colour distribution along the hb , while keeping the same mass loss for all hb stars . as in dantona & caloi ( 2008 ) we employed a gaussian mass distribution ; our adopted mean mass loss is @xmath49=0.160@xmath67 with a small ( as discussed in dantona & caloi 2008 ) gaussian spread @xmath50=0.002@xmath67 , and employed the same reddening and distance modulus as in the previous simulations . these choices allow to reproduce well the narrow period distribution ( both mean value of the logarithm of the periods and the 1@xmath52 dispersion ) whereas a comparison of the the stellar distribution in the cmd at the blue side of the strip is less satisfactory . the left and right panels of figure [ synthm3_5 ] show the observed and simulated ( @xmath71 ) cmd . notice how the observed shape of the cmd is reasonably reproduced as long as y is below @xmath450.270 . for higher values of y , synthetic stars tend to reach too bright magnitudes ( while in the @xmath141 plane they tend to be located along the more vertical blue part of the hb , at / beyond the knee ) . this reinforces our estimate of the y range as determined by our previous set of simulations that is the main purpose of our analysis and the need to increase the mass loss spread at the blue side of the instability strip instead of y in order to match the cmd location of the bluer objects .
we have derived fe abundances of 16 solar - type pleiades dwarfs by means of an equivalent width analysis of and lines in high - resolution spectra obtained with the hobby - eberly telescope and high resolution spectrograph . abundances derived from lines are larger than those derived from lines ( herein referred to as over - ionization ) for stars with k , and the discrepancy ( = [ /h ] - [ /h ] ) increases dramatically with decreasing , reaching over 0.8 dex for the coolest stars of our sample . the pleiades joins the open clusters m34 , the hyades , ic2602 , and ic2391 , and the ursa major moving group , demonstrating ostensible over - ionization trends . the pleiades abundances are correlated with infrared triplet and h chromospheric emission indicators and relative differences therein . oxygen abundances of our pleiades sample derived from the high - excitation triplet have been previously shown to increase with decreasing , and a comparison with the abundances suggests that the over - excitation ( larger abundances derived from high excitation lines relative to low excitation lines ) and over - ionization effects that have been observed in cool open cluster and disk field main sequence ( ms ) dwarfs share a common origin . curiously , a correlation between the pleiades abundances and chromospheric emission indicators does not exist . star - to - star abundances have low internal scatter ( dex ) , but the abundances of stars with k are systematically higher compared to the warmer stars . the cool star [ /h ] abundances can not be connected directly to over - excitation effects , but similarities with the and triplet trends suggest the abundances are dubious . using the [ /h ] abundances of five stars with k , we derive a mean pleiades cluster metallicity of [ fe / h ] .
1005.3619
studies of galactic chemical evolution are dependent on accurately derived abundances of stars spanning all ages , populations , kinematics , masses , and metallicities . stars with masses @xmath7 are especially important given their dominance of the initial mass function ( imf ; e.g. , * ? ? ? abundance studies utilizing high - resolution spectroscopy and local thermodynamic equilibrium ( lte ) analyses of near - solar metallicity g and k dwarfs in open clusters and in the disk field , however , have revealed that the observed abundances of at least some elements derived for these cool main sequence ( ms ) dwarfs may be spurious . in particular , studies have found evidence of over - ionization and over - excitation , i.e. , larger abundances are derived from lines of singly ionized species compared to neutral species and from high - excitation lines of neutral species compared to low - excitation lines , respectively . the first indication that some abundances derived for cool ms stars are problematic may have come from @xcite , who found the over - ionization of sc , ti , cr , and fe for a sample of 10 k dwarfs ( @xmath8 @xmath0@xmath9 k ) in the solar neighborhood . after a careful analysis of the procedures and stellar parameters used in the abundance derivations , the author was unable to account for the overabundances of the ionized species . found similar over - ionization results for sc , v , cr , fe , and y in five field k dwarfs ( @xmath10 @xmath0 @xmath11 k ) . the authors could not exclude an inaccurate temperature scale that is several hundred k too low as a possible source of the anomalous abundances , but in the end , they suggest that non - lte ( nlte ) effects are the more likely cause . open clusters have been important to the identification and continued study of the over - excitation / ionization effects , because the presumed internal chemical homogeneity of the clusters provides a baseline with which anomalous abundances can be compared . @xcite derived o abundances from the high - excitation ( @xmath12 ev ) near - ir triplet of a k dwarf in each of the the pleiades and ngc 2264 open clusters , and in both cases , the abundances were highly enhanced : [ o / h ] @xmath13 and @xmath14 , respectively . such high o abundances are not expected for clusters with nominal metallicities of [ fe / h ] @xmath15 @xcite and -0.15 @xcite , respectively . following @xcite , @xcite derived the abundances of ms dwarfs in the m34 cluster ; over - ionization of fe and over - excitation of si ( the abundances of which were derived from lines with excitation potentials in the range @xmath16 ev ) are seen in the coolest stars of the sample . oxygen abundances of the cool m34 dwarfs , as well as cool pleiades dwarfs , derived from the high - excitation triplet are also highly enhanced @xcite , confirming the earlier results of @xcite . subsequent to these early open cluster studies , the over - ionization of fe has been confirmed in the hyades @xcite and ursa major ( uma ) moving group @xcite , and of ti in the young pre - ms clusters ic2602 and ic2391 . overabundances of o derived from the triplet have been reported for the uma moving group @xcite , the hyades @xcite , and ic4665 @xcite . over - excitation effects have been reported for other elements , as well , including s in the pleiades @xcite , si , ti , ni , and cr in ic4665 @xcite , ni in the hyades @xcite , and ca , ti , and na in ic2602 and ic2391 . recent abundance analyses of cool field stars have also identified over - excitation / ionization effects , confirming the findings of earlier work . the over - excitation / ionization abundance anomalies are not thought to represent real photospheric overabundances ; rather , we believe that they are a signal that our knowledge of cool dwarf atmospheres and/or spectral line formation therein is incomplete . as of yet , the source or cause of the effects has not been identified . systematically erroneous stellar parameters , e.g. , an inaccurate @xmath0 scale , could lead to the observed abundance trends , but in general , unrealistically large parameter errors would have to be present ( e.g. , * ? ? ? * ; * ? ? ? furthermore , parameter changes made in response to the overabundances of one element often increase those of another ( e.g. , * ? ? ? nlte effects have been suggested as the cause , but in general , the over - excitation / ionization effects seen in cool dwarfs are in stark contrast to extant nlte calculations . for instance , lte analyses of the high - excitation triplet in the spectra of ms dwarfs are predicted to result in increasingly discrepant abundances with increasing @xmath0 for stars with @xmath0 @xmath17 k , requiring negative nlte corrections up to 0.4 0.5 dex at 6500 k for solar metallicity dwarfs . below 6000 k , the nlte corrections are predicted to be @xmath18 dex and essentially zero below 5500 k . chromospheric emission and photospheric activity ( spots , plages , and faculae ) have also been suggested sources for the abundance anomalies ( e.g. , * ? ? ? these inhomogeneities could produce apparent over - excitation / ionization effects within a strict lte framework . continuing our efforts to delineate and understand the observed over - excitation / ionization effects in cool ms dwarfs , we have derived and abundances of 16 pleiades dwarfs , 15 of which have had o abundances derived from the high - excitation triplet @xcite . the abundances evince a steep increase , reaching [ o / h ] @xmath19 dex near 5000 k , and star - to - star dispersion below 5500 k. we use the newly derived pleiades fe abundances to investigate if the over - excitation and over - ionization effects observed in cool ms dwarfs are related and indeed manifestations of the same phenomenon . future observational studies that could place stringent constraints on these effects and bring us closer to discovering the source of the anomalous abundances are also discussed . echelle spectra of 17 pleiades ms dwarfs were obtained with the 9.2-m hobby - eberly telescope ( het ) and high resolution spectrograph ( hrs ) at the mcdonald observatory in queue - mode on 22 separate nights between 2002 august 23 ut and 2003 february 17 ut . these spectra have been used previously by @xcite and @xcite who analyzed the @xmath20 high - excitation triplet and the @xmath21 line , respectively , in these pleiades dwarfs ; please consult these papers for detailed descriptions of the observations and spectra . briefly , the het / hrs detector is a 4096 x 4100 side - by - side ccd mosaic of two 2048 x 4100 ccds with 15 @xmath22 m pixels . the spectra are characterized by a high resolution of r @xmath23 = 60,000 , and they have typical signal - to - noise ( s / n ) ratios of 80 100 . the spectra cover the wavelengths 5095 to 8860 . data reduction followed the typical practice of using standard iraf routines to remove the bias pattern , subtract scattered light , flat - field , and wavelength - calibrate the spectra . the stars in our sample are listed in table 1 . nineteen and seven lines spanning 5793 to 7462 have been analyzed in the spectra of 16 of the seventeen pleiads in our sample ; note , however , that not all of the lines were measurable in all of the stars . equivalent widths ( ews ) were determined by fitting each line with a gaussian profile using the one - dimensional spectrum analysis software package spectre @xcite . the ew measurements , along with the wavelength , excitation potential ( @xmath24 ) , and transition probabilities ( @xmath25 ) of each line , are given in tables 2 4 . atomic data of the lines were obtained by email query to the vienna atomic line database ( vald ) . one star included in our li study @xcite but not considered here is h ii 152 . this star was observed on two nights separated by approximately ten months . as discussed by king et al . , target misidentification on one night is a concern . on the night where it is not a concern , the spectra were obtained with a grating setting distinct from the setting used for the other stars in our sample , giving a spectral coverage of 6100 to 9800 . many of the and lines in our linelist fall outside of the spectral coverage of these data , and consequently , h ii 152 was not included in our analysis here . the and abundances were derived using the lte stellar line analysis package moog @xcite and model stellar atmospheres with convective overshoot interpolated from the standard atlas9 grids of r. kurucz . in previous analyses of open cluster dwarfs ( e.g. , * ? ? ? * ; * ? ? ? * ) , we have shown that model atmospheres with convective overshoot produce consistent results as those without the overshoot approximation ( nover ) . presently , we have also tested for a subsample of our stars the more updated kurucz models without overshoot that include the most recent opacity distribution functions ( the odfnew models ) . the resulting [ /h ] and [ /h ] abundances are consistent with those derived using the overshoot models ; the differences range from 0 0.05 dex , with the odfnew - based abundances generally lower , similar to what was found previously for the nover models . the overshoot models used here are the same ones used by @xcite , from which the adopted stellar parameters are also taken . one exception is h ii 298 ; for this star , we have used the updated dereddened @xmath26 color of @xcite to calculate new @xmath0 , @xmath27 , and microturbulent velocity ( @xmath28 ) values using the relations described in @xcite and interpolate a new model from the kurucz grids . the adopted stellar parameters for our sample are provided in table 1 . line - by - line abundances derived for each of our pleiades stars are listed in tables 2 4 , and the stellar mean abundances can be found in table 5 . the final mean abundances are given relative to solar values ( table 2 ) derived via an ew analysis of sky spectra obtained with the het / hrs as part of our observational program . the relative abundances are determined on a line - by - line basis before the mean is taken ; this strict line - by - line abundance analysis ensures that the final relative abundances are independent of the adopted oscillator strengths . the adopted solar parameters are included in table 1 . the mean [ /h ] abundances and the differences in the mean [ /h ] and [ /h ] abundances ( @xmath29 [ /h ] - [ /h ] ) are plotted in figure 1 along with errorbars for three representative stars h ii 263 , h ii 2284 , and h ii 3179 that run the @xmath0 range of our sample . the errorbars denote the total internal uncertainties ( @xmath30 ) in the derived abundances . the @xmath30 uncertainties are the quadratic sum of the abundance uncertainties resulting from errors in the adopted stellar parameters @xcite and the uncertainty in the mean abundance ( table 5 ) . abundance sensitivities to the stellar parameters were determined by individually altering effective temperature ( @xmath31 ) , surface gravity ( @xmath32 ) , and microturbulent velocity ( @xmath33 ) and are given in table 6 . the @xmath30 uncertainties in the abundances for the three representative stars are @xmath34 , @xmath35 , and @xmath36 dex for h ii 263 , h ii 2284 , and h ii 3179 , respectively . for , the @xmath30 uncertainties are @xmath37 , @xmath38 , and @xmath39 , respectively . the errorbars for the [ /h ] - [ /h ] abundances shown in figure 1 represent the quadratically combined and individual @xmath30 uncertainties . the star - to - star [ /h ] abundances fall within a narrow range of 0.11 dex and have a standard deviation in the mean of 0.04 dex . however , as can been seen in the left panel of figure 1 , the [ /h ] abundances of stars with @xmath0 @xmath1 k are systematically higher than those of stars at higher @xmath0 . the discord is verified by the spearman rank correlation coefficient ( @xmath40 ) at the 99.97% confidence level . in the right panel of figure 1 , @xmath2 abundances evince a dramatic increase at about the same @xmath0 , 5400 k. the @xmath2 abundances result from large overabundances of among the cool dwarfs ( table 5 ) , and the @xmath2 vs. @xmath0 trend for the pleiades is similar to those seen in m34 @xcite , the hyades @xcite , and the uma moving group @xcite . the increase in @xmath2 with decreasing @xmath0 presented here resembles the trend of increasing pleiades o abundances derived from the high - excitation triplet @xcite . in particular , the increase in the triplet abundances also begins to become significant at approximately 5400 k. in figure 2 @xmath2is plotted against the triplet abundances ( [ o / h]@xmath41 ) , and it is seen that a strong correlation between these two abundance anomalies does exist . according to the linear correlation coefficient ( @xmath42 ) , @xmath2 and [ o / h]@xmath41 are correlated at a greater than 99.9% confidence level . also in figure 2 we plot the _ residuals _ in the @xmath2 and [ o / h]@xmath41 abundances . the residuals are differences between the observed abundances and @xmath0-dependent fitted values calculated by fitting low - order ( second or third ) polynomials to the abundance versus @xmath0 relations ( figure 3 ) ; the fitted values are determined at each stellar @xmath0 . this procedure effectively removes the global mass dependence of the abundances so that the residuals are a measure of star - to - star abundance scatter at a given @xmath0 @xcite . similar to the @xmath2 and [ o / h]@xmath41 abundances , their residuals are correlated , but at a slightly lower confidence level , 97% ( @xmath43 ) . the strong relationship between the @xmath2 and [ o / h]@xmath41 abundances and especially their residuals suggests the anomalous abundances share a common origin . inaccurate @xmath0 scales can give rise to @xmath0-dependent abundance trends if the scales are in error in a systematic way . @xcite have raised concern as to the accuracy of color - temperature relations , like the one used for our pleiades sample , arguing that disagreements between observed open cluster color - magnitude diagrams ( cmds ) and theoretical isochrones based on color - temperature relations likely arise from systematic errors in the latter . @xcite are able to obtain near - perfect agreement between the observed cmds of four nearby open clusters and isochrones using empirical corrections to the color - temperature relations as suggested by @xcite . inaccurate @xmath0 scales , however , do not appear to be at the root of the over - excitation / ionization effects observed among cool open cluster dwarfs . the pleiades and hyades [ o / h]@xmath41 abundances of @xcite and @xcite , respectively , were derived using color - temperature relations _ and _ empirically corrected isochrones , and in both cases , the steep trends of increasing abundances with decreasing @xmath0 are present . the @xmath0 from the empirically corrected isochrones for the majority of dwarfs in both clusters are higher than those from the color - temperature relations , with the differences reaching a maximum of about 190 k. temperature corrections of this magnitude also do not alleviate the large @xmath2 abundances of the coolest stars in our sample . according to the abundance sensitivities given in table 6 , the @xmath0 of h ii 263 ( @xmath0 @xmath44 k ) would have to be higher by approximately 750 k in order to bring its and abundances into agreement ; such errors in our adopted temperature scale are not expected @xcite . furthermore , increasing the @xmath0 of h ii 263 by 750 k would result in an 0.15 dex increase in its abundance and exacerbate the disagreement in the [ /h ] abundances of the cool and warm dwarfs . whatever the cause of the anomalous @xmath2 abundances , the phenomenon may also be affecting the [ /h ] abundances of the pleiads with @xmath0 @xmath1 k. @xcite found that the abundances of the two coolest m34 dwarfs ( @xmath45 k ) they analyzed are higher by about 0.1 dex than those of the rest of the sample ; the lines have excitation potentials of @xmath46 ev . similarly , hyades abundances derived from lines with excitation potentials of approximately 4.25 ev increase with decreasing @xmath0 for dwarfs with @xmath0 @xmath47 k @xcite . more interestingly , the abundances of a single cool hyades dwarf ( @xmath0 @xmath48 k ) derived from lines with excitation potentials of @xmath49 ev were approximately 0.15 dex higher than the abundance derived from a line with a low excitation potential ( @xmath50 ev ) . for a warmer dwarf ( @xmath0 @xmath51 k ) , consistent abundances were obtained from all of the lines ( please see figure 7 of schuler et al . ) . similar behavior is not seen here in the line - by - line abundances of individual pleiads ( figure 4 ) . however , identifying such excitation potential - related effects for a given cool pleiades star is difficult , because the standard deviation , a measure of the dispersion in the line - by - line abundances , in the mean [ /h ] abundance of each pleiad ranges from 0.04 to 0.13 dex and has an average of 0.07 dex . this is of the order of the effect seen among the cooler stars in m34 and hyades dwarfs , for which the effect is expected to be more severe . thus , a direct connection between the heightened mean [ /h ] abundances and the over - excitation phenomenon can not be made , but nonetheless , the fact that the [ /h ] abundances increase at the same @xmath0 as @xmath2 and the triplet abundances is intriguing and suggests that they all are the result of the same effect . intercluster comparisons of the cool cluster dwarf abundance anomalies can provide valuable insight into the nature of the over - excitation / ionization effects by potentially linking differences in abundance trend morphologies to differences in the physical characteristics of the clusters , such as age and metallicity . in figure 5 the pleiades @xmath2 values along with those of the hyades from @xcite are plotted versus @xmath0 . the @xmath2 abundances of these two clusters follow the same trend down to a @xmath0 of about 5200 k , below which the pleiades abundances clearly diverge . similar behavior is seen in the abundances of dwarfs in the hyades , pleiades and uma moving group @xcite . the abundances of stars in all three associations increase at similar rates down to a @xmath0 of about 5200 k , below which the pleiades trend becomes much steeper than both of those of the hyades and uma , which continue to track each other . the uma moving group has an age that is comparable to that of the hyades @xcite and a metallicity that is lower than both pleiades and hyades @xcite . the divergence of the pleiades triplet abundances from those of both the hyades and uma suggests that the abundance trends may undergo an age - related diminution ; the @xmath2 abundances of the pleiades and hyades are consistent with this conclusion . chromospheric emission and photospheric spots are two age - related phenomena that have been discussed in the literature as possible sources of the observed over - excitation / ionization effects . @xcite demonstrated using multicomponent model atmospheres that spotted photospheres can plausibly account for the triplet abundances of the cool hyades dwarfs . results from efforts investigating a possible connection between chromospheric activity and the anomalous triplet abundances of cool dwarfs , on the other hand , have been mixed . no correlation between h@xmath3 and infrared triplet emission measures and triplet abundances of pleiades dwarfs nor m34 dwarfs exists @xcite . however , found a strong correlation between x - ray activity indicators and pleiades triplet abundances taken from the literature . there is no correlation between h+k emission indicators and the hyades triplet abundances @xcite . for the young cluster ic4665 , @xcite show that the abundances of the cool dwarfs are highly correlated with both h@xmath3 and infrared triplet emission indicators . it is important to note that shen et al . is the only of these studies that derived the abundances and chromospheric emission levels from the same spectra . for the others , the measurements were made or taken from different sources , and thus the actual chromospheric emission level may have been different when the spectral lines were formed . as can be seen from these various studies , it is not clear if there is a connection between chromospheric emission and the over - excitation of . furthermore , chromospheric emission is often correlated with @xmath0 , so any correlation between chromospheric emission and abundances may be masking some other @xmath0-dependent effect @xcite . using the chromospheric emission data from @xcite , we have plotted the pleiades @xmath2 versus infrared triplet chromospheric emission indicators ( @xmath52 ) in figure 6 and find a correlation that is significant at greater than the 99.9% confidence level according to the linear correlation coefficient ( @xmath53 ) . a similar correlation is found for the h@xmath3 chromospheric emission ( @xmath54 ) . these correlations , while suggestive of a connection between chromospheric emission and the over - ionization of fe , should be viewed with caution , because @xmath54 and @xmath52 are also correlated with @xmath0 at approximately the 93% and 98% confidence levels . this degeneracy makes it unclear if the @xmath2 abundances , like those of , are affected by chromospheric emission or some other @xmath0-dependent effect . more importantly , residual @xmath2 abundances and residual chromospheric emission indicators ( calculated in the same manner as the abundance residuals , i.e. , they are the differences between observed and @xmath0-dependent fitted values ; please see figure 3 ) are correlated at the 93% and 99.9% confidence levels for h@xmath3 and infrared triplet , respectively ( figure 6 ) . these relationships are more indicative of a true correlation between chromospheric emission and the over - ionization of fe . we remind the reader , however , that the @xmath2 abundances and chromospheric emission indicators were not measured using the same spectra and are thus not cotemporal . @xcite have used the same het / hrs spectra analyzed here to examine the long - standing problem of the large li abundance dispersion observed among cool pleiades dwarfs . these authors find evidence that at least a portion of the dispersion is due to real li depletion and suggest that the differential depletion may be a consequence of stellar radius modulations induced by surface magnetic activity , i.e. , spots , during pre - ms evolution . it is also suggested that such spot - induced effects could be related to the over - excitation / ionization effects observed today . @xmath2 residuals are plotted versus li abundance residuals in figure 7 . li abundances are derived from @xmath55 line strengths and are taken from @xcite . the @xmath2 and li residuals relation has a correlation coefficient of @xmath56 , corresponding to a @xmath5791% confidence level . while only marginally significant at best , the mild correlation still means that a substantial fraction ( nearly - half ) of the variance in @xmath2 is related to that in li . the abstract picture painted by this result is consistent with the conjecture presented by @xcite : the considerable li abundance dispersion in cool pleiades dwarfs has a real pre - ms depletion component , a portion of which may be driven by the same mechanism ( the influence of spots ) but perhaps by different physics ( the influence on stellar structure versus the influence on line formation in addition to or independent of stellar structure ) that is also possibly responsible for the versus differences seen in these stars . this would explain the overlap in the variance of pleiades @xmath2 and li abundances but allow for only marginally significant present - day correlations between these observables . observational tests , using the same type of pleiades spectroscopic / photometric monitoring program proposed in the conclusions section below , of this possibility are discussed in @xcite . the mean abundance of the pleiades stars above 5400 k derived from lines is [ fe / h ] @xmath6 ( uncertainty in the mean ) compared to [ fe / h ] @xmath58 ( uncertainty in the mean ) for stars below 5400 k. while a direct connection between the high [ /h ] abundances and the over - excitation effects manifested in the [ o / h]@xmath41 abundances of cool open cluster dwarfs can not be made here , the similarities are suggestive and raise doubt as to the accuracy of the cool star [ /h ] abundances . for this reason , and because of the anomalously high abundances , we feel that the mean cluster metallicity is best estimated by the [ /h ] abundances of the warmer stars only , and adopt the value of [ fe / h ] @xmath6 for the pleiades . the pleiades is one of the most well studied galactic open clusters , and fe abundances of member f , g , and k dwarfs have been derived using high - resolution spectroscopy by a handful of groups , which are summarized in table 7 . @xcite found a mean abundance of [ fe / h ] @xmath59 from one f and three g dwarfs , though the abundances are characterized by a large dispersion ( 0.02 0.26 dex ) . the spectra of three of the stars have s / n ratios of @xmath60 . in the same year , @xcite report a cluster mean abundance of [ fe / h ] @xmath61 for 13 f stars that have standard deviations in their individual [ fe / h ] abundances @xmath62 dex ; the mean abundance of their entire 17 star sample is [ fe / h ] @xmath63 . subsequent to that , @xcite and @xcite find similar values , [ fe / h ] @xmath64 and -0.02 , respectively . @xcite derived the abundances of two cool pleiades k dwarfs and obtained a mean abundance of [ fe / h ] @xmath65 from an analysis of lines . this value is almost identical to that of the stars with @xmath0 @xmath1 k in our sample , and we suspect that the abundances of the two k dwarfs from king et al . are similarly suspect . however , @xcite also derived a mean abundance of [ fe / h ] @xmath65 but for five f dwarfs . these authors noted that their result is slightly larger than that of @xcite and suggested differences in the analyses , i.e. , spectral lines used and adopted stellar parameters , as a possible cause . finally , @xcite have recently reported a cluster abundance of [ fe / h ] @xmath66 based on 22 a , f , and g stars . excluding the results of @xcite due to poor data quality and those of @xcite due to the uncertainty in the abundances of the two cool k dwarfs studied , the mean pleiades cluster abundance of the six remaining studies , including ours , is [ fe / h ] @xmath67 with a standard deviation of @xmath68 . this is identical to the value from the five stars with @xmath0 @xmath69 k in our sample , and by our assessment , represents the best estimate of the pleiades cluster metallicity . we have derived fe abundances via an ew analysis of and lines in high - resolution and moderate - s / n spectra of 16 ms dwarfs in the pleiades open cluster . the [ /h ] abundances increase dramatically relative to [ /h ] at @xmath0 below 5400 k , with the difference reaching over 0.8 dex in the coolest stars . this behavior is akin to what is seen in m34 , the hyades , and the uma moving group . comparison of the @xmath2 abundance patterns in the pleiades and hyades , as well as the [ o / h]@xmath41 abundances in the pleiades , hyades , and uma moving group , suggests that the trends may relax with age , though metallicity may yet prove to be a factor . abundances of cool dwarfs in additional open clusters or other stellar associations , especially those older than the hyades , are needed to determine if either age or metallicity are related to these anomalous abundances . the [ /h ] abundances are also higher in pleiads below 5400 k , but they show no evidence of an increase with decreasing @xmath0 . the inability to attribute the high [ /h ] abundances of the cool stars to the over - excitation effects illustrates the difficulty of quantifying this phenomenon . with lines of exceptionally high excitation potential such as the triplet , the over - excitation effect is clearly seen ( e.g. , * ? ? ? * ; * ? ? ? * ) , but for lines with excitation potentials @xmath70 ev , the effect is more difficult to pinpoint . our linelist includes transitions ranging in excitation potential from 2.18 to 5.10 ev , but for the individual pleiades stars , no increase in the line - by - line abundances as a function of excitation potential , like that seen in abundances of cool hyades dwarfs ( 1.83 ev @xmath71 4.42 ev ; * ? ? ? * ) , is evident ( figure 4 ) . line - to - line sensitivities to the over - excitation / ionization effects have yet to be clearly delineated , and it needs to be determined if there is an excitation potential threshold above which the abundances derived from these lines become enhanced by these effects . similarly , it needs to be determined if there is an excitation potential threshold _ below _ which the opposite occurs , the abundances derived from the low - excitation lines are lower due to these effects . such behavior would be expected if the overabundances of high - excitation ( singly ionized ) lines are due to the overpopulation of high - excitation ( singly ionized ) electronic states at the expense of depopulating low - excitation states . whether or not the over - excitation effects impact the spectroscopic derivation of stellar parameters ( @xmath0 , @xmath27 , and @xmath28 ) , an approach not adopted here , also needs to be determined . future investigations of these effects will require high - quality high - resolution spectroscopy so that accurate line - by - line abundances can be derived , even from features of just a few m in strength . a strong correlation between the @xmath2 and [ o / h]@xmath41 abundances of the pleiades dwarfs is evident in figure 2 , suggesting that the over - excitation / ionization effects share a common cause or origin . chromospheric emission and photospheric spots have been shown to be promising culprits , but to this point , the data are inconclusive . whereas strong correlations between the pleiades @xmath2 and chromospheric emission indicators and their residuals exist , they do not exist between [ o / h]@xmath41 and chromospheric emission . these contradictory results complicate the interpretation of the observed over - excitation / ionization effects and will have to be addressed by future studies . also , comparing abundances and chromospheric emission indicators measured using different spectra may not provide an accurate test of a true correlation because of potential temporal changes in chromospheric activity levels . future investigations into the over - excitation / ionization effects in cool open cluster dwarfs should make every effort to derive chromospheric activity levels from the same spectra so that any possible relation between the two can be more definitively delineated . determining the influence of photospheric spots on abundance derivations is more arduous . multicomponent model atmospheres- simulating photospheres with different areal coverages of hot , cool , and quiescent spots- have been shown to able to reproduce the measured ews of the triplet in a sample of hyades stars , but , while such exercises are useful and demonstrate the plausibility of the photospheric spot hypothesis , the results are only suggestive . observationally , a simultaneous photometric and spectroscopic monitoring program could be used to identify any correlated changes in spot coverage and spectral line strengths . such observational constraints would be helpful to determine if spotted photospheres affect high - resolution abundance derivations . despite the challenges , the possible connection between spots , over - excitation / ionization effects , and pre - ms li depletion should provide sufficient motivation for future efforts . a final conclusion that can be drawn from this study is that those carrying out spectroscopic abundance analyses of open clusters should heed caution when their samples include cool dwarfs , particularly those with @xmath0 @xmath1 k. including the abundances of these stars may skew cluster mean abundances . a similar caution may be needed for those studying cool ms dwarfs in the disk field , as well . further investigations into the sensitivity of the over - excitation / ionization effects to excitation potential , first ionization potential , stellar age , and stellar metallicity are needed in order to identify the extent and ubiquity of these effects . gratefully acknowledges her support for summer research from the national science foundation through their research experience for undergraduates program at cerro tololo inter - american observatory ( grant ast-0647604 ) . a.l.p . also thanks the c.v . starr - middlebury school in latin america for additional support . s.c.s . is grateful for support provided by the noao leo goldberg fellowship ; noao is operated by the association of universities for research astronomy , inc . , under a cooperative agreement with the national science foundation . j.r.k . gratefully acknowledges support for this work by grants ast 00 - 86576 and ast 02 - 39518 to j.r.k . from the national science foundation and by a generous grant from the charles curry foundation to clemson university . allende prieto , c. , barklem , p. s. , lambert , d. l. , & cunha , k. 2004 , , 420 , 183 an , d. , terndrup , d. m. , pinsonneault , m. h. , paulson , d. b. , hanson , r. b. , & stauffer , j. r. 2007 , , 655 , 233 asplund , m. 2005 , , 43 , 481 boesgaard , a. m. 1989 , , 336 , 798 boesgaard , a. m. , budge , k. g. , & ramsay , m. e. 1988 , , 327 , 389 boesgaard , a. m. & friel , e. d. 1990 , , 351 , 467 castelli , f. & kurucz , r. l. 2004 , arxiv : astro - ph/0405087 cayrel , r. , cayrel de strobel , g. , & campbell , b. 1988 , the impact of very high s / n spectroscopy on stellar physics , 132 , 449 chen , y. q. , zhao , g. , izumiura , h. , zhao , j. k. , liu , y. j. , honda , s. , & ohkubo , m. 2008 , , 135 , 618 dorazi , v. & randich , s. 2009 , , 501 , 553 fabbian , d. , asplund , m. , barklem , p. s. , carlsson , m. , & kiselman , d. 2009 , , 500 , 1221 feltzing , s. & gustafsson , b. 1998 , , 129 , 237 fitzpatrick , m. j. & sneden , c. 1987 , , 19 , 1129 funayama , h. , itoh , y. , oasa , y. , toyota , e. , hashimoto , o. , & mukai , t. 2009 , , 61 , 930 gebran , m. & monier , r. 2008 , , 483 , 567 king , j. r. , krishnamurthi , a. , & pinsonneault , m. h. 2000b , , 119 , 859 king , j. r. & schuler , s. c. 2005 , , 117 , 911 king , j.r . , schuler , s.c . , hobbs , l.m . , & pinsonneault , m.h . 2010 , , 710 , 1610 king , j. r. , soderblom , d. r. , fischer , d. , & jones , b. f. 2000a , , 533 , 944 kroupa , p. 2002 , science , 295 , 82 kupka , f. , piskunov , n. , ryabchikova , t. a. , stempels , h. c. , & weiss , w. w. 1999 , , 138 , 119 morel , t. & micela , g. 2004 , , 423 , 677 noyes , r. w. , hartmann , l. w. , baliunas , s. l. , duncan , d. k. , & vaughan , a. h. 1984 , , 279 , 763 oinas , v. 1974 , , 27 , 405 pinsonneault , m. h. , terndrup , d. m. , hanson , r. b. , & stauffer , j. r. 2004 , , 600 , 946 piskunov , n. e. , kupka , f. , ryabchikova , t. a. , weiss , w. w. , & jeffery , c. s. 1995 , , 112 , 525 ramrez , i. , allende prieto , c. , & lambert , d. l. 2007 , , 465 , 271 ryabchikova , t. a. , piskunov , n. e. , stempels , h. c. , kupka , f. , & weiss , w.w . 1999 , phys . t83 , 162 schuler , s. c. , hatzes , a. p. , king , j. r. , krster , m. , & the , l .- s . 2006a , , 131 , 1057 schuler , s. c. , king , j. r. , fischer , d. a. , soderblom , d. r. , & jones , b. f. 2003 , , 125 , 2085 schuler , s. c. , king , j. r. , hobbs , l. m. , & pinsonneault , m. h. 2004 , , 602 , l117 schuler , s. c. , king , j. r. , terndrup , d. m. , pinsonneault , m. h. , murray , n. , & hobbs , l. m. 2006b , , 636 , 432 shen , z .- x . , jones , b. , lin , d. n. c. , liu , x .- w . , & li , s .- l . 2005 , , 635 , 608 shen , z .- x . , liu , x .- w . , zhang , h .- w . , jones , b. , & lin , d. n. c. 2007 , , 660 , 712 sneden , c. 1973 , , 184 , 839 soderblom , d. r. , jones , b. f. , balachandran , s. , stauffer , j. r. , duncan , d. k. , fedele , s. b. , & hudon , j. d. 1993 , , 106 , 1059 takeda , y. 2003 , , 402 , 343 yong , d. , lambert , d. l. , allende prieto , c. , & paulson , d. b. 2004 , , 603 , 697 @xmath72 & & 0.65 & & 5715 & & 4.55 & & 0.96 + @xmath73 & & 0.75 & & 5339 & & 4.61 & & 0.58 + @xmath74 & & 0.65 & & 5715 & & 4.55 & & 0.96 + @xmath75 & & 0.84 & & 5048 & & 4.64 & & 0.30 + @xmath76 & & 0.89 & & 4899 & & 4.66 & & 0.16 + @xmath77 & & 0.75 & & 5373 & & 4.60 & & 0.61 + @xmath78 & & 0.73 & & 5407 & & 4.60 & & 0.65 + @xmath79 & & 0.82 & & 5098 & & 4.64 & & 0.35 + @xmath80 & & 0.73 & & 5407 & & 4.60 & & 0.65 + @xmath81 & & 0.81 & & 5142 & & 4.63 & & 0.39 + @xmath82 & & 0.74 & & 5363 & & 4.61 & & 0.60 + @xmath83 & & 0.78 & & 5239 & & 4.62 & & 0.48 + @xmath84 & & 0.78 & & 5239 & & 4.62 & & 0.48 + @xmath85 & & 0.72 & & 5442 & & 4.59 & & 0.68 + @xmath86 & & 0.80 & & 5174 & & 4.63 & & 0.42 + @xmath87 & & 0.82 & & 5117 & & 4.64 & & 0.37 + @xmath88 & & 0.53 & & 6172 & & 4.48 & & 1.42 + @xmath89 & & & & 5777 & & 4.44 & & 1.38 + & & 5793.92 & & 4.22 & & -1.70 & & 33.9 & 7.52 & & 46.0 & 7.63 & & 37.0 & 7.59 & & 47.5 & 7.61 & & 45.4 & 7.56 & & 39.3 & 7.51 + & & 5856.10 & & 4.29 & & -1.64 & & 36.6 & 7.57 & & 42.1 & 7.56 & & 30.8 & 7.46 & & 47.8 & 7.62 & & 51.1 & 7.67 & & 43.0 & 7.59 + & & 5927.80 & & 4.65 & & -1.09 & & 45.7 & 7.52 & & 54.0 & 7.58 & & 41.5 & 7.46 & & 60.3 & 7.63 & & 63.9 & 7.67 & & 56.5 & 7.63 + & & 6089.57 & & 5.02 & & -0.94 & & 41.8 & 7.64 & & 47.1 & 7.63 & & 35.3 & 7.52 & & 49.5 & 7.64 & & 56.6 & 7.74 & & 43.0 & 7.57 + & & 6093.65 & & 4.61 & & -1.50 & & 32.0 & 7.63 & & 39.6 & 7.67 & & 30.4 & 7.60 & & 45.9 & 7.76 & & 42.5 & 7.69 & & 39.4 & 7.68 + & & 6096.67 & & 3.98 & & -1.93 & & 41.2 & 7.64 & & 52.8 & 7.75 & & 42.4 & 7.68 & & 52.9 & 7.70 & & 55.1 & 7.72 & & 49.2 & 7.69 + & & 6151.62 & & 2.18 & & -3.29 & & 52.5 & 7.45 & & 66.5 & 7.58 & & 53.6 & 7.50 & & 76.0 & 7.64 & & & & & 61.4 & 7.48 + & & 6165.36 & & 4.14 & & -1.47 & & 48.3 & 7.46 & & 58.3 & 7.56 & & 47.2 & 7.47 & & 58.1 & 7.49 & & 57.6 & 7.47 & & 53.0 & 7.46 + & & 6270.23 & & 2.86 & & -2.71 & & 56.2 & 7.59 & & 70.8 & 7.77 & & 57.6 & 7.67 & & 68.8 & 7.62 & & 75.3 & 7.69 & & & + & & 6627.56 & & 4.55 & & -1.68 & & 30.9 & 7.71 & & 30.6 & 7.59 & & 30.0 & 7.69 & & 37.0 & 7.69 & & 39.1 & 7.74 & & 36.6 & 7.73 + & & 6806.86 & & 2.73 & & -3.21 & & 36.4 & 7.57 & & 53.1 & 7.72 & & 37.8 & 7.60 & & & & & & & & 47.1 & 7.61 + & & 6839.84 & & 2.56 & & -3.45 & & 37.2 & 7.65 & & & & & 35.8 & 7.62 & & & & & 58.4 & 7.76 & & 38.5 & 7.48 + & & 6842.69 & & 4.64 & & -1.32 & & 41.8 & 7.64 & & & & & 43.5 & 7.68 & & & & & & & & 44.1 & 7.60 + & & 6857.25 & & 4.07 & & -2.15 & & 23.4 & 7.54 & & 35.0 & 7.66 & & 23.8 & 7.54 & & & & & 37.1 & 7.66 & & 28.5 & 7.52 + & & 6861.94 & & 2.42 & & -3.89 & & 21.5 & 7.61 & & & & & & & & & & & 42.5 & 7.69 & & 32.5 & 7.63 + & & 6862.50 & & 4.56 & & -1.57 & & 29.9 & 7.58 & & & & & & & & & & & 45.9 & 7.76 & & 40.2 & 7.69 + & & 7284.84 & & 4.14 & & -1.75 & & 38.8 & 7.51 & & & & & 46.0 & 7.66 & & & & & 57.9 & 7.64 & & 51.0 & 7.65 + & & 7461.53 & & 2.56 & & -3.58 & & 29.2 & 7.59 & & 37.4 & 7.53 & & & & & 48.5 & 7.67 & & 50.4 & 7.68 & & 42.0 & 7.64 + & & 7547.90 & & 5.10 & & -1.35 & & 22.6 & 7.67 & & 27.5 & 7.70 & & & & & 33.3 & 7.81 & & & & & 29.4 & 7.74 + & & 5264.81 & & 3.23 & & -3.13 & & 43.8 & 7.40 & & 42.1 & 7.82 & & 42.9 & 7.54 & & 49.9 & 8.31 & & & & & 44.0 & 7.84 + & & 5414.05 & & 3.22 & & -3.65 & & 30.7 & 7.61 & & & & & & & & & & & & & & & + & & 5425.25 & & 3.20 & & -3.39 & & 39.4 & 7.53 & & 39.7 & 7.99 & & 40.8 & 7.71 & & 41.3 & 8.32 & & 39.6 & 8.45 & & 40.5 & 7.97 + & & 6084.10 & & 3.20 & & -3.88 & & 22.0 & 7.59 & & & & & & & & & & & & & & & + & & 6247.56 & & 3.89 & & -2.44 & & 52.6 & 7.49 & & 47.4 & 7.92 & & 56.8 & 7.77 & & 52.5 & 8.37 & & 47.2 & 8.43 & & 50.5 & 7.96 + & & 6432.68 & & 2.89 & & -3.69 & & 42.0 & 7.55 & & 39.8 & 7.96 & & 41.8 & 7.70 & & 48.1 & 8.48 & & 41.5 & 8.50 & & 37.2 & 7.86 + & & 6456.39 & & 3.90 & & -2.19 & & 63.4 & 7.45 & & 55.7 & 7.86 & & 69.6 & 7.79 & & 67.1 & 8.41 & & 53.0 & 8.32 & & 63.0 & 7.98 + & & 5793.92 & & 4.22 & & -1.70 & & 40.4 & 7.54 & & 48.0 & 7.62 & & 40.5 & 7.54 & & 49.5 & 7.66 & & 48.2 & 7.69 & & 46.5 & 7.62 + & & 5856.10 & & 4.29 & & -1.64 & & 40.9 & 7.56 & & 46.7 & 7.61 & & 40.9 & 7.56 & & 49.4 & 7.66 & & 41.0 & 7.23 & & 44.3 & 7.58 + & & 5927.80 & & 4.65 & & -1.09 & & 53.0 & 7.57 & & 58.0 & 7.60 & & 51.2 & 7.54 & & 60.5 & 7.64 & & 53.2 & 7.56 & & 55.5 & 7.58 + & & 6089.57 & & 5.02 & & -0.94 & & 42.4 & 7.56 & & 49.5 & 7.64 & & 41.6 & 7.55 & & 47.0 & 7.60 & & 46.7 & 7.63 & & 54.4 & 7.73 + & & 6093.65 & & 4.61 & & -1.50 & & 38.5 & 7.67 & & 41.0 & 7.67 & & 39.0 & 7.68 & & 40.7 & 7.66 & & 40.1 & 7.69 & & & + & & 6096.67 & & 3.98 & & -1.93 & & 51.0 & 7.74 & & 54.0 & 7.72 & & 46.8 & 7.65 & & 55.2 & 7.76 & & 56.1 & 7.83 & & 51.3 & 7.70 + & & 6151.62 & & 2.18 & & -3.29 & & 63.9 & 7.55 & & 69.5 & 7.53 & & 61.7 & 7.50 & & 80.0 & 7.75 & & 70.9 & 7.68 & & 72.9 & 7.66 + & & 6165.36 & & 4.14 & & -1.47 & & 53.5 & 7.48 & & 62.3 & 7.57 & & 49.9 & 7.41 & & 56.6 & 7.48 & & 51.6 & 7.43 & & 57.0 & 7.51 + & & 6270.23 & & 2.86 & & -2.71 & & 63.5 & 7.64 & & 67.0 & 7.60 & & 64.2 & 7.66 & & 73.1 & 7.73 & & 71.2 & 7.54 & & 69.5 & 7.70 + & & 6627.56 & & 4.55 & & -1.68 & & 35.1 & 7.70 & & 41.1 & 7.78 & & 35.1 & 7.70 & & 38.6 & 7.73 & & 42.2 & 7.84 & & 39.6 & 7.76 + & & 6806.86 & & 2.73 & & -3.21 & & 46.3 & 7.60 & & 55.1 & 7.68 & & 44.6 & 7.57 & & 58.7 & 7.77 & & 49.8 & 7.66 & & 53.7 & 7.70 + & & 6839.84 & & 2.56 & & -3.45 & & 42.5 & 7.58 & & 52.0 & 7.67 & & 39.1 & 7.51 & & 46.9 & 7.57 & & 43.8 & 7.59 & & 49.6 & 7.66 + & & 6842.69 & & 4.64 & & -1.32 & & 51.0 & 7.73 & & 48.5 & 7.64 & & 48.6 & 7.69 & & & & & 49.9 & 7.70 & & 48.1 & 7.65 + & & 6857.25 & & 4.07 & & -2.15 & & 33.2 & 7.64 & & 33.0 & 7.57 & & 31.1 & 7.59 & & 35.3 & 7.63 & & 30.2 & 7.56 & & 37.6 & 7.69 + & & 6861.94 & & 2.42 & & -3.89 & & 26.6 & 7.51 & & 33.4 & 7.53 & & 26.3 & 7.50 & & & & & 32.0 & 7.61 & & & + & & 6862.50 & & 4.56 & & -1.57 & & 36.4 & 7.62 & & 42.5 & 7.70 & & 35.0 & 7.59 & & & & & 43.0 & 7.74 & & 40.3 & 7.66 + & & 7284.84 & & 4.14 & & -1.75 & & 50.8 & 7.65 & & 55.6 & 7.69 & & 48.8 & 7.61 & & & & & 48.8 & 7.60 & & & + & & 7461.53 & & 2.56 & & -3.58 & & 45.0 & 7.72 & & 48.4 & 7.68 & & 36.6 & 7.54 & & 42.2 & 7.55 & & 45.9 & 7.72 & & 47.6 & 7.71 + & & 7547.90 & & 5.10 & & -1.35 & & & & & & & & & & & 31.7 & 7.78 & & 27.0 & 7.69 & & 31.7 & 7.78 + & & 5264.81 & & 3.23 & & -3.13 & & 37.7 & 7.64 & & 36.6 & 7.91 & & 41.9 & 7.76 & & 32.7 & 7.75 & & 34.7 & 7.60 & & 37.7 & 7.79 + & & 5414.05 & & 3.22 & & -3.65 & & & & & & & & & & & & & & & & & 27.0 & 7.98 + & & 5425.25 & & 3.20 & & -3.39 & & 37.3 & 7.86 & & & & & 38.0 & 7.88 & & 32.1 & 7.96 & & 39.6 & 7.96 & & 37.3 & 8.01 + & & 6084.10 & & 3.20 & & -3.88 & & & & & & & & 21.3 & 7.88 & & & & & 14.5 & 7.66 & & & + & & 6247.56 & & 3.89 & & -2.44 & & 51.0 & 7.93 & & 38.4 & 7.96 & & 48.5 & 7.87 & & 40.7 & 7.97 & & 44.0 & 7.81 & & 43.8 & 7.93 + & & 6432.68 & & 2.89 & & -3.69 & & 36.8 & 7.82 & & 30.5 & 7.95 & & 38.6 & 7.86 & & 29.9 & 7.88 & & 35.2 & 7.82 & & & + & & 6456.39 & & 3.90 & & -2.19 & & 62.4 & 7.94 & & 51.2 & 8.03 & & 58.7 & 7.86 & & 50.6 & 7.97 & & 54.5 & 7.82 & & 57.8 & 8.02 + & & 5793.92 & & 4.22 & & -1.70 & & 47.8 & 7.65 & & 43.1 & 7.61 & & 48.0 & 7.64 & & 43.5 & 7.54 & & 26.3 & 7.58 + & & 5856.10 & & 4.29 & & -1.64 & & 46.8 & 7.63 & & 38.9 & 7.52 & & 46.6 & 7.62 & & 46.8 & 7.61 & & 24.8 & 7.55 + & & 5927.80 & & 4.65 & & -1.09 & & 57.3 & 7.61 & & 50.5 & 7.54 & & 58.3 & 7.61 & & 55.2 & 7.55 & & 34.2 & 7.52 + & & 6089.57 & & 5.02 & & -0.94 & & 51.4 & 7.69 & & 40.2 & 7.53 & & 49.5 & 7.65 & & 48.6 & 7.63 & & 30.4 & 7.60 + & & 6093.65 & & 4.61 & & -1.50 & & 47.2 & 7.80 & & 32.6 & 7.55 & & 41.2 & 7.68 & & 43.5 & 7.72 & & 24.1 & 7.66 + & & 6096.67 & & 3.98 & & -1.93 & & 49.8 & 7.67 & & 45.5 & 7.64 & & 56.1 & 7.78 & & 55.1 & 7.75 & & 29.5 & 7.65 + & & 6151.62 & & 2.18 & & -3.29 & & 72.5 & 7.66 & & 55.4 & 7.38 & & 70.5 & 7.59 & & 73.2 & 7.61 & & 36.8 & 7.51 + & & 6165.36 & & 4.14 & & -1.47 & & 57.3 & 7.51 & & 51.3 & 7.45 & & 56.5 & 7.48 & & 55.9 & 7.46 & & 37.3 & 7.50 + & & 6270.23 & & 2.86 & & -2.71 & & 71.3 & 7.74 & & 61.2 & 7.61 & & 72.9 & 7.74 & & 65.9 & 7.59 & & 44.8 & 7.70 + & & 6627.56 & & 4.55 & & -1.68 & & 34.8 & 7.66 & & 36.0 & 7.73 & & 39.6 & 7.75 & & 30.2 & 7.55 & & 21.5 & 7.70 + & & 6806.86 & & 2.73 & & -3.21 & & 56.8 & 7.76 & & 43.5 & 7.56 & & 56.3 & 7.73 & & 57.1 & 7.73 & & 23.7 & 7.62 + & & 6839.84 & & 2.56 & & -3.45 & & 51.6 & 7.71 & & 42.5 & 7.60 & & 47.9 & 7.60 & & 49.3 & 7.62 & & & + & & 6842.69 & & 4.64 & & -1.32 & & 48.5 & 7.65 & & 44.5 & 7.62 & & 49.5 & 7.67 & & 52.8 & 7.72 & & 32.5 & 7.67 + & & 6857.25 & & 4.07 & & -2.15 & & 38.2 & 7.71 & & & & & 33.3 & 7.59 & & 35.5 & 7.63 & & 17.2 & 7.60 + & & 6861.94 & & 2.42 & & -3.89 & & 41.1 & 7.76 & & & & & 36.9 & 7.64 & & 36.6 & 7.61 & & 11.0 & 7.60 + & & 6862.50 & & 4.56 & & -1.57 & & & & & & & & 44.5 & 7.74 & & 38.4 & 7.62 & & 25.3 & 7.68 + & & 7284.84 & & 4.14 & & -1.75 & & & & & & & & 57.5 & 7.73 & & & & & & + & & 7461.53 & & 2.56 & & -3.58 & & 42.4 & 7.59 & & 34.2 & 7.51 & & 44.8 & 7.62 & & 42.0 & 7.54 & & 18.9 & 7.67 + & & 7547.90 & & 5.10 & & -1.35 & & & & & 24.3 & 7.64 & & & & & 30.1 & 7.74 & & & + & & 5264.81 & & 3.23 & & -3.13 & & 34.0 & 7.69 & & 37.8 & 7.61 & & 36.9 & 7.84 & & 38.9 & 7.95 & & 51.4 & 7.48 + & & 5414.05 & & 3.22 & & -3.65 & & 23.5 & 7.86 & & & & & 21.6 & 7.86 & & & & & 30.3 & 7.53 + & & 5425.25 & & 3.20 & & -3.39 & & 33.8 & 7.91 & & 36.1 & 7.79 & & 30.8 & 7.88 & & 34.8 & 8.07 & & 48.1 & 7.63 + & & 6084.10 & & 3.20 & & -3.88 & & & & & & & & & & & 39.4 & 7.97 & & 26.5 & 7.64 + & & 6247.56 & & 3.89 & & -2.44 & & 41.0 & 7.86 & & 51.3 & 7.90 & & 38.0 & 7.86 & & 34.5 & 8.05 & & 66.6 & 7.64 + & & 6432.68 & & 2.89 & & -3.69 & & & & & 42.4 & 7.93 & & 31.5 & 7.90 & & 52.0 & 8.03 & & 48.9 & 7.61 + & & 6456.39 & & 3.90 & & -2.19 & & 51.2 & 7.87 & & 60.8 & 7.86 & & 50.8 & 7.93 & & & & & 74.1 & 7.53 + h ii 193 & & 0.06 & & 0.02 & & 0.43 & & 0.01 & & + 0.37 + h ii 250 & & 0.01 & & 0.02 & & 0.22 & & 0.05 & & + 0.21 + h ii 263 & & 0.07 & & 0.02 & & 0.89 & & 0.04 & & + 0.82 + h ii 298 & & 0.09 & & 0.01 & & 0.92 & & 0.02 & & + 0.83 + h ii 571 & & 0.02 & & 0.02 & & 0.44 & & 0.04 & & + 0.42 + h ii 746 & & 0.04 & & 0.02 & & 0.35 & & 0.06 & & + 0.31 + h ii 916 & & 0.06 & & 0.01 & & 0.49 & & 0.05 & & + 0.43 + h ii 1593 & & 0.00 & & 0.01 & & 0.35 & & 0.02 & & + 0.35 + h ii 2126 & & 0.08 & & 0.03 & & 0.42 & & 0.04 & & + 0.34 + h ii 2284 & & 0.05 & & 0.03 & & 0.28 & & 0.06 & & + 0.23 + h ii 2311 & & 0.08 & & 0.01 & & 0.45 & & 0.04 & & + 0.37 + h ii 2366 & & 0.09 & & 0.02 & & 0.34 & & 0.04 & & + 0.25 + h ii 2406 & & -0.02 & & 0.01 & & 0.33 & & 0.05 & & + 0.35 + h ii 2462 & & 0.08 & & 0.02 & & 0.37 & & 0.04 & & + 0.29 + h ii 2880 & & 0.04 & & 0.02 & & 0.53 & & 0.02 & & + 0.49 + h ii 3179 & & 0.03 & & 0.01 & & 0.06 & & 0.03 & & + 0.03 + h ii 263 @xmath90 & & @xmath91 & & @xmath92 & & @xmath93 + & & @xmath94 & & @xmath95 & & @xmath96 + & & @xmath97 & & @xmath98 & & @xmath99 + h ii 2284@xmath100 & & @xmath31 & & @xmath101 & & @xmath102 + & & @xmath103 & & @xmath104 & & @xmath105 + & & @xmath33 & & @xmath106 & & @xmath107 + h ii 3179@xmath108 & & @xmath31 & & @xmath109 & & @xmath106 + & & @xmath103 & & @xmath110 & & @xmath111 + & & @xmath33 & & @xmath98 & & @xmath112 + @xmath113 & & 0.06 & & @xcite + @xmath114 & & 0.02 & & @xcite + @xmath115 & & 0.03 & & @xcite + @xmath116 & & 0.02 & & @xcite + @xmath117 & & 0.03 & & @xcite + @xmath117 & & 0.01 & & @xcite + @xmath118 & & 0.05 & & @xcite + @xmath119 & & 0.02 & & this work + & & & & + @xmath119 & & 0.01 & & mean +
we study a multi - field inflationary theory with separable lagrangian , which has different speed of sound for each field . we find that the fields always coupled at perturbative level through gravitational interaction . we show that if the coupling terms among the perturbation fields are weak enough , these fields can be treated as a combination of decoupled fields , which are similar to normal modes in coupled oscillation . by virtue of such fields , the curvature perturbation at the horizon - crossing can be calculated up to the leading order of slow variation parameters via formalism . explicitly , we consider a model of multi - speed dbi inflation , and calculate the power spectrum in detail . the result depends on the ratio of different speeds of sound , and shows an apparent amplification when the ratio deviates from unity .
1107.0813
inflationary cosmology has become the prevalent paradigm to understand the early stage of our universe , with its advantages of resolving the flatness , homogeneity and monopole problems @xcite , and predicting a approximately scale - invariant primordial power spectrum consistent with current cosmological observations @xcite very well . however , a single field inflation model often suffers from fine tuning problems on the parameters of its potential , such as the mass and the coupling constant . in recent years , people has noticed that , when a number of scalar fields are involved in the inflationary stage , they can relax many limits on the single scalar inflation model @xcite . usually , these fields are able to work cooperatively to give an inflationary stage long enough , even none of them can sustain inflation separately . models of this type have been considered later in refs . the main results showed that both the e - folding number @xmath1 and the curvature perturbation @xmath2 are approximately proportional to the number of the scalars @xmath3 . later , the model of n - flation was proposed by dimopoulos _ @xcite , which showed that a number of axions predicted by string theory can give rise to a radiatively stable inflation . this model has explored the possibility for an attractive embedding of multi - field inflation in string theory . over the past several years , based on the recent developments in string theory , there have been many studies on its applications to the early universe in inflationary cosmology . an interesting inflation model , which has a non - canonical kinetic term inspired by string theory , was studied intensively in the literature . due to a non - canonical kinetic term , the propagation of field fluctuations in this model is characterized by a sound speed parameter and the perturbations get freezed not on hubble radius , but on the sound horizon instead . one specific realization of this type of models can be described by a dirac - born - infeld - like ( dbi ) action @xcite . based on brane inflation @xcite , the model with a single dbi field was investigated in detail@xcite , which has explored a window of inflation models without flat potentials . in this model , a warping factor was applied to provide a speed limit which keeps the inflaton near the top of a potential even if the potential is steep . motivated by the effective description of multiple d - brane dynamics in string theory @xcite , an interesting inflation model involving multiple sound speeds with each sound speed characterizing one field fluctuation was proposed @xcite . the authors of refs.@xcite suggest this scenario can be realized by a number of general scalar fields with arbitrary kinetic forms , and these scalars have their own sound speeds respectively . therefore , this model is dubbed as _ multi - speed inflation _ " ( msi ) . in this model , the propagations of field fluctuations are individual , and the usual conceptions in multi - field inflation models might be not suitable in this scenario . for example , in a usual generalized n - flation model , the length scale for perturbations being freezed takes the unique sound horizon ; however , in our model it corresponds to the maximum sound horizon . it is worth emphasizing that the model of msi is different from the usual dbi n - flation in which only multiple moduli fields are involved in one dbi action @xcite , but msi is constructed by multiple kessence type actions . for example , an explicit model of msi is made of two dbi fields in ref @xcite , and its primordial perturbations including curvature and entropy modes and their non - gaussianities were considered . in the current paper we extend the model into a much more generic case by relaxing the form of the lagrangian . we find that even there are no coupling terms among the inflaton fields , their perturbation modes are coupled due to the nonlinear gravitational interaction . we also find that if these couplings are weak , the independent components of the fields can be treated as a combination of normal modes , by which the curvature perturbation at the horizon - crossing can be calculated in detail up to leading order of slow variation parameters via @xmath0 formalism . as a specific example , we consider a model of multi - speed inflation involving two dbi fields . we show that in the relativistic limit , the coupling between two fields is mainly reflected by the damping terms in the perturbation equations . the difference between the sound speeds of these two fields is able to generate a considerable amplification on the amplitude of the primordial curvature perturbation . this is greatly different from the usual analysis on the inflationary model constructed by two canonical fields@xcite . the paper is organized as follows . in section ii , we study the cosmological perturbation theory of the msi model in a generic case , then show that the inflaton fluctuations could be decoupled through a redefinition of fields at leading order under the assumption of weak couplings . in section iii , we focus on a model of msi involving two fields , and study the detailed field transformation matrix to illustrate that the decoupling process of the field fluctuations is reliable under the weak coupling approximation . specifically , we analyze a specific model constructed by two dbi fields and give its curvature perturbation . section iv presents a summary and discussion . in the paper we take the normalization @xmath4 and the sign of metric is adopted as @xmath5 in the following . an inflation model constructed with a single kessence was originally proposed by @xcite and later its perturbation theory was developed in @xcite . in the literature this type of model has been widely studied , and one of the most significant features is that there is an effective sound speed describing the propagation of the perturbations . in the model of msi , for each k - essence field there is one sound speed correspondingly . therefore , the field fluctuations in this model do not propagate synchronously due to different sound speed parameters , but get coupled because of gravitational interactions . in the current paper , our main interests focus on the effects of multiple sound speeds and mode - couplings at the level of linear perturbation . before studying the perturbations , we first take an investigation on the background equations . the action is the summation of hilbert - einstein action and the action of @xmath3 different fields minimally coupled with gravity@xcite @xmath6,\ ] ] with @xmath7 which is the kinetic term of the inflaton field @xmath8 . from now on the subscript of capital latin letter @xmath9 always denote the different inflaton fields . in this model , the klein - gordorn equations in a homogeneous background are @xmath10 here and thereafter , the subscript `` @xmath11 '' denotes the derivative with respect to the field @xmath8 , and @xmath12 " denotes the derivative with respect to @xmath13 , for simplicity . assuming a flat friedman - robertson - walker background , the friedman equations are given by @xmath14 where @xmath15 denotes the hubble parameter @xmath16 at a given time . to get analytical results , it is helpful to define some parameters to characterize the kinetic behavior of the fields . in single - field canonical inflation , a set of slow - roll parameters are necessary . to ensure their smallness is not only the requirement of inflation to occur and endure for long enough , but also some simplification in analytical calculation . in a generic case , just as in @xcite , we define a similar set of slow - variation parameters . recall that in the simplest single - field inflation with canonical kinetic energy and flat potential , one can define two slow - roll parameters @xmath17 here the @xmath18 symbol denotes the approximate equality which only holds in slow - variant environment and neglect the possible numerical factor . the second equality defines the slow - roll parameter via the kinetic side , whereas the third equality defines it in the potential viewpoint . if one only deal with the parameters under slow - roll condition , for instance in a model with canonical kinetic terms , these definitions are almost equal , and assuming one set is small can immediately assure the smallness of the other sets . this is just because the dynamic behavior simply relies on the shape of potential . but in a generic case , for instance in dbi or k - inflation model , the dependence may be much more complex and these different definitions are no more equivalent . since they are all very important in parameterizing our final result , we have to define two sets of slow - variation parameters from different viewpoints . in general , we should replace the slow - variation parameters by some slow - variation matrixes that can reflect the correlations between different fields . since we are dealing with a separable lagrangian , we only need different parameters for each field . this is equivalent to the case with a diagonal slow - variation matrix . as is mentioned above , one way to trace the motion is to define the parameter by each kinetic term of one specific field : @xmath19 in dbi model the field may roll down the potential very fast , but the parameter @xmath20 defined here is still small and slowly variant . the summation of such parameters gives the original version of them in canonical case : @xmath21 note that the magnitude of @xmath20 and @xmath22 obeys @xmath23 in general . to accomplish our calculation we need another version of slow - variation parameters defined in the potential aspect : @xmath24 this is an analogue of @xmath25 , and for future convenience we take a squared root . it is not necessarily small in an infrared ( ir ) dbi model , in which the value of the inflaton field may be very small . so for convenience let us focus on the ultraviolet ( uv ) dbi model where @xmath26 moves from uv side of the potential to the ir side . we should remind us that @xmath27 has the same order with @xmath20 in a model with canonical kinetic energy , but is very close to 1 in multi - dbi model . we refer to it as a half - order slow variation parameter in the former case , and tried to preserve terms higher than @xmath27 . besides , @xmath28 as another extension of @xmath29 should be defined . according to the speed of sound in the single - field case , we can define the speed parameter of each field : @xmath30 by this definition , the sound speed is the actually the effective propagation speed of the perturbations , which is different from the adiabatic sound speed which is defined as @xmath31 when the cosmic fluid is described by scalar fields@xcite . notice that in a multi - dirac - born - infeld model we have an extra relationship @xmath32 which will connect this parameter with @xmath20 . in general , @xmath33 needs not to be the same for all the fields , which makes this model different from the ordinary multi - dbi model and may give us some new phenomenon that we are about to investigate . for an analysis of @xmath34 as an inverse lorentz vector in a single - field dbi inflation , see @xcite . after having defined such parameters , the main quantities involving @xmath35 can be expressed in terms of these parameters directly from the inverse relations above . for instance , @xmath36 besides the @xmath37 s , another important quantity is the time derivative of one field . by using the klein - gordon equation ( [ kg equation ] ) we have @xmath38 in this subsection we perform a generic analysis on the linear perturbations of the model introduced previously . since we are working in the frame of a cosmological system , the metric perturbations ought to be included as well as the field fluctuations . we would like to expand the action to the second order by virtue of the arnorwitt - deser - misner ( adm ) formalism @xcite . to start , the spacetime metric in the adm formalism is written as , @xmath39 the action can reads @xmath40 with @xmath41 . when expanding the action up to the second order , we can decompose the fields and lagrangian multipliers as @xmath42 first , the primary hamiltonian and the momentum constrains gives two relations @xmath43 then we can make the perturbative expansion of the action up to second order by using the lagrangian constraints : @xmath44\right\},\end{aligned}\ ] ] with @xmath45 as the perturbation of @xmath13 up to first order . this result is in consistency with the one derived in the msi model @xcite and see @xcite for a general case . the lagrangian ( [ s_2 ] ) contains a lot of coupled terms as @xmath46 and @xmath47 . these terms imply that the perturbations of inflaton fields depend on others during their evolution , and the equations of motion would be difficult to be solved directly . in this note we aim at making the transformation of the inflaton perturbations as to obtain approximately decoupled equations of motion for new fields . in principle this is difficult , if not impossible to realize . we will see in which case can we do such transformations and see what will happen by such a method . this will require some constraints on the slow - variation parameters . we will see this in detail for a double - field toy model . after utilizing the equations of motion for background and taking some integration by parts , one can write the lagrangian in a more compact form : @xmath48 with @xmath49 and @xmath50 are time - dependent damping and mass terms , @xmath51 this result is consistent with that obtained in multi - field inflation@xcite . to distinguish the diagonal and off - diagonal parts of the matrices above , we define @xmath52 after assuming that all the slow - varying parameters defined in sec . [ sect - intro ] are small , we have @xmath53 , \\ \mathcal{n}_{ij } & = & -h\left(\frac{1}{\csi^2}-1\right)e_ie_j\left[1+\frac{1}{3}\epsilon\left(\frac{\eta_i}{\epsilon_i}+\frac{\eta_j}{\epsilon_j } \right)\right ] , \\ \mathcal{m}_i & = & 3h^2\iota_i , \\ \mathcal{m}_{ij } & = & -3h^2e_ie_j\left [ 1+\frac{1}{6}\sum_k\epsilon_k\left(\frac{1}{\csk^2}-1\right)\right].\end{aligned}\ ] ] here , note that all these quantities are of order @xmath54 if @xmath55 . in such a case , as we will see below , the damping term has no correlations . for ordinary terms we have only preserved the leading order to slow - variation parameters , but if a term involves @xmath56 , we preserve one higher order . this is to ensure the consistency when @xmath55 fails , as in a model with multiple dbi actions . under the condition when @xmath57 , all the terms should be preserved . to investigate further we can derive the equations of motion for the fields @xmath58\delta_{ij}\right . \\\nn & + & \left.\frac{\csi^2}{h}\sum_j\left(\frac{1}{\csj^2}-1\right)p_j,_{x_j}\dot\phi_i\dot\phi_j\right\}\dot{\delta\phi}_j\\\label{eom : general } & + & \left[\left(\frac{k^2\csi^2}{a^2}+\frac{\csi^2}{p_i,_{x_i}}\mathcal{m}\right)\delta_{ij } + \frac{\csi^2}{p_i,_{x_i}}\sum_j\mathcal{m}_{ij}\right]\delta\phi_j=0.\end{aligned}\ ] ] note there is no summation over index @xmath59 " even if it repeats in one term . here and throughout the paper einstein s summation rule is not applied for the index of fields ( but still valid for repetition of spatial sub / superscripts ) and all the summations over capital latin letters @xmath9 " are written explicitly . in a generic case , we can rewrite the equations of motion as follows , @xmath60 with @xmath61 which are also small in inflation . we see there are two different couplings in the damping and effective mass term respectively . besides , they both depend on time via the parameters . in principle , this kind of equation is difficult to solve , if not impossible . of course we see that the coupling terms between different fields are all very weak if we admire the slow - variation parameters to be small , which may maintein the final solutions almost the same as those of @xmath3 independently evolving fields . further more , two special cases which have specific limits of sounds of speed are most interesting . one is that all fields @xmath8 have the canonical kinetic energy , with @xmath62 . this is a decoupled multi - slow - roll model , which is called assisted inflation by @xcite and received much focus@xcite . in such a case , @xmath63 and therefore can be totally neglected in the action , while @xmath64 . we notice that only the correlations in mass term are preserved , which are easy to manipulate . on the other hand , when all the fields are dbi - type , as is discussed in @xcite , @xmath33 is very small in the relativistic limit . for simplicity , we can suppose that the sound speeds are as small as @xmath22 , or even smaller . we will shortly confirm that , the latter case is naturally suitable for analytical calculation , while for former case one should impose further assumptions on the difference of @xmath65 s . but whatever case it is , we see @xmath66 which is the main part of coupling , and @xmath67 thus can be neglected . one possible method is to consider the smallness of couplings , and rotate the perturbative fields in field - space into a instantaneous orthogonal basis which can generate the adiabatic perturbation which is in the direction of the trajectory of fields in background , along with @xmath68 entropy perturbations perpendicular to the direction of field trajectory . after doing that , if we suppose the coupling terms which are proportional to the derivatives of the rotating angles with respect to time of the trajectory in field - space is small at the moment of horizon crossing , we can neglect the coupling and treat the adiabatic and entropy perturbations independently as free streaming quantum fields . after some proper quantization , all the discussions are similar to a @xmath3 decoupled multi - field theory@xcite . next if we want to consider the effect of couplings that have been omitted , one can treat such couplings as interacting vertices to the quantum fields , and see how can the fields transfer to each other in the quantum level@xcite . actually , if the fields have different speeds of sound , it is difficult to do such rotations as to make the adiabatic perturbation laid on the direction of trajectory . this is because the angle we need to take our rotation is different from that of the slope of trajectory , since sound speeds will appear in the transformation and make the `` rotation '' anisotropic . further more , different sound speeds correspond to an anisotropic rescaling in the rotation of basis . we will come back and explain this assert in a two - field case by illustration later . if we abandon the attempts to project the field into adiabatic perturbation and entropy perturbations , we may find another route to get our results . a reasonable method is to transform the fields in order to eliminate the couplings , if possible , and then treat the decoupled perturbations as free quantum fields , with a final result of curvature perturbation @xmath2 gained by @xmath0 formalism . before doing this , let us digress to take a look at the solution if we discard all the couplings and see the free fields limit of our theory . the result obtained here is also useful for future convenience . under this situation , the equations contains no coupling , while the only remaining task is to find the solution of each independent field , and do quantization in ultraviolet limit . we can consider the equations of motion as @xmath69 where the parameter @xmath70 and @xmath71 have been defined in ( [ kappa ] ) and ( [ mi ] ) . this is a little different from the standard form of single - field inflation because of an additional term in the damping effect . we consider the fields moving in an inflationary quasi - de sitter phase , which means the comoving time has a simple form@xcite @xmath72 and we can postulating a rescaling on the fields , @xmath73 note that , our rescaling contains an extra power of @xmath74 which is a small quantity of order @xmath54 comparing with the standard definition of mukhanov - sasaki variable@xcite due to the small shift in the diagonal damping term of the lagrangian . this extra power can be determined by requiring the coefficients of @xmath75 term in the equation of motion vanish . the mukhanov - sasaki variable , as in the standard process before taking the quantization of the fields , is actually defined via the normalizer before the kinetic terms in the action([action:2nd order ] ) . therefore , we can convert the equation into a standard form which is familiar to that in single field theory : @xmath76\chi_i=0.\ ] ] this is just the bessel equation of order @xmath77 which will be determined below . the solution to this equation is a linear combination of two independent bessel functions . however , to have the appropriate approximation behavior which approaches the planar wave in bunch - davies vacuum with correct normalization at early time @xmath78 @xcite , we need to choose the proper coefficients of the bessel functions , which corresponds to such a state in quantum theory that it can minimize the energy density . one possible result is to choose@xcite @xmath79 where @xmath80 is the hankel function of the first type , with @xmath81 as its order . note that , @xmath82 is a small quantity of order @xmath54 , which is irrelevant of the value of @xmath33 in a multi - dbi model when @xmath83 . therefore , the amplitude of perturbation to the original field @xmath8 is @xmath84 again we use the fact that both @xmath77 and @xmath85 are very small , thus we get the power spectrum of each field @xmath86 \left(\frac{\epsilon_i}{e_i^2}\right)^{\beta_i}\left(\frac{h}{k}\right)^{2\beta_i}(1-\epsilon)^{2(1-\sigma_i)},\\\label{spectrum i background } & \approx&\left(\frac{h}{2\pi}\right)^2\frac{2}{\csi}\left(\frac{h}{k}\right)^{2\beta_i}\frac{\epsilon_i}{e_i^2 } \left[1 - 2\epsilon+\beta_i\ln2+\sigma_i\left(\gamma-2\right)+\beta_i\ln\frac{\epsilon_i}{e_i^2}\right],\end{aligned}\ ] ] where the second approximate equality holds at the moment when the wavelength of the mode considered exits the sound horizon , i.e. @xmath87 . @xmath88 is the digamma function and relates to euler - mascheroni constant @xmath89 as @xmath90 . the dependence on the speed of sound is superficially different from that of single - field dbi inflation . but when we take an extra relationship @xmath91 which holds in dbi - type action , a new relation between the parameters @xmath33 , @xmath92 and @xmath20 , holds as to leading order , @xmath93 therefore the dependence on @xmath33 in ( [ spectrum i background ] ) is canceled , and ( [ spectrum i background ] ) will be coincident up to leading order with the result in @xcite . by virtue of @xmath0 formalism @xcite . @xmath94 where @xmath1 is the local e - folding number along a trajectory from a spatially flat slice at a moment @xmath95 soon after the relevant scale has passed outside the horizon during inflation , to a uniform - density slice at another moment @xmath96 after complete reheating when @xmath97 has become a constant . so the power spectrum curvature of perturbation @xmath97 can be calculated by @xcite . @xmath98 where @xmath99 can be determined via @xmath100 in a general case . here for simplicity , we should consider the case when the @xmath3 fields moves almost in the same manner . this is proved by @xcite to be a late - time attractor for some appropriate initial conditions . later on we will demonstrate our analytical calculation for a system with different sound speeds is valid when @xmath101 , which implies a field configuration like this . under such assumptions , to order @xmath54 one have @xmath102 in a generic case . therefore , together with ( [ extra relation in dbi ] ) we get @xmath103.\end{aligned}\ ] ] the superscript 0 denotes this power spectrum is the one for a theory neglecting couplings . we can see if all the fields have the same speed of sound , and @xmath104 , ( [ p zeta uncoupled ] ) gives the results in multifield dbi inflation@xcite which contains @xmath3 identifying speeds of sound up to first order of slow variation parameters . from ( [ spectrum i background ] ) , the scalar spectral index @xmath105 is @xmath106 and the contribution to spectral index of @xmath97 mainly comes from the largest one of @xmath107 s , which is just the power spectuum of curvature perturbation . to consider the coupling terms as a perturbations to equations of motion ( [ eom : unperturbed ] ) , we could take a redefinition of inflaton fields through a linear transformation which , up to leading order in slow - variation parameters , can give a set of @xmath3 decoupled equations of motion , with different coefficients appearing as new speeds of sound and new effective mass terms . in general , this is difficult to be realized . fortunately , we are dealing with a theory that is just perturbed from a decoupled one . as the @xmath3 fields have tendency to move together , they have an even smaller difference in the parameters . intuitively the change in eigenvalue should be small if the damping and mass matrix is not far apart from a diagonal one . actually hoffman - wielandt theorem in matrix analysis tells us that the smallness of `` perturbation '' to a diagonal matrix will cause small variation to the eigenvalues . to make it clear let us denote @xmath108 as the eigenvalues of a diagonal matrix corresponding to a theory with no couplings , whose diagonal entries are just the coefficients before each @xmath109 term in ( [ eom : unperturbed ] ) , and @xmath110 to be eigenvalues of the `` perturbed '' matrices corresponds to the coupled coefficients ( [ eom : general ] ) then , there exists a permutation @xmath111 for the new eigenvalues @xmath112 that satisfies@xcite @xmath113^{1/2}\leq\parallel\delta m\parallel_2\sim\scro(n\epsilon).\ ] ] here @xmath114 is the euclidean norm of a perturbation matrix @xmath115 which under our circumstance has the magnitude of order @xmath54 . thus , roughly speaking , the variation in eigenvalue is a small quantity of order less than @xmath54 , if @xmath3 is not very large . this fact ensures our validity of our upcoming discussions on estimating the eigenvalues after field redefinition . let us analyze the coupled equations under the fact that all the coupling terms are small quantities of order @xmath54 . in this case we expect the solutions will differ from the decoupled ones very little . later on we will see that , even under such case , different sound speeds will bring a fruitful physical imprint . if we impose a transformation in field space such that @xmath116 where @xmath117 is a non - singular transformation matrix . if we require the transformation to be a representation of @xmath118 we can define a new basis of perturbations which can be interpreted as adiabatic / entropy directions , this matrix should be set orthogonal . but here it is not possible in our case . our purpose of transforming the fields is to decouple . and unless for some special cases , the new fields has nothing to do with adiabatic / entropy perturbations . actually , if all the fields have the same speed of sound , the fields redefinition ( [ field rotation ] ) could be an orthogonal transformation of @xmath118 group . and one of the new basis would lay on the direction of tangent line of trajectory in field space , which just duplicate the result in @xcite . however , for fields with different speeds of sound , we do not have such clear geometric explanation . in such a case , we will see that the transformation is even only orthogonal up to leading order @xmath119 . later on we will see how we can explain the new transformation in a field space . to make our discussion clear we could require @xmath117 is also slow - variant , which means @xmath120 . this is reasonable in a quasi - de sitter background , and its validity will be confirmed after we get the concrete form of @xmath121 . thus , one can write the time derivative of new field transformed from @xmath122 as @xmath123 substitute into the equations of motion ( [ eom : general ] ) , we have the equations for new variable @xmath124 . left multiply an inverse transformation @xmath125 and take the summation over index @xmath59 , we get @xmath126\dot\vp_m \\\nn & + & \sum_{m , i , j}(r^{-1})_n^i\left[\left(\frac{k^2\csi^2}{a^2}r_j^m+m_i^2r_j^m+3h(1+\kappa_i)\dot r_j^m\right)\delta_{ij}\right.\\ & + & \left.m_{ij}r_j^m+h\xi_{ij}\dot r_j^m\right]\dot\vp_m=0.\end{aligned}\ ] ] if we need to decouple these equations into the form of ( [ eom : neglect couple ] ) and get a set of free propagating waves in ultraviolet , we should find a way to diagonalize the two coefficients in the square brackets before @xmath127 and @xmath124 for each equation at the same time . this is the simultaneous diagonalization of two matrices , and is in principle impossible unless the two matrices are commutative . the coefficient matrices does not commute each other since @xmath128 is even asymmetric . all the diagonal matrices are commutative . so the only remaining matrices that may be non - commutative are those with @xmath129 and @xmath128 as their elements . since their main part are diagonal elements that surely commutes , we will justify that the complete matrices with a little variation in off - diagonal parts are believed to simultaneously diagonalizable up to @xmath54 . if we can find a way to do this , then we can write @xmath130 & = & 3hk_n\delta_n^m,\\\nn \sum_i\left[3h(r^{-1})_n^i(1+\kappa_i)\dot r^m_i+\frac{k^2}{a^2}(r^{-1})_n^i\csi^2r_j^m\right . & & \\\label{diagonalize mass}\left.+(r^{-1})_n^im_i^2r_i^m+\sum_j(r^{-1})_n^im_{ij}^2r_j^m\right ] & = & \left(\frac{k^2}{a^2}c_{\mathrm{s}n}+m_n\right)\delta_n^m.\end{aligned}\ ] ] here we have neglect a @xmath131 term since @xmath132 is of order @xmath54 . under such transformation the equations of motion are completely decoupled , @xmath133 now we use the hoffman - wielandt theorem which ensures us that the solution to such eigenvalue equation of a perturbed matrix is a small variation from the eigenvalue before . according to the fundamental theorem of algebra , there are @xmath3 sets of such solutions to ( [ diagonalize damping ] ) and ( [ diagonalize mass ] ) , each has a new speed of sound @xmath134 . the @xmath135 s are the speeds of propagation of the new fields , akin to those of normal modes in coupled oscillation system or phonon representation in statistical system . to investigate whether the new representation will change qualitatively the physics of an decoupled system is an interesting question . for instance when @xmath33 s are all close to 1 , can we find a mode that has a very small speed of sound @xmath135 that can possibly amplify the power spectrum and non - gaussianity ? we should emphasize that although this is possible , it will not appear in an weakly coupled system we are just dealing with since the reason we ve mentioned above . on the contrary , if the couplings are indeed strong , the physics could be much more different . for example , in hybrid inflation@xcite where the two fields have complicated coupling which is not small at the waterfall stage of inflation , there are plentiful physical phenomena far beyond our results here @xcite . even in a weakly coupled system we are considering , analytical calculation is difficult to do . but after we can diagonalize the damping and mass matrices and take a set of new fields @xmath136 which can simplify the equations of motion ( [ eom : general ] ) to ( [ eom : decoupled ] ) , the remaining task is very simple , just parallel to the discussion under ( [ eom : neglect couple ] ) since the form of the equation is the same , only with a different definition of parameters @xmath137 , @xmath138 and @xmath139 . the most difficult part of our method is to find the transformation @xmath121 . in the next section we will calculate in detail for an example of two - field inflation and get a taste of such scheme . to calculate the curvature perturbation at horizon - crossing in detail we need the quantization of the fields in ultraviolet , which requires a set of independent wave equations . as we mentioned above , to simultaneously diagonalize so many matrices in ( [ diagonalize damping ] ) and ( [ diagonalize mass ] ) seems to be an impossible task . but after we notice all the off - diagonal entries are very small and superadd an assumption that the diagonal entries are almost the same up to @xmath54 , we can decouple the equations in a natural manner . in a two - field model , the fields which drive inflation are @xmath140 and @xmath141 , and a new definition of fields provides two equations for the eigenvector of the coefficient matrix , ( [ diagonalize damping])([diagonalize mass ] ) . we have seen that for a generic value of @xmath33 the off - diagonal damping and mass term will coexist . but for a model with canonical kinetic energy , @xmath62 and there is only coupling in mass term@xcite . on the other hand if @xmath33 is small as in the multi - dbi model , the coupling in damping term survives . let us work in the multi - dbi case , leaving aside the matrix @xmath129 which has been proved to be up to @xmath142 , as has been discussed in the paragraph under ( [ mij ] ) . therefore the only remaining matrix to be diagonalized is @xmath128 , which has the eigenvalues @xmath143 with the eigenvalues @xmath144 these eigenvalues corresponds to a transformation matrix @xmath145 actually the eigenvector can only determine this matrix up to two normalization constants . here we adopt a normalization such that it will preserve the determinant to be unity . this is not important in the quantization process , but will affect the calculation of power spectrum later . we know the similarity transformation @xmath121 will diagonalize the matrix with elements @xmath128 . next we will show that it can also diagonalize the diagonal matrices with different diagonal entries , say @xmath146 , under some reasonable assumptions about the model . when the similarity matrix @xmath121 operates on @xmath147 , @xmath148 so a diagonal matrix which has different diagonal entries can remain diagonal after the similarity transformation @xmath121if @xmath149 with @xmath85 s and @xmath20 s in the matrix are still of order @xmath22 . we will see , that for a specific multi - dbi model , the assumption that @xmath150 as well as @xmath151 and @xmath152 are of order @xmath153 is reasonable , since the difference between two small positive quantities of order @xmath54 must be even smaller , if the fields have the tendency to move together under an appropriate initial condition . besides , we will suppose that the differences between other parameters like @xmath154 and @xmath155 , are also as small as @xmath153 for future convenience . note , that this do not implies @xmath156 which is a much more strict constraint on the evolution of fields . we will come back to this issue and make some necessary justification to a specific two - field - dbi model later . at this moment , we just emphasize that this assumption preserve the diagonal property of the diagonal matrices in ( [ diagonalize damping ] ) and ( [ diagonalize mass ] ) involving @xmath85 , @xmath157 and @xmath158 automatically up to @xmath153 under a similarity transformation by any matrix @xmath121 . next let us check whether this matrix can make @xmath159 diagonal . the entries of the matrix are @xmath160,\\ ( \mathbb{r}^{-1}\dot{\mathbb{r}})_{12 } & = & \frac{h\epsilon_1}{\epsilon_1+\epsilon_2 } \frac{e_2c_{\mathrm{s}1}^2}{e_1c_{\mathrm{s}2}^2 } \left[(h_1-h_2)-2(s_1-s_2)\right],\\ ( \mathbb{r}^{-1}\dot{\mathbb{r}})_{21 } & = & \frac{h\epsilon_2}{\epsilon_1+\epsilon_2 } \frac{e_1c_{\mathrm{s}2}^2}{e_2c_{\mathrm{s}1}^2 } \left[(h_1-h_2)-(\eta_1-\eta_2)-2(s_1-s_2)\right].\end{aligned}\ ] ] we see all the entries are proportional to the differences among some slow - variation parameters of the two fields , which are of order @xmath153 as we mentioned before . then we can assert that up to @xmath54 the time derivative of @xmath121 multiplied by its inverse in ( [ diagonalize damping ] ) is negligible . we will pause for a while and see the possible geometric explanation of transformation @xmath121 . we have seen that it is only orthogonal up to @xmath54 under the assumption that @xmath156 . when @xmath161 as is the case of ordinary assisted inflation , one have that the transformation @xmath121 is just rotating the basis by @xmath162 and @xmath163 is just the adiabatic perturbation with @xmath164 the entropy perturbation which is negligible , as is depicted in fig.[pert1](a ) . a more interesting case is the theory with different speeds of sound , as in ( b ) . is depicted by a transformation in field space of the perturbations . ( a ) denotes the trivial case when @xmath165 , @xmath166 and @xmath161 . we see for a totally symmetric configuration @xmath121 poses an so(2 ) rotation by an angle @xmath162 , and @xmath140 is just the adiabatic perturbation in usual case . ( b ) shows a more interesting result when @xmath167 , @xmath168 and @xmath169 . note this is to ensure @xmath170 such that field are moving along a diagonal trajectory in fields space . we see that @xmath121 represents a transformation that rotates the field and rescales it at the same time with different rotating angle and anisotropic rescaling factor for two fields , i.e. the red arrow denoting the basis of new fields are not orthogonal , and the new fieds @xmath124 can not preserve the unitarity.,title="fig:",scaledwidth=45.0% ] is depicted by a transformation in field space of the perturbations . ( a ) denotes the trivial case when @xmath165 , @xmath166 and @xmath161 . we see for a totally symmetric configuration @xmath121 poses an so(2 ) rotation by an angle @xmath162 , and @xmath140 is just the adiabatic perturbation in usual case . ( b ) shows a more interesting result when @xmath167 , @xmath168 and @xmath169 . note this is to ensure @xmath170 such that field are moving along a diagonal trajectory in fields space . we see that @xmath121 represents a transformation that rotates the field and rescales it at the same time with different rotating angle and anisotropic rescaling factor for two fields , i.e. the red arrow denoting the basis of new fields are not orthogonal , and the new fieds @xmath124 can not preserve the unitarity.,title="fig:",scaledwidth=45.0% ] + after the transformation , one can find that two equations ( [ eom : general ] ) are decoupled , which is just ( [ eom : decoupled ] ) , @xmath171 with @xmath172 to go further let us study the change of the normalizer @xmath173 under the similarity transformation by @xmath121 . note that , for example , when a normalizer matrix @xmath174 , which is also diagonal , is multiplied by another matrix , say @xmath175 , the product is transformed under @xmath121 by @xmath176 which is just multiplying each diagonal entries after the similarity transformation respectively . denote by @xmath177 the diagonal entries of @xmath178 , @xmath179 after doing this , we can turn to the same process after equation ( [ eom : decoupled ] ) . then we define new field variables @xmath180 which satisfy the equations of motion @xmath181\chi_i=0.\ ] ] now the equations have been decoupled . for each equation , the solution is a hankel function with order @xmath182 , where @xmath183 finally , we obtain the power spectra of the new fields @xmath184 as follows , @xmath185 , \\\nn \mathcal{p}_{\varphi_2 } & = & \left(\frac{h}{2\pi}\right)^2\frac{2}{c_{\mathrm{s}2}}\left(\frac{kc_{\mathrm{s}2}}{ha}\right)^{2\varsigma_2 } \frac{c_{\mathrm{s}1}^2(\epsilon_1+\epsilon_2)^2}{(c_{\mathrm{s}1}e_2\epsilon_2\epsilon_1^{-1/2 } + c_{\mathrm{s}2}e_1\epsilon_1\epsilon_2^{-1/2})^2 } \\\nn & \cdot&\!\!\!\!\left[1 - 2\epsilon-2\varsigma_2\left(\ln2+\psi\left(3/2\right)\right)+\left(3k_2 - 2\bigtriangleup_2\right ) \ln\frac{c_{\mathrm{s}1}c_{\mathrm{s}2}(\epsilon_1+\epsilon_2 ) } { c_{\mathrm{s}1}e_2\epsilon_2\epsilon_1^{-1/2}+c_{\mathrm{s}2}e_1\epsilon_1\epsilon_2^{-1/2}}\right].\\\end{aligned}\ ] ] the power spectra above are calculated at the moment when the wavelength exceeds the sound horizon , respectively . when one field , say @xmath164 have exceeded the sound horizon , the power spectrum of @xmath164 is frozen , and so the curvature perturbation calculated later should involve the value of @xmath186 calculated at @xmath187 which satisfies @xmath188 . the power spectrum of curvature perturbation should be calculated at the moment when the wavelength has exceeded all the sound horizons corresponding to different @xmath189 s . the power spectra are nearly scale - invariant , with @xmath190 as their spectral index . the spectral index of curvature perturbation @xmath97 is thus @xmath190 which has the largest absolute value . using @xmath0 formalism , we can calculate the curvature perturbation after the wavelength being stretched out of the largest sound horizon . note that to leading order @xmath191 there is no correlation between different @xmath124 s since they are independent kinetic modes as we defined . after using the additional relation ( [ extra relation in dbi ] ) we can substitute @xmath192 in the expression by @xmath193 . therefore we can simplify the final result @xmath194 the derivatives of the e - folding number can be calculated in a specific model in detail , for instance in a model with standard kinetic energy and separable potential@xcite . since we only want to emphasize the effects caused by different sound speeds , again we use the assumption that both fields are moving in the same manner , i.e. in the diagonal trajectory in field space as in ( b ) of fig.[pert1 ] , and ( [ n , i ] ) holds . this kind of motion is nearly adiabatic , which will always suppress the isocurvature perturbation inside the sound horizon . the final result is a little lengthy and instead of writing it here , we would rather focus on the dependence on different sound speeds by calculating @xmath195 . by defining @xmath196 one gets @xmath197 now the power spectrum degenerates to @xmath198 which now just contains quantities up to leading order , and is in coincidence with ( [ p zeta uncoupled ] ) when @xmath199 , i.e. @xmath161 . this normalized power spectrum seems to be divergent when @xmath200 or @xmath201 . however , since we require @xmath202 , the corresponding physical region is the neighborhood of @xmath203 . we depict the dependence on @xmath204 in fig.[p - c ] . of the power spectrum . both the power spectra of @xmath124 are normalized by @xmath205 . and @xmath206 is normalized by the un - coupled one ( [ p zeta uncoupled ] ) . there is severe divergence of @xmath206 when one of the sound speed vanish , but the only valid domain under our assumption is the case when @xmath207 , i.e. in the neighborhood of @xmath208 . we see the amplification by the reciprocal sound speed is much larger than that in a trivial @xmath199 case of ordinary unique - speed model . ] we can see from fig.[p - c ] that , there is a minimum of @xmath206 when @xmath209 , where @xmath210 . but the most interesting aspect is that if one of the speed of sound is a few times larger than the other ( with themselves still as small as @xmath153 so as to satisfy our condition required before ) , the power spectrum of curvature perturbation increases magnificently . we see from fig.[p - c ] and ( [ p(x ) ] ) that if @xmath211 , the contribution from @xmath212 exceeds that from @xmath213 , and vis visa if @xmath214 . therefore we have encountered an interesting feature different from our intuition out of a theory without coupling , that the field with smaller speed of sound will not in general contribute more to the curvature perturbation . for instance , if @xmath215 , it is the power spectrum from auxiliary field @xmath212 which contributes more to the final curvature perturbation , not @xmath216 . we also see that if @xmath217 deviates from 1 , the power spectrum of curvature perturbation is amplified significantly , compared with the one without couplings . this just tells us that the effect given by the coupling terms may be large and can not be neglect in previous calculations . since the power spectrum of curvature perturbation can be observed to rather high accuracy , this result actually tells us that we can loosen the constraints on the slow - variation parameters of individual inflaton even further . in the frame of inflationary cosmology involving multiple inflaton fields , it is possible to allow these fields possess their own sound speed parameters which characterize the propagation of field fluctuations correspondingly . it is important to study how these modes are related to the cosmological perturbation we may observe in cmb experiments . in the present paper we analyzed the dynamics of field fluctuations in a general scenario of msi . assuming no direct coupling terms among the inflaton fields , we noticed that these perturbation modes are generically coupled both through their effective mass terms and cosmic damping terms . fortunately , we found that if these couplings are weak , the perturbation theory can be treated as a combination of normal modes , by which the curvature perturbation at the horizon - crossing can be calculated in detail up to leading order of slow - variation parameters via @xmath0 formalism for a specific model . specifically we studied a model of multi - speed inflation involving two dbi fields with their sound speeds being small and slow - varying . our analysis showed that in the relativistic limit , the coupling between two field fluctuations is mainly contributed by the damping terms in their perturbation equations . we further studied the curvature perturbation in this model , and verified their primordial power spectra are nearly scale - invariant . this has interesting implications to the curvaton scenario@xcite , which suggests a light field in inflationary phase could seed entropy perturbation and convert it into curvature perturbation at the end of inflation@xcite . namely , as shown in @xcite , a light dbi field can realize the curvaton scenario and generate sizable non - gaussianities of local and equilateral types after suitable fine - tuning on the model parameters . our study also showed that , when the sound speed parameters for the inflaton fields are not equal , the perturbation modes would get frozen at different sound horizons . an important phenomenon is that the curvature perturbation of msi could obtain an enhancement which depends on the ratio of the sound speeds . the scenario of msi is a very important branch of various models of multiple field inflation . in recent years , the inflationary models involving multiple field components have been studied extensively in the literature , namely , the analysis on the dynamics of n - flation @xcite ; multiple field inflation with particle decays @xcite ; the n - flation model in the frame of braneworld @xcite ; the model of staggered nflation in the stringy landscape @xcite ; dynamics of the model of coupled n - flation with canonical fields @xcite ; the inflation model constructed by a dbi field coupled to radiation @xcite ; and see @xcite for comprehensive reviews on this field . it has been well realized that , for the model of multi - field inflation , when the sound speeds for the fields are the same , the field fluctuations are able to be rotated orthogonally to a basis in which the curvature and entropy fluctuations decouple explicitly , at least up to leading order . however , when we relax the scenario by allowing the sound speed parameters are different , our analysis implied that the rotation of the field fluctuations are no longer orthogonal but we still are able to find a transformation to decouple the fields under some conditions and obtain the curvature perturbation via @xmath218 formalism . we are grateful to antonio de felice , xian gao and tower wang for useful discussions . we thank yi - fu cai and bin chen for thoroughly reading the manuscript and providing valuable comments on our work . this work is supported by the nsfc grant no.10975005 and rfdp . sp is supported by scholarship award for excellent doctoral student granted by ministry of education of china . a. h. guth , phys . d * 23 * , 347 ( 1981 ) ; a. d. linde , phys . b * 108 * , 389 ( 1982 ) ; a. albrecht and p. j. steinhardt , phys . lett . * 48 * , 1220 ( 1982 ) . for some early attempts we refer to : a. a. starobinsky , phys . b * 91 * , 99 ( 1980 ) ; k. sato , mon . not . soc . * 195 * , 467 ( 1981 ) . a. r. liddle , a. mazumdar and f. e. schunck , dimopoulos:2005ac phys . d * 58 * , 061301 ( 1998 ) [ arxiv : astro - ph/9804177 ] . k. a. malik and d. wands , phys . d * 59 * , 123501 ( 1999 ) [ arxiv : astro - ph/9812204 ] . p. kanti and k. a. olive , phys . d * 60 * , 043502 ( 1999 ) [ arxiv : hep - ph/9903524 ] . e. j. copeland , a. mazumdar and n. j. nunes , phys . d * 60 * , 083506 ( 1999 ) [ arxiv : astro - ph/9904309 ] . a. m. green and j. e. lidsey , phys . d * 61 * , 067301 ( 2000 ) [ arxiv : astro - ph/9907223 ] . s. dimopoulos , s. kachru , j. mcgreevy and j. g. wacker , jcap * 0808 * , 003 ( 2008 ) [ arxiv : hep - th/0507205 ] . o. aharony , s. s. gubser , j. m. maldacena , h. ooguri and y. oz , phys . rept . * 323 * , 183 ( 2000 ) [ arxiv : hep - th/9905111 ] . r. c. myers , jhep * 9912 * , 022 ( 1999 ) [ arxiv : hep - th/9910053 ] . g. r. dvali and s. h. h. tye , phys . lett . b * 450 * , 72 ( 1999 ) [ arxiv : hep - ph/9812483 ] . e. silverstein and d. tong , phys . d * 70 * , 103505 ( 2004 ) [ arxiv : hep - th/0310221 ] ; x. chen , phys . d * 71 * , 063506 ( 2005 ) [ arxiv : hep - th/0408084 ] ; x. chen , jhep * 0508 * , 045 ( 2005 ) [ arxiv : hep - th/0501184 ] . s. thomas and j. ward , phys . d * 76 * , 023509 ( 2007 ) [ arxiv : hep - th/0702229 ] . d. a. easson , r. gregory , d. f. mota , g. tasinato and i. zavala , jcap * 0802 * , 010 ( 2008 ) [ arxiv:0709.2666 [ hep - th ] ] . m. x. huang , g. shiu and b. underwood , phys . rev . d * 77 * , 023511 ( 2008 ) [ arxiv:0709.3299 [ hep - th ] ] . j. ward , jhep * 0712 * , 045 ( 2007 ) [ arxiv:0711.0760 [ hep - th ] ] . d. langlois and s. renaux - petel , jcap * 0804 * , 017 ( 2008 ) [ arxiv:0801.1085 [ hep - th ] ] . c. r. contaldi , g. nicholson and h. stoica , arxiv:0807.2331 [ hep - th ] . d. langlois , s. renaux - petel and d. a. steer , arxiv:0902.2941 [ hep - th ] . l. p. chimento , r. lazkoz and m. g. richarte , phys . rev . d * 83 * , 063505 ( 2011 ) [ arxiv:1011.2345 [ astro-ph.co ] ] . d. wands , n. bartolo , s. matarrese and a. riotto , phys . d * 66 * , 043520 ( 2002 ) [ arxiv : astro - ph/0205253 ] . c. gordon , d. wands , b. a. bassett and r. maartens , phys . d * 63 * , 023506 ( 2001 ) [ arxiv : astro - ph/0009131 ] . l. p. chimento , r. lazkoz and m. g. richarte , phys . d * 83 * , 063505 ( 2011 ) [ arxiv:1011.2345 [ astro-ph.co ] ] . c. armendariz - picon , t. damour and v. f. mukhanov , phys . b * 458 * , 209 ( 1999 ) [ arxiv : hep - th/9904075 ] . j. garriga and v. f. mukhanov , phys . b * 458 * , 219 ( 1999 ) [ arxiv : hep - th/9904176 ] . x. chen , m. x. huang , s. kachru and g. shiu , jcap * 0701 * , 002 ( 2007 ) [ arxiv : hep - th/0605045 ] . a. j. christopherson and k. a. malik , phys . b * 675 * , 159 ( 2009 ) [ arxiv:0809.3518 [ astro - ph ] ] . a. j. christopherson and k. a. malik , phys . b * 675 * , 159 ( 2009 ) [ arxiv:0809.3518 [ astro - ph ] ] . f. arroja , s. mizuno and k. koyama , jcap * 0808 * , 015 ( 2008 ) [ arxiv:0806.0619 [ astro - ph ] ] . x. gao , jcap * 0806 * , 029 ( 2008 ) [ arxiv:0804.1055 [ astro - ph ] ] . d. langlois , s. renaux - petel , d. a. steer and t. tanaka , phys . * 101 * , 061301 ( 2008 ) [ arxiv:0804.3139 [ hep - th ] ] . d. langlois , s. renaux - petel , d. a. steer and t. tanaka , phys . rev . d * 78 * , 063523 ( 2008 ) [ arxiv:0806.0336 [ hep - th ] ] . x. gao , jcap * 1002 * , 019 ( 2010 ) [ arxiv:0908.4035 [ hep - th ] ] . n. bartolo , e. komatsu , s. matarrese and a. riotto , phys . rept . * 402 * , 103 ( 2004 ) [ arxiv : astro - ph/0406398 ] . v. f. mukhanov , h. a. feldman and r. h. brandenberger , phys . rept . * 215 * , 203 ( 1992 ) . t. s. bunch and p. c. w. davies , proc . a * 360 * , 117 ( 1978 ) . m. sasaki and e. d. stewart , prog . * 95 * , 71 ( 1996 ) [ arxiv : astro - ph/9507001 ] . d. h. lyth , k. a. malik and m. sasaki , jcap * 0505 * , 004 ( 2005 ) [ arxiv : astro - ph/0411220 ] . a. d. linde , phys . d * 49 * , 748 ( 1994 ) [ arxiv : astro - ph/9307002 ] . d. h. lyth , arxiv:1005.2461 [ astro-ph.co ] . d. h. lyth , arxiv:1012.4617 [ astro-ph.co ] . a. a. abolhasani , h. firouzjahi and m. h. namjoo , class . * 28 * , 075009 ( 2011 ) [ arxiv:1010.6292 [ astro-ph.co ] ] . j. o. gong and m. sasaki , jcap * 1103 * , 028 ( 2011 ) [ arxiv:1010.3405 [ astro-ph.co ] ] . d. langlois , phys . d * 59 * , 123512 ( 1999 ) [ arxiv : astro - ph/9906080 ] . n. bartolo , s. matarrese and a. riotto , phys . d * 64 * , 123504 ( 2001 ) [ arxiv : astro - ph/0107502 ] . c. t. byrnes and d. wands , phys . d * 74 * , 043529 ( 2006 ) [ arxiv : astro - ph/0605679 ] . f. vernizzi and d. wands , jcap * 0605 * , 019 ( 2006 ) [ arxiv : astro - ph/0603799 ] . d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) [ arxiv : hep - ph/0110002 ] . s. mollerach , phys . d * 42 * , 313 ( 1990 ) . a. d. linde and v. f. mukhanov , phys . d * 56 * , 535 ( 1997 ) [ arxiv : astro - ph/9610219 ] . k. enqvist and m. s. sloth , nucl . b * 626 * , 395 ( 2002 ) [ arxiv : hep - ph/0109214 ] . t. moroi and t. takahashi , phys . b * 522 * , 215 ( 2001 ) [ erratum - ibid . b * 539 * , 303 ( 2002 ) ] [ arxiv : hep - ph/0110096 ] . y. f. cai , r. brandenberger and x. zhang , jcap * 1103 * , 003 ( 2011 ) [ arxiv:1101.0822 [ hep - th ] ] . s. li , y. f. cai and y. s. piao , phys . b * 671 * , 423 ( 2009 ) [ arxiv:0806.2363 [ hep - ph ] ] . j. zhang , y. f. cai and y. s. piao , jcap * 1005 * , 001 ( 2010 ) [ arxiv:0912.0791 [ hep - th ] ] . y. f. cai and y. wang , phys . d * 82 * , 123501 ( 2010 ) [ arxiv:1005.0127 [ hep - th ] ] . s. a. kim and a. r. liddle , phys . d * 74 * , 023513 ( 2006 ) [ arxiv : astro - ph/0605604 ] . k. y. choi and j. o. gong , jcap * 0706 * , 007 ( 2007 ) [ arxiv:0704.2939 [ astro - ph ] ] . g. panotopoulos , phys . d * 75 * , 107302 ( 2007 ) [ arxiv:0704.3201 [ hep - ph ] ] . d. battefeld , t. battefeld and a. c. davis , jcap * 0810 * , 032 ( 2008 ) [ arxiv:0806.1953 [ hep - th ] ] . a. ashoorioon , a. krause and k. turzynski , jcap * 0902 * , 014 ( 2009 ) [ arxiv:0810.4660 [ hep - th ] ] . y. f. cai , j. b. dent and d. a. easson , phys . d * 83 * , 101301 ( 2011 ) [ arxiv:1011.4074 [ hep - th ] ] . d. wands , lect . notes phys . * 738 * , 275 ( 2008 ) [ arxiv : astro - ph/0702187 ] . d. langlois , j. phys . ser . * 140 * , 012004 ( 2008 ) [ arxiv:0809.2540 [ astro - ph ] ] .
we present a study of the expected properties of the low redshift damped absorber population determined from a sample of hi - selected galaxies in the local universe . because of a tight correlation between the hi mass and hi cross - section , which we demonstrate spans all galaxy types , we can use our hi - selected sample to predict the properties of the absorption line systems . we use measurements of the number density and hi cross - section of galaxies to show that the total hi cross - section at column densities sufficient to produce damped absorption is consistent with no evolution of the absorber population . we also find that the distribution is dominated by galaxies with hi masses near 10 . however , because of the large dispersion in the correlation between hi mass and stellar luminosity , we find that the distribution of as a function of l is fairly flat . additionally , we examine the line widths of the hi - selected galaxies and show that there may be evolution in the kinematics of hi - rich galaxies , but it is not necessary for the higher redshift population to contain a greater proportion of high mass galaxies than we find locally .
astro-ph0202216
damped lyman-@xmath3 absorption - line systems ( dlas ) seen against background quasars are the highest column density absorbers , n@xmath4 @xmath5 , and are thought to be associated with galaxies . often they are assumed to be the disks of large spiral galaxies or their progenitors ( wolfe 1995 ) , but there is growing evidence that not all dlas are associated with bright spirals ( e.g. cohen 2001 , lanzetta et al . 1997 , miller et al . 1999 , colbert & malkan 2001 ) . the difficulty in studying the properties of these systems arises , in large part , because there are not many systems known at low-@xmath6 : only 9 dlas have been found with @xmath7 ( rao & turnshek 2000 , bowen et al . 2001a , steidel et al . 1994 , lanzetta et al . 1997 , and le brun et al . 1997 ) , and only 2 of those are at @xmath8 where detailed observations are possible . absorption line studies do not provide adequate statistical samples at low-@xmath6 because their pencil beams survey only small volumes of local space . many of the identified dla candidates have low luminosity or low surface brightness , so it has been suggested that the absorption line population may be biased against detecting bright spiral galaxies . for example , bright spiral galaxies might contain enough dust to obscure the background qso ( fall & pei 1993 ) . however , no low redshift comparison sample exists to provide an indication of what we should expect to see . in this study we construct a local comparison sample from a blind 21 cm survey we carried out at the arecibo observatory ( rosenberg & schneider 2001 ) and use it to predict the characteristics of the low redshift dla population . several previous attempts have been made to predict the properties of dlas in the local universe based on known properties of nearby galaxies . rao & briggs ( 1993 ) and rao , turnshek & briggs ( 1995 ) combined optical luminosity functions for different morphological types with a linear relationship between hi mass and optical luminosity . they found that nearly all of the galaxy cross - section comes from large , bright spirals and that the hi cross - section implied by these galaxies would indicate a substantial decline in total hi cross - section since @xmath6=0.5 . salucci & persic ( 1999 ) also attempted to account for the local population of dlas by combining hi observations with optical luminosity functions . they found that large spirals contain most of the hi mass density in the local universe and that their mass density is similar to what is found in the damped absorbers . in parallel with this work , zwaan , briggs , & verheijen ( 2001 ) studied a volume limited , optically selected sample of galaxies in the ursa major cluster . the results they obtain are similar those those found here despite concerns one may have about the optical selection and overdensity in the region . because galaxies with relatively high hi masses contribute most of the hi content of the local universe ( rosenberg & schneider 2001 ; zwaan et al . 1997 ) , and because most luminous spirals have large hi masses , it may seem reasonable to conclude that luminous spirals must produce most of the absorption . the failing of this argument is that the correlation between hi mass and stellar luminosity is quite weak a large fraction of high hi - mass systems have very low stellar luminosities , as we show in 4.2 . the recent completion of large 21 cm hi surveys provides a galaxy measurement that is more directly related to hi absorption that can be used to predict the characteristics of dlas . preliminary studies of these `` blind '' surveys ( rosenberg & schneider 2001 ; zwaan et al . 2001 ) suggested a significantly larger total hi cross - section for the nearby galaxy population than was suggested by the earlier optically - selected samples . the present paper expands on those preliminary results with a more detailed analysis of the predicted hi and optical properties of the dlas . in particular , we analyze near - infrared measurements of our sample of galaxies from the 2-micron all sky survey ( 2mass ) to predict the stellar properties of the dlas . in section 2 we show that an hi - selected sample makes a good basis for predicting the properties of dlas . in addition to selecting galaxies within the same range of gas column densities , hi mass selection is nearly equivalent to selection by cross - sectional area . section 3 contains a discussion of how @xmath0 is calculated for our hi - selected sample and the resulting value of @xmath0 at @xmath6=0 . in section 4 we discuss our predictions for the properties of local dlas : hi masses ( 4.1 ) , j - band luminosities ( 4.2 ) , and kinematics ( 4.3 ) . studying the properties of damped absorbers is difficult because it requires the systems to be nearby where absorption - line studies requires space based uv spectroscopy and the volume surveyed by each line of sight is small . to study the properties of low-@xmath6 absorbers we must construct a surrogate sample for which we have better statistics . we use a blind , 21 cm emission line survey , the arecibo dual - beam survey ( adbs ) as a local comparison sample for dla properties . the adbs selected galaxies on the basis of their 21-cm hi flux . the survey covered @xmath9 430 deg@xmath10 in the arecibo main beam with a velocity coverage of 654 to 7977 and identified 265 galaxies , with hi masses ranging from @xmath11 to @xmath12 . only a third of the adbs galaxies had previously identified optical counterparts , partly because we detected some galaxies in the zone of avoidance , but also because there is a substantial population of low surface brightness galaxies that are not easily identified in magnitude - limited optical surveys . to relate the adbs detection statistics to those of dla studies , we need to determine the cross - sectional area for the galaxies above a column density of @xmath13 @xmath5 . only a few of the adbs sources were resolved in the arecibo observations and , therefore , only had a measure of the total hi flux . we made spatially - resolved d - array follow - up observations from the very large array ( vla ) for 84 of the adbs galaxies . for this study , we use the 50 sources with a major axis diameter ( at 2@xmath14 @xmath5 ) at least 40% larger than the beam to determine the hi cross - section properties of our sample . we find that the hi cross - section is closely correlated with the hi mass of each source , as shown in figure [ fig : massarea ] . the hi mass is determined from the source flux , distance , and linewidth using the standard equation : @xmath15 where @xmath16 is the integrated line flux in jy , and d is the distance in mpc , corrected for large scale velocity flows ( tonry et al . 2000 ) , adopting h@xmath17 = 75 mpc@xmath18 . the area of the galaxy , @xmath19 , where @xmath20 and @xmath21 are the major and minor axis diameters at an hi column density of @xmath13 @xmath5 . the measurements for each galaxy are provided in the appendix . = 0.9 because most of our vla measurements were not highly resolved , one might question whether our cross - section estimates are accurate . we have therefore consulted the literature , collecting high - resolution hi measurements for a wide variety of galaxies ( see appendix ) . the 53 literature sources with minor - axis dimensions at least 5 times larger than synthesized beam are shown in figure [ fig : massarea ] with open triangles , and are in excellent agreement with the adbs sample , shown as solid circles . the highest - resolution literature sources ( with a minor axis diameter at least 10 times larger than the beam size ) are indicated by large triangles . we find no significant differences between any of these samples , leading us to conclude that the correlation seen in figure [ fig : massarea ] is unrelated to resolution effects . in any case , the hi masses are based on the total hi emission from the galaxy , so resolution effects could only shift points up or down in the figure and could not artificially generate the observed correlation . distance errors _ do _ lead to errors parallel to the observed correlation ; however , the largest conceivable errors lead to small shifts on the scale of this figure , which spans four orders of magnitude in hi mass . we averaged forward and backward least squares fits to the high - resolution data , yielding the relationship : @xmath22 where @xmath23 is the cross - section in kpc@xmath10 and @xmath24 is the hi mass in solar masses ( shown in the figure by the dotted line ) . we find similar relationships for the adbs data alone ( although with a slope slightly less than unity ) , and for subsets of the high - resolution data . none of these subsets have slopes significantly different from 1 . if we combine all of the data , the slope is @xmath25 , with a correlation coefficient of 97.7% . if we assume the relationship has slope 1 , the resulting fit is : @xmath26 and the intercept value ( @xmath27 ) is consistent for the various subsets to within @xmath28 . a similar correlation between hi size and hi mass was previously noted for high - hi - mass galaxies by giovanelli & haynes ( 1983 ) using the arecibo telescope . their sample included galaxies in the virgo cluster , yet they found no significant differences between the fits for hi - deficient and hi - rich galaxies . verheijen & sancisi ( 2001 ) likewise observed a correlation for ursa major cluster galaxies ( with @xmath29 measuring them at a slightly lower hi column density . our results demonstrate that the correlation spans a very wide range of masses , and the correlation is consistent whether the sample is hi - selected ( adbs galaxies ) or optically selected ( literature data ) . this indicates that the relationship between hi mass and cross - section is fundamental . _ the tightness of this relationship indicates that our hi mass dependent galaxy selection is nearly equivalent to the cross - sectional area selection of the dla population . _ the strong correlation we find between hi cross - section and hi mass is probably due , in part , to measuring the galaxies at a high column density . we expect a larger dispersion at lower column densities because in the outer parts of galaxies tidal debris and gas - stripping processes have a strong effect on the measured extent . for both the adbs and literature galaxies , the hi above the @xmath30 @xmath5 isophote is generally confined to a roughly disk - shaped region , but we see emission with a wide variety of extents and morphologies at lower column densities . another reason for the good correlation is that the hi cross - section has a weak dependence on galaxy inclination . there is a partial cancellation between the smaller area presented by an inclined disk , and the larger radius at which the column density remains above @xmath30 @xmath5 because of the longer path length through an inclined disk . we find this to be demonstrated empirically for those disk galaxies that authors have constructed inclination - corrected models of the radial dependence of hi surface density ( see appendix ) . haynes & giovanelli ( 1984 ) found a similar correlation between hi mass and the _ optical _ sizes of galaxies for a sample spanning a similar range in hi mass . we also find that the dimensions of adbs galaxies as measured on the palomar sky survey closely correlate with their hi mass . this suggests that the size of the optically visible region is closely related to the dense regions of the hi gas being studied here . finally , we note that in comparing emission and absorption line studies we are comparing gas on very different angular scales : a background quasar s extent of perhaps 0.25 - 1 for absorption versus @xmath91060 for synthesis array studies . the resolution of the synthesis data may overlook the small scale properties of these systems . if the surface filling factor for n@xmath31 @xmath5 is small , we would overestimate the size of the damped region for each of these galaxies . on the other hand , there may be small regions of damped emission diluted with low column density gas beyond our measured hi isophote which contribute to a larger hi covering area . detailed studies of the hi covering fraction are available for very few galaxies , and from what has been done to date , there are disparities in the covering fractions that have been derived . braun & walterbos ( 1992 ) , studying the small scale hi structure in m31 with a resolution of 33 pc , find that the hi surface filling fraction approaches 1 where the emission brightness temperature exceeds @xmath95 k ( equivalent to a column density of 2 @xmath32 @xmath5 for an asymptotic temperature t@xmath33 k and a velocity width of 21.5 ) . these numbers suggest that the correction for the covering fraction in these galaxies should be relatively small . we can estimate the probability of detecting a galaxy in an absorption line study if we know its cross - section and its relative contribution to the overall population of absorbers . for each adbs galaxy we use the correlation found in the previous section to estimate its cross - section @xmath34 , and we determine its relative contribution based on the relative sensitivity to each source detected . we characterize the latter by the total volume @xmath35 within which each source could have been detected . the sensitivity criteria and completeness of the adbs has been carefully evaluated and tested and is a function of the galaxy s flux and linewidth , the signal - to - noise ratio of the detection , and properties of the instrumental response ( see rosenberg & schneider 2002 for a complete discussion ) . to find the number of systems expected per unit redshift ( @xmath0 ) , we sum over all sources : @xmath36 where @xmath37 gives the volume density for each source . thus , summing over all of the adbs galaxies , our predicted total cross section at zero redshift is @xmath38 . we assume an error on the adbs result of 25% as an estimate of potential systematics in the mass cross - section relationship and the covering factor to actually produce damped absorption . the formal statistical uncertainty associated with our estimate is only a few percent . = 0.9 figure [ fig : dndz ] shows @xmath0 as a function of redshift for dlas . the highest-@xmath6 points , indicated by stars , are from storrie - lombardi & wolfe ( 2000 ) who combined their survey of 40 quasars with previous surveys . the resulting sample contains 646 quasars and 85 dlas with column densities @xmath39 @xmath5 covering the redshift range @xmath40 . the open squares are from rao & turnshek ( 2000 ) whose sample consisted of 12 dlas detected in 87 mg ii absorber lines of sight . at @xmath41 , the solid triangle is from the study of rao et al . ( 1995 ) , based on the properties of optically - selected galaxies at low redshift . the filled circle is our result described above . our value for @xmath0 at @xmath41 is consistent with the @xmath42 = @xmath43 value found by churchill ( 2001 ) for mg ii systems with strong fe ii absorption , which are good candidates for damped absorption systems ( rao & turnshek 2000 ) . these results are also consistent with a similar study of the hi properties of galaxies by zwaan et al . ( 2001 ) . our adbs result is consistent with the @xmath44 , no evolution model shown by the solid line in figure [ fig : dndz ] . because of the large error bars on these data , we can not entirely rule out evolution in the dla population . however , our result rules out strong evolution of the absorber population between @xmath41 and @xmath45 as was implied by the rao et al . ( 1995 ) result . these results differ because the @xmath0 distribution in the local universe is not dominated by large disks , but has substantial contribution from dwarf galaxies as we explore in the next section . we have identified a sample of galaxies that we believe is representative of dlas at @xmath46 . we use the properties of this sample to predict the nature of the dla population locally by splitting up the contribution to @xmath0 in equation 2 according to various properties of the adbs galaxies . figure [ fig : avgarea ] shows the contribution of galaxies to @xmath0 according to their hi mass . the errors for each bin have been determined from small number counting statistics using the method described by gehrels ( 1986 ) , and include no estimate of systematic errors . the vertical line indicates m@xmath47 , the knee of the schechter function fit to the hi mass function , as determined by rosenberg & schneider ( 2002 ) . the dotted histogram corresponds to the determination of hi cross - section from the averaged forward and backward least squares fit , while the solid line if from the fit assuming that the slope is unity . = 0.9 this figure shows that the central 50% of the cross - sectional area comes from galaxies in the range @xmath48 . galaxies above @xmath49 account for only 9.5% of the area . therefore , this central 50% of sources cover a range in hi mass from @xmath50 to @xmath51 , relatively low hi mass sources compared to the knee of the schechter function . because @xmath52 galaxies are a major source of the light in the universe , and @xmath53 galaxies are a major source of the hi mass , it is sometimes assumed that the optical counterparts to the dlas are hi - rich @xmath52 galaxies bright spirals ( wolfe 1995 ) . instead , many of the low-@xmath6 dlas appear to be low luminosity or are not detected ( colbert & malkan 2001 , steidel et al . 1994 , lebrun et al . 1997 , lanzetta et al . 1997 , rao & turnshek 1998 , pettini et al . most of the dla candidates do not have confirmed redshifts so the actual luminosity distribution of dlas is hard to determine . we use the adbs sample to predict the distribution of luminosities for dlas in the local universe by calculating the contribution to @xmath0 in @xmath54-band luminosity bins . we obtained the near infrared data used in these analyses from the 2-micron all - sky survey ( 2mass , jarrett et al . 2mass used 2 identical 1.3 m telescopes in tucson and chile which provide simultaneous @xmath54 , @xmath55 , and @xmath56 observations . although the observing time per point is only 7.8 seconds , we were able to detect all but 27 of the galaxies in our sample on the 2mass images . eighteen of these were undetected while the remaining 9 did not have image data available at the time of this study . the automated 2mass galaxy detection algorithms missed 52 faint galaxies that are visible on the images so we measured them individually within elliptical isophotes at 21 mag arcsec@xmath57 . in addition , 13 galaxies had magnitudes measured by the 2mass galaxy extraction algorithms , but for various reasons they did not have magnitudes measured at a @xmath54-band 21 mag arcsec@xmath57 isophote . for the 2 galaxies which have @xmath54-band magnitudes measured within the @xmath56-band 20 mag arcsec@xmath57 isophote , we substitute these values since these two types of magnitudes are well correlated . for the remaining 11 galaxies , we use the correlation between the @xmath54-band magnitude at 21 mag arcsec@xmath57 and the magnitude in a 10 circular aperture to calculate the magnitude . ( see rosenberg et al . 2002 for details of these galaxy measurements . ) = 0.9 the distribution of hi mass as a function of @xmath54-band luminosity is shown in figure [ fig : massmag ] . the solid line in the figure shows what a one - to - one relation between @xmath58 and @xmath24 would look like , while the dashed line shows the actual linear least - squares fit between the values ( @xmath59 ) . although there is a correlation between the the hi mass and the j - band luminosity , the relationship is nearly vertical , so that a galaxy with a given hi mass could have a j - band luminosity spanning four orders of magnitude . there is also a wide scatter in hi cross - sections for any given j - band luminosity . the dotted lines show @xmath53 and @xmath60 , the knees in the schechter function fit to , respectively , the hi mass function ( rosenberg & schneider 2002 ) and j - band lumiunosity function ( kochanek et al . 2001 ) . = 0.9 figure [ fig : dndzj ] shows the predicted contribution to @xmath0 according to galaxy magnitude along the line of sight . the histograms assume the unity - slope fit between hi mass and cross - section ( see 2 ) , but we show several alternatives for the distribution depending on the estimates of the @xmath54-band luminosities of the undetected galaxies ( see below ) . no matter what assumptions we make about the missing galaxies , the luminosity distribution is quite flat . even if we make a conservative assumption that the missing galaxies are relatively bright , 50% of the galaxies lie between luminosities of @xmath61 and @xmath62 , spanning a factor of @xmath9110 in luminosity . only 4.5% of the galaxies are brighter than @xmath63 ( @xmath64 ) ( cole et al . 2001 ) . in making various alternative assumptions for generating the histograms in figure [ fig : dndzj ] , we have attempted to explore what biases may have resulted from the 27 galaxies for which we have no 2mass fluxes . the dotted - line histogram shows the result when we simply exclude the sources that were not detected . the solid - line histogram assumes that the missing galaxies had @xmath54-band luminosities determined by the fit to the mass - luminosity relation in figure [ fig : massmag ] . this is probably an upper limit to the luminosities of the galaxies that were too faint to be detected by 2mass . the dashed - line histogram assumes that the missing galaxies have a @xmath54-band magnitude of 19 , slightly fainter than the faintest apparent magnitude we detect . the differences in the results for the 3 assumptions is small , however , if the non - detections are all very faint sources then the distribution is even flatter down to lower j - band luminosities . the definition of dwarf " galaxy often varies from paper to paper , but systems that are fainter than 10% of @xmath52 are often considered dwarfs . by this definition , 63% of dlas in the local universe should be dwarf galaxies . prochaska & wolfe ( 1997 ) argue that the kinematics of dlas rule out the possibility that they are dwarf galaxies at the 97% confidence level based on simulating dwarfs as slowly rotating hot disk systems . however , their definition of a dwarf is not a luminosity definition as is often used observationally , but instead is a galaxy with circular velocity less than 50 . our results above indicate that most dlas might well be classified as dwarfs in terms of near infrared luminosity and hi mass . significant differences between the kinematics of the low-@xmath6 adbs sample , and the @xmath65 systems studied by prochaska & wolfe ( 2001 , hereafter pw ) could indicate evolution of the population despite the fairly constant number density of sources found above . one complication in comparing the absorption and emission kinematics is that absorption studies probe a single line - of - sight through a galaxy s kinematic profile while emission studies probe the full extent of the system . we can make a comparison between the linewidths for the high and low redshift systems , but with the difference in the methods of measurement , we will not be able to show definitively that there has been no evolution of the population . pw have analyzed the kinematics of 46 low - ionization lines and 32 high - ionization lines in dlas and used them to constrain models of the origin of these systems ( prochaska & wolfe 2001 , prochaska & wolfe 1997 ; prochaska & wolfe 1998 ; wolfe & prochaska 2000 ) . pw use the low - ionization transitions ( al ii , cr ii , fe ii , ni ii , and si ii ) as tracers of the disk component and the high - ionization transition ( c iv ) as a tracer of an infalling component . the velocity width of the system is defined to be 90% of the optical depth of the line . we compare our emission line widths to twice the low - ionization line widths calculated in these studies to account for the full rotation of the galaxy . = 0.9 = 0.9 for the adbs , we subdivide the calculated @xmath0 values ( 3 ) according to the linewidth ( measured at 50% peak ) . the left panel in figure [ fig : dndzw ] shows the fraction of sources with each linewidth in the adbs sample as compared with the low - ionization species ( open histograms ) in the pw sample ( 46 dlas ) . the pw widths have been multiplied by two in order to compare the single sightline velocity dispersion with the full velocity width measured by the hi . we find that the adbs velocity widths are not significantly smaller than the dla widths as is found for comparison with pure dwarf galaxy samples . we note that the velocity widths of the dlas should also be corrected for the effect of only sampling a small portion of the disk , but we do not know what that correction is so we do not make it . the upper panel of figure [ fig : dndzwcum ] shows the cumulative distribution comparison for the same data sets . a ks test indicates that the kinematics can be rejected as being from the same sample at the 9% confidence level . because we do not know how to correct for the differences in disk sampling we can not show that there has been no evolution of the population , but the statistics do not require it to explain the observations . the right panel of figure [ fig : dndzw ] shows the comparison between the adbs data and the high - ionization species in the pw sample ( 32 dlas ) . here , we have corrected the adbs linewidths for inclination and compared then with the high - ionization lines ( which have not been multiplied by two ) . if the high - ionization species are infalling , they should be a good probe of the galactic potential , and should measure gas from both the near and far sides of the galaxy . here the difference between the adbs data and the dla data would be due to the probes not penetrating a radial line of sight through the galaxy s center . figure [ fig : dndzwcum ] shows the cumulative distribution and the ks - test probability that they are drawn from different samples is 31% . there is a high line width tail for the adbs sources that could indicate that there has been evolution , or that the absorption lines are just not probing the center of the potential . the k - s test does not preclude that the kinematics of the adbs galaxies and the dlas are drawn from similar samples . it is not clear whether the kinematic differences are due to small number statistics , differences in the way the quantities are measured , or evolution of the populations . however , the adbs line widths are , in general , as large or larger than the dla linewidths indicating that the dla sample does not require significantly more massive disks than those identified in an hi selected sample locally . we have predicted the properties of the low-@xmath6 dla population from an hi - selected galaxy survey of the local universe . the strong linear correlation between hi cross - section and hi mass means that selection effects should be very similar and allows us to use the statistics available for the properties of adbs galaxies to predict those of low-@xmath6 dlas . we find a much higher value of @xmath0 at @xmath6=0 from the adbs than was found for the optically - selected , spiral dominated sample of rao et al . the rao et al . data have been used as evidence that there was evolution of the dla population between @xmath6=0.5 and @xmath6=0 . however , the adbs prediction for the total cross - section at @xmath46 is consistent with @xmath0 for mg ii absorbers with fe ii absorption , systems that are good candidates for damped absorption and both of these data sets indicate that no evolution is required to explain the dla population . with the large error bars on all of the @xmath0 points , we can not rule out some evolution on the dla population , but it is not required . the @xmath0 distribution of galaxies as a function of hi mass is peaked around @xmath66 , and 50% of the galaxies should fall within a factor of @xmath93 of this value . to date only 2 dla absorbers have been studied in emission at 21 cm : one has a m@xmath67 of 1.3 @xmath68 ( bowen et al . 2001b ) , the other was not detected with a 3@xmath69 upper limit of 2.25 @xmath68 ( kanekar et al . 2001 ) . these results do not provide a statistical test of our hi mass distribution , but the results are consistent with what we would expect . the adbs sample indicates that the stellar luminosities of dlas should be distributed fairly evenly over several orders of magnitude . as with the hi masses , there are few galaxy / dla associations for which the galaxy redshift is confirmed and a luminosity is measured . the absorber candidates that have been identified ( steidel et al . 1994 , lebrun et al . 1997 , lanzetta et al . 1997 , rao & turnshek 1998 , pettini et al . 2000 ) span a wide range of luminosities . this range of dla luminosities is a warning for high redshift dla candidate searches the absorber candidates are not necessarily the nearest l@xmath70 galaxy to the line of sight . we have compared the internal kinematics of the adbs galaxies with those for high redshift dlas ( a low redshift sample does not exist ) . prochaska & wolfe ( 1997 ) , through simulations of gas disks , rule out galaxies with small velocity widths as the primary source of damped emission at a 97% confidence level . the adbs galaxies that we associate with the dlas locally have larger line widths than the prochaska & wolfe dwarf " galaxies , but the sample consists of a range of galaxy types dominated by galaxies with hi masses of @xmath90.1 and luminosities below l@xmath71 . we find that we can not rule out evolution in the kinematics of hi - rich galaxies , but it is not required and more massive hi disks then found in our local sample are not necessary to account for the kinematics . we thank jason x. prochaska for helpful discussions about the kinematics of dlas and the referee for a thoughtful reading of this paper . we would also like to thank the vla staff for their assistance in the observations and reduction of this data , and the 2mass team for all of the hard work that went into making the survey a reality . in this appendix we present vla data for the adbs sample of galaxies , as well as high - resolution literature data used as a comparison sample to confirm our results . in addition , we provide some further analysis of the hi cross - sections of the galaxies beyond the analysis in the main text . the galaxies in the adbs were observed with the vla in d - array , with relatively short duration ( 10 to 15 min ) integrations . these were usually split into two time intervals in order to improve the @xmath72-@xmath73 coverage of the observations . further details of the observational procedures can be found in rosenberg & schneider ( 2002 ) . we present the basic information on the 50 sources detected with the vla that have a major axis diameter ( at @xmath74 @xmath5 ) at least 40% larger than the beam .. all were reduced using the observatory s _ aips _ routine _ imagr _ , setting the _ robust _ parameter to 0 . this provides a good compromise , providing nearly as good noise characteristics as natural weighting and nearly as small a beam as uniform weighting . table a.1 lists the fwhm dimensions of the beam along with other measured parameters for the galaxies . the elliptical dimensions of the galaxies at a column density of @xmath30 @xmath5 were fit to the total hi maps using the _ iraf stsdas _ package _ ellipse_. the mean velocities and total hi masses determined from the vla observations are all consistent with our arecibo detection measurements . we also provide a flow - corrected distance estimate based on tonry et al . ( 2000 ) , assuming @xmath75 km s@xmath18 mpc@xmath18 . we compiled higher - resolution hi measurements from the literature for comparison with our results . we consulted sources listed in martin s ( 1998 ) bibliographic compilation of hi maps in order to identify synthesis observations with beam sizes at least ten times smaller than the reported hi dimensions . we further restricted our sample to sources that martin indicated reached column densities below @xmath30 @xmath5 and which had both hi contour maps and radial density profiles . this yielded about 100 candidate sources . we found that the resulting list contained few low - mass sources , so we added several more - recent papers that concentrated on dwarf galaxies . the resulting sample spans all galaxy types from giant ellipticals to dwarf irregulars . table a.2 compiles the information collected from these papers in a similar format as for our adbs galaxies . none of the papers directly reported the dimensions of the galaxy at @xmath30 @xmath5 , so we estimated them from the hi column density contour maps . the beam dimensions are those reported in each paper for the contour maps . in a few cases ( noted by the word `` ring '' or `` mult '' in the notes column of the table ) the distribution of hi above @xmath30 @xmath5 was not well described as a disk . in these cases , we give major and minor axis dimensions that approximately indicate ( a ) the combined length of the regions of hi that exceeded our fiducial level , and ( b ) the narrower dimension of these regions . these dimensions should yield a fairly accurate total cross - section for the sources . we note that many other sources also showed irregularities around the edges , or small holes or small regions outside the main disk , but these amount to only slight deviations in the overall dimensions of the sources . for papers that displayed inclination - corrected , azimuthally - averaged radial density profiles , we again used the published plots to estimate the radius at which the hi column density reached our fiducial level . in a few cases the radial profile was plotted , but did not exceed the @xmath30 @xmath5 level . in all of these cases , the hi contour map had regions above this level , but the azimuthal averaging had produced a lower mean level . we note that these radial profiles are quite sensitive to the inclination estimates . as a result , they tend to be more accurate for disk galaxies , and much less so for dwarf irregulars . the redshifts and morphological types listed in the table were taken from the nasa / ipac extragalactic database ( ned ) . the distances are again based on the flow - corrected tonry model , except when the redshift was less than 250 km s@xmath18 , in which case we used the distance quoted by the author . for two of these nearby galaxies , ugc 8091 ( dohm - palmer et al . 1998 ) and ugc 12613 ( gallagher et al . 1998 ) , recent distance estimates were significantly different from those reported in the hi paper , and we have adopted the newer values . the strong , one - to - one correlation between hi mass and cross - section over such a wide range of galaxy sizes and masses is somewhat surprising . at first glance , it suggests that all types of galaxies share the same surface density of hi . actually the data are a little more complicated to interpret , since the relationship is between the total hi , and the area of the disk that exceeds the relatively high surface density of @xmath30 @xmath5 . if all galaxies had their hi in exponential disks with varying scale lengths but the same central surface brightness , such a relationship would pertain , but clearly that is not a realistic description of galaxy hi distributions . we present here a few exploratory comments about the data presented in table a.2 to examine this correlation in greater detail . for the central purpose of this paper , we focused on the relationship between the hi mass and the _ observed _ cross - section of hi on the sky . however , as we briefly noted in 2 , the observed cross - section is relatively independent of inclination . we show this in figure [ fig : axrat ] , where we plot the ratio of observed to face - on cross sections above @xmath30 @xmath5 as a function of the observed axis ratio . the sample shown is the subset of galaxies from figure [ fig : massarea ] that also had inclination - corrected radial profile models . the cross - section does not simply decline as @xmath76 ( curve in figure ) , as would occur for a sharp - edged disk . because of the longer path length through the disk , the @xmath30 @xmath5 isophote shifts outward to a physically larger radius by an amount that depends on the scale - length of the hi distribution . the two effects nearly cancel for most inclinations , although edge - on galaxies do finally exhibit a significantly smaller cross - section . overall , the observed cross - sections are about 25% smaller , on average , than the inclination - corrected models predict for galaxies with axis ratios larger than @xmath77 . the scatter about the fit between mass and cross - sectional area improves slightly when we use azimuthally - averaged , inclination - corrected model results . using these face - on cross - sections yields a similar slope ( @xmath78 ) as we found using the observed cross - sections in a linear least squares fit , although the intercept shifts because of the larger cross - section found on average for the face - on models . we will assume both actually have slope unity so that we can just compare the distribution of @xmath79 $ ] . for the 64 galaxies with measurements of both the face - on , @xmath80 , and observed dimensions , @xmath81 , the rms scatter of @xmath79 $ ] is 0.194 and 0.227 respectively . there seems to be little dependence with morphological type . if we use observed cross - sections for galaxies which are not highly inclined ( @xmath77 ) , we find the following results by morphological type : the earliest - type galaxies ( sa sb ) may have slightly lower values of @xmath79 $ ] than late - types , but the difference is marginal . this appears to be true for the s0 s as well , but our few examples have peculiar hi distributions , so any conclusions depend on whether the model or observed cross - sections are used , and whether outliers are rejected . while there may be some dependence on morphological type , the remarkable thing is how little variability there is between types over an enormous range of masses . lc@ c@ c@c@ c@ rrr 002526 + 2136 & [email protected] & [email protected] & 5001 & 66.7 & 9.44 + 014729 + 2719 & [email protected] & [email protected] & 743 & 9.9 & 9.14 + 015434 + 2312 & [email protected] & [email protected] & 5298 & 70.6 & 9.60 + 020320 + 1837 & [email protected] & [email protected] & 2727 & 36.4 & 9.50 + 020320 + 2345 & [email protected] & [email protected] & 3081 & 41.1 & 9.06 + 020405 + 2412 & [email protected] & [email protected] & 989 & 13.2 & 8.11 + 022859 + 2808 & [email protected] & [email protected] & 1341 & 17.9 & 8.69 + 023323 + 2810 & [email protected] & [email protected] & 1316 & 17.5 & 8.23 + 023622 + 2526 & [email protected] & [email protected] & 987 & 13.2 & 9.45 + 025537 + 1938 & [email protected] & [email protected] & 893 & 11.9 & 8.38 + 025726 + 1008 & [email protected] & [email protected] & 1092 & 14.6 & 8.56 + 030546 + 2212 & [email protected] & [email protected] & 4365 & 58.2 & 9.94 + 031420 + 2409 & [email protected] & [email protected] & 1523 & 20.3 & 9.12 + 040344 + 2209 & [email protected] & [email protected] & 6307 & 84.1 & 10.09 + 040411 + 2207 & [email protected] & [email protected] & 6320 & 84.3 & 10.13 + 053017 + 2233 & [email protected] & [email protected] & 2555 & 34.1 & 8.64 + 055517 + 2526 & [email protected] & [email protected] & 1547 & 20.6 & 8.64 + 062054 + 2008 & [email protected] & [email protected] & 1279 & 17.1 & 8.86 + 065406 + 0834 & [email protected] & [email protected] & 2420 & 32.3 & 9.42 + 070920 + 2038 & [email protected] & [email protected] & 4879 & 65.1 & 10.27 + 071225 + 2342 & [email protected] & [email protected] & 4138 & 55.2 & 9.76 + 071352 + 1031 & [email protected] & [email protected] & 304 & 4.1 & 7.69 + 071553 + 1207 & [email protected] & [email protected] & 1972 & 26.3 & 8.98 + 081726 + 2110 & [email protected] & [email protected] & 2036 & 27.1 & 8.92 + 090024 + 2536 & [email protected] & [email protected] & 1806 & 24.1 & 9.36 + 090526 + 2533 & [email protected] & [email protected] & 2869 & 38.3 & 9.44 + 101421 + 2207 & [email protected] & [email protected] & 1686 & 22.5 & 8.49 + 103143 + 2518 & [email protected] & [email protected] & 1205 & 16.1 & 9.00 + 103439 + 2305 & [email protected] & [email protected] & 1142 & 15.2 & 8.46 + 104722 + 1404 & [email protected] & [email protected] & 264 & 3.5 & 7.34 + 104916 + 1226 & [email protected] & [email protected] & 1238 & 16.5 & 8.73 + 110710 + 1834 & [email protected] & [email protected] & 1082 & 14.4 & 8.94 + 112914 + 2035 & [email protected] & [email protected] & 1392 & 18.6 & 8.82 + 115004 + 2628 & [email protected] & [email protected] & 2096 & 27.9 & 9.21 + 115040 + 2531 & [email protected] & [email protected] & 2149 & 28.7 & 8.91 + 120033 + 2004 & [email protected] & [email protected] & 1254 & 16.7 & 9.14 + 121516 + 2038 & [email protected] & [email protected] & 743 & 9.9 & 8.89 + 123606 + 2602 & [email protected] & [email protected] & 1453 & 19.4 & 10.17 + 124516 + 2708 & [email protected] & [email protected] & 1050 & 14.0 & 8.87 + 125223 + 2138 & [email protected] & [email protected] & 479 & 6.4 & 8.08 + 131652 + 1232 & [email protected] & [email protected] & 689 & 9.2 & 8.41 + 141556 + 2303 & [email protected] & [email protected] & 195 & 2.6 & 7.21 + 144842 + 1226 & [email protected] & [email protected] & 1950 & 26.0 & 9.35 + 145647 + 0930 & [email protected] & [email protected] & 3221 & 42.9 & 9.59 + 154540 + 2805 & [email protected] & [email protected] & 2353 & 31.4 & 9.27 + 163122 + 2010 & [email protected] & [email protected] & 2560 & 34.1 & 9.53 + 214731 + 2209 & [email protected] & [email protected] & 2033 & 27.1 & 8.90 + 214813 + 2209 & [email protected] & [email protected] & 1967 & 26.2 & 9.06 + 225557 + 2610 & [email protected] & [email protected] & 2979 & 39.7 & 8.72 + 230433 + 2709 & [email protected] & [email protected] & 1439 & 19.2 & 8.82 + lrrc@ rc@ c@ lll ngc 55 & [email protected] & [email protected] & 31.9 & 129 & 1.6 & 8.96 & sb(s)m : & & puche et al . ( 1991 ) + ngc 247 & [email protected] & [email protected] & 22.0 & 160 & 2.5 & 8.90 & sab(s)d & & carignan & puche ( 1990b ) + ngc 300 & [email protected] & [email protected] & 27.0 & 144 & 1.8 & 8.84 & sa(s)d & & puche et al . ( 1990 ) + ic 1613 & [email protected] & [email protected] & & -234 & 0.7 & 7.78 & ib(s)m & ring & lake & skillman ( 1989 ) + ugc 891 & [email protected] & [email protected] & 5.1 & 643 & 14.6 & 9.11 & sabm : & & van zee et al . ( 1997 ) + m 74 & [email protected] & [email protected] & 16.5 & 657 & 14.5 & 10.32 & sa(s)c & & wevers et al . ( 1986 ) + ugca 20 & [email protected] & [email protected] & & 498 & 12.2 & 8.58 & i m & mult & van zee et al . ( 1996 ) + ngc 925 & [email protected] & [email protected] & 12.7 & 553 & 11.9 & 9.93 & sab(s)d & & wevers et al . ( 1986 ) + ngc 1058 & [email protected] & [email protected] & 7.3 & 518 & 11.1 & 9.46 & sa(rs)c & & dickey et al . ( 1990 ) + ngc 1073 & [email protected] & [email protected] & 8.7 & 1211 & 21.0 & 9.82 & sb(rs)c & & england et al . ( 1990 ) + ugc 2259 & [email protected] & [email protected] & 4.2 & 583 & 11.8 & 8.80 & sb(s)dm & & carignan et al . ( 1988 ) + ngc 1140 & [email protected] & [email protected] & 4.7 & 1501 & 25.1 & 9.72 & ibm : & & hunter et al . ( 1994 ) + ngc 1291 & [email protected] & [email protected] & & 839 & 15.5 & 9.41 & ( r)sb(l)0/a & ring & van driel et al . ( 1988b ) + ngc 1300 & [email protected] & [email protected] & 7.0 & 1568 & 25.7 & 10.00 & ( r)sb(s)bc & & england ( 1989 ) + ugc 2684 & [email protected] & [email protected] & 3.1 & 350 & 8.9 & 8.38 & i m ? & & van zee et al . ( 1997 ) + ngc 1343 & [email protected] & [email protected] & & 2215 & 31.1 & 9.72 & sab(s)b : & & taylor et al . ( 1994 ) + ngc 1398 & [email protected] & [email protected] & 8.3 & 1407 & 23.2 & 9.74 & ( r)sb(rs)ab & & moore & gottesman ( 1995 ) + ic 342 & [email protected] & [email protected] & 52.0 & 31 & 2.0 & 9.53 & sab(rs)cd & & newton ( 1980 ) + ngc 1560 & [email protected] & [email protected] & 15.0 & -36 & 3.0 & 8.91 & sa(s)d & & broeils ( 1992 ) + ugc 3174 & [email protected] & [email protected] & 4.1 & 670 & 10.9 & 8.60 & iab(s)m : & & van zee et al . ( 1997 ) + ugca 116 & [email protected] & [email protected] & 2.8 & 789 & 10.8 & 8.70 & bcd / sbc & mult & van zee et al . ( 1998 ) + ngc 2366 & [email protected] & [email protected] & 11.7 & 100 & 3.3 & 8.72 & ib(s)m & & wevers et al . ( 1986 ) + ngc 2403 & [email protected] & [email protected] & 29.0 & 131 & 3.2 & 9.51 & sab(s)cd & & wevers et al . ( 1986 ) + ugc 3974 & [email protected] & [email protected] & 7.0 & 272 & 3.5 & 8.25 & ib(s)m & & walter & brinks ( 2001 ) + ugc 4483 & [email protected] & [email protected] & 2.1 & 178 & 3.4 & 7.57 & dwarf & & van zee et al . ( 1998 ) + ngc 2787 & [email protected] & [email protected] & & 696 & 10.9 & 8.73 & sb(r)0 + & ring & shostak ( 1987 ) + ngc 2841 & [email protected] & [email protected] & 12.2 & 638 & 8.3 & 9.37 & sa(r)b : & & bosma ( 1981 ) + ngc 2903 & [email protected] & [email protected] & 12.8 & 556 & 4.6 & 9.07 & sb(s)d & & wevers et al . ( 1986 ) + holmberg i & [email protected] & [email protected] & & 143 & 3.6 & 8.04 & iab(s)m & & ott et al . ( 2001 ) + leo a & [email protected] & [email protected] & & 20 & 2.2 & 7.87 & ibm & & young & lo ( 1996 ) + ngc 3079 & [email protected] & [email protected] & 9.7 & 1125 & 16.4 & 9.86 & sb(s)c & & irwin & seaquist ( 1991 ) + ngc 3198 & [email protected] & [email protected] & 14.0 & 663 & 7.7 & 9.53 & sb(rs)c & & begeman ( 1989 ) + ic 2574 & [email protected] & [email protected] & 15.0 & 57 & 3.0 & 8.78 & sab(s)m & & martinbeau et al . ( 1994 ) + ugc 5716 & [email protected] & [email protected] & 3.1 & 1277 & 16.3 & 8.76 & sm : & & van zee et al . ( 1997 ) + ugc 5764 & [email protected] & [email protected] & 3.0 & 586 & 5.0 & 7.79 & ib(s)m : & & van zee et al . ( 1997 ) + ugc 5829 & [email protected] & [email protected] & & 629 & 5.7 & 8.68 & i m & & taylor et al . ( 1994 ) + kdg 73 & [email protected] & [email protected] & & 113 & 3.6 & 6.32 & i m & mult & ott et al . ( 2002 ) + ngc 3510 & [email protected] & [email protected] & & 705 & 5.9 & 8.59 & sb(s)m & & taylor et al . ( 1994 ) + ngc 3626 & [email protected] & [email protected] & & 1493 & 19.7 & 8.89 & ( r)sa(rs)0 + & mult & van driel et al . ( 1989 ) + ngc 3628 & [email protected] & [email protected] & 10.1 & 843 & 6.2 & 9.38 & sab : & & wilding et al . ( 1993 ) + ngc 3726 & [email protected] & [email protected] & 7.8 & 866 & 11.6 & 9.54 & sab(r)c & & wevers et al . ( 1986 ) + ugc 6578 & [email protected] & [email protected] & 1.2 & 1099 & 9.7 & 7.92 & pair & & van zee et al . ( 1998 ) + ngc 3900 & [email protected] & [email protected] & 5.8 & 1798 & 28.5 & 9.50 & sa(r)0 + & & van driel et al . ( 1989 ) + um 461 & [email protected] & [email protected] & 2.1 & 1039 & 7.9 & 7.96 & bcd / irr & & van zee et al . ( 1998 ) + ugc 6850 & [email protected] & [email protected] & 1.5 & 1055 & 8.2 & 7.96 & pec & & van zee et al . ( 1998 ) + ngc 3941 & [email protected] & [email protected] & & 928 & 11.7 & 8.76 & sb(s)@xmath83 & ring & van driel & van woerden ( 1989 ) + m 109 & [email protected] & [email protected] & 8.7 & 1048 & 15.7 & 9.67 & sb(rs)bc & & gottesman et al . ( 1984 ) + ngc 4151 & [email protected] & [email protected] & 9.7 & 995 & 13.5 & 9.53 & ( r)sab(rs)ab : & & pedlar et al . ( 1992 ) + ugc 7178 & [email protected] & [email protected] & 3.2 & 1339 & 16.0 & 8.70 & iab(rs)m : & & van zee et al . ( 1997 ) + m 98 & [email protected] & [email protected] & 7.0 & -142 & 17.0 & 9.72 & sab(s)ab & & cayatte et al . ( 1990 ) + ngc 4216 & [email protected] & [email protected] & 5.5 & 131 & 17.0 & 9.35 & sab(s)b : & & cayatte et al . ( 1990 ) + ngc 4222 & [email protected] & [email protected] & & 230 & 17.0 & 8.95 & sc & & cayatte et al . ( 1990 ) + ngc 4242 & [email protected] & [email protected] & 5.5 & 517 & 5.6 & 8.44 & sab(s)dm & & wevers et al . ( 1986 ) + m 99 & [email protected] & [email protected] & 6.3 & 2407 & 34.6 & 10.33 & sa(s)c & & cayatte et al . ( 1990 ) + m 61 & [email protected] & [email protected] & 7.3 & 1566 & 19.5 & 9.87 & sab(rs)bc & & cayatte et al . ( 1990 ) + m 100 & [email protected] & [email protected] & 6.0 & 1571 & 20.4 & 9.68 & sab(s)bc & & cayatte et al . ( 1990 ) + ngc 4395 & [email protected] & [email protected] & 15.2 & 319 & 3.0 & 8.81 & sa(s)m : & & wevers et al . ( 1986 ) + ngc 4402 & [email protected] & [email protected] & 1.8 & 232 & 17.0 & 8.68 & sb & & cayatte et al . ( 1990 ) + m 88 & [email protected] & [email protected] & 4.7 & 2281 & 33.4 & 9.82 & sa(rs)b & & cayatte et al . ( 1990 ) + ngc 4535 & [email protected] & [email protected] & 6.8 & 1961 & 27.9 & 10.11 & sab(s)c & & cayatte et al . ( 1990 ) + ic 3522 & [email protected] & [email protected] & & 668 & 8.1 & 8.19 & imiii - iv & & skillman et al . ( 1987 ) + ngc 4568 & [email protected] & [email protected] & & 2255 & 33.5 & 9.50 & sa(rs)bc & & cayatte et al . ( 1990 ) + m 58 & [email protected] & [email protected] & 4.0 & 1519 & 19.6 & 8.88 & sab(rs)b & mult & cayatte et al . ( 1990 ) + ngc 4654 & [email protected] & [email protected] & 5.3 & 1037 & 13.3 & 9.33 & sab(rs)cd & & cayatte et al . ( 1990 ) + ugc 7906 & [email protected] & [email protected] & & 1010 & 12.7 & 8.19 & imiv & & skillman et al . ( 1987 ) + ngc 4725 & [email protected] & [email protected] & 11.0 & 1206 & 16.7 & 9.85 & sab(r)ab & & wevers et al . ( 1986 ) + m 94 & [email protected] & [email protected] & 5.7 & 308 & 3.5 & 8.43 & ( r)sa(r)ab & & mulder & van driel ( 1993 ) + ngc 4731 & [email protected] & [email protected] & & 1495 & 17.3 & 9.87 & sb(s)cd & & gottesman et al . ( 1984 ) + ngc 4789a & [email protected] & [email protected] & 7.2 & 374 & 3.1 & 8.21 & ib(s)miv - v & & carignan & beaulieu ( 1989 ) + ugc 8091 & [email protected] & [email protected] & & 214 & 2.2 & 6.99 & imv & & lo et al . ( 1993 ) + ngc 5033 & [email protected] & [email protected] & 11.8 & 875 & 11.4 & 9.83 & sa(s)c & & wevers et al . ( 1986 ) + m 63 & [email protected] & [email protected] & 24.7 & 504 & 5.4 & 9.57 & sa(rs)bc & & bosma ( 1981 ) + ugc 8333 & [email protected] & [email protected] & & 935 & 11.3 & 8.60 & i m : & & lake et al . ( 1990 ) + ngc 5101 & [email protected] & [email protected] & 6.8 & 1861 & 20.3 & 9.48 & ( r)sb(rl)0/a & ring & van driel et al . ( 1988b ) + ngc 5102 & [email protected] & [email protected] & & 467 & 2.8 & 7.97 & sa0- & & van woerden et al . ( 1993 ) + ngc 5371 & [email protected] & [email protected] & 4.3 & 2553 & 38.9 & 9.97 & sab(rs)bc & & wevers et al . ( 1986 ) + ngc 5585 & [email protected] & [email protected] & 8.6 & 305 & 5.0 & 8.98 & sab(s)d & & cote et al . ( 1991 ) + ngc 6503 & [email protected] & [email protected] & 13.2 & 60 & 3.8 & 8.82 & sa(s)cd & & wevers et al . ( 1986 ) + eso 594-g004 & [email protected] & [email protected] & & -77 & 1.1 & 6.94 & ib(s)m : & mult & young & lo ( 1997 ) + ngc 6814 & [email protected] & [email protected] & 5.7 & 1563 & 20.6 & 9.47 & sab(rs)bc & & liszt & dickey ( 1995 ) + ngc 6946 & [email protected] & [email protected] & 20.0 & 48 & 7.0 & 10.01 & sab(rs)cd & & tacconi & young ( 1986 ) + ugc 11820 & [email protected] & [email protected] & 5.2 & 1104 & 18.7 & 9.50 & sm & & van zee et al . ( 1997 ) + ngc 7331 & [email protected] & [email protected] & 13.6 & 816 & 15.8 & 10.02 & sa(s)b & & bosma ( 1981 ) + ic 5267 & [email protected] & [email protected] & 1.7 & 1713 & 27.5 & 9.63 & ( r)sa(rs)0/a & ring & van driel et al . ( 1988b ) + ugc 12613 & [email protected] & [email protected] & & -183 & 0.8 & 6.42 & dirr / dsph & & lo et al . ( 1993 ) + ngc 7793 & [email protected] & [email protected] & 13.2 & 230 & 3.4 & 8.88 & sa(s)d & & carignan & puche ( 1990a ) + cole , s. , norberg , p. , baugh , c. m. , frenk , c. s. , bland - hawthorn , j. , bridges , t. , cannon , r. , colless , m. , collins , c. , couch , w. , cross , n. , dalton , g. , de propris , r. , driver , s. p. , efstathiou , g. , ellis , r. s. , glazebrook , k. , jackson , c. , lahav , o. , lewis , i. , lumsden , s. , maddox , s. , madgwick , d. , peacock , j. a. , peterson , b. a. , sutherland , w. , taylor , k. 2001 , mnras , 326 , 255 .
we show that there are informationally complete joint measurements of two conjugated observables on a finite quantum system , meaning that they enable to identify all quantum states from their measurement outcome statistics . we further demonstrate that it is possible to implement a joint observable as a sequential measurement . if we require minimal noise in the joint measurement , then the joint observable is unique . if the dimension is odd , then this observable is informationally complete . but if is even , then the joint observable is not informationally complete and one has to allow more noise in order to obtain informational completeness .
1111.3509
the general aim in quantum tomography is to identify quantum states from measurement outcome statistics . a collection of observables with this property is called _ informationally complete _ @xcite . even a single observable can be informationally complete , but then it must be a noncommutative positive operator valued measure ( povm ) @xcite . we will study a class of informationally complete povms in dimension @xmath0 with the minimal number of @xmath1 outcomes and we will explain how they can be implemented as sequential measurements of two @xmath0-outcome measurements . one can interpret the generated joint observable as a phase space measurement in the discrete phase space @xmath2 @xcite . there are some particularly interesting approaches to finite dimensional quantum tomography , and one of them is based on complete collections of mutually unbiased bases ( mubs ) @xcite . in a @xmath0-dimensional hilbert space one needs @xmath3 mubs in order to be able to identify all quantum states , but it is not known if a complete set of mubs exists in all dimensions . in fact , there is evidence that for @xmath4 there is no complete set of mubs @xcite . in our scheme we start from two mutually unbiased bases connected by the finite fourier transform . they define a pair of complementary observables , which can not be measured jointly . however , it is possible to realize their joint measurement if some additional noise is allowed . we show that their joint measurement can be chosen to be informationally complete , and that this can be realized as a sequential measurement where we first perform a ` weak measurement ' in one basis and then another successive measurement in the other basis . compared to the fact that one would need @xmath3 complementary observables in order to reach informational completeness in separate measurements , it is remarkable that in the sequential scheme only two observables suffice . the price to have a joint measurement is that the marginal observables are not the original complementary observables but their unsharp versions . we will analyze the required additional noise and characterize the optimal joint observable from this point of view . the qubit case has been first studied in @xcite , and our work generalizes those results to arbitrary finite dimension . the covariant phase space observables , i.e. , povms covariant under the finite weyl - heisenberg group , play a special role in our investigation . we prove that if a pair of conjugate observables have a joint measurement , then they also have a joint measurement which is a covariant phase space observable . since every covariant phase space observable arises from a sequential measurement of two conjugate observables @xcite , the covariant phase space observables are an outstanding choice for finite dimensional quantum tomography . the weyl - heisenberg group has also a pivotal role in the investigations of symmetric informationally complete ( sic ) observables @xcite . it is generally believed that a weyl - heisenberg covariant sic observable exists in every finite dimension and their existence is numerically tested in all dimensions up to 67 @xcite . our results show that any such observable has a neat sequential realization scheme . there is an interesting difference between the even and odd dimensional hilbert spaces . if we require minimal noise in both marginal observables , then their joint observable is unique . if @xmath0 is odd , then this observable is informationally complete . but if @xmath0 is even , then the joint observable is not informationally complete . this result gives an additional aspect to the common observation that quantum tomography is different in even and odd dimensions @xcite . in this section we fix some notations and introduce the basic concepts . let @xmath5 be a finite dimensional hilbert space , with @xmath6 . we denote by @xmath7 the vector space of all linear operators on @xmath5 . a positive operator @xmath8 having trace one is a _ state _ , and we denote by @xmath9 the set of all states . observables are generally described by _ positive operator valued measures _ ( povms ) @xcite . in this work we only consider observables with finite number of outcomes . therefore , an observable can be defined as a function @xmath10 , where each @xmath11 is a positive operator and @xmath12 . here the sum runs over all @xmath13 , where the set @xmath14 is the collection of all possible measurement outcomes . if a system is prepared in a state @xmath15 , then a measurement of an observable @xmath16 will lead to an outcome @xmath17 with the probability @xmath18}$ ] . an observable @xmath16 is _ informationally complete _ if its measurement outcome probability distribution is sufficient to identify a unique state @xcite . in other words , two different states must give rise to different probability distributions : for all pairs of states @xmath19 , @xmath20 } = { { \rm tr}\left[\varrho_2 { \mathsf{a}}(j)\right ] } \quad \forall j\in\omega_{\mathsf{a}}\quad \rightarrow \quad\varrho_1=\varrho_2 \ , .\ ] ] the informational completeness of an observable @xmath16 is equivalent to the property that the linear span of the set @xmath21 is @xmath7 @xcite . given two observables @xmath16 and @xmath22 , we say that they are _ jointly measurable _ if there exists a third observable @xmath23 with @xmath24 and satisfying @xmath25 in other words , @xmath16 and @xmath22 correspond to the ` marginals ' of @xmath23 . any observable satisfying is called a _ joint observable _ of @xmath16 and @xmath22 @xcite . we recall that joint measurability is equivalent to the following @xcite : there exists an observable @xmath26 and stochastic matrices @xmath27 $ ] , @xmath28 $ ] such that @xmath29 hence , two observables are jointly measurable iff they can be ` post - processed ' from a single observable . we will use several times the following simple fact : if @xmath23 and @xmath30 are joint observables of @xmath16 and @xmath22 , then also all their convex combinations @xmath31 , @xmath32 , are joint observables of @xmath16 and @xmath22 . it follows that two jointly measurable observables have either a unique joint observable or infinitely many of them . another useful fact is related to unitary transformations . let @xmath33 be a unitary operator on @xmath5 . two observables @xmath16 and @xmath22 are jointly measurable if and only if the observables @xmath34 and @xmath35 are jointly measurable . indeed , it is easy to see that @xmath23 is a joint observable of @xmath16 and @xmath22 if and only if @xmath36 is a joint observable of @xmath34 and @xmath37 . an observable describes the statistics of the outcomes of a measurement but leaves open how the measurement disturbs the input state . in order to discuss this we need the concept of an _ instrument _ @xcite . an instrument with finitely many outcomes is a mapping @xmath38 such that each @xmath39 is a completely positive linear map on @xmath7 and @xmath40}=1 $ ] for all states @xmath15 . the _ adjoint map _ @xmath41 of @xmath39 is defined via the usual trace duality @xmath42}={{\rm tr}\left[{\mathcal{i}}_x^\ast(s)t\right ] } \qquad \forall s , t\in{\mathcal{l(h)}}\ , .\ ] ] in other words , @xmath41 and @xmath39 correspond to the heisenberg and schrdinger pictures , respectively . suppose that @xmath16 is an observable . then we say that an instrument @xmath43 is _ @xmath16-compatible _ if @xmath44 for every @xmath17 . every @xmath16-compatible instrument describes some particular kind of measurement of @xmath16 @xcite . an example of an @xmath16-compatible instrument is the _ lders instrument _ @xmath45 , defined by @xmath46 any other @xmath16-compatible instrument @xmath43 is of the form @xmath47 for some collection @xmath48 of completely positive trace preserving maps on @xmath7 @xcite . by a _ sequential measurement _ we mean a setting where two measurements are combined into a third measurement by performing them one after the other @xcite . generally , the order in which the measurements are performed is crucial @xcite . suppose we have two @xmath49-outcome observables @xmath50 and we measure them subsequently ; first @xmath16 and then @xmath22 . as a result , we have in total @xmath51 possible measurement outcomes . generally , we do not obtain a joint measurement of @xmath16 and @xmath22 since the first measurement distrubs the input state . in fact , the overall measurement depends on the way we measure @xmath16 . if the first measurement is described by an @xmath16-compatible instrument @xmath43 , then the overall observable @xmath23 is given by @xmath52 } = { { \rm tr}\left[{\mathsf{b}}(k){\mathcal{i}}_j(\varrho)\right]}\ ] ] for all input states @xmath15 , or equivalently , @xmath53 let us notice that first marginal of @xmath23 is always @xmath16 , while the second marginal is a perturbed version of @xmath22 and depends on the instrument @xmath43 . we start with a preliminary example , which is mainly a collection of well known facts . it hints the forthcoming developments and clarifies the aims of the later sections . we refer to @xcite for more details and further references . fix @xmath56 , and let @xmath16 and @xmath22 be the two observables corresponding to the measurements of spin-@xmath57 components in the directions @xmath17 and @xmath58 , respectively . thus , @xmath59 where @xmath60 are the pauli spin matrices . since @xmath16 and @xmath22 consist of projections and they do not mutually commute , it is not possible to measure them jointly . moreover , if we measure them separately on two similarly prepared ensembles , we still can not infer the unknown state . an alternative way is to perform a sequential measurement . the first measurement has to be a weak measurement , meaning that we do not measure @xmath16 but its unsharp version . we define an unsharp version @xmath61 of @xmath16 by @xmath62 here @xmath63 $ ] is a parameter quantifying the noise or imprecision . we can write @xmath61 in the form @xmath64 in a similar way we define an unsharp version @xmath65 of @xmath22 by @xmath66 we want to study the disturbance of the first measurement on the system , and for this reason we define an instrument related to @xmath61 . a class of @xmath61-compatible instruments can be defined by @xmath67 where @xmath68 are arbitrary unitary operators . if the subsequent measurement is a @xmath22-measurement , then the overall statistics of the sequential measurement is given by the observable @xmath69 the properties of @xmath23 obviously depend on @xmath70 and @xmath71 . in the following we consider two different choices of @xmath72 . if we choose @xmath73 , then we obtain @xmath74 in particular , the marginals are @xmath75 the joint observable @xmath23 is an optimal approximate joint measurement of @xmath54 and @xmath55 . this means that the unsharp parameters @xmath76 and @xmath77 saturate the inequality @xmath78 indeed , it is known that this inequality is a necessary and sufficient criterion for two observables @xmath61 and @xmath65 to be jointly measurable @xcite . let us also notice that the joint observable of @xmath61 and @xmath65 is unique if @xmath79 @xcite . another interesting option is to choose @xmath80 for some fixed angle @xmath81 . in this case we obtain @xmath82 and the marginals are @xmath83 it is easy to see that the linear span of the four operators @xmath84 , @xmath85 , is the set of all @xmath86 - complex matrices . it follows that the joint observable @xmath23 is informationally complete . the unsharpness parameters @xmath76 and @xmath87 do not saturate the inequality . altering the parameter @xmath88 we can make the sum @xmath89 as close to @xmath90 as we want , hence we conclude that @xmath61 and @xmath65 admit an informationally complete joint observable if and only if @xmath91 finally , we remark that with the choices @xmath92 and @xmath93 the joint observable @xmath23 is a symmetric informationally complete ( sic ) observable . we start by recalling the usual definition of complementary observables in a finite @xmath0-dimensional hilbert space and some related basic facts @xcite , @xcite . we denote @xmath94 . let @xmath95 and @xmath96 be _ mutually unbiased bases _ ( mubs ) , i.e. , they are orthonormal bases in @xmath5 and @xmath97 we define two @xmath0-outcome observables @xmath16 and @xmath22 corresponding to @xmath95 and @xmath96 , respectively . hence , @xmath98 obviously , two orthonormal bases @xmath95 and @xmath99 define the same observable @xmath16 iff @xmath100 for some complex numbers @xmath101 of modulus one . the mutual unbiasedness condition can be rephrased by saying that @xmath16 and @xmath22 are _ complementary _ observables , meaning that in any state @xmath15 where the outcome of @xmath16 is predictable , the @xmath22-distribution is uniform ( and vice versa ) . this entails that the following implications are valid for any state @xmath15 and all outcomes @xmath102 , @xmath103 } & = 1 \quad\rightarrow\quad { { \rm tr}\left[\varrho{\mathsf{b}}(k)\right ] } = 1/d \\ { { \rm tr}\left[\varrho{\mathsf{b}}(k)\right ] } & = 1 \quad\rightarrow\quad { { \rm tr}\left[\varrho{\mathsf{a}}(j)\right ] } = 1/d \ , .\end{aligned}\ ] ] since @xmath104 } = 1 $ ] iff @xmath105 , it is easy to see that the complementarity of @xmath16 and @xmath22 is indeed equivalent to the mutual unbiasedness of the bases @xmath95 and @xmath96 . there is a canonical way to produce two mutually unbiased bases . in the following , suppose an orthonormal basis @xmath106 of @xmath5 is fixed . denoting @xmath107 , we define the following unitary representations @xmath33 and @xmath108 of the cyclic group @xmath109 in @xmath5 : @xmath110 for all @xmath111 . in the above formulas and in the rest of the paper , addition and multiplication of elements in @xmath109 are understood modulo @xmath0 . ( for instance , we will often use @xmath112 ) . it is easy to verify that @xmath113 the fourier transform ( with respect to the basis @xmath106 ) is the unitary operator @xmath114 defined by @xmath115 the adjoint operator @xmath116 of @xmath117 is given by @xmath118 and we have @xmath119 . we denote @xmath120 and it is immediate to check that @xmath121 and @xmath96 are mubs , with @xmath122 . the fourier transform has the intertwining properties @xmath123 from which it follows that @xmath124 the observables @xmath16 and @xmath22 related to @xmath121 and @xmath96 , respectively , satisfy the following conditions for all @xmath125 : @xmath126 and @xmath127 in other words , @xmath16 is @xmath33-covariant and @xmath108-invariant , while @xmath22 is @xmath33-invariant and @xmath108-covariant . we also note that @xmath16 and @xmath22 are conjugated by @xmath117 , i.e. , @xmath128 for all @xmath129 . it is customary to say that @xmath16 and @xmath22 are _ canonically conjugated _ observables . the conditions are analogous to the symmetry properties of the usual position and momentum observables on the real line @xmath130 ( see e.g. @xcite ) . in some situations the covariance properties may have some physical meaning or motivation . however , for our purposes they are just useful features that can be utilized later in our calculations . measurements of two complementary observables are incompatible and therefore have to be performed separately . this means that their measurements require different settings . however , it is possible to perform a simultaneous measurement of two complementary observables if we allow some additional imprecision or noise . in other words , we can measure jointly _ unsharp versions _ of @xmath16 and @xmath22 . we define an unsharp version @xmath61 of @xmath16 by @xmath131 here @xmath63 $ ] is a parameter quantifying the noise . this type of noise is equivalent to the situation where an input state @xmath15 is first depolarized into a state @xmath132 and then a measurement of @xmath16 is performed . more generally , if @xmath133 is a probability distribution on @xmath109 , then we define an unsharp version @xmath134 of @xmath16 by @xmath135 the special case @xmath136 corresponds to the probability distribution @xmath133 defined as @xmath137 we can also write this probability distribution in the form @xmath138 where @xmath139 is the point distribution at @xmath140 and @xmath141 is the uniform distribution on @xmath109 , i.e. , @xmath142 in a similar way a probability distribution @xmath143 on @xmath109 defines an unsharp version @xmath144 of @xmath22 by @xmath145 a special class is , again , characterized by noise parameters @xmath146 $ ] and we denote @xmath147 naturally , there are also other type of approximations of @xmath16 and @xmath22 than the previously defined @xmath134 and @xmath144 . the usefulness of @xmath134 and @xmath144 is that they satisfy the same covariance and invariance relations than @xmath16 and @xmath22 , respectively . namely , the observables @xmath134 and @xmath144 satisfy the following conditions : @xmath148 and @xmath149 thus , @xmath134 and @xmath144 are _ conjugated observables _ although they need not be complementary anymore @xcite . as we will see , two observables @xmath134 and @xmath144 can have a joint observable even if they do not commute . suppose that @xmath150 is a @xmath0-outcome observable satisfying @xmath151 for all @xmath152 . then @xmath153 for some probability distribution @xmath133 . namely , it follows from the second condition in that @xmath150 commutes with @xmath16 ( since @xmath154 ) and hence @xmath155 for some real numbers @xmath156 . the first condition in then implies that @xmath157 . we recall that two observables @xmath134 and @xmath144 are jointly measurable if they have a joint observable , i.e. , an observable @xmath23 on @xmath2 such that @xmath158 for all @xmath102 . a special class of joint observables turns out to be crucial for our developments . we say that an observable @xmath23 on @xmath2 is a _ covariant phase space observable _ if @xmath159 for all @xmath125 . the covariant phase space observables have a simple form @xcite . namely , if @xmath23 is a covariant phase space observable , then there is unique operator @xmath160 such that @xmath161 , where we have denoted @xmath162 also , each @xmath160 defines a covariant phase space observable by this formula . the correspondence @xmath163 is therefore one - to - one and the elements in @xmath9 parametrize the covariant phase space observables . the marginals of a covariant phase space observable @xmath164 are conjugated observables on @xmath109 . indeed , a direct calculation shows that @xmath164 has marginals @xmath134 and @xmath144 , with @xmath165 } \ , , \qquad \gamma ( k ) = { { \rm tr}\left[{\mathsf{b}}(-k)t\right ] } \ , .\ ] ] ( this calculation can be found in @xcite ) . the essential role of covariant phase space observables in our discussion becomes clear in the following observation . [ prop : if - jm - then - cov ] if @xmath134 and @xmath144 are jointly measurable , then they have a joint observable which is a covariant phase space observable . suppose that @xmath23 is a joint observable of @xmath134 and @xmath144 . for each @xmath166 , we define an observable @xmath167 by @xmath168 using the covariance and invariance properties it is straightforward to verify that @xmath167 is a joint observable of @xmath134 and @xmath144 . we then define @xmath169 to be the uniform mixture of all @xmath167 , i.e. , @xmath170 since every @xmath167 is a joint observable of @xmath134 and @xmath144 , also @xmath169 is their joint observable . a direct calculation , using , shows that @xmath169 is a covariant phase space observable . we conclude from proposition [ prop : if - jm - then - cov ] that two observables @xmath134 and @xmath144 are jointly measurable iff their related probability distributions @xmath133 and @xmath143 are of the form for some @xmath160 . equations can also be rewritten in a slightly different form . namely , observe that , if @xmath160 , then there exists a unit vector @xmath171 such that @xmath172 $ ] , where @xmath173 is the partial trace with respect to the second factor . ( a vector state giving @xmath174 via the partial trace is often called a _ purification _ of @xmath174 ) . conversely , if @xmath171 is a unit vector , then @xmath172 $ ] is a state . inserting this form into we obtain @xmath175 note that if a vector @xmath171 satisfies the above two equations for some probability densities @xmath176 and @xmath143 , then the normalization @xmath177 is automatic . we thus have the following characterization of jointly measurable observables . [ prop : joint->phi ] let @xmath176 , @xmath143 be probability densities on @xmath109 . the following facts are equivalent : * the observables @xmath134 and @xmath144 are jointly measurable . * there exists a state @xmath178 such that the probability densities @xmath133 and @xmath143 satisfy . * there exists a vector @xmath179 such that the probability densities @xmath133 and @xmath143 satisfy . let us note that any @xmath134 is jointly measurable with some @xmath144 . namely , for each @xmath133 we can define a state @xmath180 as @xmath181 then @xmath182}$ ] and @xmath134 is thus jointly measurable with @xmath144 , where @xmath143 is defined as @xmath183}$ ] . proposition [ prop : joint->phi ] can be seen as a trade - off relation between the probability distributions @xmath133 and @xmath143 that describe the deviations of @xmath134 and @xmath144 from @xmath16 and @xmath22 , respectively . for instance , if @xmath184 , then necessarily @xmath185 . hence , we recover the fact that @xmath16 is jointly measurable only with the trivial observable and no other @xmath144 . we end this subsection with some additional observations . _ if two observables @xmath134 and @xmath144 are jointly measurable , they can have several different covariant phase space observables as their joint observables . _ for instance , let @xmath186 be an orthonormal basis which is mutually unbiased with respect to both orthonormal bases @xmath95 and @xmath96 . then , for each @xmath187 , we have @xmath188 therefore , the marginals of the covariant phase space observables @xmath189 are the same although @xmath190 whenever @xmath191 . _ if two observables @xmath134 and @xmath144 are jointly measurable , they can have a joint observable which is not a covariant phase space observable . _ for instance , let @xmath192 $ ] be a bivariate probability distribution with uniform margins . then the observable @xmath193 is a joint observable of @xmath194 and @xmath195 . it is clear that @xmath23 is a covariant phase space observable only if @xmath196 is a uniform distribution . however , a bivariate probability distribution with uniform margins need not be uniform . for instance , if we set @xmath197 then @xmath198 , but @xmath196 is not uniform . the existence of non - covariant joint observables is not limited to the trivial observables @xmath194 and @xmath195 . namely , suppose that @xmath23 is a joint observable of @xmath194 and @xmath195 and @xmath30 is a joint observable of @xmath61 and @xmath65 . then the convex combination @xmath199 , @xmath32 , is a joint observable of @xmath200 and @xmath201 . it is easy to see that if @xmath30 is a covariant phase space observable but @xmath23 is not , then their convex combination @xmath199 can not be a covariant phase space observable . [ prop : unique ] _ suppose that @xmath202 and @xmath65 have a unique joint observable @xmath164 among the covariant phase space observables and that @xmath203 . then @xmath164 is a unique joint observable of @xmath61 and @xmath65 . _ to prove this claim , let @xmath23 be a joint observable of @xmath61 and @xmath65 . we need to show that @xmath161 . we define @xmath167 and @xmath169 as in . since @xmath169 is by its construction a covariant phase space observable , we must have @xmath204 by the assumption on uniqueness . in particular , each operator @xmath205 is rank-1 . since @xmath206 by , it follows that there exists a real constant @xmath207 such that @xmath208 , hence @xmath209 by . suppose @xmath210 . since @xmath23 has @xmath202 as its first marginal , then @xmath211 the right hand side of this equation is a rank-@xmath0 operator , while on the left hand side we have the sum of @xmath0 operators with rank-@xmath90 . it then follows that the set @xmath212 is linearly independent in @xmath7 . since @xmath23 and @xmath169 have the same marginals , @xmath213 for all @xmath17 , hence @xmath214 for all @xmath215 by linear independence . the case @xmath216 is treated in a similar way , by taking the marginal @xmath65 in the place of @xmath202 ( and now necessarily @xmath217 ) . therefore , @xmath164 is the unique joint observable of @xmath202 and @xmath65 . in this subsection we apply proposition [ prop : joint->phi ] to the cases where @xmath136 and @xmath218 . these special type of marginal observables are interesting as we can quantify their unsharpnesses by single numbers @xmath76 and @xmath219 . in particular , we can ask how small @xmath76 and @xmath219 must be in order for @xmath61 and @xmath65 to become jointly measurable . we first notice that this question is , indeed , meaningful . [ prop : ord . makes sense ] let @xmath220 $ ] . the following conditions are equivalent : * @xmath61 and @xmath65 are jointly measurable . * @xmath221 and @xmath222 are jointly measurable for all @xmath223 and @xmath224 . * @xmath221 and @xmath222 are jointly measurable for all @xmath225 and @xmath226 . suppose that ( i ) holds and @xmath227 . we denote @xmath228 and hence @xmath229 . we have @xmath230 meaning that @xmath222 is a convex combination of @xmath231 and the trivial observable @xmath195 . by the assumption @xmath61 is jointly measurable with @xmath231 and @xmath61 is also jointly measurable with the trivial observable @xmath195 ( since they commute ) . if @xmath232 is a joint observable of @xmath61 and @xmath231 and @xmath233 is a joint observable of @xmath61 and @xmath195 , then the convex combination @xmath234 is a joint observable of @xmath61 and @xmath222 . therefore , @xmath61 and @xmath222 are jointly measurable . we can interchange the roles of @xmath61 and @xmath231 and run the same argument , hence we obtain ( ii ) . it is clear that ( ii ) implies ( iii ) . hence , to complete the proof we need to show that ( iii ) implies ( i ) . suppose that ( iii ) holds . we choose sequences @xmath235 and @xmath236 such that @xmath237 , @xmath238 and @xmath239 , @xmath240 . for each @xmath241 , we fix a state @xmath242 such that the corresponding covariant phase space observable @xmath243 is a joint observable of @xmath244 and @xmath245 . the set of states @xmath9 is compact in the operator norm topology , hence the sequence @xmath246 has a convergent subsequence . we denote by @xmath174 the limit of this convergent subsequence . using we see that the covariant phase space observable @xmath164 is a joint observable of @xmath61 and @xmath65 . thus , ( i ) holds . we conclude from proposition [ prop : ord . makes sense ] that for every @xmath63 $ ] , there is a number @xmath247 such that @xmath202 and @xmath231 are jointly measurable iff @xmath248 . similarly , for every @xmath249 $ ] , there is a number @xmath250 such that @xmath202 and @xmath231 are jointly measurable iff @xmath251 . we also know that @xmath252 ( since a trivial observable is jointly measurable with any other observable ) and that @xmath253 ( see the discussion after proposition [ prop : joint->phi ] ) . the equality @xmath254 holds for all @xmath255 $ ] . it is enough to show that @xmath202 and @xmath65 are jointly measurable if and only if @xmath256 and @xmath257 are such . joint measurability of @xmath202 and @xmath65 means that there exists an observable @xmath23 on @xmath258 having marginals @xmath202 and @xmath65 , respectively . we set @xmath259 for every @xmath102 . then @xmath260 we conclude that @xmath261 is a joint observable of @xmath256 and @xmath257 , hence the latter two are jointly measurable . we will now find out the function @xmath262 , or , equivalently , @xmath263 . suppose that @xmath264 $ ] are such that @xmath61 and @xmath65 are jointly measurable . by proposition [ prop : joint->phi ] this is equivalent to the existence of a vector @xmath179 satisfying @xmath265 for all @xmath102 ( see for the definition of @xmath139 and @xmath141 ) . we now give a condition on the parameters @xmath266 which is necessary and sufficient for the existence of a vector @xmath179 satisfying the above two equations . moreover , we show that , for the extreme values of @xmath266 , the vector @xmath267 is essentially unique . [ lemma : bound ] let @xmath264 $ ] . suppose there exists a vector @xmath171 satisfying . then @xmath268 \ , .\ ] ] for any choice of a unit vector @xmath269 , the vector @xmath270 with @xmath271 \ , , \qquad \beta_{\lambda}= \sqrt{1-{\lambda } } \ , , \end{aligned}\ ] ] satisfies with equality in . hence , the right hand side in is equal to @xmath262 . if @xmath272 is a vector satisfying with @xmath273 , then @xmath274 for some unit vector @xmath275 . suppose @xmath171 satisfies . we write @xmath267 in the form @xmath276 where @xmath277 are vectors in @xmath5 . from it follows that @xmath278 hence there exist unit vectors @xmath279 such that @xmath280 on the other hand , we have @xmath281 so , by , we must have @xmath282 this equation , evaluated at @xmath283 , gives @xmath284 the maximum value of @xmath219 is then achieved when the right hand side of this equation is maximal , i.e. , when there exists a unit vector @xmath285 such that @xmath286 @xmath287 . the corresponding maximum value @xmath288 of @xmath219 is given by @xmath289 i.e. , @xmath290 \ , .\ ] ] in order to show that , if the sequence @xmath291 is chosen as above , then the corresponding vector @xmath292 satisfies also with @xmath293 ( and thus the maximum is indeed achieved by @xmath267 ) , we evaluate @xmath294 which is . as a consequence of the above discussion , we obtain an inequality for the unsharpnesses of two jointly measurable observables @xmath61 and @xmath65 . for @xmath295 ( red color ) and for @xmath296 ( orange color).,width=151 ] [ prop : gammamax ] two observables @xmath61 and @xmath65 are jointly measurable if and only if @xmath297 \ , , \ ] ] ( or , equivalently , its modified form under the exchange @xmath298 ) . if @xmath273 , then @xmath61 and @xmath65 have a unique joint observable . this unique joint observable is the covariant phase space observable @xmath164 defined by the state @xmath299 with @xmath300 \ , , \qquad \beta_\lambda = \sqrt{1-\lambda } \ , .\end{aligned}\ ] ] if @xmath202 and @xmath65 are jointly measurable , then the inequality follows from proposition [ prop : joint->phi ] and lemma [ lemma : bound ] . conversely , if @xmath288 is given by the right hand side of , then the pair @xmath202 and @xmath301 are jointly measurable again by an application of proposition [ prop : joint->phi ] and lemma [ lemma : bound ] . then , @xmath202 and @xmath65 are jointly measurable by proposition [ prop : ord . makes sense ] . now suppose @xmath302 and @xmath219 achieve the bound , and let @xmath164 be a covariant joint observable of @xmath202 and @xmath65 . pick @xmath171 such that @xmath172 $ ] . as @xmath267 satisfies with @xmath303 , by lemma [ lemma : bound ] it must be given by for some choice of a unit vector @xmath269 , and @xmath172 = { |\chi_{\lambda}\rangle\langle\chi_{\lambda}|}$ ] , with @xmath304 as in the statement of the proposition . finally , we need to prove that @xmath164 is the unique joint observable ( and not only unique among covariant phase space observables ) . we notice that @xmath203 , and the claim thus follows from remark [ prop : unique ] . the graph of the function @xmath305 is a part of an ellipse . in fig . [ fig : boundary ] we have depicted it for @xmath306 . suppose that two jointly measurable observables @xmath61 and @xmath65 are ` equally unsharp ' but as close to @xmath16 and @xmath22 as possible , i.e. , @xmath307 in this case proposition [ prop : gammamax ] gives @xmath308 the observables @xmath61 and @xmath65 then have a unique joint observable , which is the covariant phase space observable @xmath164 associated to the state @xmath309 , with @xmath310 the vector state @xmath311 is hence an equal superposition of the vector states @xmath312 and @xmath313 . by a direct calculation one can verify that the inequality can be rewritten in the following equivalent form which is symmetric in @xmath76 and @xmath219 . [ prop : symmetric ] two observables @xmath61 and @xmath65 are jointly measurable iff @xmath314 the second inequality in describes a full ellipse . therefore , the first condition @xmath315 is needed to ignore the lower part of the ellipse , which is not a correct boundary for joint measurability . this is depicted in fig . [ fig : ellipse ] . . only the upper side of the ellipse is relevant for the joint measurability.,width=151 ] let us notice that for @xmath316 the latter inequality in becomes @xmath317 , and then the first condition is redundant . we thus recover the single condition stated in and first proved in @xcite . also , for @xmath318 and @xmath319 a direct calculation shows that the the first condition in is redundant and the latter quadratic inequality is necessary and sufficient for the joint measurability . for @xmath320 we need both conditions in . by inspecting the function @xmath305 we see that for every @xmath321 , there is a pair of observables @xmath61 and @xmath65 such that they are not jointly measurable and @xmath322 . thus , the criterion @xmath315 is the best _ sufficient _ condition for joint measurability which is linear and symmetric in @xmath76 and @xmath219 . the best linear and symmetric _ necessary _ condition for joint measurablity is achived by taking the tangent of the boundary curve in the point where it crosses the line @xmath323 . in this way , we obtain the following conclusion . [ prop : linear ] if @xmath61 and @xmath65 are jointly measurable , then @xmath324 one can also see proposition [ prop : linear ] in the opposite order ; if @xmath325 , then @xmath61 and @xmath65 are not jointly measurable . in fig . [ fig : lines ] we have depicted the linear necessary and sufficient conditions in the case @xmath326 . . the gray regions represent the necessary and sufficient linear conditions . in the white stripe one has to invoke the quadratic ellipse criterion , whereas otherwise the joint measurability can be deduced from the simple linear criteria.,width=151 ] so far , we have concentrated on covariant observables @xmath134 and @xmath144 . let us have a short look on a class of non - covariant observables . let @xmath196 and @xmath327 be two probability distributions on @xmath109 . we define @xmath328 and @xmath329 it is straightforward to verify that @xmath330 is @xmath33-covariant and @xmath108-invariant iff @xmath196 is the uniform distribution on @xmath109 , in which case @xmath331 . ( analogous statement holds for @xmath332 ) . the following result is a generalization of proposition 5 in @xcite . [ prop : non ] if @xmath330 and @xmath332 are jointly measurable , then @xmath202 and @xmath231 are jointly measurable . suppose that @xmath330 and @xmath332 are jointly measurable and let @xmath23 be their joint observable . as in the proof of proposition [ prop : if - jm - then - cov ] , we define the observable @xmath169 , given by @xmath333 for each @xmath334 , we have @xmath335 in a similar way we obtain @xmath336 for every @xmath337 . therefore , @xmath169 is a joint observable for @xmath202 and @xmath231 . as a consequence of propositions [ prop : gammamax ] and [ prop : non ] we conclude the following necessary criterion for joint measurability . [ cor : nec ] if two observables @xmath330 and @xmath332 are jointly measurable , then @xmath338 \ , .\ ] ] a necessary and sufficient inequality for the joint measurability of @xmath330 and @xmath332 must contain also @xmath196 and @xmath327 in a form or in another . it is thus clear that corollary [ cor : nec ] does not give a sufficient condition . a necessary and sufficient condition in the case @xmath316 has been obtained in @xcite . we remark that a general necessary condition for the joint measurability of two observables on a finite dimensional system has been presented in @xcite . a comparison to proposition [ prop : gammamax ] shows that this condition is not sufficient . we leave it as an open problem to find a necessary and sufficient condition for the joint measurability of @xmath330 and @xmath332 . we will now study the informational completeness of joint observables of @xmath202 and @xmath65 . let us first recall that the informational completeness of a covariant phase space observable @xmath164 is equivalent to the criterion @xmath339 } \neq 0 \quad \forall x , y\in{\mathbb z}_d \ , .\ ] ] this result has been discussed e.g. in @xcite . for completeness , we provide a proof in appendix . [ prop : icmax ] suppose @xmath202 and @xmath65 are two observables with @xmath340 and @xmath341 . then @xmath202 and @xmath65 have a unique joint observable @xmath23 . the observable @xmath23 is informationally complete if and only if @xmath0 is odd . from proposition [ prop : gammamax ] we know @xmath61 and @xmath65 have a unique joint observable @xmath164 , generated by the state @xmath342 , with @xmath343 . the informational completeness of @xmath164 is equivalent to the condition , and a straightforward calculation gives @xmath344 } = { \left\langle\,\chi_{\lambda}\,|\,u_xv_y\chi_{\lambda}\,\right\rangle}= \alpha_{\lambda}^2 \delta_{x,0 } + \beta_{\lambda}^2 \delta_{y,0 } + \frac{\alpha_{\lambda}\beta_{\lambda}}{\sqrt{d } } \left ( \omega^{-xy } + 1 \right ) \ , .\ ] ] let us first notice that @xmath345 and @xmath346 since both @xmath76 and @xmath219 are nonzero ( see ) . hence , @xmath347}=0 $ ] exactly when @xmath348 . the latter condition is equivalent to @xmath349 . we conclude that the informational completeness of @xmath164 is equivalent to the fact that the equation @xmath350 has no solution @xmath351 , and this holds if and only if @xmath0 is odd . in proposition [ prop : icmax ] the crucial assumption is that @xmath341 . this guarantees that @xmath61 and @xmath65 have a unique joint observable . if we have @xmath352 , then @xmath61 and @xmath65 have infinitely many joint observables . in this case it is always possible to choose an informationally complete joint observable , as we prove in the following . [ prop : icnonmax1 ] suppose @xmath202 and @xmath65 are two observables with @xmath340 and @xmath353 . then they have an informationally complete covariant joint observable . let @xmath354 be the intersection of the half line @xmath355 with the boundary of the domain in @xmath356 , and let @xmath357 such that @xmath358 . let @xmath359 . we treat separately the cases of odd and even @xmath0 . \1 ) suppose that @xmath360 is even . for all @xmath129 , we denote @xmath361 the linear maps @xmath362 are selfadjoint trace zero operators for every @xmath337 , and it is easy to check that @xmath363 we introduce the selfadjoint operators @xmath364 and , for @xmath365 , @xmath366 if @xmath367 , then @xmath368 . moreover , the associated covariant phase space observable @xmath369 has trivial marginals @xmath194 and @xmath195 . a straightforward calculation gives @xmath370 } = \delta_{x,0}\delta_{y,0 } + i \kappa ( \omega^{-xy } - 1 ) \ , .\ ] ] the covariant phase space observable @xmath371 associated to the state @xmath372 , with @xmath373 , has margins @xmath202 and @xmath65 . moreover , @xmath374 } & = & ( 1-\tau ) { \left(}\alpha_{{\lambda}_0}^2 \delta_{x,0 } + \beta_{{\lambda}_0}^2 \delta_{y,0 } { \right)}+ \tau \delta_{x,0 } \delta_{y,0 } \notag\\ & & + ( 1-\tau ) \frac{\alpha_{{\lambda}_0 } \beta_{{\lambda}_0}}{\sqrt{d } } ( \omega^{-xy } + 1 ) + i\kappa \tau ( \omega^{-xy } - 1 ) \ , .\end{aligned}\ ] ] let @xmath375 for @xmath376 , the right hand side of is nonzero for all @xmath166 , which proves informational completeness of @xmath371 by the criterion . \2 ) suppose that @xmath0 is odd . then , for @xmath377 , the associated covariant phase space observable @xmath164 has margins @xmath202 and @xmath65 , and @xmath344 } = ( 1-\tau ) \left [ \alpha_{{\lambda}_0}^2 \delta_{x,0 } + \beta_{{\lambda}_0}^2 \delta_{y,0 } + \frac{\alpha_{{\lambda}_0 } \beta_{{\lambda}_0}}{\sqrt{d } } \left ( \omega^{-xy } + 1 \right ) \right ] + \tau \delta_{x,0 } \delta_{y,0 } \ , , \ ] ] which is nonzero for all @xmath378 . the informational completeness of @xmath164 then follows from the criterion . the two trivial cases @xmath379 or @xmath380 are not very interesting , but for completeness we make the following observation . [ prop : icnonmax2 ] suppose @xmath202 and @xmath65 are two observables with @xmath381 or @xmath380 . then they have no informationally complete joint observable . we consider only the case @xmath379 , the case @xmath380 being similar . suppose that @xmath23 is a joint observable of @xmath382 and @xmath257 . we have @xmath383 hence @xmath384 and then , for @xmath385 , @xmath386 thus , @xmath23 is not informationally complete . in this section we discuss the sequential implementation of joint observables of @xmath134 and @xmath144 in the light of the recent results obtained in @xcite and @xcite . for illustrative purposes , we point out that two naive methods do not work . suppose that @xmath134 and @xmath144 are jointly measurable , i.e. , they satisfy the condition stated in proposition [ prop : joint->phi ] . the most uncomplicated way to realize their joint measurement would be to perform an @xmath134-measurement without disturbing the subsequent @xmath144-measurement . in terms of instruments , this would mean that we choose an @xmath134-compatible instrument @xmath43 such that @xmath387 for all @xmath337 . however , this type of measurement is typically not possible since a quantum measurement necessarily disturbs the input state . let us first notice that @xmath61 and @xmath65 commute if and only if @xmath388 , meaning that one of them is a trivial observable . generally , a non - disturbing measurement can be possible even if two observables do not commute . but applying proposition 3 from @xcite we see that this possibility is excluded whenever @xmath65 is informationally equivalent with @xmath22 in the sense that the linear spans of the sets @xmath389 and @xmath390 are equal . this property is satisfied by any @xmath65 with @xmath391 . therefore , whenever both observables are nontrivial , then @xmath61-measurement disturbs the subsequent @xmath65-measurement and the resulting observable is not a joint measurement of @xmath61 and @xmath65 . suppose we have a measurement setup for @xmath16 and that the corresponding instrument is @xmath43 . we can implement an unsharp observable @xmath61 by performing the @xmath16-measurement in a randomly chosen @xmath76-part of the ensemble and doing nothing for the rest @xmath392-part . the corresponding @xmath61-compatible instrument @xmath393 is then @xmath394 this is clearly a very direct way to decrease the disturbance that an @xmath16-measurement would cause . by measuring the observable @xmath22 after the first measurement , one could expect to have a useful joint measurement of @xmath61 and some approximate version of @xmath22 . however , this type of method does not yield an informationally complete joint measurement . the observable @xmath16 consists of rank-1 operators , and any @xmath16-compatible instrument @xmath43 is of the form @xmath395 } \xi_j\ ] ] for some set of states @xmath396 @xcite . if we insert this form into , we see that a sequential measurement consisting of @xmath393 followed by a @xmath22-measurement leads to the joint observable @xmath397 } { \mathsf{a}}(j ) + \frac{1-\lambda}{d } \ , { \mathsf{b}}(k ) \ , , \qquad j , k\in{\mathbb z}_d \ , .\ ] ] the linear span of the set @xmath398 is contained in the linear span of the union @xmath399 . the latter is strictly smaller than @xmath7 , hence @xmath23 is not informationally complete . we also see that this kind of approach can not give more information than separate measurements of @xmath16 and @xmath22 would give . we recall from @xcite that every joint observable of @xmath134 and @xmath144 can be implemented as a sequential measurement of @xmath134 followed by a measurement of @xmath22 . namely , suppose that @xmath23 is a joint observable of @xmath134 and @xmath144 . we define an instrument @xmath43 by @xmath400 } { \mathsf{b}}(k ) \ , .\ ] ] this is an @xmath134-compatible instrument , and from @xmath401 it follows that @xmath402 } = { { \rm tr}\left[\varrho { \mathsf{c}}(j , k)\right ] } \ , .\ ] ] hence , @xmath403 , and we conclude that @xmath23 is implemented as a sequential measurement of @xmath134 followed by a measurement of @xmath22 , as claimed . the instrument defined in may look quite artificial and before we know the structure of @xmath23 , the formula does not give us any hint on the structure of @xmath43 . in contrast , every covariant phase space observable can implemented as sequential measurement of @xmath134 and @xmath22 in a very specific form . as explained in @xcite , every covariant @xmath134-compatible instrument gives rise to a covariant phase space observable . covariance of an instrument @xmath43 here means that @xmath404 for all @xmath405 and @xmath406 . it is straightforward to verify that the joint observable @xmath407 is a covariant phase space observable . we demonstrate this method by choosing the @xmath61-compatible lders instrument @xmath45 , defined as @xmath408 it is straightforward to see that @xmath45 satisfies . the covariant joint observable is then @xmath409 and its associated state is @xmath410 since @xmath411 and @xmath412 we see that @xmath413 , hence by proposition [ prop : gammamax ] the marginal @xmath65 is such that @xmath273 . in conclusion , this type of sequential measurement of @xmath61 and @xmath22 is effectively a joint measurement of @xmath61 and @xmath65 with minimal unsharpnesses . in our investigation we have concentrated on canonically conjugated pairs of observables , i.e. , the orthonormal bases @xmath95 and @xmath96 have been assumed to be fourier connected with respect to the fourier transform of the cyclic group @xmath109 ; see . equivalently , we have assumed that the two bases satisfy @xmath122 for all @xmath102 . as a consequence the observables @xmath16 and @xmath22 , both defined on @xmath414 as @xmath415 and @xmath416 , satisfy the covariance and invariance conditions , which turn out to be very useful in our calculations . our approach covers more cases than it may seem at the first sight . namely , we recall that two orthonormal bases @xmath417 and @xmath418 define the same observable iff there are complex numbers @xmath101 with @xmath419 such that @xmath100 . to illustrate an application of this many - to - one correspondence , suppose that the dimension @xmath0 is an odd prime number , say @xmath420 ( the generalization to the case @xmath421 , with @xmath327 positive integer , is straightforward ) . in this case it is easy to give a full set of @xmath422 mubs @xcite ; fix an orthonormal basis @xmath423 and define @xmath196 orthonormal bases @xmath424 , each one labeled by @xmath425 , by @xmath426 the fact that these are mubs follows from the gauss summation formula @xmath427 where @xmath428 is the legendre symbol ( see e.g. @xcite ) . it is immediate to see that the orthonormal basis @xmath424 is fourier connected to the orthonormal basis @xmath429 given by @xmath430 i.e. , @xmath431 . moreover , for @xmath432 , with @xmath433 , define the rescaled orthonormal bases @xmath434 and @xmath435 , given by @xmath436 ( here @xmath437 means ` @xmath438 to the inverse of @xmath17 in the field @xmath439 ' and should not be confused with @xmath440 ) . then , an easy computation using the gauss formula yelds @xmath441 which shows that also @xmath434 and @xmath435 are fourier connected . more generally , one can start from a complementary pair of observables , which means that @xmath417 and @xmath442 are mutually unbiased but not necessarily fourier connected . obviously , we can still ask similar questions on joint measurements . especially , it would be interesting to know whether proposition [ prop : symmetric ] is still valid under this more general setting . in other words , the question is whether all complementary pairs are essentially similar with respect to joint measurability even if we leave this question open in the general case , we can see that our approach generalizes to a larger domain than we have explicitly used it for . indeed , all our results are still valid ( and with only very slight modifications in some of the proofs ) if we consider fourier transform with respect to a generic abelian group @xmath443 with order @xmath0 , i.e. , @xmath444 for @xmath445 and @xmath446 , with @xmath447 prime and @xmath448 integer for all @xmath449 . in this case , @xmath450 with @xmath451 , a basis @xmath452 is chosen in each factor hilbert space @xmath453 , and the @xmath443-fourier transform of @xmath5 is just the tensor product @xmath454 , where each @xmath455 is the @xmath456-fourier transform in @xmath453 with respect to the basis @xmath452 , as defined in . the mutually unbiased bases @xmath121 and @xmath96 are replaced by the bases @xmath457 and @xmath458 of @xmath5 , given by @xmath459 their associated complementary observables @xmath16 and @xmath22 are now both defined on @xmath443 , and given by @xmath460 and @xmath461 . they still satisfy the analogues of the covariance and invariance conditions , if the representations @xmath33 and @xmath108 are replaced by suitable tensor products . to demonstrate that we can now handle larger class of complementary observables , let @xmath462 , choose an orthonormal basis @xmath463 of @xmath464 and set @xmath465 then the observables @xmath415 and @xmath416 are complementary . they are equally defined by any two orthonormal bases @xmath466 and @xmath467 , where @xmath468 are complex numbers with unit modulus . if some pair of these orthonormal bases were connected by the @xmath469-fourier transform , then the matrix of their scalar products @xmath470 $ ] should be equal to the @xmath469-fourier matrix @xmath471 or to a matrix obtained from the above by some permutations of its rows and columns . it is straightforward to verify that the deriving set of equations for @xmath101 and @xmath472 has no solution . however , the matrix of scalar products @xmath473 $ ] is just the fourier matrix of @xmath474 , i.e. , @xmath475 in other words , the two orthonormal bases are connected by the fourier transform of @xmath476 . [ th : infocom ] let @xmath164 be a covariant phase space observable . then @xmath164 is informationally complete if and only if @xmath477 } \neq 0 \quad \forall x , y\in{\mathbb z}_d \ , .\ ] ] our proof of the above theorem relies on the following well known reconstruction formula for the weyl - heisenberg group , which is just a special case of orthogonality relations for irreducible representations of compact groups . [ prop : recons ] the following reconstruction formula holds for every @xmath478 : @xmath479 } u_x v_y = a \ , .\ ] ] for all @xmath480 , we have @xmath481 . thus , for all @xmath482 , we obtain @xmath483 } u_x v_y \right ] { \varphi}_h\,\right\rangle } = \\ & & \qquad \qquad \qquad \qquad = \sum_{x , y\in{\mathbb z}_d } { \left\langle\,u_x v_y { \varphi}_n\,|\,{\varphi}_m\,\right\rangle } { \left\langle\,{\varphi}_k\,|\,u_x v_y { \varphi}_h\,\right\rangle } \\ & & \qquad \qquad \qquad \qquad = \sum_{x , y\in{\mathbb z}_d } \omega^{(h - n)y } \ , \delta_{h+x , k } \ , \delta_{n+x , m } \\ & & \qquad \qquad \qquad \qquad = \sum_{y\in{\mathbb z}_d } \omega^{(h - n)y } \ , \delta_{h - n , k - m } \\ & & \qquad \qquad \qquad \qquad = d \ , \delta_{h , n } \ , \delta_{h - n , k - m } \\ & & \qquad \qquad \qquad \qquad = d \ , \delta_{h , n } \ , \delta_{k , m } \\ & & \qquad \qquad \qquad \qquad = d { \left\langle\,{\varphi}_k\,|\,{\varphi}_m\,\right\rangle } { \left\langle\,{\varphi}_n\,|\,{\varphi}_h\,\right\rangle } \ , , \end{aligned}\ ] ] which proves for @xmath484 . since every @xmath478 is a linear combination of this type of operators , the claim follows . let @xmath485 be the linear space of complex functions on @xmath486 . we recall that , by proposition 5.1 in @xcite , @xmath164 is informationally complete if and only if the linear map @xmath487 ( x , y ) = { { \rm tr}\left[{\mathsf{c}}_t ( x , y ) a\right]}\ ] ] is injective . since the dimensions of @xmath7 and @xmath488 are both @xmath1 , we conclude that @xmath164 is informationally complete if and only if @xmath489 is an isomorphism . we define the following three linear maps @xmath490 } \ , { \varphi}_x\otimes{\varphi}_y \\ m_t & : { \mathcal{h}}\otimes{\mathcal{h}}{\to}{\mathcal{h}}\otimes{\mathcal{h}}\ , , \qquad m_t ( { \varphi}_x\otimes{\varphi}_y ) = { { \rm tr}\left[tu_x v_y\right ] } \ , { \varphi}_x\otimes{\varphi}_y \\ r & : { \mathcal{h}}\otimes{\mathcal{h}}{\to}\ell ( { \mathbb z}^2_d ) \ , , \qquad r \phi ( x , y ) = { \left\langle\,{\varphi}_y\otimes{\varphi}_x\,|\,\phi\,\right\rangle } \ , .\end{aligned}\ ] ] the map @xmath491 is clearly a linear isomorphism , @xmath492 is a linear isomorphism by proposition [ prop : recons ] , and @xmath493 is a linear isomorphism if and only if holds . we now evaluate the composition map @xmath494 . for all @xmath478 , we obtain @xmath495(h , k ) = \\ & & \qquad = \frac{1}{d } \sum_{x , y\in{\mathbb z}_d } { { \rm tr}\left[tu_x v_y\right ] } { { \rm tr}\left[av^*_y u^*_x\right ] } { \left\langle\,{\varphi}_k\otimes{\varphi}_h\,|\,({\mathcal{f}}\otimes{\mathcal{f}}^*)({\varphi}_x\otimes{\varphi}_y)\,\right\rangle } \\ & & \qquad = \frac{1}{d^2 } \sum_{x , y\in{\mathbb z}_d } { { \rm tr}\left[tu_x v_y\right ] } { { \rm tr}\left[av^*_y u^*_x\right ] } \omega^{yh - xk } \\ & & \qquad = \frac{1}{d^2 } \sum_{x',y'\in{\mathbb z}_d } \omega^{y'h - x'k } { { \rm tr}\left[tu^*_h u_{x ' } v^*_k v_{y'}\right ] } { { \rm tr}\left[av_k v^*_{y ' } u_h u^*_{x'}\right ] } \\ & & \qquad = \frac{1}{d^2 } \sum_{x',y'\in{\mathbb z}_d } { { \rm tr}\left[tu^*_h v^*_k u_{x ' } v_{y'}\right ] } { { \rm tr}\left[av_k u_h v^*_{y ' } u^*_{x'}\right ] } \\ & & \qquad = \frac{1}{d } { { \rm tr}\left[tu^*_h v^*_k av_k u_h\right ] } \\ & & \qquad = [ v_t ( a ) ] ( h , k)\end{aligned}\ ] ] ( in the third equality we set @xmath496 , @xmath497 , in the fourth we used the commutation relation for @xmath33 and @xmath108 , and in the fifth we applied the reconstruction formula ) . as the map @xmath494 is an isomorphism if and only if holds , the same is true for the map @xmath489 , and the theorem follows . t.h . is grateful to cosmo lupo and mario ziman for illuminating discussions on qubit tomography . t.h . acknowledges financial support from the academy of finland ( grant no . 138135 ) and the magnus ehrnrooth foundation .
we study a contracting universe composed of cold dark matter and radiation , and with a positive cosmological constant . as is well known from standard cosmological perturbation theory , under the assumption of initial quantum vacuum fluctuations the fourier modes of the comoving curvature perturbation that exit the ( sound ) hubble radius in such a contracting universe at a time of matter - domination will be nearly scale - invariant . furthermore , the modes that exit the ( sound ) hubble radius when the effective equation of state is slightly negative due to the cosmological constant will have a slight red tilt , in agreement with observations . we assume that loop quantum cosmology captures the correct high - curvature dynamics of the space - time , and this ensures that the big - bang singularity is resolved and is replaced by a bounce . we calculate the evolution of the perturbations through the bounce and find that they remain nearly scale - invariant . we also show that the amplitude of the scalar perturbations in this cosmology depends on a combination of the sound speed of cold dark matter , the hubble rate in the contracting branch at the time of equality of the energy densities of cold dark matter and radiation , and the curvature scale that the loop quantum cosmology bounce occurs at . importantly , as this scenario predicts a positive running of the scalar index , observations can potentially differentiate between it and inflationary models . finally , for a small sound speed of cold dark matter , this scenario predicts a small tensor - to - scalar ratio .
1412.2914
observations of the cosmic microwave background ( cmb ) most recently @xcite have clearly established that scalar perturbations in the early universe were nearly scale - invariant . it is thus necessary for any realistic cosmological model to generate , in some fashion , scale - invariant perturbations . to achieve this , many cosmological models rely on the presence of matter fields ( typically scalar fields ) that have not yet been observed in nature . the new matter fields are necessary in these models as they play an essential role in the generation of scale - invariant perturbations . while it is of course a requirement for any cosmological scenario to predict near scale - invariance in order to be potentially viable , there are some cosmological scenarios where it is possible to avoid the weakness of postulating the existence of unknown matter fields and nonetheless obtain scale - invariance . we shall study one such model in this paper . this cosmological model consists of a spatially flat friedmann - lematre - robertson - walker ( flrw ) universe , with a positive cosmological constant , cold dark matter ( cdm ) , and radiation . these are three ingredients known to be present in our universe , and we will not assume the existence of any other matter fields . we also assume that the initial conditions are such that the space - time curvature is small and the universe is large and contracting . as the universe contracts , the space - time curvature will increase , and quantum gravity effects are expected to become important at some point , likely when the space - time curvature nears the planck scale . in this work , we will assume that loop quantum cosmology ( lqc ) captures the salient non - perturbative quantum gravity effects in the very early universe . lqc , a mini - superspace approach to quantum cosmology motivated by loop quantum gravity , predicts that a bounce occurs near the planck scale and that , once these quantum gravity effects are included , the space - time is free of the singularities that appear in classical general relativity @xcite . thus , this model will be that of a bouncing universe , with a matter content of radiation and cold dark matter and a positive cosmological constant . now , it is well known in cosmological perturbation theory that , for perturbations that are initially in the quantum vacuum state , the fourier modes that reach the long wavelength limit in a contracting space - time whose dynamics are dominated by a pressureless matter field become scale - invariant @xcite . ( note that the long - wavelength limit of a fourier mode does not always coincide with the mode exiting the hubble radius here the relevant length scale is the sound hubble radius , as explained in more detail later . ) furthermore , if the pressure is slightly negative , for example due to the presence of a positive cosmological constant , then the long wavelength perturbation modes will be almost scale - invariant with a slight red tilt . therefore , in the model considered here , we expect the modes that become large during the epoch of the universe that is dominated by cold dark matter to be almost scale - invariant , and those that become large when the effective equation of state is slightly negative to have a small red tilt . various realizations of the matter bounce scenario suggested in @xcite have been considered in the literature , but in most studies the matter content is taken to be scalar fields with specific potentials that can mimic pressureless matter fields for some specific initial conditions . to the best of our knowledge , this is the first study of a matter bounce scenario where the pressureless matter field is taken to be cold dark matter and that the effect of a positive cosmological constant is also included . we will compare the predictions of this scenario with other realizations of the matter bounce scenario in sec . [ s.disc ] . in this paper we calculate the spectrum of the cosmological perturbations for this @xmath0cdm bounce scenario . by the use of some approximations , it is possible to complete the calculations entirely analytically ; we also solve the equations numerically in order to provide a check on the validity of the approximations . in sec . [ s.hom ] we study the dynamics of the background , first analytically and then numerically . then in sec . [ s.scal ] we calculate the evolution of the scalar perturbations , from their initial quantum vacuum state to their final form after the bounce in the background flrw space - time . once again , this calculation is first done analytically with the help of some approximations , and is solved numerically afterwards . we continue in sec . [ s.tens ] by determining the spectrum of the primordial gravitational waves , and end in sec . [ s.disc ] with a discussion . we use units where @xmath1 , but keep @xmath2 and @xmath3 explicit except where stated otherwise , typically in the sections devoted to the numerical studies . as explained in the introduction , we are interested in studying the dynamics of a flat flrw cosmology with a positive cosmological constant @xmath4 and whose matter content is composed of radiation and cold dark matter , which are modeled as perfect fluids . classically , the dynamics are given by the friedmann equations , while in lqc there exists a hamiltonian constraint operator that generates the evolution of the wave function representing the quantum cosmology state . while the full quantum evolution is in general rather complicated , for sharply peaked states ( i.e. , states that admit a clear semi - classical interpretation at low curvature scales ) the full quantum dynamics are very well approximated by a set of effective equations @xcite . the key point is that since the state is sharply peaked ( and it remains sharply peaked throughout the entire evolution , including at the bounce point ) , it is meaningful to speak of an effective geometry , with an effective scale factor , and to ask what equations of motion govern the dynamics of this effective scale factor ; these equations are called the effective equations . as radiation will dominate the dynamics in the high - curvature regime , it is enough to consider the effective equations for a radiation - dominated flat flrw space - time , which are given by [ lqc - fr ] h^2 = ( 1 - ) , where @xmath5 is the hubble rate ( in proper time ) , @xmath6 is the energy density of the radiation matter field , and @xmath7 is the critical energy density , which is of the order of the planck energy density . in addition , the matter field satisfied the continuity equation [ rad - cont ] + 4 h = 0 , where we have used the fact that the pressure of a radiation perfect fluid is @xmath8 . note that the classical friedmann equations are obtained in the limit @xmath9 . in this paper , we will restrict our analysis to the effective equations of lqc , but a full quantum treatment of a radiation - dominated space - time in lqc is given in @xcite . in the first part of this section , we will use some reasonable approximations in order to derive some analytical results , and in the second section we present some numerical results that in part complement the analytic results and in part provide a check on the validity of the approximations . to be specific , we shall make three approximations in the analytical treatment of the background : first , we shall assume that quantum gravity effects are negligible already a few orders of magnitude away from the lqc bounce . this has been verified in numerical simulations @xcite , and can also be seen from studying the effective equations . this will allow us to solve the classical friedmann equations away from the bounce , and the lqc corrections will only become relevant when the space - time curvature nears the planck scale during the radiation - dominated epoch . second , we will assume that the evolution of the background universe can be broken into two distinct eras : the first one which is dominated by the combination of the cosmological constant and cold dark matter , and another era which is dominated by radiation . we assume a discontinuous change in the equation of state between these two eras , and impose continuity in the scale factor and in the ( conformal ) hubble rate during this transition . this approximation is supported by results in sec . [ ss.hom-n ] that show that the transition between the matter- and radiation - dominated epochs occurs very rapidly . finally , recall that the goal of this paper is to calculate the power spectrum of the perturbations . as is well known , especially from calculations in inflation , the key ingredient that determines the scale - dependence of the perturbations is the equation of state of the background _ at the time that the mode reaches the long wavelength limit_. therefore , in order to simplify calculations , we will assume a constant equation of state during the time - frame that the perturbation modes of interest reach the long wavelength limit ( i.e. , when the dynamics of the space - time are dominated by the cdm , but the cosmological constant provides a small correction to the effective equation of state ) . this approximation is justified by the effective equation of state being nearly constant during the period of interest , as seen in sec . [ ss.hom-n ] . nonetheless , one should keep in mind that the effective equation of state due to the combination of @xmath4 and cdm is in fact changing in time , and in general will be slightly different for different modes . as we shall see later , this effect leads to a running of the scalar index @xmath10 . using the effective equations , we shall first determine the dynamics around the bounce point , and then solve for the scale factor at earlier pre - bounce times when the space - time curvature is much smaller . it is easy to see that implies that [ rho_r ] ( t ) = , where @xmath11 is a constant of integration , and this can be used to solve , giving a(t ) = ( ( t - t_o)^2 + ) ^1/4 , where @xmath12 is another constant of integration . it is of course possible to choose any values for @xmath12 and @xmath11 ; for convenience we shall set @xmath13 so that the bounce occurs at @xmath14 , and @xmath15 so that the value of the scale factor at the bounce point is 1 . then , [ a - lqc ] a(t ) = ( t^2 + 1 ) ^1/4 . well before and after the bounce @xmath16 , the space - time curvature is much smaller than the planck scale and the scale factor is very well - approximated by the classical solution [ a - rad - t ] a(t ) = a_o^1/4 , where we have defined [ a0 ] a_o = for later convenience , and the hubble rate is given by [ hubble - t ] h = . it is also easy in the classical regime to change to conformal time @xmath17 via the relation @xmath18 , which gives |t| = ^2 , which in turn shows that the scale factor in terms of conformal time is given by [ a - rad ] a ( ) = || , and the conformal hubble rate , again in the classical regime , is given by [ hubbles ] = = - ( ) ^1/4 ; the second equality holds for the contracting epoch of the cosmology where @xmath19 and @xmath20 . now we shall consider the earlier epoch where cold dark matter and the cosmological constant dominate the dynamics ( the @xmath4cdm era ) . in this regime , the classical friedmann equations in conformal time can be written as [ eq_frw1 ] ^2 = a^2 ( _ cdm + _ ) = a^2 _ tot , where @xmath21 , and the combined continuity equation ( also in conformal time ) for cold dark matter and the cosmological constant is given by [ eq_continuity ] _tot + 3 ( _ tot + p_tot ) = 0 . since @xmath22 and @xmath23 , it follows that the effective equation of state for cold dark matter and the cosmological constant combined with @xmath24 , but here we are interested in the situation where the small positive contribution to @xmath25 from the cold dark matter and the small negative contribution to @xmath25 from the cosmological constant combine to give a slightly negative @xmath25 . ] is @xmath26 , with @xmath27 . in order to solve these two equations exactly , we shall assume that @xmath28 is a constant , and furthermore , since we are interested in the regime where the dynamics are dominated by the cold dark matter , we also take @xmath29 . of course , the effective equation of state does not remain constant in this setting , but recall that we are interested in calculating the power spectrum of cosmological perturbations , and their scale - dependence depends most sensitively on the effective equation of state at the time when they reach the long wavelength limit . therefore , the calculations where the spectra of the scalar and tensor perturbations are determined are to be understood as being for the modes that reach the long wavelength limit when the effective equation of state is given by @xmath28 , and so the specific value of @xmath30 will vary from one mode to another . note that this variation will be monotonic with @xmath30 becoming closer and closer to zero for shorter and shorter wavelengths , or for larger and larger @xmath31 . the exact rate at which this occurs will depend on the relative contributions of cold dark matter and the cosmological constant to the total matter energy density . in the approximation that @xmath30 is constant , the total energy density behaves as [ rho_tot_cdm ] _ tot = , where @xmath32 is a constant of integration , and the scale factor is given by a ( ) = ^2/(1 - 3 ) , where @xmath33 is also a constant of integration . it follows that the conformal hubble rate is = . in order to determine the values of @xmath32 and @xmath33 , let us assume that the transition between the radiation - dominated epoch and the @xmath4cdm era occurs at the equality conformal time @xmath34 . then , imposing that the scale factor and the conformal hubble rate be continuous at the transition time , we find that @xmath35 and @xmath36 $ ] , where @xmath37 and @xmath38 respectively . then , the scale factor can be rewritten as [ a - cdm ] a ( ) = a_e ( ) ^2/(1 - 3 ) . thus , for early times @xmath39 , the scale factor is given by ; then for @xmath40 the relation holds so long as quantum - gravity effects are negligible . when quantum gravity effects become important , it is necessary to use to describe the dynamics of the scale factor . in this subsection we numerically study the background solution presented in the @xmath4cdm model within the context of lqc . to be explicit , we study the background universe by separating the evolution into two periods . in the first stage , we apply the realistic data from the planck results and numerically solve for the period that the contracting universe evolves from dark energy domination to the matter - dominated phase . in particular , we assume that the density parameters of all matter components are the same as what we observed today , which are given by @xmath41 , @xmath42 and a deduced value @xmath43 , respectively @xcite . during this phase , the numerical result of the background evolution is provided in fig . [ fig1 ] . one particularly important result is shown in the inset of fig . [ fig1a ] : the effective equation of state evolves very slowly as the era of matter - domination is approached . this shows that , for modes that reach the long wavelength limit in this time - frame , any effect due to an evolving effective equation of state will be very small , and it is justified to make the approximation that @xmath30 is constant in the analytic calculations . then , in the second stage , we consider a toy model of a universe filled with a dark matter component and a radiation component with their energy densities evolving as follows , @xmath44 respectively , where @xmath45 and @xmath46 are their initial values when @xmath47 . note that while it would be nice to simultaneously include both matter fields and a cosmological constant , this is significantly more expensive from a computational point of view . these numerical studies presented here already provide strong support for the approximations used in the analytic section , and we leave more detailed numerical studies for future work . in order to relate this stage of the dynamics with the previous one , the pressure is taken to have a very small ( but non - vanishing ) negative value . also , to make the comparison between the analytic and numerical results as simple as possible , we adopt the conformal time @xmath17 and set the value of the scale factor @xmath48 to be unity at the bouncing moment @xmath49 in the numerical calculation . the dynamics are calculated by numerically solving the second friedmann equation @xmath50 instead of the first one in , which is given by @xmath51 in the case of lqc . in addition , the continuity equation is applied so that the background equations of motion are self - complete after an initial value of the background energy density has been imposed . for the numerics , we work in units of the reduced planck mass @xmath52 ( with @xmath53 ) for all model parameters with dimensions . as an explicit example ( although not a realistic one ) , we choose the values of the energy densities , the critical density @xmath54 and an effective equation of state parameter for the cdm at the initial moment to be @xmath55 and our numerical results are shown in fig . [ fig2 ] . the evolution of the scale factor , which is depicted by a blue solid curve in the left panel , explicitly shows that a non - singular bouncing solution is obtained in our model due to the quantum gravity effects captured by lqc , in particular , the minimal value of the scale factor is non - zero . from the middle panel , one can read more details about the background evolution . for example , the absolute value of @xmath56 is increasing when @xmath17 is about less than @xmath57 . after that , @xmath56 becomes approximately a linear function of the conformal time and correspondingly , the universe enters the bouncing phase with @xmath56 evolving from the negative valued regime to the positive valued one . eventually , the value of @xmath56 decreases after @xmath58 and hence the universe naturally connects to a regular thermal expanding phase after the bounce ; there is no need for reheating . the right panel of fig . [ fig2 ] characterizes the evolutions of the density parameters of the dark matter and radiation in our model , which are defined by @xmath59 with the subscripts @xmath60 " and @xmath61 " representing dark matter and radiation respectively . one can see that the universe was originally dominated by the cold dark matter with @xmath62 , as described the green dash - dotted curve . during the cosmic contraction the contribution of radiation , which is depicted by the orange dashed line , grows faster than that of dark matter and then dominates over the background evolution before the bounce . after the bounce , the universe would have experienced a period of radiation - dominated expanding phase and eventually enters the cdm era and hence is in qualitative agreement with cosmological observations . note that the sharp transition between these two eras provides justification for the assumption of a discontinuous transition between matter- and radiation - domination used in the analytic calculations . also , it is important to keep in mind that a more careful choice of the initial parameters would make the model more precisely consistent with experimental data . in this section , we will calculate the final spectrum of scalar perturbations after the bounce , assuming they begin in the quantum vacuum state in the distant past of the pre - bounce epoch . in cosmological perturbation theory , it is convenient to use the gauge - invariant mukhanov - sasaki variable @xcite v = z , where @xmath63 is the comoving curvature perturbation and z = . linear perturbations can be handled in lqc by following the ` separate universe ' approach presented in @xcite , and from the resulting quantum theory it is possible to derive effective equations that can be used to calculate expectation values of sharply - peaked states . the lqc effective equations for the mukhanov - sasaki variable are @xcite [ lqc - ms ] v - c_s^2 ( 1 - ) ^2 v - v = 0 , and it is easy to see that the standard classical expression is recovered in the limit @xmath9 . the effective equation is expected to provide a good approximation to the full quantum dynamics for modes that always remain large compared to the planck length @xcite , which is the case for the observationally relevant modes in the matter bounce scenario . in this section , using we will determine how the fourier modes @xmath64 evolve , from their initial quantum vacuum state , to their form as they exit the sound hubble radius . note that it is equivalent to state that a given mode reaches the long wavelength limit when its ` sound wavelength ' exits the hubble radius . ] @xmath65 and finally how they propagate through the bounce . as we shall see , the fourier modes that exit the sound hubble radius during the period of matter domination in the contracting branch become scale - invariant and therefore these modes are of particular interest . this is why in this paper we will only consider this family of the fourier modes @xmath64 , and ignore the modes that exit the hubble radius either before ( during the epoch dominated by the cosmological constant ) or after ( during the radiation - dominated phase ) . in the first part , we present analytical calculations , and in the second , numerical simulations . as in the previous section , it is necessary to make certain approximations in order to make the analytical calculations tractable and the numerical studies in the second part serve in part to check that the approximations are valid . we will begin by solving the dynamics of the perturbations in the @xmath4cdm era and the modes of interest are those that reach the long wavelength limit during this era and then determine their evolution during the radiation - dominated epoch , including through the bounce . in order to solve the equations of motion for @xmath64 , it is necessary to make the following approximations : ( i ) we only consider the modes that reach the long wavelength limit during the @xmath4cdm era , where the background matter field is modelled as a perfect fluid with a small and negative constant equation of state as explained in sec . [ sss.cdm ] ; ( ii ) we assume continuity in @xmath64 and @xmath66 at the spatial slice where we approximate the equation of state changing discontinuously to the radiation - dominated epoch ; and ( iii ) we work in the long wavelength limit during the bounce period . recall from sec . [ ss.hom-n ] that the equation of state changes very slowly in the regime where the effective equation of state is slightly negative . this supports the first approximation , since the key ingredient in determining the long wavelength spectrum of scalar perturbations is the equation of state of the background at the time that the given mode exits the sound hubble radius . since the effective equation of state is changing slowly , we expect corrections to the approximation of a constant equation of state to be subleading . finally , the validity of approximations ( ii ) and ( iii ) is verified numerically in sec . [ ss.scal-n ] . for the contracting portion of the space - time where the scale factor is given by ( and safely neglecting quantum gravity effects at this stage ) , is v_k + c_s^2 k^2 v_k - v_k = 0 . since the sound speed of cold dark matter is unknown , we set @xmath67 which we assume to be constant . we expect @xmath68 to be a small positive number . the solutions to this differential equation are @xmath69 \nn \\ & + a_2 h_n^{(2)}[- \ep k ( \eta-\eta_o ) ] \big),\end{aligned}\ ] ] where @xmath70 and @xmath71 are the hankel functions , and n = + 6 + o(^2 ) , where after the last equality we drop terms of order @xmath72 and higher ( recall that @xmath29 ) . choosing the initial conditions to be quantum vacuum fluctuations sets @xmath73 and @xmath74 . then , as @xmath17 approaches @xmath34 , some modes satisfy @xmath75 . these modes are said to be in the long wavelength limit , and in this limit it is possible to use the small argument expansion of the hankel functions to show that @xmath76 \\ & = \sqrt \f{8 \hbar}{9 } ( \ep k)^{3/2 + 6\de } \mh^{-(2 + 6\de ) } \nn \\ & \qquad - i \sqrt \f{\hbar}{4 } ( \ep k)^{-3/2 - 6\de } \mh^{1 + 6\de},\end{aligned}\ ] ] where in the second equality the time dependence has been rewritten in terms of the conformal hubble rate , and the exponents are accurate to first order in @xmath30 , while the numerical prefactors are only accurate to zeroth order in @xmath30 . it is straightforward to determine higher order corrections in @xmath30 , but this will not be necessary here . during the radiation - dominated epoch , the scale factor is proportional to @xmath17 while the sound speed is given by @xmath77 , and so the mukhanov - sasaki variable satisfies the equation v_k + v_k = 0 , at least in the classical regime where quantum gravity effects are negligible . note that due to the drastic change in the speed of sound , some modes that were in the long wavelength regime may at first be in the short wavelength limit at the onset of the radiation - dominated epoch . therefore , the relevant solutions are [ v - rad1 ] v_k = b_1 + b_2 , and @xmath78 and @xmath79 can be determined from by demanding that @xmath64 and @xmath80 be continuous at @xmath34 . note that as the bounce is approached @xmath81 and therefore the second term with the prefactor @xmath79 will dominate , so we can drop @xmath78 . imposing continuity in @xmath64 and @xmath80 gives @xmath82 some modes will have rebecome short wavelength modes due to the drastic change in the sound speed . for these modes , the small argument expansion for the trigonemetric functions can not be used , and these terms will not be scale - invariant . thus , we expect that scale - invariance will only be obtained for the modes that satisfy @xmath83 . we will return to this point later . in the contracting phase , as the bounce is approached ( but before quantum gravity effects become important ) we have @xmath84 and in this limit the solution for the mukhanov - sasaki variable tends to [ v - rad - bef ] v_k = b_2 , with @xmath79 given in . during the bounce , all of the modes of cosmological interest remain in the long - wavelength limit , and therefore the equation of motion for @xmath64 is v_k - v_k = 0 , where @xmath85 [ recall that @xmath86 during radiation - domination and @xmath87 is defined in ] , and the solution is v_k = c_1 z + c_2 z _ . note that @xmath88 is not simply proportional to @xmath48 due to the quantum gravity effects that modify the friedmann equation at high curvatures as seen in . the integral can be evaluated by rewriting it in terms of the proper time via the relation @xmath89 , giving [ v-2f1 ] v_k = c_1 z + c_2 z t , where @xmath90 has been redefined in order to absorb some numerical factors . in the classical pre - bounce era ( @xmath91 ) , this expression must agree with and this uniquely determines [ c_i ] c_1 = b_2 , c_2 = - b_2 . note that during this calculation it is important to keep in mind that @xmath92 in the pre - bounce era . it is also easy to calculate the form of the scalar perturbations in the classical post - bounce era by taking the limit @xmath93 in , which gives in terms of the comoving curvature perturbation [ r - fin ] _ k = = 2 c_1 + ( t^-1 ) , where we have only kept the dominant contribution , namely the constant mode which is the only one that does not decay with time . the amplitude of the comoving curvature perturbation of @xmath94 after the bounce depends on @xmath79 via and so it is easy to check whether the resulting scalar perturbations are scale - invariant or not . since the only dependence of @xmath31 in @xmath95 resides in @xmath79 , a quick examination of suffices to determine the scale - dependence of @xmath63 . as one can readily verify , one obtains near scale - invariance only in the limit of @xmath96 ( otherwise the dominant contribution would be oscillations superimposed over a red spectrum ) , in which case b_2 = - ( k)^--6 _ e^1 + 6 ( 1 + ( k^2 _ e^2 ) ) . therefore , a necessary condition for scale - invariance is that [ k - lim ] | k _ e | 1 , that is to say that the modes that become ( nearly ) scale - invariant during the matter - dominated contracting era must remain outside the sound hubble radius during the entire contracting radiation - dominated epoch and the bounce in order to remain ( nearly ) scale - invariant . for these modes , the power spectrum is @xmath97 where @xmath98 and @xmath99 , and the tilt is given by [ tilt ] n_s = 1 - 12 . thus , the observation of the tilt to be @xmath100 @xcite sets @xmath101 , which means that when the wavelength @xmath102 exits the sound hubble radius @xmath103 the effective equation of state must have been @xmath104 . also , in this model we predict a small running of the scalar index . this is due to the following two effects : ( i ) the departure from scale - invariance in a small interval of @xmath31 depends on the background effective equation of state , and ( ii ) the background equation of state is dynamical . the smallest values of @xmath31 reach the long - wavelength limit first , at a time when @xmath0 contributes slightly more to the background dynamics than it does at later times when larger values of @xmath31 reach the long - wavelength limit . therefore , as @xmath31 increases , the background equation of state at the ` sound - hubble - crossing ' time also increases , > 0 , and since @xmath105 and @xmath106 , it follows that [ running ] > 0 . therefore another prediction of this realization of the matter bounce scenario is for the scalar index @xmath10 to increase with @xmath31 . although the presence of this effect is clear in this model , its amplitude is not known . in order to calculate the expected amplitude of @xmath107 , it would be necessary to know quite precisely the energy densities corresponding to cold dark matter and the cosmological constant during the contracting phase , which is not an easy task especially since we do not expect the universe to be symmetric around the bounce point , as we shall discuss next . nonetheless , the qualitative result is a clear prediction for this realization of the matter bounce scenario . finally , as stated above , the results and only hold for fourier modes that satisfy the condition . in order to better understand this condition , it is useful to rewrite it in terms of the physical wave number @xmath108 and of the hubble rate at equality @xmath109 via , and , [ k - phys ] 1 . a good choice to ensure that @xmath110 corresponds to modes that are observed in the cosmic microwave background today , is to choose , here we simply choose a reasonable value of @xmath31 in order to better understand the consequences of imposing . ] @xmath111 , which lies roughly in the middle of the logarithmic range of the scales probed by the planck telescope @xcite . the value of @xmath112 at the time of equality is given by the relation @xmath113 . with our choice of conventions of setting @xmath114 at the bounce , it follows that @xmath115 . this can be calculated from the fact that ( i ) the scale factor increased by a factor of @xmath116 after matter - radiation equality until today , and ( ii ) that in the expanding branch matter - radiation equality is known to occur at @xmath117 giving a scale factor of @xmath118 ( assuming @xmath7 ) at the time of matter - radiation equality in the expanding branch . ( recall that matter - radiation equality occurs before recombination , and the superscript ` + ' on @xmath119 denotes the matter - radiation equality in the _ expanding _ post - bounce branch . ) if we assume a symmetric bounce , then it follows that the times of matter - radiation equality before and after the bounce are symmetric around @xmath14 , in which case @xmath120 . from this and the relation , it is easy to check that @xmath121 is of the order of unity rather than much smaller than 1 . this shows that a symmetric bounce is not viable in this model . to make this conclusion explicit , we rewrite as 1 , which in turn , since the second term is of order unity and via and , gives [ asymmetry ] 1 . this relation shows that the bounce must be significantly asymmetric . indeed , in order for the condition to hold , @xmath122 must be much larger ( by at least a few orders of magnitude ) than it would be in a model with a symmetric bounce , in which case it is necessary for the matter - radiation equality to occur at much higher curvature scales in the contracting branch than it does in the expanding branch . one possible way for this to happen would be if a very large number of additional quanta of radiation are created during the bounce ; as an aside note that such a process would generate a significant amount of entropy . interestingly , it has recently been suggested that particle production may play an important role during the bounce in lqc and this effect would cause the bounce to be significantly asymmetric in precisely the manner outlined here @xcite . that being said , it is not yet clear whether particle production could generate the amount of asymmetry that is required for this realization of the matter bounce to be viable . we leave this question for future work . there are also other reasons why an asymmetric bounce is necessary in this scenario . assuming the hubble parameter at @xmath123 to be of the order of @xmath124 as would be the case for a symmetric bounce then for the amplitude of the scalar perturbations to match the observed value of @xmath125 , it would be necessary to have a very small value of the sound speed of cold dark matter of @xmath126 ( assuming @xmath54 to be of approximately the same order of magnitude as @xmath127 ) ; such a small value of the sound speed parameter typically leads to primordial perturbations with over - large non - gaussianities . in order to have a larger ( although still small ) value of @xmath68 and thus sufficiently small non - gaussianities , it is again necessary to have an asymmetric bounce . this likely requires even more asymmetry than is needed to satisfy the condition . finally , many of the modes observed today reentered the sound hubble radius during radiation - domination . however , in order for them to be scale - invariant , in the contracting branch of the universe , they must have exited the sound hubble radius during matter - domination . this is yet another reason that an asymmetric bounce is necessary in the @xmath4cdm bounce scenario . to complete the analysis of cosmological perturbations , in this subsection we perform a numerical computation of the evolution of the primordial curvature perturbations and gravitational waves in the model under consideration . to be consistent with the background numerics , we consider the universe filled with a dust matter field with its energy density evolving as , of which the initial value is the same as the one given in . furthermore , we impose the initial conditions of the cosmological perturbations to be vacuum fluctuations during the matter dominated contracting phase ( in units where @xmath53 ) , [ para_pert ] r_k^ini , h_k^ini . in addition , we take @xmath128 and fix @xmath129 as an example in the detailed calculation . our numerical results are presented in fig . [ fig3 ] . in the left panel of fig . [ fig3 ] , one can see how primordial cosmological perturbations evolve from the sub - hubble scale to the super - hubble region in the contracting phase . for a fixed comoving wave number @xmath31 , the curvature perturbation exits the hubble radius much earlier than the gravitational wave since its oscillation gets squeezed when @xmath130 at the sound horizon which is much smaller than the hubble radius . consequently , one expects that there ought to be more oscillations in the power spectrum of primordial gravitational waves than that of primordial curvature perturbations . this expectation is exactly verified in the middle panel of the figure , which displays the evolutions of primordial power spectra of both scalar and tensor perturbations . in the middle part of fig . [ fig3 ] , one can see that @xmath131 experiences only one oscillation and then becomes squeezed very soon with its amplitude increasing until the nonsingular bounce takes place . however , @xmath132 experiences several oscillations during the contracting phase and only becomes squeezed when the universe is very near the bouncing phase . they both become conserved at super - hubble scales after the bounce , which can be read from the right regime of fig . [ fig3b ] . interestingly , one can observe that the magnitudes of the two spectra are comparable ( though the amplitude of the scalar mode is slightly larger ) during the contracting phase but the amplitude of the scalar spectrum becomes significantly larger than that of tensor spectrum after the bounce , and hence the tensor - to - scalar ratio is suppressed to a small value that is consistent with observations . this can also be seen in the right panel of fig . [ fig3 ] which depicts the dynamics of the tensor - to - scalar ratio @xmath133 throughout the cosmological bouncing evolution . in figs . [ fig3b ] and [ fig3c ] , we see that while the scalar perturbation mode passes through the bouncing phase in a relatively smooth fashion , the magnitude of the gravitational wave mode is significantly damped . while the amplitude of the gravitational wave increases somewhat after the bounce , it remains significantly lower than before the bouncing phase . in the specific example with @xmath134 being considered , we read approximately @xmath135 from the numerical computation . as we will analyze in the next section on the tensor - to - scalar ratio , this is roughly in the same order of the analytical estimate ( @xmath136 ) given by , though the detailed values are not in exact agreement with each other . we will comment further on this slight numerical discrepancy at the end of this section . another important point is that in the numerical computation we have chosen the critical density @xmath54 to be very small ( @xmath137 ) so that the bounce scale is of order @xmath138 planck mass ( this choice for @xmath54 significantly lowers the computational cost of the numerics as this causes the bounce to occur at a lower curvature scale ) . as a consequence , the amplitude of the power spectrum of primordial curvature perturbations is about @xmath139 which is approximately three orders lower than the magnitude of the observed cmb spectrum . the numerical computation performed in this subsection is sufficient to demonstrate the formation of primordial power spectra in the model under consideration and to verify their relations obtained in semi - analytic analyses . however , one can improve the agreement of the amplitude of the spectrum with observations by fine - tuning the values of the parameters in this model . it is also important to keep in mind that it is possible to obtain a similar amplitude of scalar perturbations even if @xmath7 , although such a choice would require a smaller value of @xmath122 and/or a larger value of @xmath68 . finally , note that since there are two matter components used in this step of the numerical computation , the sound speed parameter depends on the background evolution approximately as c_s = & for _ m 1 + & otherwise , where @xmath68 is the small constant sound speed for cdm and @xmath140 is the time - dependent equation of state @xmath141 , which includes the contributions coming from the radiation field and therefore equals @xmath77 near the bounce when radiation dominates the dynamics of the universe . this time - dependence effect has been taken into account in the above numerical computation . finally , these numerical solutions also validate the approximations made in the analytical section . first , we see that in the near - bounce region it is justified to assume that the modes of interest are in the long - wavelength limit . furthermore , we have also numerically checked the difference between the solution obtained in the previous section under the approximation of a discontinuous change of the equation of state , and the result plotted in fig . [ fig3b ] where no such approximation is made . we do not include the graph here , as the two curves lie practically one on top of the other , showing the validity of this approximation . a careful comparison of the two curves shows that the theoretical calculation given in sec . [ ss.hom-a ] very slightly overestimates the amplitude of @xmath63 . this last point raises an important issue : the results presented in this paper rely either on analytic calculations based on several approximations that are well - motivated but certainly introduce some small errors , or on numerical simulations that , while accurate enough for our purposes here , are not high - precision numerical studies . due to the use of these approximations and numerical simulations , the predictions presented in this paper necessarily contain some small errors . that being said , these small errors are not expected to affect the reliability of the estimates obtained for the predicted observables in the @xmath4cdm bounce scenario we consider here , which we estimate ( by comparisons between the analytical and numerical results ) to be accurate up to an overall factor of approximately 2 . it is possible to also calculate the spectrum of tensor perturbations after the bounce , again assuming that the initial state was the quantum vacuum . the lqc effective mukhanov - sasaki equation for tensor perturbations is @xcite - ( 1 - ) ^2 - = 0 , where @xmath142 , with @xmath143 being the usual tensor perturbation modes , and z_t = . while the ` time - dependent potentials ' @xmath144 for scalar perturbations and @xmath145 for tensor perturbations are not the same , the most important difference for our purposes is the fact that the sound speed for tensor perturbations is always 1 , while the sound speed for scalar perturbations is significantly smaller than 1 during the matter - dominated phase . this effect strongly suppresses the tensor - to - scalar ratio , as we show below . with the tensor power spectrum defined as _ h^2 = 64 g |h|^2 , and the tensor - to - scalar ratio @xmath133 r = , it is possible to calculate the spectrum of the tensor perturbations in a fashion analogous to sec . [ s.scal ] . while the calculation is a little long , it is not particularly illuminating as it follows exactly the same steps as the one for scalar perturbations . while there are a few numerical factors that are different , the procedure is identical and therefore here we will simply state the results . the power spectrum of the tensor perturbations is found to be almost scale - invariant , although the tilt depends on the value of the effective equation of state at the time that the mode exits the hubble radius . recall that the scalar modes reach the long wavelength limit well before the tensor modes since , during the cold - dark - matter - dominated era , the sound speed for scalar modes is @xmath146 while @xmath147 for the tensor modes . therefore , while the effective equation of state when the scalar perturbations reach the long wavelength limit is @xmath148 , the effective equation of state when the tensor modes exit the hubble radius will have changed and will be larger ( as the cosmological constant contributes less to the effective equation of state for a smaller scale factor ) , though still close to zero . we denote this value of the effective equation of state by @xmath149 , and we can bound @xmath150 above by @xmath151 . then the departure from scale - invariance is given by n_t = - 12 _ t. note that @xmath150 , while expected to be small , may be negative and this would give a slight blue tilt to the spectrum of the tensor perturbations . therefore , while near - scale - invariant tensor perturbations are predicted by this model , the exact departure from scale - invariance for the tensor modes will depend on how the effective equation of state varies in time , and this will determine whether there is a small blue tilt or a small red tilt . finally , for the particular scenario studied here , the amplitude of primordial gravitational waves is predicted to be very strongly suppressed , with a tensor - to - scalar ratio of [ r_ratio ] r = 24 ^3 , where @xmath68 refers to the sound speed of cold dark matter and @xmath133 is therefore predicted to be very small . note that the tensor - to - scalar ratio is suppressed both by a contribution due to the sound speed of cold dark matter , and also by a further factor of @xmath152 during the bounce due to quantum gravity effects . there do not appear to be many estimates of the sound speed of cold dark matter in the literature ; one interesting reference is @xcite which provides a bound of approximately @xmath153 ( note however that in that paper the authors study a considerably different model from this one ) . it is likely that in the future a better estimate for @xmath68 can be found , but nonetheless the constraint @xmath153 already implies an upper bound of @xmath154 on the tensor - to - scalar ratio , a result which is in agreement with the latest observations @xcite . in this paper we have seen how in a contracting universe cosmological perturbations , assumed to be initially in their quantum vacuum state , become scale - invariant if their sound wavelength becomes larger than the hubble radius when the dynamics of the universe is dominated by cold dark matter . a small red tilt is generated when the effective equation of state is negative due to the presence of a positive cosmological constant , and a small tensor - to - scalar ratio is predicted . the scale - invariant perturbations can provide appropriate initial conditions for an expanding universe in order to seed structure formation if there is a bounce to connect the contracting branch of the universe to our current expanding branch . in the realization of the matter bounce scenario studied here , the bounce occurs due to non - perturbative quantum gravity effects , as captured by lqc , that resolve the classical singularity and provide a quantum bridge between the pre - big - bang and post - big - bang epochs . further , the only matter fields present in this model are assumed to be cold dark matter and radiation , together with a positive cosmological constant . the main predictions depend on four parameters : the scalar index @xmath10 is determined by the effective equation of state at the time when the fourier modes @xmath64 exit the sound hubble radius , while the amplitude of the scalar perturbations depend on a combination of the sound speed of cold dark matter @xmath68 , the amount of asymmetry in the universe , which can be parametrized by @xmath109 the hubble rate at matter - radiation equality in the contracting branch , and the matter energy density at the bounce @xmath54 . in particular , a symmetric bounce is ruled out and , in order to match observations , it is necessary for matter - radiation equality to occur at higher space - time curvature scales in the contracting branch than in the expanding branch . this type of asymmetry can be caused by particle production during the bounce , a process which may be important in lqc @xcite . there exist several other realizations of the matter bounce scenario , some where the bounce is caused by matter fields that violate energy conditions @xcite , and others where it is quantum gravity effects that provide the bounce @xcite . there are also some matter bounce scenarios where the gravitational sector is modified not only in the high - curvature regime , but also in the pre - bounce era in order to obtain a matter - like contracting phase without requiring the presence of any matter fields @xcite . of course , there is no _ a priori _ reason to prefer one realization of the matter bounce scenario over another ( other than perhaps the simplicity or the elegance of a particular model ) . instead , it is necessary to determine how these realizations differ in their predictions which can then be compared to observations . for this reason , it is important to point out that the predictions of the different matter bounce realizations vary in some significant aspects , concerning both cmb experiments @xcite as well as dark matter searches @xcite . the recent review @xcite explains the different predictions of many of the various realizations of the matter bounce scenario and in order to complement that paper , here we shall briefly explain how the predictions of the model studied in this paper differ from the other realizations of the matter bounce scenario . first , the @xmath4cdm bounce model studied in this paper gives a slight red tilt in a natural fashion , something which is absent in many other realizations of the matter bounce ( see for example @xcite for specific realizations ) . and second , the predicted tensor - to - scalar ratio is very small . most realizations of the matter bounce typically predict relatively large tensor - to - scalar ratios , and it is often necessary to assume the presence of a large number of fields ( and thus of entropy perturbations as well ) in order to decrease the relative amplitude of the tensor perturbations . this is not necessary here since the tensor - to - scalar ratio is naturally predicted to be small due to the small sound speed of cold dark matter . note that this effect of a small @xmath155 strongly damping the value of @xmath133 has also previously been noticed in a study of the matter bounce scenario in a bohm - de broglie quantum cosmology model @xcite . another important point is that this scenario predicts a positive running of the scalar index , @xmath156 . this result shows that it is possible for observations to differentiate between the @xmath0cdm bounce scenario and inflationary models , where the running of the scalar index is predicted to be negative . the amplitude of this effect in the @xmath0cdm bounce scenario depends on the time evolution of effective equation of state during matter domination @xmath157 , which in turn depends quite sensitively on the parameters @xmath158 , @xmath159 and @xmath160 set as initial conditions for the background in the contracting branch . we leave a detailed study of the relation between these parameters and the amplitude of @xmath107 for future work . there do remain two other important open questions regarding this model that are also left for future work . the first one concerns the importance of particle production effects during the bounce . will there be enough particle production to cause sufficient asymmetry around the bounce point for this model to be viable ? the second , more difficult , open problem is to determine how the presence of anisotropies would modify the predictions of this model . the friedmann equation for the mean scale factor in bianchi models shows that anisotropies dominate the dynamics in the high curvature limit , and we should expect them to typically become important during ( and possibly for some time before and after ) the bounce in lqc . indeed , anisotropies can modify the predictions of a number of cosmological scenarios including inflation @xcite . despite their importance , anisotropies are often neglected as cosmological perturbation theory becomes considerably more complex in their presence @xcite ; in particular , for non - vanishing anisotropies the equations of motion for the scalar , vector and tensor perturbations no longer decouple . note that if there is an ekpyrotic phase in the contracting branch of the space - time , it is possible to avoid the growth of anisotropies generated during matter contraction @xcite and then there is no need to include them in the analysis . however , it is certainly possible ( and outside of ekpyrotic models it appears more natural ) for anisotropies to grow and dominate the dynamics near the bounce . in this case , while the anisotropies will be diluted soon after the bounce ( see , e.g. , sec . iiic of @xcite ) , they may change the spectrum of the cosmological perturbations as they evolve through the anisotropy - dominated bounce , and some of these modifications may ultimately be observable today . for this reason , it is important to allow for anisotropies by generalizing the results obtained here for the case where the background is an anisotropic bianchi i space - time ( rather than flat flrw ) , and determining precisely how anisotropies may affect the predictions of this model . finally , we conclude with some comments regarding the universality and robustness of the predictions of the @xmath4cdm bounce scenario with respect to the physics of the bounce . in this paper we assumed that the bounce is caused by the quantum gravity effects captured by loop quantum cosmology . as mentioned above , it is also possible to generate a bounce via other mechanisms , either with matter fields that violate energy conditions , or by modifications to the gravitational action as occur for example in @xmath161 theories . an important property of the @xmath0cdm bounce scenario is that most of the salient characteristics of the predicted spectra of scalar and tensor perturbations are due to the pre - bounce physics , where quantum gravity and other high energy effects are negligible and therefore the majority of the predictions including the small red tilt for the modes that exit the sound hubble radius when the effective equation of state is slightly negative , as well as the positive running of the scalar index are independent of the high - curvature dynamics of the background space - time at the bounce point . however , there is one effect that is partially due to lqc : the tensor - to - scalar ratio is suppressed by a factor of 1/4 during the bounce by lqc effects ( specifically , due to the modifications that arise in the lqc effective mukhanov - sasaki equation for tensor perturbations ) . thus , while many predictions of the @xmath0cdm bounce are quite robust and are mostly independent of the physics of the bounce , one exception is the predicted value of the tensor - to - scalar ratio which in fact is affected by lqc effects and , if the sound speed of cold dark matter is known , this effect could provide a potentially important observational test for lqc . we would like to thank robert brandenberger , jim cline , parampreet singh , jun - qing xia and wei xue for helpful discussions . the work of yfc is supported in part by an nserc discovery grant and by funds from the canada research chair program . this work is supported in part by a grant from the john templeton foundation . , g. hinshaw _ et al . _ , `` nine - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological parameter results , '' astrophys . j. suppl . * 208 * ( 2013 ) 19 , http://arxiv.org/abs/1212.5226[arxiv:1212.5226 ] . , p. ade _ et al . _ , `` planck 2013 results . cosmological parameters , '' astron . astrophys . ( 2014 ) http://arxiv.org/abs/1303.5076[arxiv:1303.5076 ] . m. bojowald , `` loop quantum cosmology , '' living rev . * 11 * ( 2008 ) 4 . a. ashtekar and p. singh , `` loop quantum cosmology : a status report , '' class . * 28 * ( 2011 ) 213001 , http://arxiv.org/abs/1108.0893[arxiv:1108.0893 ] . k. banerjee , g. calcagni , and m. martn - benito , `` introduction to loop quantum cosmology , '' sigma * 8 * ( 2012 ) 016 , http://arxiv.org/abs/1109.6801[arxiv:1109.6801 ] . d. wands , `` duality invariance of cosmological perturbation spectra , '' phys . * d60 * ( 1999 ) 023507 , http://arxiv.org/abs/gr-qc/9809062[arxiv:gr-qc/9809062 ] . f. finelli and r. brandenberger , `` on the generation of a scale invariant spectrum of adiabatic fluctuations in cosmological models with a contracting phase , '' phys . rev . * d65 * ( 2002 ) 103522 , http://arxiv.org/abs/hep-th/0112249[arxiv:hep-th/0112249 ] . v. taveras , `` corrections to the friedmann equations from lqg for a universe with a free scalar field , '' phys . * d78 * ( 2008 ) 064072 , http://arxiv.org/abs/0807.3325[arxiv:0807.3325 ] . c. rovelli and e. wilson - ewing , `` why are the effective equations of loop quantum cosmology so accurate ? , '' phys . * d90 * ( 2014 ) 023538 , http://arxiv.org/abs/1310.8654[arxiv:1310.8654 ] . singh and v. taveras , `` a note on the effective equations in lqc , '' unpublished . t. pawowski , r. pierini , and e. wilson - ewing , `` loop quantum cosmology of a radiation - dominated flat flrw universe , '' http://arxiv.org/abs/1404.4036[arxiv:1404.4036 ] . v. f. mukhanov , h. feldman , and r. h. brandenberger , `` theory of cosmological perturbations , '' phys . * 215 * ( 1992 ) 203333 . d. salopek and j. bond , `` nonlinear evolution of long wavelength metric fluctuations in inflationary models , '' phys . * d42 * ( 1990 ) 39363962 . d. wands , k. a. malik , d. h. lyth , and a. r. liddle , `` a new approach to the evolution of cosmological perturbations on large scales , '' phys . * d62 * ( 2000 ) 043527 , http://arxiv.org/abs/astro-ph/0003278[arxiv:astro-ph/0003278 ] . e. wilson - ewing , `` holonomy corrections in the effective equations for scalar mode perturbations in loop quantum cosmology , '' class . * 29 * ( 2012 ) 085005 , http://arxiv.org/abs/1108.6265[arxiv:1108.6265 ] . e. wilson - ewing , `` lattice loop quantum cosmology : scalar perturbations , '' class . * 29 * ( 2012 ) 215013 , http://arxiv.org/abs/1205.3370[arxiv:1205.3370 ] . a. t. mithani and a. vilenkin , `` instability of an emergent universe , '' jcap * 1405 * ( 2014 ) 006 , http://arxiv.org/abs/1403.0818[arxiv:1403.0818 ] . t. cailleteau , j. mielczarek , a. barrau , and j. grain , `` anomaly - free scalar perturbations with holonomy corrections in loop quantum cosmology , '' class . * 29 * ( 2012 ) 095010 , http://arxiv.org/abs/1111.3535[arxiv:1111.3535 ] . t. cailleteau , a. barrau , j. grain , and f. vidotto , `` consistency of holonomy - corrected scalar , vector and tensor perturbations in loop quantum cosmology , '' phys . * d86 * ( 2012 ) 087301 , http://arxiv.org/abs/1206.6736[arxiv:1206.6736 ] . a. balbi , m. bruni , and c. quercellini , `` lambda - alpha dm : observational constraints on unified dark matter with constant speed of sound , '' phys . * d76 * ( 2007 ) 103519 , http://arxiv.org/abs/astro-ph/0702423[arxiv:astro-ph/0702423 ] . y .- f . cai , t .- qiu , r. brandenberger , and x .- m . zhang , `` a nonsingular cosmology with a scale - invariant spectrum of cosmological perturbations from lee - wick theory , '' phys . * d80 * ( 2009 ) 023511 , http://arxiv.org/abs/0810.4677[arxiv:0810.4677 ] . c. lin , r. h. brandenberger , and l. perreault levasseur , `` a matter bounce by means of ghost condensation , '' jcap * 1104 * ( 2011 ) 019 , http://arxiv.org/abs/1007.2654[arxiv:1007.2654 ] . cai , d. a. easson , and r. brandenberger , `` towards a nonsingular bouncing cosmology , '' jcap * 1208 * ( 2012 ) 020 , http://arxiv.org/abs/1206.2382[arxiv:1206.2382 ] . cai , e. mcdonough , f. duplessis , and r. h. brandenberger , `` two field matter bounce cosmology , '' jcap * 1310 * ( 2013 ) 024 , http://arxiv.org/abs/1305.5259[arxiv:1305.5259 ] . cai , `` exploring bouncing cosmologies with cosmological surveys , '' sci . china phys . * 57 * ( 2014 ) 14141430 , http://arxiv.org/abs/1405.1369[arxiv:1405.1369 ] . t. biswas , a. mazumdar , and w. siegel , `` bouncing universes in string - inspired gravity , '' jcap * 0603 * ( 2006 ) 009 , http://arxiv.org/abs/hep-th/0508194[arxiv:hep-th/0508194 ] . r. brandenberger , `` matter bounce in horava - lifshitz cosmology , '' phys . * d80 * ( 2009 ) 043516 , http://arxiv.org/abs/0904.2835[arxiv:0904.2835 ] . e. wilson - ewing , `` the matter bounce scenario in loop quantum cosmology , '' jcap * 1303 * ( 2013 ) 026 , http://arxiv.org/abs/1211.6269[arxiv:1211.6269 ] . r. h. brandenberger , c. kounnas , h. partouche , s. p. patil , and n. toumbas , `` cosmological perturbations across an s - brane , '' jcap * 1403 * ( 2014 ) 015 , http://arxiv.org/abs/1312.2524[arxiv:1312.2524 ] . p. peter , e. j. pinho , and n. pinto - neto , `` a non inflationary model with scale invariant cosmological perturbations , '' phys . * d75 * ( 2007 ) 023516 , http://arxiv.org/abs/hep-th/0610205[arxiv:hep-th/0610205 ] . s. odintsov and v. oikonomou , `` matter bounce loop quantum cosmology from @xmath162 gravity , '' http://arxiv.org/abs/1410.8183[arxiv:1410.8183 ] . g. leon and a. a. roque , `` qualitative analysis of kantowski - sachs metric in a generic class of f(r ) models , '' jcap * 1405 * ( 2014 ) 032 , http://arxiv.org/abs/1308.5921[arxiv:1308.5921 ] . k. bamba , a. n. makarenko , a. n. myagky , and s. d. odintsov , `` bouncing cosmology in modified gauss - bonnet gravity , '' phys . * b732 * ( 2014 ) 349355 , http://arxiv.org/abs/1403.3242[arxiv:1403.3242 ] . f . cai , t .- qiu , j .- q . xia , and x. zhang , `` a model of inflationary cosmology without singularity , '' phys . * d79 * ( 2009 ) 021303 , http://arxiv.org/abs/0808.0819[arxiv:0808.0819 ] . f . cai and x. zhang , `` primordial perturbation with a modified dispersion relation , '' phys . * d80 * ( 2009 ) 043520 , http://arxiv.org/abs/0906.3341[arxiv:0906.3341 ] . j. liu , y .- f . cai , and h. li , `` evidences for bouncing evolution before inflation in cosmological surveys , '' j. theor . * 1 * ( 2012 ) 110 , http://arxiv.org/abs/1009.3372[arxiv:1009.3372 ] . j .- xia , y .- f . cai , h. li , and x. zhang , `` evidence for bouncing evolution before inflation after bicep2 , '' phys . lett . * 112 * ( 2014 ) 251301 , http://arxiv.org/abs/1403.7623[arxiv:1403.7623 ] . cai , j. quintin , e. n. saridakis , and e. wilson - ewing , `` nonsingular bouncing cosmologies in light of bicep2 , '' jcap * 1407 * ( 2014 ) 033 , http://arxiv.org/abs/1404.4364[arxiv:1404.4364 ] . c. li , r. h. brandenberger , and y .- k . e. cheung , `` big bounce genesis , '' http://arxiv.org/abs/1403.5625[arxiv:1403.5625 ] . e. cheung , j. u. kang , and c. li , `` dark matter in a bouncing universe , '' jcap * 1411 * ( 2014 ) 001 , http://arxiv.org/abs/1408.4387[arxiv:1408.4387 ] . e. cheung and j. vergados , `` direct dark matter searches - test of the big bounce cosmology , '' http://arxiv.org/abs/1410.5710[arxiv:1410.5710 ] . e. elizalde , j. haro , and s. d. odintsov , `` quasi - matter domination parameters in bouncing cosmologies , '' http://arxiv.org/abs/1411.3475[arxiv:1411.3475 ] . d. bessada , n. pinto - neto , b. b. siffert , and o. d. miranda , `` stochastic background of relic gravitons in a bouncing quantum cosmological model , '' jcap * 1211 * ( 2012 ) 054 , http://arxiv.org/abs/1207.5863[arxiv:1207.5863 ] . c. pitrou , t. s. pereira , and j uzan , `` predictions from an anisotropic inflationary era , '' jcap * 0804 * ( 2008 ) 004 , http://arxiv.org/abs/0801.3596[arxiv:0801.3596 ] . m. den , `` gauge invariant perturbations in a spatially flat anisotropic universe , '' prog . * 77 * ( 1987 ) 653 . h. noh and j. hwang , `` perturbations of an anisotropic space - time : formulation , '' phys . rev . * d52 * ( 1995 ) 19701987 . k. dunsby , `` gauge invariant perturbations of anisotropic cosmological models , '' phys . rev . * d48 * ( 1993 ) 35623576 . t. s. pereira , c. pitrou , and j .- uzan , `` theory of cosmological perturbations in an anisotropic universe , '' jcap * 0709 * ( 2007 ) 006 , http://arxiv.org/abs/0707.0736[arxiv:0707.0736 ] . cai , r. brandenberger , and p. peter , `` anisotropy in a nonsingular bounce , '' class . * 30 * ( 2013 ) 075019 , http://arxiv.org/abs/1301.4703[arxiv:1301.4703 ] . f . cai and e. wilson - ewing , `` non - singular bounce scenarios in loop quantum cosmology and the effective field description , '' jcap * 1403 * ( 2014 ) 026 , http://arxiv.org/abs/1402.3009[arxiv:1402.3009 ] . s. alexander , y .- f . cai , and a. marciano , `` fermi - bounce cosmology and the fermion curvaton mechanism , '' http://arxiv.org/abs/1406.1456[arxiv:1406.1456 ] . b. gupt and p. singh , `` quantum gravitational kasner transitions in bianchi - i spacetime , '' phys . * d86 * ( 2012 ) 024034 , http://arxiv.org/abs/1205.6763[arxiv:1205.6763 ] .
in this paper , we develop a new method for finding all perfect powers which can be expressed as the sum of two rational-units , where is a finite set of primes . our approach is based upon the modularity of galois representations and , for the most part , does not require lower bounds for linear forms in logarithms . its main virtue is that it enables to carry out such a program explicitly , at least for certain small sets of primes ; we do so for and .
1603.07922
if @xmath3 is a finite set of primes , we define the set of _ @xmath0-units _ to be those integers of the shape @xmath4 , with exponents @xmath5 nonnegative integers . the arithmetic of such sets has been frequently studied due to its connections to a wide variety of problems in number theory and arithmetic geometry . in the latter direction , equations of the shape @xmath6 where @xmath7 and @xmath8 are @xmath0-units for certain specific sets @xmath0 , arise naturally when one wishes to make effective a theorem of shafarevich on the finiteness of isomorphism classes of elliptic curves over a number field @xmath9 with good reduction outside a given finite set of primes . by way of a simple example , if we wish to find all elliptic curves @xmath10 with nontrivial rational @xmath11-torsion and good reduction outside @xmath12 , we are led to consider curves @xmath13 of the shape @xmath14 where @xmath15 and @xmath16 are rational integers satisfying @xmath17 for nonnegative integers @xmath5 . writing @xmath18 , we thus seek to solve equation ( [ two ] ) , with @xmath19 and @xmath20 . an algorithm for computing all solutions to equations of the shape ( [ two ] ) , over @xmath21 , can be found in chapter 7 of de weger @xcite , where , for instance , one can find a complete characterization of the solutions to equation ( [ two ] ) in case @xmath22 . this algorithm combines lower bounds for linear forms in complex and @xmath23-adic logarithms with lattice basis reduction for @xmath23-adic lattices . more generally , for a given set of primes @xmath0 , we may consider equations of the shape @xmath24 with @xmath25 an integer , @xmath7 and @xmath8 @xmath0-units , and @xmath26 a nonzero integer . we will call such a quadruple @xmath27 a _ primitive _ solution of ( [ enn ] ) if @xmath28 is @xmath29th - power free . such equations are the main topic of discussion in chapter 9 of shorey and tijdeman @xcite , due to their connections to the problem of characterizing perfect powers in nondegenerate binary recurrence sequences of algebraic numbers . if we write @xmath30 in equation ( [ enn ] ) , with @xmath31 @xmath29th - power free ( whereby there are at most @xmath32 choices for @xmath31 ) , then it follows from the theory of thue - mahler equations that , for a fixed value of @xmath33 , the set of primitive solutions @xmath27 of ( [ enn ] ) is finite ( see ( * ? ? ? * thm . 7.2 ) ) . a stronger statement still is the following ( essentially theorem 9.2 of @xcite ) : [ shti - thm9.2 ] there are only finitely many coprime @xmath0-units @xmath34 and @xmath35 for which there exist integers @xmath36 and @xmath37 such that @xmath38 . the aim of this paper is to illustrate the use of frey - hellegouarch curves in solving equations like ( [ two ] ) and , more generally , ( [ enn ] ) . specifically , we will apply such an approach to provide a new proof of theorem [ shti - thm9.2 ] that _ a priori _ avoids the use of lower bounds for linear forms in logarithms , instead combining frey - hellegouarch curves , modularity and level - lowering , with the aforementioned theorem of shafarevich . in fairness , it must be mentioned , that effective versions of the latter result have typically depended fundamentally on linear forms in logarithms ; for recent papers along these lines , see the work of fuchs , von knel and wstholz @xcite and von knel @xcite . the benefit of our approach is it enables us to , in section [ s : all_solutions ] , explicitly solve equation ( [ enn ] ) for a pair of sets @xmath0 with cardinality @xmath39 . to the best of our knowledge , this is the first time this has been carried out . indeed , it is unclear whether the classical approach to theorem [ shti - thm9.2 ] via only lower bounds for linear forms in logarithms can be made practical with current technology , in any nontrivial situations . the outline of our paper is as follows . section [ sec2 ] introduces our basic notation . in section [ s : finiteness ] , we show how to obtain various finiteness results currently proved with techniques from diophantine approximation , via frey - hellegouarch curves over @xmath40 . philosophically , this bears a strong resemblance to recent work of von knel @xcite and of murty and pasten @xcite . section [ sec4 ] contains explicit details of the connections between frey - hellegouarch curves and modular forms . in sections [ s : n=2 ] and [ s : n=3 ] , we carry out such a `` modular '' approach quite explicitly for exponents @xmath41 and @xmath42 respectively . as an illustration of our methods , we completely solve ( [ enn ] ) for @xmath43 and @xmath44 ( hence recovering de weger s aforementioned result ) , and also for @xmath45 and @xmath44 , for every prime @xmath46 . it should be emphasized that this is not a `` serious '' application of our method , but merely meant as an illustration of a partial converse of the connection between solving equations of the shape ( [ enn ] ) and computing elliptic curves . the reader may wish to omit these sections at first ( and , for that matter , subsequent ) readings . a more interesting result along these lines is due to kim @xcite , where the connection between more general cubic thue - mahler equation and shafarevich s theorem is mapped out . section [ s : all_solutions ] contains , as previously mentioned , the main result of the paper , namely an explicit solution of equation ( [ enn ] ) for the sets @xmath1 and @xmath47 . the techniques we employ to prove these results , besides the aforementioned use of frey - hellegouarch curves and their associated modular forms , are local methods and appeal to computer algebra packages for solving thue and thue - mahler equations ; for the last of these , we rely extensively upon the computational number theory packages magma @xcite , pari @xcite and sage @xcite . we thank rafael von knel and benjamin matschke for pointing out to us a missing solution in a previous version of proposition [ prop : local_obstructions_2 - 3-p - n=2 ] . in what follows , we will let @xmath23 be a prime number and @xmath48 a nonzero integer . we denote by @xmath49 the radical of @xmath50 , i.e. the product of distinct primes dividing @xmath48 , and by @xmath51 the largest nonnegative integer @xmath52 such that @xmath53 divides @xmath48 . we write @xmath54 for the prime - to-@xmath23 part of the radical of @xmath50 , that is , the largest divisor of @xmath49 that is relatively prime to @xmath23 . we will begin by noting some basic results on modular forms and connections between them and elliptic curves . suppose that the @xmath55-expansion @xmath56 defines a weight @xmath11 , level @xmath57 ( cuspidal ) newform , with coefficients @xmath58 generating a number field @xmath59 . further , let @xmath10 be an elliptic curve of conductor @xmath60 and @xmath29 a rational prime . if @xmath61 is a prime satisfying @xmath62 , we define @xmath63 we say that @xmath13 _ arises modulo @xmath29 from the newform @xmath64 _ and write @xmath65 if there exists a prime ideal @xmath66 of @xmath9 such that , given any prime @xmath67 , we have either @xmath68 or @xmath69 throughout this section , we will let @xmath0 denote a finite set of primes and @xmath15 a positive integer . the second part of the following lemma is a classical and easy application of the theory of linear forms in logarithms ( see corollary 1.2 of @xcite ) . we here give a complete proof of the result below using instead frey - hellegouarch curves and shafarevich s theorem ( theorem ix.6.1 of @xcite ) . [ lem : sum_three_s_units ] there are only finitely many @xmath0-units @xmath70 with @xmath71 for which there exists a nonzero integer @xmath26 such that @xmath72 . in particular , if @xmath73 are @xmath0-units with @xmath71 such that @xmath74 is again a @xmath0-unit , then @xmath75 is bounded by a constant depending on @xmath0 and @xmath15 . let @xmath34 and @xmath35 be @xmath0-units and let @xmath37 be an integer such that @xmath72 . consider the elliptic curve @xmath76 with discriminant @xmath77 it follows that @xmath13 has good reduction outside @xmath78 . by shafarevich s theorem , there are only finitely many isomorphism classes of rational elliptic curves having good reduction outside a finite set of primes . this forces @xmath79 to take only finitely many values when @xmath7 and @xmath8 range over all @xmath0-units . but then , since @xmath80 , the quotient @xmath81 also takes finitely many different values and therefore @xmath82 is bounded whenever @xmath71 . we next prove a version of theorem 9.1 of @xcite through appeal to ( various ) frey - hellegouarch curves and level - lowering . [ thm : st ] let @xmath34 and @xmath35 be @xmath0-units with @xmath71 . let @xmath36 and @xmath83 be integers . if @xmath38 , then @xmath29 is bounded by a constant depending only on @xmath0 and @xmath15 . assume @xmath84 . if @xmath26 is an @xmath0-unit , then , by the previous lemma , @xmath82 and thus @xmath29 is bounded by a constant depending only on @xmath0 and @xmath15 . therefore , one may assume that @xmath26 is not a @xmath0-unit and , in particular , we may choose a prime @xmath85 dividing @xmath26 . define an elliptic curve @xmath86 as follows . if @xmath87 , then consider @xmath88 whose discriminant and @xmath89-coefficient are given by @xmath90 if however @xmath91 , we take @xmath92 with discriminant and @xmath89-coefficient given by @xmath93 in both cases , @xmath13 has multiplicative reduction at @xmath55 with @xmath94 divisible by @xmath29 . let us then further assume that @xmath95 , @xmath96 and that the mod @xmath29 representation attached to @xmath13 is absolutely irreducible . by classical bounds on conductors ( see @xcite for instance ) , modularity of elliptic curves over @xmath40 ( @xcite ) and ribet s level lowering theorem ( @xcite ) , the elliptic curve @xmath13 arises modulo @xmath29 from a weight @xmath11 newform @xmath97 of ( trivial nebentypus and ) level @xmath98 such that @xmath57 is coprime to @xmath55 . therefore , @xmath57 is bounded by a constant depending only on @xmath0 and there exists a prime ideal @xmath99 above @xmath29 in the ring of integers of the coefficient field @xmath9 of @xmath64 such that @xmath100 by deligne s bounds , @xmath101 is a nonzero algebraic integer in @xmath9 whose galois conjugates are all less than @xmath102 in absolute value and whose norm is divisible by @xmath29 . therefore we have @xmath103}$ ] . since @xmath104 $ ] is bounded by the dimension of the space of weight-@xmath11 cuspforms of level @xmath57 , it follows that @xmath29 is bounded from above by a constant depending only on @xmath0 , as desired . combining these results with the finiteness of the number of solutions to thue - mahler equations ( see , for instance , theorem 7.1 of @xcite ) allows us to prove the following . [ thm : finiteness ] there are only finitely many @xmath0-units @xmath70 with @xmath71 for which there exist integers @xmath36 and @xmath37 such that @xmath38 . by lemma [ lem : sum_three_s_units ] , the equation @xmath105 has only finitely many solutions in @xmath0-units @xmath34 and @xmath35 , with @xmath71 . further , if @xmath34 and @xmath35 are @xmath0-units with @xmath71 such that there exist integers @xmath36 and @xmath26 with @xmath83 satisfying @xmath38 , then , by the previous theorem , @xmath29 is bounded by a constant depending only on @xmath0 and @xmath15 . but , for a fixed value of @xmath36 , the finiteness of solutions to equation @xmath38 in @xmath0-units @xmath70 with @xmath71 and integer @xmath26 follows from lemma [ lem : sum_three_s_units ] and from the finiteness of the set of solutions to thue - mahler equations for @xmath41 and @xmath33 respectively . given a primitive solution @xmath27 of ( [ enn ] ) , let us denote by @xmath106 the gcd of @xmath7 and @xmath8 . then @xmath106 divides @xmath107 and , without loss of generality , we may write @xmath108 where @xmath109 is a nonzero integer and @xmath35 is a @xmath29th - power free positive @xmath0-unit . if @xmath110 and @xmath111 , then one has @xmath112 conversely , let @xmath113 be pairwise coprime @xmath0-units with @xmath35 positive satisfying the above equation and let @xmath114 be a positive @xmath0-unit such that @xmath115 . put @xmath116 , @xmath117 and @xmath118 . then , @xmath27 is a primitive solution of ( [ enn ] ) . therefore , all the primitive solutions of equation ( [ enn ] ) can be deduced from the finite set of @xmath0-units satisfying the condition of theorem [ thm : finiteness ] with @xmath119 . moreover , combining this remark with the previous results , we have the following : for a fixed value of @xmath36 , there are only finitely many triples @xmath120 such that @xmath121 with @xmath26 nonzero integer , @xmath73 @xmath0-units and @xmath28 @xmath29th - power free . moreover , there are only finitely many primitive solutions to equation ( [ enn ] ) if and only if @xmath122 . according to the discussion above , the first part of the corollary is a direct consequence of theorem [ thm : finiteness ] ( with @xmath119 ) . if , however , @xmath123 , then @xmath124 is a primitive solution to ( [ enn ] ) for any @xmath36 . conversely , if @xmath122 , consider a primitive solution @xmath27 to ( [ enn ] ) . dividing the equation by @xmath125 leads , as explained earlier , to an equation of the shape @xmath126 where @xmath113 are coprime @xmath0-units and @xmath127 . by assumption , @xmath113 are odd and therefore @xmath109 is even . in particular , we have @xmath128 and by theorem [ thm : st ] , @xmath29 is bounded independently of @xmath73 and @xmath26 . the desired finiteness result now follows from the first part of the corollary . in the proof of lemma [ lem : sum_three_s_units ] we have appealed to @xmath129-frey - hellegouarch curves and , for theorem [ thm : st ] , to @xmath130 and @xmath131-frey - hellegouarch curves . the main goal of this paper is to make the statements of this section completely explicit in a number of situations . for this purpose , we will have use of refined information on frey - hellegouarch curves of the above signatures . we recall some ( by now ) classical but useful results on frey - hellegouarch curves associated with generalized fermat equations of signature @xmath130 , @xmath129 and @xmath131 , and the newforms from which they arise . let @xmath132 and @xmath133 be @xmath29th - power free pairwise coprime nonzero integers and let @xmath134 and @xmath135 be pairwise coprime nonzero integers such that @xmath136 we make the additional simplifying assumptions ( which are not without loss of generality , but will be satisfied in the cases of interest to us ) that @xmath137 define an elliptic curve @xmath138 via @xmath139 we summarize the properties of @xmath138 that will be useful to us in the following result ( see kraus @xcite ) . [ nnn ] if @xmath140 is prime and @xmath141 , we have that @xmath142 for @xmath64 a weight @xmath11 cuspidal newform of level @xmath143 further , if @xmath61 is prime with @xmath144 , then @xmath145 next let @xmath146 and @xmath133 be nonzero integers such that @xmath147 with @xmath148 and @xmath149 pairwise coprime , @xmath133 squarefree and @xmath150 prime . without loss of generality , we may suppose that @xmath151 and @xmath152 are @xmath29th - power free and that we are in one of the following situations : 1 . [ item : nn2_case_i ] @xmath153 and @xmath154 ; + 2 . [ item : nn2_case_ii ] @xmath155 and either @xmath156 or @xmath157 ; + 3 . [ item : nn2_case_iii ] @xmath155 , @xmath158 and @xmath159 ; + 4 . [ item : nn2_case_iv ] @xmath155 , @xmath160 and @xmath161 ; + 5 . [ item : nn2_case_v ] @xmath162 and @xmath161 . in cases ( [ item : nn2_case_i ] ) and ( [ item : nn2_case_ii ] ) , we will consider the curve @xmath163 in cases ( [ item : nn2_case_iii ] ) and ( [ item : nn2_case_iv ] ) , we will instead consider @xmath164 and in case ( [ item : nn2_case_v ] ) , @xmath165 these are all elliptic curves defined over @xmath21 . the following lemma summarizes some useful facts about these curves . apart from its ( easy - to - check ) assertion ( [ item : lemma_nn2_2 ] ) and up to some slight differences of notation , this is lemma 2.1 of @xcite . [ lem : nn2_arith ] let @xmath166 or @xmath167 and @xmath168 . 1 . the discriminant @xmath169 of the curve @xmath13 is given by @xmath170 2 . [ item : lemma_nn2_2 ] the @xmath171-invariant @xmath172 of the curve @xmath13 is given by @xmath173 3 . the conductor @xmath174 of the curve @xmath13 is given by @xmath175 where @xmath176 in particular , @xmath13 has multiplicative reduction at each odd prime @xmath23 dividing @xmath177 . also , @xmath13 has multiplicative reduction at @xmath11 if @xmath178 . 4 . the curve @xmath13 has a @xmath21-rational point of order @xmath11 . for the purposes of our applications , we will have need of an analog of proposition [ nnn ] , essentially lemma 3.3 of @xcite . [ nn2 ] if @xmath95 is prime and @xmath179 , we have , for each @xmath180 , that @xmath181 for @xmath64 a weight @xmath11 cuspidal newform of level @xmath182 where @xmath183 where @xmath184 is as defined in lemma [ lem : nn2_arith ] . let us suppose that @xmath146 and @xmath133 are nonzero integers such that @xmath185 with @xmath186 prime . assume @xmath187 , @xmath188 , and @xmath189 are pairwise coprime , and , without loss of generality , that @xmath190 and @xmath191 . further , suppose that @xmath133 is cubefree and that @xmath151 and @xmath152 are @xmath29th - power free . we consider the elliptic curve @xmath192 with the above assumptions , we have the following result ( lemma 2.1 of @xcite ) . [ lem : nn3_arith ] let @xmath193 . 1 . the discriminant @xmath194 of the curve @xmath13 is given by @xmath195 2 . the @xmath171-invariant of @xmath13 is given by @xmath196 3 . the conductor @xmath174 of the curve @xmath13 is @xmath197 where @xmath198 in particular , @xmath13 has split multiplicative reduction at each prime @xmath199 dividing @xmath188 , split multiplicative reduction at each prime dividing @xmath187 congruent to @xmath200 , @xmath201 , @xmath202 , @xmath203 , @xmath204 , @xmath205 or @xmath206 modulo @xmath207 , and non - split multiplicative reduction at all other primes dividing @xmath187 , except @xmath167 . also , @xmath13 has split multiplicative reduction at @xmath167 if @xmath208 , and good reduction if @xmath209 . 4 . the curve @xmath13 has a @xmath21-rational point of order @xmath167 . in what follows , we will apply frey - hellegouarch curves of signature @xmath131 more frequently than other signatures ; the following result , essentially proposition 4.2 of @xcite ( though it is worth noting that the value @xmath210 as define in lemma 3.4 of that paper is actually stated incorrectly for the cases where @xmath211 ) , will be of particular use to us . [ lem : nn3_ll-1 ] if @xmath140 is prime , @xmath212 and @xmath213 does not correspond to the identities @xmath214 then we have that @xmath215 for @xmath216 a weight @xmath11 cuspidal newform of level @xmath217 where @xmath218 with @xmath184 as defined in lemma [ lem : nn3_arith ] unless @xmath219 in which case we have @xmath220 more precisely , if @xmath61 is a prime , coprime to @xmath221 , then @xmath29 divides @xmath222 where @xmath9 is the number field generated by the fourier coefficients of @xmath64 and @xmath223 , with @xmath224 if @xmath225 , and @xmath226 otherwise . in this section , we consider equation ( [ two ] ) ( as treated by de weger in @xcite and @xcite ) , via an @xmath129 frey - hellegouarch curve approach . according to the discussion of section [ s : finiteness ] , the corresponding equations to treat are of the shape @xmath227 where @xmath34 and @xmath35 are pairwise coprime @xmath0-units . define @xmath228 , @xmath229 , @xmath230 , @xmath231 and @xmath232 . then we have @xmath233 and may assume , without loss of generality , that we are in one of the situations ( [ item : nn2_case_i])-([item : nn2_case_v ] ) of [ ss : nn2 ] . consider the associated elliptic curve @xmath86 of lemma [ lem : nn2_arith ] . it has good reduction outside @xmath234 . therefore , if we know representatives @xmath235 of all ( the finitely many ) isomorphism classes of rational elliptic curves ( with a nontrivial two - torsion subgroup ) having good reduction outside @xmath236 , all that remains to do to solve ( [ gen ] ) is to check for an equality @xmath237 and @xmath238 denote the @xmath171-invariants of @xmath13 and @xmath239 respectively . computing such representatives is a classical but challenging problem that has only been achieved for a rather restrictive list of sets , including @xmath240 , @xmath241 , @xmath242 , @xmath243 , @xmath244 and @xmath245 ( see @xcite ) . nevertheless , the precise information on the conductor @xmath174 of @xmath13 provided by lemma [ lem : nn2_arith ] , namely @xmath246 where @xmath247 allows us to sometimes show , for some specific sets @xmath0 , that we have @xmath248 except for some precisely identified quadruples @xmath249 . in that situation , we can therefore appeal , via the sage package ` database_cremona_ellcurve ` ( @xcite , @xcite ) , to cremona s tables which , at the time of writing ( spring 2015 ) , contain representatives for all rational elliptic curves of conductor less than @xmath250 . by way of example , consider @xmath43 or @xmath251 with @xmath23 prime , @xmath252 . we stress the fact that , for most of these latter sets , we presently do not know a complete list of representatives of the isomorphism classes of rational elliptic curves having good reduction outside @xmath0 . nevertheless , the strategy outlined above does apply and this allows us to easily recover de weger s result mentioned in the introduction and to solve equations that remained unsolved in a recent paper by terai @xcite ( see corollary [ prop : terai ] ) . we carry out the details in the next two subsections ; the sage code used for our computations is available at 8ex ` http://www.math.ubc.ca/~bennett/sum_of_two_s-units.pdf ` . specializing the approach above to the case @xmath43 considered by de weger , we note that _ a priori _ we must consider conductors of the shape @xmath253 where @xmath254 , @xmath255 , and @xmath256 . however , we have the following easy result . [ lem : local_obstructions ] in each case @xmath248 . further , if @xmath257 , then either @xmath258 or @xmath259 and @xmath260 . if @xmath261 , we have a solution to an equation of the shape @xmath262 where @xmath52 is nonnegative , @xmath263 and @xmath264 . the solutions to this equation for @xmath265 correspond to those listed in the lemma moreover , if @xmath266 , none of these equations has a solution modulo @xmath267 unless we have @xmath268 or @xmath269 , that is @xmath259 . similarly , it is easy to check that @xmath270 , unless we have @xmath271 @xmath272 or @xmath273 of these , only the cases @xmath274 may correspond to solutions to ( [ gen ] ) , namely to equations of the shape @xmath275 and @xmath276 . each of these equations , however , is insoluble modulo @xmath267 . from lemma [ lem : local_obstructions ] and the above discussion , we may thus appeal to cremona s sage package to reproduce de weger s results ( with a slightly faster search provided by the restrictions on the exponents given in the above lemma ) . in particular , if @xmath43 , as noted in ( * ? ? ? * thm . 7.2 ) , there are precisely @xmath277 solutions @xmath120 to ( [ two ] ) with @xmath28 squarefree and @xmath278 and @xmath279 . we list these solutions in the file ` http://www.math.ubc.ca/~bennett/2-3-5-7.pdf ` . as an obvious byproduct , we also obtain a complete list of solutions to ( [ two ] ) with @xmath280 and @xmath281 . as we will consider these sets again in section [ s : all_solutions ] , we state these results here . the only primitive solutions to equation ( [ two ] ) with @xmath280 and with , say , @xmath282 are given by @xmath283 the only primitive solutions to equation ( [ two ] ) with @xmath281 and with , say , @xmath282 are given by @xmath284 we now turn our attention to the case @xmath251 where @xmath23 is a prime in the range @xmath285 . once again , we must _ a priori _ consider conductors of the shape @xmath286 where @xmath254 , @xmath287 , and @xmath256 ; many of these conductors exceed the limits of the current cremona database . however , we have the following result . [ prop : local_obstructions_2 - 3-p - n=2 ] in each case @xmath248 , unless @xmath249 corresponds to one of the following equations @xmath288 @xmath289 it is easy to check that we have @xmath248 unless we are in one of the following situations : @xmath290 of these , only the cases @xmath291 may actually correspond to our equations . we are therefore led to solve the following equations , where @xmath263 and @xmath52 is a positive integer : @xmath292 @xmath293 and @xmath294 we deal with each of these in turn . considering first the equation @xmath295 , we readily check that it has no solution modulo @xmath296 for odd values of @xmath52 and @xmath297 . assume therefore that @xmath52 is even . if moreover @xmath298 , then by factorization , we end up with an equation of the shape @xmath299 or @xmath300 for some integer @xmath109 . according to ( * ? ? ? 1.1 ) , this leads to a unique solution to our equation , namely @xmath301 . finally , if @xmath52 is even and @xmath302 , the equation @xmath303 has no solution modulo @xmath304 unless @xmath305 or @xmath306 . however , for @xmath307 and @xmath306 , it has no solution modulo @xmath308 where @xmath309 and @xmath310 respectively . in the remaining case , namely @xmath311 , write @xmath312 with @xmath313 and @xmath314 . then , @xmath315 is an integral point on the elliptic curve @xmath316 computing its integral points using the aforementioned sage command thus leads to the unique solution @xmath317 . we now turn our attention to the equation @xmath318 . we easily check that for @xmath23 in the range @xmath319 , this equation has no solution modulo @xmath304 unless we have @xmath320 for @xmath321 and @xmath306 , the corresponding equation has no solution modulo @xmath308 where @xmath322 and @xmath310 respectively . for the remaining case , that is @xmath323 , reducing modulo @xmath324 , we find that @xmath325 . writing @xmath326 , @xmath327 is thus an integral point on the elliptic curve @xmath328 as before , we compute its integral points on sage and deduce the unique solution @xmath329 . consider now equation @xmath330 for @xmath23 in the range @xmath331 . we check that there is no solution modulo @xmath304 unless we have @xmath332 for @xmath333 and @xmath334 , we however have that the corresponding equation has no solution modulo @xmath308 where @xmath335 and @xmath203 respectively . it remains to deal with the cases @xmath336 and @xmath337 . in the latter case , we find that @xmath338 by reducing modulo @xmath339 . writing @xmath340 , @xmath341 is necessarily an integral point on the elliptic curve @xmath342 this again leads to a unique solution , namely @xmath343 . in the former case , that is @xmath336 , write @xmath312 with @xmath313 and @xmath314 . then , @xmath344 is an integral point on the elliptic curve @xmath345 computing its integral points using the command ` integralpoints ` in magma ( @xcite ) leads to a unique solution which is @xmath346 . we note here that the corresponding routine in sage has marked difficulty with this equation . we now consider equation @xmath347 . if @xmath302 , then we have no solution modulo @xmath304 unless @xmath348 or @xmath339 . for @xmath349 or @xmath339 , however , we have that the corresponding equation has no solution modulo @xmath308 where @xmath350 or @xmath351 respectively . if @xmath352 , write @xmath312 with @xmath313 and @xmath314 . then , @xmath353 is an integral point on the elliptic curve @xmath354 computing its integral points using sage leads to the unique solution @xmath355 . if now @xmath298 , then reducing @xmath356 modulo @xmath357 shows that @xmath52 is necessarily even . write @xmath358 and @xmath359 with @xmath313 and @xmath314 . then , for some divisor @xmath360 of @xmath26 , either @xmath361 or @xmath362 is an integral point on one of the elliptic curves @xmath363 however none of their integral points ( which we have already computed ) corresponds to a solution of our equation . let us now consider the equation @xmath364 . if @xmath302 , then the corresponding equation has no solution modulo @xmath304 unless @xmath365 or @xmath334 . moreover in the former case , we have @xmath366 . if @xmath367 , we check that @xmath368 has no solution modulo @xmath369 . assume thus that @xmath365 and write @xmath370 . then @xmath371 is an integral point on the elliptic curve @xmath372 we therefore find that there is only one solution corresponding to @xmath373 . similarly , if @xmath298 , then the corresponding equation has no solution modulo @xmath304 unless @xmath374 . write @xmath312 with @xmath314 and @xmath375 . then , @xmath376 is an integral point on the elliptic curve @xmath377 this gives rise to the unique solution @xmath378 . we finally deal with the last equation , namely @xmath379 . if @xmath302 , then the corresponding equation has no solution modulo @xmath304 unless @xmath380 or @xmath306 . if @xmath380 , we find a local obstruction modulo @xmath381 . if @xmath382 , we write @xmath312 with @xmath314 and @xmath383 , whereby @xmath384 is an integral point on the elliptic curve @xmath385 we therefore conclude that there is only one solution corresponding to @xmath386 . similarly , if @xmath298 , then the corresponding equation has no solution modulo @xmath304 unless @xmath349 or @xmath365 . moreover , in that latter case , we have @xmath338 and it is easy to check that @xmath387 has no solution modulo @xmath388 . assume therefore that @xmath365 and write @xmath340 . then @xmath389 is an integral point on the elliptic curve @xmath390 this gives rise to the unique solution @xmath391 , which completes the proof of the proposition . utilizing proposition [ prop : local_obstructions_2 - 3-p - n=2 ] , we can again appeal to cremona s sage package to compute primitive solutions to equation ( [ two ] ) . for each set @xmath251 , we have tabulated these solutions in the file ` http://www.math.ubc.ca/~bennett/2-3-p.pdf ` . by quick examination of this table for @xmath392 and @xmath393 , we immediately deduce the following result about equations that remained unsolved in proposition 3.3 of terai @xcite ( but have been recently solved via rather different methods by deng @xcite ) . [ prop : terai ] the only solutions to @xmath394 and @xmath395 are @xmath396 and @xmath397 respectively . we now deal with equation ( [ enn ] ) when @xmath42 using a @xmath398 frey - hellegouarch curve approach . the corresponding equations to treat are of the shape @xmath399 where @xmath34 and @xmath35 are pairwise coprime @xmath0-units . we may assume , without loss of generality , that @xmath35 is cubefree and positive and that we have @xmath400 and @xmath401 . with the notation of [ ss : nn3 ] , we attach to such a solution the elliptic curve @xmath402 : @xmath403 as in section [ s : n=2 ] , all that remains to do to solve equation ( [ eq : n=3 ] ) is to check for an equality @xmath404 and @xmath238 denote the @xmath171-invariants of @xmath13 and @xmath239 respectively , with @xmath239 ranging over all representatives of the isomorphism classes of elliptic having good reduction outside @xmath405 ( and a nontrivial @xmath167-torsion subgroup ) . to circumvent the difficulty of computing representatives and make this approach work for a broader list of sets @xmath0 ( including some for which we do not know a complete list of such representatives ) , we also make use of the precise formula for the conductor of @xmath13 given by lemma [ lem : nn3_arith ] , namely @xmath406 where @xmath407 as in the previous section , we now apply this approach to the sets @xmath43 and @xmath251 with @xmath23 prime , @xmath408 . specializing to @xmath43 , we _ a priori _ need to consider all conductors of the shape @xmath409 with @xmath410 , @xmath411 and @xmath412 . however , we have , as before , the following result . in each case @xmath413 . it is easy to check that we have the desired inequality for @xmath174 , unless @xmath414 among these possibilities , only the case @xmath415 may correspond to solutions . since each of the equations @xmath416 is insoluble modulo @xmath267 , we obtain the stated result . as explained previously , we can therefore appeal to cremona s table of elliptic curves to solve equation ( [ eq : n=3 ] ) and thus ( [ enn ] ) with @xmath42 and @xmath43 . it turns out that there are exactly @xmath417 triples @xmath120 such that @xmath418 with @xmath7 and @xmath8 @xmath240-units , with @xmath28 cubefree and , say , @xmath278 , and @xmath26 a positive integer . they are listed in the file ` http://www.math.ubc.ca/~bennett/2-3-5-7.pdf ` . from there we easily extract the latter solutions to this equation for @xmath280 and @xmath281 . we list them here for later use . the only primitive solutions to equation ( [ enn ] ) with @xmath42 , @xmath280 and , say , @xmath282 are given by @xmath419 the only primitive solutions to equation ( [ enn ] ) with @xmath42 , @xmath281 and , say , @xmath282 are given by @xmath420 we now turn our attention to the case @xmath251 where @xmath285 is prime . once again , we _ a priori _ need to consider all conductors of the shape @xmath421 with @xmath422 , @xmath423 and @xmath412 . however , we have the following result , the proof of which , it being quite similar to that of proposition [ prop : local_obstructions_2 - 3-p - n=2 ] , we omit for the sake of concision . we have @xmath248 unless @xmath249 corresponds to a one of the following equations @xmath424 @xmath425 combining this proposition with cremona s tables of elliptic curves then allows us to compute all the solutions to ( [ enn ] ) with @xmath42 , @xmath251 and @xmath23 as above . we list them in the file ` http://www.math.ubc.ca/~bennett/2-3-p.pdf ` . in this last section , we completely solve equation ( [ enn ] ) for two specific sets of primes , namely @xmath280 and @xmath281 , using a variety of frey - hellegouarch curves , level - lowering and heavy computations involving modular forms and thue - mahler equations . it is only through careful combination of a variety of frey - hellegouarch curves , in conjunction with local arguments , that we are able to reduce these problems to a feasible collection of thue - mahler equations ( in our case , all of degree @xmath202 ) . this is , in essence , the main feature of our approach that distinguishes it from one purely reliant upon lower bounds for linear forms in logarithms . this latter method is , in our opinion , at least with current technology , impractical for explicitly solving equation ( [ enn ] ) for any set @xmath0 with at least two elements . the cases where @xmath428 were covered in the previous two sections . we may therefore assume , without loss of generality , that @xmath140 is prime . the corresponding equation to treat is @xmath429 with @xmath34 and @xmath35 coprime @xmath430-units and @xmath35 @xmath29th - power free . assume first that @xmath431 . since precisely one of @xmath34 and @xmath432 is even , we may suppose , without loss of generality , that @xmath433 and that @xmath434 we write @xmath435 or @xmath436 if @xmath8 or @xmath35 is even , respectively , and set @xmath437 , for @xmath132 and @xmath133 @xmath29th - power free integers . with the notation of [ ss : nnn ] , considering the elliptic curve @xmath438 , we thus have that @xmath29 fails to divide @xmath439 and @xmath440 , and hence , in the notation of proposition [ nnn ] , @xmath441 where @xmath64 is a weight @xmath11 newform of level @xmath442 . the resulting contradiction implies that necessarily @xmath443 . assume next that @xmath444 . without loss of generality , we may suppose @xmath400 , @xmath401 and that @xmath445 we define @xmath446 depending on whether @xmath447 or @xmath35 , respectively , where , in either case , @xmath151 and @xmath152 are chosen to be @xmath29th - power free and @xmath133 is cubefree . in both cases , we have @xmath448 and @xmath449 . if @xmath450 , then with the notation of proposition [ lem : nn3_ll-1 ] , we have @xmath451 where @xmath64 is a weight @xmath11 newform of level @xmath452 and hence a contradiction . we therefore have @xmath453 and we are led to solve @xmath454 with @xmath455 . both equations have no solution for @xmath456 and , by reducing modulo @xmath357 , we see that the equation @xmath457 has no solution for @xmath458 as well . if @xmath459 and @xmath458 , then @xmath52 is necessarily even , whereby both @xmath460 and @xmath461 are powers of @xmath11 . hence @xmath462 and this is again a contradiction . the following result is the main theorem of the paper . as we shall observe , despite the apparently small size of @xmath0 and its elements , the computations involved here are really approaching the limits of current `` off the shelf '' technology ( though coming refinements in computational tools for modular forms will alleviate this somewhat ) . [ lem : factorization ] let @xmath7 and @xmath8 be coprime integers and @xmath29 an odd prime number . if we write @xmath467 then we have that @xmath468 and , moreover , @xmath469 further , @xmath470 and if @xmath471 is a prime dividing @xmath472 , then either @xmath473 or @xmath474 . apart from its very last assertion , this lemma is proved in ( * ? ? ? * lemma 2.1 ) . suppose that @xmath471 is a prime dividing @xmath472 . then , by coprimality , @xmath471 does not divide @xmath8 . we may thus find an integer @xmath475 such that @xmath476 , whence @xmath477 . if @xmath478 , it follows that @xmath479 so that @xmath480 . otherwise , since @xmath29 is prime , @xmath481 has order @xmath29 modulo @xmath61 , so that @xmath482 , i.e. @xmath483 . assume @xmath487 and @xmath488 . then , with the above notation , we have that @xmath489 , since , if @xmath490 , then necessarily @xmath491 . it follows that there exists a prime @xmath492 such that @xmath493 and hence , by lemma [ lem : factorization ] , @xmath480 , whereby @xmath494 . we thus have @xmath495 which again has no solutions with @xmath488 . this contradiction completes the proof of the lemma . we first deal with @xmath496 . since @xmath26 is even , we have that @xmath133 is odd and by factorization ( up to permutation of @xmath23 and @xmath55 ) that either @xmath501 for some odd integer @xmath360 . by a classical result of strmer ( @xcite ) , the first case holds only for @xmath502 . similarly in the third case , we deduce that @xmath503 for some nonzero integer @xmath504 which in turn contradicts corollary 1.4 of @xcite . we therefore end up in the situation where @xmath505 for some positive integer @xmath504 . by factorization , @xmath506 and @xmath507 for some positive integers @xmath508 and @xmath509 , whereby we have @xmath510 we may thus appeal to theorem 1.1 of @xcite with ( in the notation of that paper ) @xmath511 to conclude that @xmath512 and @xmath513 , whence @xmath514 as claimed . we now turn our attention to the equation @xmath497 . by factorization , for some positive integers @xmath508 and @xmath509 , we have either @xmath515 according to _ loc . cit . _ applied to @xmath516 , the former equation has no such solutions , whereas the only solutions to the latter are with @xmath512 , @xmath517 and @xmath513 . this gives rise to @xmath518 , as claimed . with those preliminary results in hands , we now turn to the actual proof of theorem [ thm : main ] . once again , recall that we have to solve @xmath519 where @xmath70 are coprime @xmath520-units with @xmath521 positive , @xmath186 prime and @xmath26 nonzero . [ prop : y_pm_1 ] let @xmath7 and @xmath35 be coprime positive @xmath520-units , @xmath522 and let @xmath186 be a prime number such that @xmath38 for some positive integer @xmath26 . then , @xmath523 , @xmath512 , @xmath524 and @xmath525 according to lemma [ lem : factorization2 ] , we may assume , without loss of generality , that @xmath526 . similarly by applying ( * ? ? ? 1.1 ) to each subset of @xmath520 of cardinality @xmath11 , we may also assume that @xmath527 . besides , sieving modulo @xmath528 shows that we necessarily have @xmath523 and that either @xmath7 is a @xmath201th power and @xmath35 has two distinct prime factors ( in @xmath520 ) , or that @xmath7 is a square and @xmath35 has only one prime factor . we finally conclude using proposition [ prop : stormer ] . according to lemma [ lem : factorization2 ] and proposition [ prop : y_pm_1 ] above , in order to prove theorem [ thm : main ] , it remains to solve each equation of the shape @xmath529 where @xmath530 and @xmath531 are nonnegative integers , @xmath186 is prime and @xmath532 . in the remainder of this section , we prove the following precise result on these equations . combining it with proposition [ prop : y_pm_1 ] and results from sections [ s : n=2 ] and [ s : n=3 ] completes the proof of theorem [ thm : main ] . suppose that we have a solution to one of equations ( [ eq6 ] ) ( [ eq18 ] ) in nonnegative integers @xmath530 and @xmath531 , prime @xmath140 and @xmath532 , and rewrite this in the shape ( [ eq - nn3 ] ) for suitable choices of @xmath534 and @xmath135 . applying proposition [ lem : nn3_ll-1 ] , we thus have that @xmath65 for some weight @xmath11 cuspidal newform of level @xmath57 where , crudely , we have that @xmath535 . since @xmath536 and hence necessarily @xmath537 , we may apply congruence ( [ tech2 ] ) with @xmath538 to conclude that @xmath539 where @xmath540 denotes the number of cuspidal weight-@xmath11 newforms of level @xmath57 . in what follows , we will in fact show that this upper bound can , through somewhat refined arguments using various frey - hellegouarch curves , local computations and thue(-mahler ) solvers , be replaced by the assertion that @xmath541 , corresponding to the solutions noted in theorem [ boss ] . in the case @xmath512 , an approach via frey - hellegouarch curves , while theoretically of value , in practice appears to work poorly . instead , we will use magma code for solving thue - mahler equations due to k. hambrook @xcite ; documentation for this may be found at according to the shapes of the equations and the noted restrictions on @xmath544 and @xmath531 , it suffices to solve the following thue - mahler equations ( some of which may be treated locally ) @xmath545 this is easily achieved using hambrook s code and leads to the solutions mentioned . it is worth noting , at this point in the proceedings , that , at least as currently implemented , the thue - mahler solver we are using has severe difficulties with equations of degree @xmath203 and higher . in particular , we will need to work very carefully in order to avoid its use for the remaining cases under consideration . we will now deal with each of equations ( [ eq6])([eq18 ] ) in turn , assuming @xmath546 and that we have a solution satisfying the conditions mentioned . we will make repeated use of proposition [ lem : nn3_ll-1 ] , explicitly using magma to calculate newforms of all relevant levels @xmath57 . the results are as follows . we list in table [ table : levels ] the triples @xmath547 of interest to us for which the space of weight @xmath11 cuspidal newforms of level @xmath548 is nontrivial , together with the list of all primes @xmath150 for which there exists at least one form @xmath64 of level @xmath57 satisfying both ( [ tech2 ] ) with @xmath538 , and the congruences of proposition [ lem : nn3_ll-1 ] for all primes @xmath549 . the magma code used for this computation ( when @xmath550 ) is available at ` http://www.math.ubc.ca/~bennett/nn3.m ` . for the larger levels @xmath551 and @xmath552 , we have first used wiese s magma function ` decomposition ` ( from his package ` artinalgebras ` available on his homepage ) to compute the characteristic polynomials of the first few fourier coefficients . the output can be found in by reducing the equation modulo @xmath357 , we see that @xmath184 and @xmath554 have the same parity , and that @xmath184 is odd if and only if we have @xmath555 . we begin by assuming that @xmath184 and @xmath554 are even , and @xmath546 . if , further , @xmath556 , we consider the curve @xmath557 where @xmath558 , with @xmath559 . it has good reduction at @xmath202 and is of the shape @xmath560 where @xmath561 with @xmath562 and @xmath563 , whereby @xmath564 . on the other hand , proposition [ nn2 ] , @xmath565 where @xmath64 is a weight-@xmath11 newform of level @xmath566 . since there is a unique such newform and it satisfies @xmath567 , we obtain a contradiction from ( [ tech2 ] ) for @xmath568 . assume next that @xmath184 is even , @xmath569 and @xmath570 . if @xmath571 , then we have @xmath572 and , since @xmath573 , @xmath574 . this leads to the conclusion that @xmath575 and hence a contradiction . if @xmath576 , then we have @xmath577 and the equation to solve is @xmath578 . however we check that there is no value of @xmath579 such that @xmath580 and hence obtain a contradiction in this case as well . we therefore may assume that @xmath184 is even , @xmath569 and @xmath581 , and consider the curve @xmath582 where @xmath583 , with @xmath584 and @xmath585 , with @xmath586 . then , by proposition [ lem : nn3_ll-1 ] , @xmath565 where @xmath64 is a weight @xmath11 newform of level @xmath57 with @xmath587 or @xmath588 if @xmath589 or @xmath590 respectively . since @xmath591 , we therefore reach a contradiction upon appealing to table [ table : levels ] . next , suppose that @xmath184 and @xmath554 are odd . if @xmath592 and @xmath593 , we simply note that there is no solution modulo @xmath594 . if @xmath592 and @xmath576 , then @xmath595 and modulo @xmath596 , we have that @xmath597 and @xmath598 . we are therefore led to solve the thue equation @xmath599 using pari / gp . this gives rise to a unique solution to our equation , namely @xmath600 . if , however , we assume that @xmath601 , we begin by considering the following @xmath131 frey - hellegouarch curve @xmath602 where @xmath603 then , by proposition [ lem : nn3_ll-1 ] , @xmath565 where @xmath64 has level @xmath57 with @xmath604 or @xmath605 , where @xmath606 , if @xmath589 or @xmath260 respectively . according to table [ table : levels ] , since @xmath591 , we necessarily have level @xmath607 , whence @xmath608 to treat this remaining case , we now consider the following @xmath130 frey - hellegouarch curve @xmath609 where @xmath610 and @xmath611 with @xmath612 ( recall that @xmath613 ) . according to proposition [ nnn ] , @xmath13 arises from a weight-@xmath11 newform @xmath64 of level @xmath614 and @xmath64 corresponds to an elliptic curve @xmath615 such that @xmath616 this implies that @xmath239 has good reduction at @xmath310 and @xmath617 , or , in other words , that neither @xmath310 nor @xmath617 divides @xmath26 . we may thus sieve locally at these primes to show that there is no triple @xmath618 ( with @xmath554 odd and @xmath555 ) such that @xmath619 simultaneously for @xmath350 and @xmath617 . this leads to the desired contradiction , whereby we may conclude that the only solution to ( [ eq6 ] ) for @xmath546 prime corresponds to @xmath600 . by reducing modulo @xmath357 , we see that @xmath184 and @xmath554 necessarily have the same parity , and that @xmath184 is odd precisely when we have @xmath555 . assume first that @xmath184 and @xmath554 are even , and @xmath546 . let us consider @xmath621 where @xmath583 , @xmath558 , with @xmath622 . since @xmath623 , @xmath13 has good reduction at @xmath202 . moreover , if @xmath577 , then it satisfies @xmath624 . from proposition [ nn2 ] , @xmath13 arises modulo @xmath29 from a weight @xmath11 newform of level @xmath625 ( recall that @xmath626 ) . but there is a unique newform at level @xmath625 corresponding to an elliptic curve @xmath615 with @xmath627 , a contradiction . it follows that we have @xmath595 , which in turn implies @xmath531 even . we may thus consider instead @xmath628 where @xmath629 with @xmath630 . then , by proposition [ nn2 ] , @xmath13 arises from a weight @xmath11 newform of level @xmath631 . this is an obvious contradiction . next , assume that @xmath184 and @xmath554 are odd , and @xmath150 . we consider the elliptic curve @xmath193 with @xmath632 where @xmath583 , @xmath558 with @xmath622 and @xmath633 with @xmath634 . then @xmath441 where @xmath64 is a weight @xmath11 newform of level , say , @xmath57 which we may compute using proposition [ lem : nn3_ll-1 ] . if @xmath589 , we find that @xmath635 and hence reach a contradiction from consideration of table [ table : levels ] . assume therefore that @xmath260 and that @xmath636 does not correspond to the solution @xmath637 . then , necessarily , @xmath592 and @xmath638 . in particular , we have @xmath639 , i.e. , @xmath640 . we now use local arguments to conclude . indeed , if @xmath593 , then by reducing mod @xmath641 , we obtain that @xmath642 , a contradiction . if , however , @xmath576 , then @xmath595 and we check that there is no value of @xmath554 ( odd ) such that @xmath643 and @xmath644 holds for @xmath645 and @xmath646 simultaneously . by reducing modulo @xmath357 , we see that @xmath184 and @xmath554 have the same parity and that @xmath184 is odd if and only if we have @xmath555 . from the preceding two subsections , we may suppose that @xmath648 . if @xmath592 and @xmath649 , then considering the equation modulo @xmath650 , we conclude that @xmath575 , a contradiction . if , however , @xmath592 and @xmath576 , we can easily check that there is no solution modulo @xmath651 . we may thus suppose that @xmath601 . further , if @xmath544 and @xmath531 are all even ( so that @xmath595 ) , we can consider the elliptic curve @xmath652 where @xmath583 with @xmath584 . by proposition [ nn2 ] , it arises at level @xmath653 , a contradiction . we may thus suppose that at least one of @xmath544 or @xmath531 is odd . we now use an @xmath130 frey - hellegouarch curve argument to treat this remaining case . consider either the elliptic curve @xmath658 or @xmath659 where @xmath660 if @xmath531 is even or odd , respectively . then , by proposition [ nnn ] , @xmath565 where @xmath64 has level @xmath661 . it follows that the newform @xmath64 corresponds to an elliptic curve @xmath615 for which there are unique isogeny classes at level @xmath662 and five isogeny classes at level @xmath614 . for @xmath663 , we compute @xmath664 to show that @xmath13 has good reduction at @xmath55 ( i.e. , @xmath665 ) unless , perhaps , if either @xmath666 and @xmath64 corresponds to the isogeny class ` 210c ` ( in cremona s notation ) , or @xmath667 and @xmath64 corresponds to the isogeny class ` 210d ` . we finally sieve over @xmath544 and @xmath531 ( not all even ) to show that we do not have @xmath668 for @xmath669 and @xmath306 simultaneously if the isogeny class of @xmath239 is not ` 210c ` and for @xmath669 and @xmath670 simultaneously otherwise . assume now @xmath671 . we basically follow the same strategy we used for the case @xmath589 applying first an @xmath131 and then an @xmath130 argument , though the details are somewhat simpler . indeed , we first consider @xmath672 or @xmath673 where @xmath674 if @xmath260 or @xmath590 respectively . then , by proposition [ lem : nn3_ll-1 ] , @xmath565 where @xmath64 has level @xmath675 , where @xmath676 . from appeal to table [ table : levels ] , we conclude that @xmath677 ( and @xmath678 ) . we now use an @xmath130 frey - hellegouarch curve argument to deal with the case @xmath677 . as before , we consider the elliptic curve @xmath679 or @xmath680 with @xmath681 as above , if @xmath531 is even or odd , respectively . then @xmath13 arises mod @xmath205 from an elliptic curve @xmath239 of conductor @xmath661 and for @xmath682 , we check , by computing @xmath683 that @xmath13 has good reduction at @xmath55 ( i.e. , that @xmath665 ) . for @xmath671 , we finally sieve over @xmath554 and @xmath531 to show that @xmath684 does not hold for @xmath685 and @xmath686 simultaneously . this finishes the proof that equation ( [ eq15 ] ) has no solution for @xmath150 . considering the equation @xmath357 , we conclude that @xmath184 and @xmath554 have the same parity and that @xmath184 is odd if and only if we have @xmath577 . if @xmath184 and @xmath554 are odd , then @xmath577 and , since @xmath688 , we have @xmath689 and hence a contradiction . next , assume that @xmath184 and @xmath554 are even and @xmath150 . the equation to treat is now @xmath690 we thus consider the elliptic curve @xmath691 where @xmath692 by proposition [ lem : nn3_ll-1 ] , we have @xmath565 where @xmath64 has level @xmath693 with @xmath694 and @xmath695 using table [ table : levels ] , it then follows that @xmath590 and @xmath696 . we first consider the @xmath131 frey - hellegouarch curve @xmath697 where @xmath629 with @xmath698 . by proposition [ lem : nn3_ll-1 ] , the curve @xmath13 arises modulo @xmath29 from a newform @xmath64 of level @xmath699 , where @xmath676 . hence , using table [ table : levels ] , we deduce that @xmath700 and , in particular , that @xmath701 . if , however , @xmath592 , then we use local arguments to get a contradiction . if @xmath593 , we check that there is no solution modulo @xmath702 . similarly , if @xmath576 , then there is no solution modulo @xmath703 . this shows that equation ( [ eq16 ] ) has no solution for @xmath150 prime . by reducing mod @xmath705 , we conclude that @xmath706 , @xmath707 and @xmath595 . write @xmath708 and @xmath709 . then , there exist nonzero integers @xmath710 with @xmath504 odd such that either @xmath711 adding these equations and recalling that @xmath712 is even yields either @xmath713 if @xmath591 , we consider the @xmath129 frey - hellegouarch curve @xmath714 according to whether we are in the first or second case above , respectively . then , by proposition [ nn2 ] , the curve @xmath13 arises at level @xmath631 , an immediate contradiction . we may therefore assume that @xmath592 and sieve over @xmath544 and @xmath531 to show that there is no solution modulo @xmath715 . this shows that equation ( [ eq17 ] ) has no solution for @xmath150 prime . by reducing mod @xmath717 , we find that @xmath718 , @xmath719 and @xmath595 . we then consider the @xmath129 frey - hellegouarch curve @xmath720 where @xmath629 with @xmath721 . it has good reduction at @xmath203 and satisfies @xmath722 . on the other hand , the curve @xmath13 arises from the ( unique ) newform @xmath64 at level @xmath723 , which satisfies @xmath724 . this gives us the desired contradiction and hence proves that equation ( [ eq18 ] ) has no solution for @xmath150 prime .
the temperature and nuclear composition of the crust and ocean of an accreting neutron star depend on the mix of material ( the ashes ) that is produced at lower densities by fusion of the accreting hydrogen and helium . the hydrogen / helium burning is thermally stable at high accretion rates , a situation encountered in weakly magnetic ( ) neutron stars accreting at rates and in most accreting x - ray pulsars , where the focusing of matter onto the magnetic poles results in local accretion rates high enough for stable burning . for a neutron star accreting at these high rates , we calculate the steady state burning of hydrogen and helium in the upper atmosphere ( ) , where . since the breakout from the `` hot '' cno cycle occurs at a temperature comparable to that of stable helium burning ( ) , the hydrogen is always burned via the rapid proton capture ( rp ) process of wallace and woosley . the rp process makes nuclei far beyond the iron group , always leading to a mixture of elements with masses . the average nuclear mass of the ashes is set by the extent of helium burning via (,p ) reactions , and , because these reactions are temperature sensitive , depends on the local accretion rate . nuclear statistical equilibrium , leading to a composition of mostly iron , occurs only for very high local accretion rates in excess of 50 times the eddington rate . we briefly discuss the consequences of our results for the properties of the neutron star . the wide range of nuclei made at a fixed accretion rate and the sensitivity of the ash composition to the local accretion rate makes it inevitable that accreting neutron stars have an ocean and crust made up of a large variety of nuclei . this has repercussions for the thermal , electrical and structural properties of the neutron star crust . a crustal lattice as impure as implied by our results will have the conductivity throughout most of its mass set by impurity scattering , allowing for more rapid ohmic diffusion of magnetic fields than previously estimated for mono - nuclear mixes . * to appear in the astrophysical journal *
astro-ph9905274
neutron stars in mass - transferring binaries accrete hydrogen and helium rich material from their companions at rates ranging from @xmath9@xmath10 . this matter undergoes thermonuclear fusion within hours to days of reaching the neutron star surface , releasing @xmath11 per nucleon for solar abundances . the nuclear burning is thermally unstable on weakly magnetic neutron stars ( @xmath12 ) accreting at @xmath13 and produces energetic ( @xmath14 ) type i x - ray bursts when @xmath15 ( see lewin , van paradijs & taam 1995 and bildsten 1998b for recent reviews ; observationally , the nature of the time dependent burning in the regime @xmath16 is still not understood ) . the composition of the ashes from the unstable burning is still uncertain , but most certainly consists of heavy nuclei , potentially beyond the iron group ( hanawa , sugimoto & hashimoto 1983 ; wallace & woosley 1984 ; schatz et al . 1997 ; schatz et al . 1998 , koike et al . 1999 ) . in this paper , we calculate for the first time the mix of nuclei made during _ thermally stable _ hydrogen / helium burning in the upper atmosphere of an accreting neutron star . this is appropriate for most accreting x - ray pulsars , where the local accretion rate is high enough for stable burning on the magnetic polar cap ( joss & li 1980 ; bildsten & brown 1997 ) , and for the bright `` z '' sources of hasinger and van der klis ( 1989 ) that are not regular type i bursters and accrete globally at @xmath17 . whereas accurately calculating the ashes from unstable burning requires time - dependent , three - dimensional modelling of the ignition and propagation of the flame , stable burning is time - independent and therefore calculation of the ashes is much more straightforward . the ashes are forced by accretion into the neutron star ocean and crust , replacing what was there at birth . the ensuing electron captures , neutron emissions and pycnonuclear reactions release energy locally and drive the matter neutron - rich ( bisnovatyi - kogan & chechetkin 1979 ; sato 1979 ; haensel & zdunik 1990 ; blaes et . al . 1990 ; bildsten 1998a ; brown , bildsten & rutledge 1998 ) . thus the composition of an accreted or accreting neutron star ocean and crust is very different from the primordial one ( see pethick and ravenhall 1995 for a recent discussion of the primordial crust ) and depends critically on the range and type of nuclei made during the h / he burning in the upper atmosphere . previous studies of accreting neutron star crusts presumed that iron is the sole product of nuclear burning in the upper atmosphere and is the only nucleus entering the neutron star crust . we find that the hydrogen burning on these rapidly accreting neutron stars is mostly via the rp process of wallace and woosley ( 1981 ) , resulting in a complicated mix of elements , _ nearly all much heavier than iron . _ the thin outer crust of a neutron star ( before neutron drip at @xmath18@xmath19 ) is replaced by accretion of only @xmath20 of material . thus all of the neutron stars in low - mass x - ray binaries should have an accreted outer crust . indeed these objects accrete enough material to unambiguously replace their _ whole _ crust ( down to the crust / core interface ) , which is typically a few percent of the total stellar mass ( ravenhall & pethick 1994 ) . the x - ray pulsars typically accrete from the wind of a massive companion at a rate @xmath21@xmath22 ( see bildsten et al . 1997 for a recent overview ) and so are capable of replacing their outer crusts . a few of these objects ( smc x-1 , lmc x-4 , her x-1 ) are accreting at high enough rates or for long enough times ( as inferred by the @xmath23 mass companion to her x-1 ) to replace their whole crust . knowledge of the composition and thermal properties of an accreting neutron star s ocean and crust is important for many studies . for example , the temperature and composition of the crust affects the thermal conductivity and the ohmic diffusion time there ( see brown & bildsten 1998 for a recent application / overview of this problem ) . the crustal composition determines the amount of heat deposited directly in the crust ( haensel & zdunik 1990 ; miralda - escude et al . 1990 ) and the rate of neutrino emission from the crust ( haensel , kaminker & yakovlev 1996 ) , both of which are important for finding the equilibrium core temperature of an accreting neutron star ( fujimoto et al . 1984 ; brown & bildsten 1998 ) . the predicted frequencies of the so - far unobserved ocean g - modes depend directly on the average nuclear mass ( bildsten & cutler 1995 ; bildsten & cumming 1998 ) . we begin in 2 with an introduction to the basic equations we solve and a summary of the input microphysics . in 3 , we discuss the high accretion rate burning regime and explain why the burning is thermally stable at high accretion rates and why the helium ignites in a hydrogen - rich environment , thus providing an excellent site for the rp process . in 4 , we explain the overall thermal and compositional structure of the burning layer and provide convenient analytic expressions for the temperature and depth of the burning . section 5 contains an in - depth discussion of the nature of the rp process . we explain how the burning depends on the local accretion rate , with particular emphasis on the important role of the @xmath8p process in determining the final average mass of the nuclei . we conclude in 6 with a summary of our work and some speculations about how these results will impact studies of the neutron star crust and ocean . in the appendix , we describe how we calculate the radiative and conductive opacities for the complex mixtures produced by the hydrogen / helium burning . the plane - parallel nature of the neutron star atmosphere means that the physics of nuclear burning depends on the accretion rate per unit area , @xmath24 ( fujimoto , hanawa & miyaji 1981 , hereafter fhm ) . this parameter determines the local behavior on the neutron star and need not be the same everywhere , especially on accreting x - ray pulsars . the local zero metallicity eddington rate ( the accretion rate at which the outgoing radiation exerts a force comparable to gravity ) is @xmath25 where @xmath26 is the thomson scattering cross - section , @xmath27 is the proton mass , @xmath28 is the speed of light , @xmath29 is the hydrogen mass fraction , and @xmath30 is the stellar radius . in this paper , we use the eddington accretion rate for solar composition ( @xmath31 ) and @xmath32 km , @xmath33 , as our basic unit for the local accretion rate . steady accretion modifies the equations of particle continuity and entropy for a hydrostatic settling atmosphere ( fhm ; brown & bildsten 1998 ) . we write these equations in a coordinate system of fixed pressure @xmath34 , so that the accreted matter flows through the coordinates as it is compressed by accretion of fresh material from above . hydrostatic balance ( the ram pressure of the accretion flow is negligible ) yields @xmath35 , where the column depth @xmath36 ( in g @xmath37 ) is defined by @xmath38 , and @xmath39 ( we neglect general relativistic corrections ) . the continuity equation for an element @xmath40 ( with number density @xmath41 ) is @xmath42 where @xmath43 is the summed rates of particle creation and destruction processes . for these high accretion rates , there is not time for differential settling and all elements co - move downward at the speed needed to satisfy mass continuity , @xmath44 ( wallace , woosley & weaver 1982 ; bildsten , salpeter & wasserman 1993 ) . we define a mass fraction @xmath45 where @xmath46 is the baryon number of species @xmath40 and expand the continuity equation to obtain @xmath47 the equation for the entropy is @xmath48 where @xmath49 is the energy release rate from nuclear burning and @xmath50 is the heat flux . we write the entropy as @xmath51 , where @xmath52 for an adiabatic change and @xmath53 is the specific heat at constant pressure . since the temperature can depend on both time and pressure , we find @xmath54 equations ( [ eq : contin ] ) and ( [ eq : entrop ] ) describe the hydrostatic evolution of the neutron star atmosphere while constantly accreting . in our steady - state calculations , we neglect the time - dependent terms in these equations . we solve the continuity and entropy equations in connection with a nuclear reaction network to determine the nuclear energy generation rate @xmath49 in equation ( [ eq : entrop ] ) and the nuclear abundances @xmath55 . the nuclear reaction network used here is described in detail in schatz et al . ( 1998 ) and the references therein ( see also herndl et al . 1995 , van wormer et al . 1994 , and wiescher et al . 1986 for discussions of the reaction rates ) . the network includes 631 nuclei between hydrogen and @xmath56sn covering the range from stability to the proton drip line . the types of reactions considered are all proton , neutron and @xmath8 induced reactions as well as photodisintegration , @xmath57 decay and electron capture . for most nuclei in the range @xmath58 and @xmath59 the temperature and density dependent weak interaction rates are taken from the compilation of fuller , fowler & newman ( 1980 , 1982a , 1982b ) . these weak interaction rate data include energy losses due to emission of neutrinos in electron captures and @xmath57 decays . since similar data for heavier nuclei are not available , the @xmath60-decay rates of nuclei with @xmath61 were approximated using temperature and density independent earth rates ( for a justification see schatz et al . energy loss via neutrinos is neglected for these nuclei . electron captures do not play an important role for the density regime in which we work . electron screening is treated according to graboske et al . ( 1973 ) and itoh et al . ( 1979 ) . for the equation of state , we assume the ions behave as an ideal gas ( coulomb corrections to the ion equation of state are unimportant at the column depths of interest to us ) and we use the interpolation formulae of paczynski ( 1983 ) to account for the partially degenerate electrons . the heat flux through the atmosphere is given by @xmath62 where @xmath63 is the radiation constant . the opacity @xmath64 is set by electron scattering , free - free absorption and conduction . in the appendix , we discuss in detail how we calculate each of these contributions for the complex mixtures produced by the hydrogen / helium burning . we assume an initial solar composition ( anders and grevesse 1989 ) of the accreted material and a neutron star with @xmath65 and @xmath66 . the influence of these parameters on our results is discussed later . the only boundary condition that has to be chosen is the radiation flux exiting the top of the atmosphere . for these steady - state models at high accretion rates , the flux is dominated by the nuclear energy release from the conversion of hydrogen and helium to heavy elements ( @xmath67 5 mev per nucleon ) . there is also energy release from gravitational settling and energy generated in the crust due to electron captures , neutron emission and pycnonuclear reactions ( see brown & bildsten 1998 for a discussion of this issue ) . however , these contributions to the flux are small ( @xmath68 ) compared to the nuclear energy release , and we do not include them here . in particular , we drop the terms proportional to @xmath53 on the right - hand side of equation ( [ eq : entrop ] ) . our approach is to guess the flux at the top of the atmosphere , integrate to the base of the burning layer and then compare the flux generated by the nuclear burning to our initial estimate . we find that this procedure converges to give the correct upper boundary condition in only a few iterations . for the high accretion rates of interest to us , the temperature at the depth where burning occurs is always in excess of @xmath69 k , so the hydrogen burns either via the hot cno cycle or , at the higher temperatures at later stages , via the rp process . during the hot cno cycle the timescale for proton captures is shorter than that for subsequent @xmath60 decays . the time to go around the catalytic loop is set by the @xmath60 decay lifetimes of @xmath70o ( @xmath71 s ) and @xmath72o ( @xmath73 s ) and is _ temperature independent_. the @xmath60-decays fix the hydrogen burning rate at ( hoyle & fowler 1965 ) @xmath74 where @xmath75 is the mass fraction of cno in the accumulating matter . the timescale to consume all of the hydrogen is then @xmath76 or about one day for solar metallicity . for high accretion rates ( especially when the burning is stable ) , the matter reaches high enough temperatures for helium ignition before the hydrogen has been exhausted . the helium burning then occurs in a hydrogen rich environment ( lamb & lamb 1978 ; taam & picklum 1978 , 1979 ; fhm ; taam 1982 ) . for normal or sub - solar metallicities and accretion rates in the regime @xmath77 , the helium ignition is thermally unstable and results in a type i x - ray burst . at higher accretion rates , the helium burns at a temperature ( @xmath5 ) where the burning is _ thermally stable _ because the nuclear energy generation rate is less temperature sensitive than the radiative cooling ( fhm ) . this allows a rough estimate of the critical accretion rate @xmath78 where the burning becomes stable ( bildsten 1998b ) , @xmath79 in which we take a nominal value @xmath80 , where @xmath81 is the thomson scattering opacity . bildsten ( 1998b ) recently reviewed the accretion rate regimes originally introduced by fhm . the high local accretion rates of the x - ray pulsars are the standard argument as to why they do nt burst ( joss & li 1980 , bildsten & brown 1997 ) . the simple calculation which gives equation ( [ eq : mdotsted ] ) is not accurate enough to say whether @xmath82 is above or below the eddington rate . fhm estimated @xmath83 ( they called it @xmath84 in their table 1 ) for 3 different neutron stars ( here @xmath85 ) . they found @xmath86 for @xmath87 and @xmath88 . equation ( [ eq : mdotsted ] ) agrees to within 2030% , giving @xmath89 for the same three masses and radii . to accurately determine if the burning is thermally stable requires solving the time - dependent equations numerically . ayasli & joss ( 1982 ) and taam , woosley & lamb ( 1996 ) performed time - dependent calculations at accretion rates comparable to the eddington limit and found that the burning is thermally unstable when @xmath90 . ayasli & joss ( 1982 ) also carried out some calculations with super - eddington accretion rates ( by a factor of a few ) and found thermal stability . however , there has not been a thorough survey of the dependence of the critical accretion rate @xmath91 on @xmath92 , @xmath30 and composition . our focus in this paper is calculating the nucleosynthesis when the burning is thermally stable . in this regime , the helium is consumed over a time interval of less than 20 minutes and the resulting enhancement of the number of cno seed nuclei together with breakout from the cno cycle leads to a rapid consumption of hydrogen via the rp process . steady - state burning of hydrogen and helium takes place in three phases : hydrogen burning via the hot cno cycle at low densities ( [ burn ] ) , mixed hydrogen and helium burning via the 3@xmath8 reaction and the @xmath8p and rp processes when helium ignites ( [ burn_ign ] and [ burn_hhe ] ) , and pure helium burning via @xmath8 captures after the exhaustion of hydrogen ( [ burn_he ] ) . for all accretion rates where the burning is in steady - state , consumption of hydrogen by the hot cno cycle is unimportant before the helium ignites . it is helium ignition which acts as the trigger for hydrogen burning . to understand this , we write the depth at which hydrogen would be consumed by the hot cno cycle as @xmath93 , or , using equation ( [ eq : cno_burn ] ) , @xmath94 for the high accretion rates of interest here , @xmath95 is always much greater than the column depth at which helium ignites , @xmath96 ( see the estimate of @xmath96 in eq . [ [ eq : yburn ] ] below ) . thus the helium ignites before significant amounts of hydrogen are burned by the hot cno cycle that when breakout occurs at high accretion rates , the limited number of available seed nuclei and the short time remaining before the onset of the 3@xmath8 reaction limits the hydrogen consumption via breakout reactions prior to helium ignition to only 10% of the available fuel . ] . rapid consumption of hydrogen by the rp process then follows . we have calculated steady - state models for local accretion rates between @xmath97 and 60 . the temperature , density and composition as a function of column depth are shown in figure [ fig : profiles ] for four accretion rates , @xmath98 and 50 , which we will discuss in detail . the steep rise in the mean molecular weight indicates the location of most of the hydrogen burning . figure [ fig : opac ] shows the opacity @xmath64 for these models . the opacity `` bump '' , particularly prominent at low accretion rates , is due to free - free absorption . this dominates electron scattering after the heavy elements are made and until the opacity is set by electron conduction ( see appendix for a full discussion ) . figure [ fig : trho ] shows the temperature - density profiles . the region to the left of the dashed line is where the temperature is high enough so that the positron fraction exceeds 10% . we have safely stayed in the regime where we can neglect positrons , confining our calculations to @xmath99 . we show later that this limit coincides with the onset of nuclear statistical equilibrium , where the outcome of the nucleosynthesis is simply determined by thermodynamics and does not need to be calculated in detail . figure [ fig : abut ] shows the abundances of hydrogen and helium as a function of time and column depth . it is clear that nearly all the hydrogen burning occurs where the helium ignites . we make a simple estimate of this location by matching the helium lifetime to the @xmath100 reaction to the time it takes to cross a scale height . using the energy generation rate for helium burning via the 3@xmath8 reaction ( @xmath101 ) , we write @xmath102 ( taam 1981 , fushiki & lamb 1987 , bildsten 1995 ) where @xmath103 is the energy release from @xmath104c . using the temperature - density relation for an atmosphere with pure thomson scattering and a constant flux @xmath105 , we find that the helium ignition temperature is given by the solution to @xmath106 \exp\left({-44\over t_8}\right)\ ] ] ( bildsten 1998b ) , where we write the energy released per gram of accreted material in units of @xmath107 as @xmath108 and the accretion rate in units of @xmath109 as @xmath110 . solving this transcendental equation for the temperature gives the pressure at which the helium burns and , most importantly , how these quantities depend on @xmath24 , @xmath92 , @xmath30 and the abundances . in order to simplify the transcendental , we expand the exponential about the temperature of @xmath111 ( in the middle of the range we are interested in ) so that @xmath112 . some accuracy is compromised because of this approximation , but not much . we thus find @xmath113 and @xmath114 as the temperature and column density at the helium burning location . these estimates compare well with our numerical results ( we take @xmath115 , @xmath116 , @xmath117 , @xmath31 , and @xmath118 ) . equation ( [ eq : yburn ] ) accurately gives the location where @xmath119% of the helium is burned . the temperature estimate is consistently 20% too high ( for the same location ) . we have also checked the dependence of the burning conditions on gravity for a fixed @xmath120 model . we find that the burning takes place at higher temperatures , lower densities and higher column depths for a lower gravity , confirming the gravity dependence of equations ( [ eq : tburn ] ) and ( [ eq : yburn ] ) . thus our simple estimate accurately reproduces the scalings seen in the numerical simulations . when the helium ignites , mixed hydrogen and helium burning occurs via the 3@xmath8 reaction , the @xmath8p process and the rp process . the ignition of @xmath100-burning and subsequent carbon production in a hydrogen rich environment provides additional seeds which accelerate the burning of hydrogen . the @xmath8p process is a series of ( @xmath8,p ) reactions and proton captures starting with the @xmath70o(@xmath8,p)@xmath121f reaction . this is a pure helium burning process since for each proton released in an ( @xmath8,p ) reaction a proton is captured in a subsequent ( p,@xmath122 ) reaction ( wallace and woosley 1981 ) . hydrogen must be present for the @xmath8p process to play a role , since proton capture reactions provide the link between the @xmath123c produced by the 3@xmath8 reaction , and the @xmath70o at the beginning of the @xmath8p process . most of the hydrogen is burned by the rp process ( rapid proton capture process ) , a series of fast proton captures and slow @xmath60-decays close to the proton drip line ( wallace and woosley 1981 ) . we show later that the typical mass of nuclei at the endpoint of the rp process ( @xmath124 ) is around 60100 and is determined both by the exhaustion of hydrogen and the prevalence of the @xmath8p process . before describing our detailed results , we start with a simple picture of the mixed hydrogen and helium burning . imagine that helium burns only via the 3@xmath8 reaction and breakout from the hot cno cycle , for example via the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction . then , @xmath125ne is the seed nucleus for hydrogen burning via the rp process . neglecting the small amount of hydrogen burned before helium ignition , for solar abundance there are 38 protons per @xmath125ne seed nucleus . if there are no catalytic loops , the ensuing rp process would produce nuclei with a typical mass @xmath126 after all the hydrogen is burned . we generalize this simple estimate to the case where helium is burning via the @xmath8p process , by assuming that only one endpoint nucleus is produced from the @xmath8p process ( mass number @xmath127 ) , which again serves as the seed nucleus for the rp process . the typical mass @xmath124 of the endpoint of the rp process then depends on the mass fraction of helium burned into seed nuclei for the rp process ( @xmath128 ) and the endpoint of the helium burning ( @xmath127 ) , @xmath129 ( schatz et al . thus the strong temperature sensitivity of the @xmath130,p ) reactions in the @xmath8p process is imprinted on the rp process ashes . in this section , we first describe the physics that determines @xmath127 and show that the temperature at the depth of helium ignition is the most important parameter . we then discuss the rp process and the resulting distribution of elements , and show that equation ( [ eq : arp ] ) is a good estimate of our numerical results . crucial for the rp process is the extent to which @xmath8-induced reactions other than the 3@xmath8 reaction can take place . the major limiting factor is the coulomb barrier for the @xmath8 particles , which increases rapidly towards heavier nuclei . in figure [ tr ] , the dotted line shows the temperatures and densities where the @xmath72o(@xmath8,@xmath122)@xmath125ne rate equals the @xmath72o @xmath57 decay rate . to the right of this line , breakout from the hot cno cycle can occur . the dashed lines show where the ( @xmath8,p ) reaction rate on @xmath70o , @xmath131ne , @xmath132 mg , @xmath133si , @xmath134s and @xmath135ar equals the rate of destruction ( by @xmath57 decay and proton capture ) of each of these isotopes . the @xmath136 process will proceed via a particular @xmath130,p ) reaction if the temperature and density of a fluid element reach the corresponding threshold ( dashed line ) during helium burning . otherwise , the @xmath136 process will end and the reaction flow towards heavier nuclei will continue via the rp process . for nuclei above @xmath131ne proton captures can in principle compete with @xmath8-induced reactions as the proton capture rates are typically much faster than the ( @xmath8,p ) reaction rates . however , at the temperatures where the helium burning occurs , the ( @xmath122,p ) photodisintegration reactions are in balance with the ( p,@xmath122 ) reactions , which strongly hampers effective proton captures . this does not mean , as it is often concluded , that proton captures are always negligible . in fact , the deviations of the thresholds shown in figure [ tr ] from the smooth behavior at low densities are due to proton captures . for example , let s consider @xmath132 mg . in this case , the nuclei @xmath132 mg and @xmath137al are in ( p,@xmath122)-(@xmath122,p ) equilibrium determined by the saha equation . however , the @xmath137al(p,@xmath122)@xmath138si reaction operating on the low @xmath137al equilibrium abundance establishes a net reaction flow from @xmath132 mg to @xmath138si . this is similar to the 2p capture reactions discussed in grres et al . ( 1995 ) and schatz et al . this type of proton capture is a weak process but goes with @xmath139 , and therefore there is a critical density above which the ( @xmath8,p ) reaction competes with proton captures rather than with the density independent @xmath60 decay . for @xmath132 mg this happens at densities above @xmath140 . as can be seen in figure [ tr ] , the threshold for the ( @xmath8,p ) reaction behaves very differently in that regime because of the very different density and temperature dependence of the competing reaction . for @xmath132 mg , @xmath133si , @xmath134s , and @xmath135ar , the critical density is such that proton captures play an important role in determining the @xmath8p process path at these isotopes . for example , proton captures move the border for the @xmath132mg(@xmath8,p)@xmath141al reaction to higher temperatures , which is why the @xmath8p process needs accretion rates above 20@xmath142 to proceed beyond @xmath132 mg . proton captures also lead to an inversion of the thresholds for the @xmath133si(@xmath8,p)@xmath143p and the @xmath134s(@xmath8,p)@xmath144cl reactions for densities above @xmath145 . on the other hand , in the cases of @xmath70o and @xmath131ne the next isotone is proton unbound and the necessary densities are much higher , roughly @xmath146 ( grres et al . 1995 ) . the solid lines in figure [ tr ] are the evolutionary tracks for fluid elements in the temperature - density plane at various accretion rates . the fat line segment indicates where hydrogen burning occurs ( from 90% to 10% of the initial hydrogen abundance ) and clearly shows that the temperature - density track for all accretion rates crosses the threshold for the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction before exhausting the hydrogen . thus at all accretion rates , there is a breakout of the hot cno cycle . there are uncertainties in the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction rate by factors of 510 ( mao et al . 1995 , 1996 , wiescher , grres & schatz 1999 ) . however , the strong temperature dependence of the reaction rate means that changes in the reaction rate within the uncertainties have hardly any impact on the reaction flow and the breakout of the hot cno cycle . to confirm this , we have performed a test calculation with @xmath97 and the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction rate reduced by a factor of ten . the only difference we find in this case is that the column depth of the burning zone is deeper by 8% resulting in a slightly longer cno burning phase before the breakout . as we discuss in [ burn_he ] , this slightly increases the amount of helium that survives hydrogen burning . _ therefore we conclude that steady state hydrogen burning must always take place via the rp process . _ the time - integrated reaction flows for accretion rates of @xmath147 , 5 , 20 , and 50 are shown in figure [ flow ] . at accretion rates below @xmath148 , the only important @xmath8-induced reaction ( apart from the 3@xmath8 process ) is @xmath72o(@xmath8,@xmath122)@xmath125ne , which triggers the breakout of the hot cno cycle into the rp process starting at @xmath125ne . the @xmath70o(@xmath8,p)@xmath121f reaction requires somewhat higher temperatures and starts to occur at @xmath149 . this reaction does not lead to a breakout of the hot cno cycle but instead opens up another loop that proceeds via @xmath121f(p,@xmath122)@xmath131ne(@xmath57)@xmath131f(p,@xmath8)@xmath72o . this loop just provides another path from @xmath70o to @xmath72o having the net effect of a @xmath60 decay and a proton capture . the major hydrogen burning at @xmath150 still proceeds via the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction and the subsequent synthesis of heavier elements via the rp process . the picture changes at higher accretion rates , however , where the @xmath131ne(@xmath8,p)@xmath151na reaction becomes the dominant breakout of the hot cno cycle . the reaction flow from the cno region towards heavier elements is then dominated by the @xmath8p process starting with the @xmath70o(@xmath8,p)@xmath121f reaction . as can be seen in figure [ tr ] , as @xmath24 increases the ignition temperature gets higher and more ( @xmath8,p ) reactions become possible . this is confirmed in figure [ mrange ] , which shows the mass number of the last isotope made in the @xmath8p process @xmath152 as a function of @xmath24 . this mass increases with accretion rate and for @xmath153 the @xmath8p process does not end until after @xmath154sc . the rapid increase in @xmath152 at @xmath155 reflects the small temperature differences between the thresholds for the various ( @xmath8,p ) reactions at @xmath156 ( figure [ tr ] ) . the inversion of the @xmath133si(@xmath8,p)@xmath143p and the @xmath134s(@xmath8,p)@xmath144cl reaction thresholds is reflected in the jump in @xmath152 from 26 to 34 just above 20@xmath157 . as mentioned earlier , these effects are a consequence of the competition between proton captures and ( @xmath8,p ) reactions . figure [ abut2 ] shows the abundances of hydrogen , helium , carbon , oxygen , nickel and the last major isotope produced in the rp process for several accretion rates as a function of time . also shown is the abundance of one of the isotopes that dominates the composition at the end of our calculation . at @xmath158 , the cno cycle opens only at @xmath72o , which causes the slight drop in the @xmath72o abundance and the build up of @xmath159ni after about 200 s ( @xmath159ni is the first major waiting point in the rp process path ) . the destruction of @xmath72o is then delayed by the conversion of @xmath70o into @xmath72o via the hot cno cycle . at higher accretion rates the opening of the cno cycle is reflected in the rapid depletion of @xmath70o and @xmath72o with the simultaneous production of heavier nuclei like @xmath159ni . there is a brief period of time between the breakout of the cno cycle and the onset of the 3@xmath8 reactions supplying additional cno nuclei . this is best seen for @xmath160 , where the first peak in the @xmath159ni abundance comes from processing of the initial cno abundances , while the second peak at 20 s indicates the onset of the 3@xmath8 reaction supplying more seed nuclei . the time between the breakout from the cno cycle and helium ignition ranges from roughly 260 s at @xmath158 to 20 s at @xmath161 . however , the amount of hydrogen burned in this period via the rp process is only about @xmath162% of the initial hydrogen abundance because of the relatively long timescales for rp processing . therefore , helium ignition always takes place in a hydrogen - rich environment . the build up of the last isotope in the rp process path ( dot - dashed line in figure [ abut2 ] ) coincides with the depletion of hydrogen . the peak abundances reached for these isotopes are quite low as they are determined by the competition between production by the rp process and rapid destruction via fast @xmath60 decays . once the hydrogen is gone , the isotopes along the rp process path quickly decay into more stable isobars via a series of @xmath60 decays . the resulting build up of the most abundant nuclei at the end of our calculation is shown as dotted lines in figure [ abut2 ] . some of these nuclei are still unstable and will @xmath60 decay on longer timescales at greater depths . the nuclei produced by helium burning are the seeds for burning the hydrogen via the rp process . as discussed above , the burning conditions are set by the point of helium ignition . for each accretion rate , the thick line segment in figure [ tr ] shows the region of the temperature - density plane where the hydrogen burns . the density is roughly the same for all accretion rates and increases from about @xmath145 to @xmath163 during the burning . on the other hand , the burning temperature depends strongly on the accretion rate and increases from @xmath164 k at @xmath165 to @xmath166@xmath167 k at @xmath168 . the timescale for hydrogen burning is given in table [ endpoints ] and varies between 100 and 500 s. figure [ mrange ] shows the mass number of the rp process endpoint nucleus , @xmath124 , as a function of @xmath24 . the corresponding reaction flows for some selected cases are in figure [ flow ] . clearly , @xmath124 increases with @xmath24 and , for @xmath169 , the rp process reaches the end of our reaction network near @xmath170 . as discussed in [ summary ] we expect that for accretion rates above @xmath171 ( but less than @xmath172 , see discussion below ) significant amounts of nuclei heavier than @xmath170 are produced . the rp process reaction path in figure [ flow ] is characterized by a sequence of proton captures and @xmath60-decays and proceeds above @xmath159ni along the @xmath173 line as discussed in schatz et al . ( 1998 ) . however , we find that two - proton - capture reactions on @xmath174se and @xmath175kr do not play a role in steady state burning owing to the somewhat lower densities and temperatures compared to the x - ray burst peak conditions discussed in schatz et al . some branchings occur when proton captures are slowed down by photodisintegration or , at lower accretion rates and temperatures , because of small reaction rates . at accretion rates above @xmath176 the temperatures get so high ( @xmath177 ) that photodisintegration starts to severely hamper the rp process . at @xmath178 , photodisintegration of @xmath179mo inhibits further proton captures on @xmath180nb . as a consequence , the @xmath180nb(p,@xmath8)@xmath181zr reaction becomes the dominant destruction mechanism of @xmath180nb and the rp process ends in the zr - nb cycle . this causes the drop in @xmath124 in figure [ mrange ] at high @xmath24 s . as discussed in schatz et al . ( 1998 ) the zr - nb cycle occurs at high temperatures because of the very low @xmath8 binding energy of @xmath179mo . this low @xmath8 binding energy is predicted by the frdm ( 1992 ) mass model ( mller et al . 1995 ) used in our study . an experimental confirmation of this would be highly desirable . at accretion rates beyond @xmath182 , the temperatures get so high that photodisintegration drives the material into nuclear statistical equilibrium . the nuclei produced in the rp process are then quickly converted into iron peak isotopes , mostly @xmath159ni . _ it is only in this very high accretion rate regime ( @xmath183 ) that we find a pure nickel solution that will later decay to make the ashes nearly pure iron . _ what determines the endpoint of the rp process for accretion rates @xmath184 ? the rp process ends either when all the hydrogen is consumed or when the proton capture rates become too slow because of the increasing coulomb barrier . the effect of the coulomb barrier can be estimated from the formula given in woosley and weaver ( 1984 ) @xmath185 \frac{t_9^{1/3}}{4.25 - 1.33 t_9^{1/3}},\ ] ] where @xmath186 is the maximum charge number that can be synthesized in a time @xmath187 at a density @xmath188 , a temperature @xmath189 ( in 10@xmath190 k ) and a proton abundance @xmath29 . we find that , for typical conditions during hydrogen burning and using our timescale for hydrogen burning from table 1 , the @xmath186 is always higher than the heaviest nucleus produced in our calculations : for @xmath191 , 5 , 20 , 50 we find @xmath192 , 45 , 61 , and 92 . _ therefore , the coulomb barrier does not limit the rp process in steady state burning at high accretion rates . _ this is also confirmed from the reaction flows in figure [ flow ] . the flows are always close to the proton drip line all the way up to the heaviest nucleus synthesized , indicating that it is the @xmath60-decays and not the proton captures that set the timescale . the deviation of the rp process path from the proton drip line by 1 or 2 mass units is in most cases due to photodisintegration of the weakly - bound proton - rich nuclei . we find that hydrogen is completely consumed by the rp process near the depth where helium ignites . no hydrogen reaches deeper regions of the atmosphere , and electron capture on hydrogen does not occur when the burning is in steady - state , as earlier speculated by taam et al . the only exception is at very high accretion rates , typically beyond @xmath193 , where photodisintegration hampers efficient hydrogen burning via the rp process beyond @xmath159ni . indeed , figure [ flow ] shows that , at @xmath168 , there is a very weak reaction flow via hydrogen electron capture . but even at @xmath194 the hydrogen mass fraction that is left after the rp process burning is 0.4% . these results contradict the speculations of taam et al . ( 1996 ) about the possible stabilizing effects of hydrogen electron captures at large depths . a likely explanation for this is that the network of taam et al . did not include nuclei with @xmath195 however , from the simple arguments giving equation ( 13 ) , it is clear that the rp process during steady - state burning always leads to production of nuclei beyond iron . thus the omission of nuclei beyond @xmath196 would result in under - consumption of hydrogen , and lead one to the incorrect conclusion that hydrogen survives the nuclear burning . since it is hydrogen consumption that limits the rp process , the endpoint is roughly given by equation ( [ eq : arp ] ) . the important parameters are the endpoint of the @xmath8p process and the amount of helium burned into seed nuclei during hydrogen burning ( @xmath197 ) . we have shown that @xmath152 can in principle be obtained from figure [ tr ] , but @xmath128 is more difficult to estimate , since some mass fraction of helium ( @xmath198 ) survives the hydrogen burning phase ( see table [ endpoints ] ) . this helium is not available for the production of rp process seed nuclei and burns later in the pure helium burning phase discussed in [ burn_he ] . table [ endpoints ] shows that the amount of helium surviving hydrogen burning is 23% of the initial helium abundance at @xmath199 . helium survives the hydrogen burning because the timescale to burn helium via the 3@xmath8-reaction , @xmath200 , is longer than the timescale for reaching the last isotope in the rp process , @xmath201 , which is the sum of the lifetimes of all the nuclei along the calculated reaction paths . figure [ tau ] shows @xmath200 and @xmath201 as a function of mass number for @xmath202 and 20 . when calculating @xmath201 , we neglect the influence of photodisintegration and of the change in hydrogen abundance during the burning . this is a reasonable approximation since the rp process timescale is mostly set by @xmath57 decays . only the proton captures at @xmath203fe and @xmath159ni contribute to the rp process timescale at low accretion rates . at low @xmath24 , the timescale for helium burning is indeed significantly longer than the timescale for the rp process to reach its endpoint and burn all the hydrogen . at higher accretion rates , the two timescales become comparable , and the amount of helium remaining unburned after hydrogen exhaustion is less . the final abundance distribution of the rp process ashes from steady state burning is shown in figure [ abu ] for @xmath204 , 5 , 20 , 50 , and 60 . as discussed above , the higher the accretion rate , the heavier the nuclei produced until , at @xmath205 , the rp process gets stuck in the zr - nb cycle at @xmath206 . at this accretion rate , nuclear statistical equilibrium starts to play a role during the final burning stage , and drives some nuclei back into @xmath159ni . these two effects lead to the `` double peak '' structure of the abundance pattern with maxima around @xmath196 and @xmath206 . at @xmath207 , nuclear statistical equilibrium dominates and all nuclei are converted into iron peak nuclei , mainly @xmath159ni . the abundance distribution for @xmath208 is determined by the nuclei with the slowest reaction rates in the rp process path , the so - called `` waiting points '' ( wallace and woosley 1981 ) . some fraction of the material is locked into these waiting points until the burning is over and this leads to the production of a wide range of isotopes . the important waiting points for the conditions discussed here have been identified in schatz et al . ( 1998 ) and are the even - even @xmath173 nuclei @xmath159ni , @xmath209ge , @xmath174se , @xmath175kr , @xmath210sr and @xmath181zr . for these nuclei , proton captures are inefficient since photodisintegration or proton decay can remove a captured proton quickly and @xmath60-decay rates are relatively low . this explains the local maxima in the abundance distributions ( for @xmath211 ) at the mass numbers of these waiting point isotopes . additional peaks occur at the nuclei that are constructed by @xmath8 capture on @xmath123c as discussed in [ burn_he ] . at @xmath212 , an additional peak at @xmath213 occurs because of the relatively slow @xmath203fe(p,@xmath122)@xmath214co reaction rate at low temperatures ( this can also be seen in figure [ tau ] , where for low accretion rates the first major increase of the rp process timescale occurs at @xmath213 ) . the flat and structureless abundance distribution above @xmath215 in the @xmath216 case is caused by the fact that the rp process barely leaks beyond the major waiting points @xmath174se and @xmath175kr . as can be seen in figure [ flow ] , the drastically reduced processing timescale beyond @xmath175kr ( see schatz et al . 1998 ) distributes the reaction flow over a wide range of nuclei as hydrogen is rapidly depleted . in figure [ mrange ] , we show the mass number of the last isotope reached in the @xmath8p process ( @xmath127 , open circles ) and in the rp process ( @xmath124 , open squares ) as a function of @xmath217 . also shown is the average mass number ( @xmath218 , filled circles ) and the average charge number ( @xmath219 , filled triangles ) of the final composition . we now test equation ( [ eq : arp ] ) for accretion rates @xmath220 by assuming @xmath221 and by taking @xmath198 from table [ endpoints ] . the resulting estimated endpoints of the rp process are shown in figure [ mrange ] as open triangles . as expected the estimated endpoint is a few mass numbers too high , since some hydrogen is burned during the beginning of the burning phase when temperatures are lower and the @xmath8p process can not yet reach its final endpoint . thus the estimate of equation ( [ eq : arp ] ) gives an upper limit to the rp process endpoint in the accretion rate regime between 15 and 45 @xmath148 where the reaction flow reaches the end of our network . figure [ mrange ] shows that the heaviest nuclei that can be produced by the rp process in steady state burning are estimated to be in the @xmath222 region . at these mass numbers the proton drip line is around @xmath223 and applying equation ( [ eq : zmax ] ) we find that the coulomb barrier should not inhibit the rp process even for these heavy nuclei . the endpoint of the rp process might be considerably below this upper limit because of loops in the reaction path that might occur above @xmath56sn as the reaction path enters a region of nuclei with very low @xmath8 binding energies . _ to summarize , the synthesis of much heavier isotopes at higher accretion rates is not because of faster proton capture rates , but due to enhanced ( @xmath8,p)-reactions that lead to a longer @xmath8p process and therefore to heavier seed nuclei and a larger fuel to seed ratio for the rp process . _ the final abundance distributions as a function of mass number are shown in figure [ abu ] for various accretion rates . _ generally , not just a single nucleus but a whole range of isotopes is produced by steady state burning_. even for our lowest accretion rate of @xmath224 significant amounts of nuclei heavier than @xmath159ni are synthesized . the mixed hydrogen and helium burning is responsible for nearly all of the released nuclear energy . the energy released by the cno cycle before helium ignition ( [ burn_ign ] ) and by the helium burning after hydrogen consumption ( [ burn_he ] ) is negligible . table [ endpoints ] lists the energy produced by nuclear burning for different accretion rates . hydrogen and helium are completely burned in all of our models . because of the weak dependence of nuclear binding energy with mass number in the range of the rp process endpoints reached in this study , the released energy depends only weakly on accretion rate and burning conditions . this energy is predominantly produced by hydrogen burning . helium burning does not contribute more than 0.6 mev / nucleon ( 3@xmath8 only ) to 1.2 mev / nucleon ( @xmath8p process up to @xmath225k ) . it was pointed out previously ( in [ sec : rp ] ) that the relatively long timescale for helium burning allows some helium to survive the hydrogen burning phase . the mass fractions of unburned helium , @xmath198 , for various accretion rates are listed in table [ endpoints ] . this residual helium burns later in a hydrogen - free environment via the 3@xmath8 reaction and , depending on the temperature , via a series of @xmath8 captures starting at @xmath123c . figure [ abut2 ] shows the build up of @xmath123c right after the hydrogen is burned , and , at accretion rates above @xmath226 , the depletion of @xmath123c by the subsequent @xmath123c(@xmath8,@xmath122)@xmath227o reaction . following the @xmath123c(@xmath8,@xmath122)@xmath227o reaction , a sequence of @xmath8 captures occurs that can be identified in figure [ flow ] . this @xmath8 capture chain is longer for the higher temperatures at higher accretion rates , although less helium is then available ( see table [ endpoints ] ) . at @xmath228 , the @xmath8 capture chain reaches @xmath229ti . the nuclei produced during the pure helium burning phase can be identified in the abundance patterns shown in figure [ abu ] and are @xmath123c and @xmath138 mg for @xmath226 , @xmath123c , @xmath230si , and @xmath231s for @xmath232 and @xmath233ca and @xmath229ti for @xmath228 . the final mass fraction of @xmath123c for accretion rates of @xmath234 and 5 is 4.1% and 0.23% , and is essentially zero for higher accretion rates . the final amount of @xmath123c is much larger at low accretion rates , because more helium is available after hydrogen exhaustion ( table [ endpoints ] ) and at lower temperatures much less @xmath123c is destroyed by the @xmath123c(@xmath8,@xmath122)@xmath227o reaction . these conclusions depend somewhat on the rates for the cno breakout reactions . for low accretion rates , where most @xmath123c is produced , this is the @xmath72o(@xmath8,@xmath122)@xmath125ne reaction . in principle , a lower breakout rate leads to more hydrogen burning via the hot cno cycle , in which helium is produced ( in contrast to the rp process ) , giving a higher final helium abundance after hydrogen burning . however , we find that a factor of ten lower @xmath72o(@xmath8,@xmath122)@xmath125ne reaction rate increases the final @xmath123c mass fraction by just 10% ( a factor of ten higher rate reduces it by 25% ; the uncertainty in the reaction rate is a factor of 510 , see [ sec : break ] ) . we conclude that the final mass fraction of @xmath123c produced in steady state burning will never exceed 6% . it thus seems unlikely that enough @xmath123c is produced to release additional energy via explosive carbon burning deeper in the atmosphere as discussed by brown & bildsten ( 1998 ) . we have fully explored the complicated nucleosynthesis from thermally stable mixed hydrogen / helium burning at high accretion rates on accreting neutron stars . our major finding is that the rp process produces a mixture of very heavy elements , the average mass of which depends on the local accretion rate . the important reaction sequences during the mixed hydrogen and helium burning are the 3@xmath8 reaction , the @xmath8p process and the rp process . for all accretion rates where the nuclear burning is in steady state , a breakout of the hot cno cycles into the rp process takes place . it is the endpoint of the rp process that determines the energy generation and final composition of the ashes . in contradiction to taam et al . ( 1996 ) , we find that no hydrogen survives the steady state burning and consequently deep hydrogen burning by electron capture does not take place ( the possibility still remains that there is substantial residual hydrogen from unstable burning in x - ray bursts ) . for @xmath123c the situation is similar : for accretion rates of a few times eddington , some @xmath123c survives the burning , but never enough to trigger carbon flashes via exposive burning in deeper layers of the atmosphere . the most important nuclear physics input parameters for our steady state burning calculations are 1 ) the 3@xmath8 reaction rate , which triggers the burning ; 2 ) the breakout reactions from the hot cno cycles @xmath72o(@xmath8,@xmath122)@xmath125ne and @xmath131ne(@xmath8,p)@xmath151na , as discussed in 5.1 ; 3 ) the ( @xmath8,p ) and ( p,@xmath122)- reaction rates on @xmath70o , @xmath132 mg , @xmath133si , and @xmath134s together with the proton separation energies and the proton capture rates of @xmath137al , @xmath235p , @xmath236cl , and @xmath237k ( these data affect the extent of helium burning and therefore the endpoint of the rp process ) ; 4 ) the proton capture rates on the waiting points @xmath203fe and @xmath159ni ; and 5 ) proton capture q - values and @xmath60-decay half lives of the even @xmath238 , @xmath173 and the even @xmath238 , @xmath239 nuclei between @xmath159ni and @xmath56sn as discussed in schatz et al . while the 3@xmath8 reaction is known with sufficient accuracy , most of the other data are completely or partially based on theoretical data , for which extrapolation to the very neutron deficient nuclei in the rp process is often doubtful ( see schatz et al . 1998 for a more detailed discussion ) . however , while these uncertainties may affect the detailed abundance pattern for a given local accretion rate , they have no influence on our general conclusions concerning the nature of steady state burning on accreting neutron stars ( see 5.1 ) . nevertheless , more experimental information on the nuclear data mentioned above would certainly be desirable . our calculations of the composition of the nuclear ashes in steady - state burning show that the ocean and crust of an accreting neutron star do not consist of pure iron , as assumed in previous work . instead , the final composition consists of a wide range of nuclei . this is characteristic of the rp process in which some fraction of nuclei is locked at a large number of waiting points with long lifetime . our results will have interesting consequences for studies of the crust of accreting neutron stars . for example , our discovery of the large range of nuclei present will directly impact estimates of the thermal and electrical conductivity of the crust . to illustrate this point , we have calculated the `` impurity parameter '' @xmath240 for our models , where @xmath241 are the nuclear abundances and the subscript max indicates the most abundant species . we find that @xmath242 is typical of the mixture from the ashes of steady state burning . even at very high accretion rates , ( @xmath243 ) when nuclear statistical equilibrium favors @xmath159ni as the sole product of nucleosynthesis , @xmath244 . brown & bildsten ( 1998 ) showed that for @xmath245 the thermal and electrical conductivity is dominated by impurity scattering , which is strongly composition dependent . in previous work which assumed a crust of pure iron ( before electron captures ) , impurity scattering was unimportant . thus in accreting neutron stars , the thermal and electrical crust conductivities will be in general much lower than previously assumed , leading to a different thermal structure and faster ohmic diffusion of magnetic fields in the crust . we thank ed brown and felix rembges for many discussions during this work . we are especially grateful to f .- k . thielemann and t. rauscher for providing the reaction network solver and many of the reaction rates . this research was supported by nasa via grants nag 5 - 2819 and nagw-4517 . l. b. was supported by the hellman family faculty fund award ( ucb ) and the alfred p. sloan foundation . electron scattering , free - free absorption and conduction all play a role in setting the opacity at different depths in the neutron star atmosphere . in this appendix , we describe how we calculate each of these contributions . prior to h / he burning , the flux in the atmosphere is carried by radiation and the opacity is set by thomson scattering , @xmath246 where @xmath29 is the hydrogen mass fraction . there are corrections to @xmath247 due to degeneracy and high temperatures for which we use paczynski s ( 1983 ) fit to the results of buchler & yueh ( 1976 ) . this formula is valid for @xmath248 , where @xmath249 is the electron fermi energy excluding the rest mass . we find in our models that when electron scattering dominates the opacity we are safely in the @xmath250 regime . as the hydrogen and helium are burned , electrons are consumed and the average nuclear charge of the nuclei increases . this results in free - free absorption becoming much more important than electron scattering . the free - free opacity is given by @xmath251 ( clayton 1983 ) , where the sum is over all nuclear species and @xmath252 . the dimensionless free - free gaunt factor @xmath253 takes into account the dependence of the opacity on coulomb wavefunction corrections , degeneracy and relativistic effects . itoh et al . ( 1991 ) have calculated @xmath253 for pure hydrogen , helium , carbon and oxygen plasmas . so as to implement their work , we devised the fitting formula @xmath254 \left[{1-\exp(-2\pi\gamma/\sqrt{\pi+u})\over 1-\exp(-2\pi\gamma/\sqrt{\pi})}\right ] \left[1+\left({t_8\over 7.7}\right)^{3/2}\right],\ ] ] where @xmath255 , @xmath256 , @xmath257 @xmath258 and @xmath259^{2/3}.\ ] ] in the regime @xmath260 , our fitting formula agrees with their table to better than 10% for @xmath261 and 20% for @xmath262 . in the regime @xmath263 and for @xmath264 , the agreement is better than 20% for hydrogen and within a factor of two for oxygen . this accuracy is adequate for our application as the @xmath265 regime is only relevant when the composition is mostly hydrogen , in which case electron scattering dominates . the highest @xmath238 element considered by itoh et al . ( 1991 ) is oxygen ( @xmath266 ) . we assume in our work that our formula , which includes the important scalings , is applicable to elements with greater @xmath238 . to calculate @xmath249 , we use the analytic fitting formulae of antia ( 1993 ) and chabrier & potekhin ( 1998 , eq . [ 24 ] ) . the three terms in equation ( [ eq : gff ] ) are physically motivated . the first is the degeneracy correction to the electron velocity . we follow cox & giuli ( 1968 , 16.2 ) , who corrected for the electron degeneracy by integrating the free - free absorption cross - section over the fermi - dirac distribution . the second term is the elwert factor ( elwert 1939 ) ; a correction to the electron wavefunction normalization due to the coulomb potential ( pratt & tseng 1975 ; lee et al . 1976 ; bethe & salpeter 1977 ) . the elwert factor depends on the ratio of the coulomb energy to the electron energy , parameterized by @xmath267 , and the energy of the photon involved in the transition . we write an average photon energy @xmath268 , and find that @xmath256 gives a good fit . the parameter @xmath269 tracks the transition between @xmath270 and @xmath249 as the energy scale which sets the electron velocity . we chose the particular form of the function @xmath269 that gave the best fit . the third term includes the effect of relativity on the coulomb scattering . bethe & salpeter ( 1977 ) show that in the extreme relativistic limit , the free - free cross section @xmath271 is proportional to the electron energy @xmath272 , whereas in the non - relativistic case @xmath273 . the gaunt factor rises rapidly at high temperatures to allow for this different scaling , as found in the numerical calculations of itoh and coworkers ( itoh et al . 1985 , 1990 , 1991 ; nakagawa et al . 1987)@xmath274 pair production , increasing the number density of scatterers . we never encounter this regime in our steady - state models and so do not include it here in our fitting formula . ] . we thus added a term @xmath275 at high temperatures , choosing the transition temperature @xmath276 which gave the best fit . the heat transport is dominated by electron conduction once the material becomes degenerate . yakovlev & urpin ( 1980 , hereafter yu ) wrote the conductivity as @xmath277 where @xmath278 is the electron collision frequency and @xmath279 ( @xmath249 is the electron fermi energy not including the rest mass ) . we calculate electron - electron collisions using the fit of potekhin et al . ( 1997 ) , but these make a small contribution to the total collision frequency ( typically @xmath280% ) . the conductivity is mainly determined by electron - ion collisions , for which we have generalized yu s results to an ionic mixture . the scalings become evident by writing @xmath281 , where we sum independently over each ion and @xmath282 is the coulomb logarithm . since @xmath283 and @xmath284 , we find @xmath285 a generalized form of the familiar yu result , where @xmath286 and we have inserted the correct prefactor as found by yu . the coulomb logarithm is @xmath287 ( yu ) , where the second term is the relativistic correction to the cross - section , and @xmath288 and @xmath289 are the limits of the integral over impact parameters . the lower limit is set by the electron wavelength @xmath290 . yu give a good fit for the upper limit as @xmath291 , where the debye screening length @xmath292 for a mixture of ions is @xmath293 ( shu 1991 ) and @xmath63 is the average spacing between ions , defined by @xmath294 . if we also define @xmath295 for the mixture as @xmath296 ( hubbard & lampe 1969 ) , then we find @xmath297.\ ] ] equation ( [ eq : cond1 ] ) , together with equations ( [ eq : cond2 ] ) , ( [ eq : cond3 ] ) and ( [ eq : cond4 ] ) , gives a general expression for the electron - ion collision frequency for an ion mixture . for the case of a single species of ion , @xmath298 and @xmath299 and these equations reduce to the expressions of yu . note that the scalings in the mixed case are _ not _ those obtained by substituting the mean value of z into the expressions of yu . in figure [ fig : appkappa ] , we show the resulting opacity for pure @xmath159fe in the @xmath188@xmath300 plane . the solid lines show where the dominant opacity source changes from electron scattering to free - free absorption , and from free - free opacity to conduction , i.e. where @xmath301 and @xmath302 , where @xmath303 is the conductive opacity . the line @xmath302 agrees well with figure 2 of gudmundsson , pethick & epstein ( 1983 ) , who used the los alamos opacities for pure @xmath159fe . the dotted lines show contours of total opacity @xmath64 , given by @xmath304 . to the left of the long dashed line , significant @xmath305@xmath274 pair production occurs . we do not include this in our calculations as we never encounter this regime in our steady - state models ( see figure [ fig : trho ] ) , thus our opacity calculations are not valid in this region . we choose iron here as an illustrative example to compare to previous work . in practice , as we show in this paper , there is a rich mixture of elements produced by the nuclear burning . we now compare our opacity calculations to those of previous workers . the main difference is in the treatment of free - free opacity . taam , woosley & lamb ( 1996 ) in their time - dependent simulations used the kepler code of weaver , zimmerman & woosley ( 1978 ) ( see also woosley & weaver 1984 ) which calculates opacities from the analytic fits of iben ( 1975 ) . iben ( 1975 ) provides fitting formulae to the radiative opacities of cox & stewart ( 1970a , b ) and the conductive opacities of hubbard & lampe ( 1969 ) ( for non - relativistic electrons ) and canuto ( 1970 ) ( for relativistic electrons ) . fujimoto and coworkers ( fhm ; fujimoto et al . 1984 ; hanawa & fujimoto 1984 , 1986 ) also used this same fit . joss ( 1977 ) , joss & li ( 1980 ) and ayasli & joss ( 1982 ) used iben s fit to the conductive opacity , but a fit by stellingwerf ( 1975 ) to the radiative opacities of cox , king & tabor ( 1973 ) . more recently , potekhin , chabrier & yakovlev ( 1997 ) used the opal opacity library ( rogers , swenson & iglesias 1996 ) for their study of cooling neutron stars with accreted envelopes with pure h , he , c , o and fe layers . as an example , we have compared our calculation of the total opacity @xmath64 with that obtained using the fit of iben ( 1975 ) for the free - free contribution . we find that for pure @xmath159fe at @xmath306 and @xmath307 ( the @xmath188 and @xmath300 regime for which free - free opacity becomes important in our models ) , the total opacity we calculate is 30% to 50% larger than that using iben s fit .
a method is presented to derive electron temperatures and densities of planetary nebulae ( pne ) simultaneously , using the observed hydrogen recombination spectrum , which includes continuum and line emission . by matching theoretical spectra to observed spectra around the balmer jump at about 3646 , we determine electron temperatures and densities for 48 galactic pne . the electron temperatures based on this method [ hereafter(bal ) ] are found to be systematically lower than those derived from [ ] and [ ] ratios [ hereafter ,([] ) and([] ) ] . and the electron densities based on this method are found to be systematically higher than those derived from [ ] , [ ] , [ ] , [ ] and [ ] ratios . these results suggest that temperature and density fluctuations are generally present within nebulae . the comparison of([] ) and(bal ) suggests that the fractional mean - square temperature variation ( ) has a representative value of 0.031 . a majority of temperatures derived from the [ ] ratio are found to be higher than([] ) , which is attributed to the existence of dense clumps in nebulae those [ ] ir fine - structure lines are suppressed by collisional de - excitation in the clumps . by comparing([] ) ,([] ) and(bal ) and assuming a simple two - density - component model , we find that the filling factor of dense clumps has a representative value of 7 . the discrepancies between([] ) and(bal ) are found to be anti - correlated with electron densities derived from various density indicators ; high - density nebulae have the smallest temperature discrepancies . this suggests that temperature discrepancy is related to nebular evolution . in addition , he / h abundances of pne are found to be positively correlated with the difference between([] ) and(bal ) , suggesting that he / h abundances might have been overestimated generally because of the possible existence of h - deficient knots . electron temperatures and densities deduced from spectra around the paschen jump regions at 8250 are also obtained for four pne , ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 . electron densities derived from spectra around the paschen jump regions are in good agreement with the corresponding values derived from spectra around the balmer jump , whereas temperatures deduced from the spectra around the paschen jump are found to be lower than the corresponding values derived from spectra around the balmer jump for all the four cases . the reason remains unclear .
astro-ph0403371
accurate determinations of metal abundances in planetary nebulae ( pne ) are important for understand the chemical evolution of galaxies and the nucleosynthesis and mixing processes in low- and intermediate - mass stars . a long - standing problem in nebular abundance studies has been that the heavy - element abundances derived from optical recombination lines ( orls ) are systematically higher than those derived from collisionally excited lines ( cels ; see liu 2001 , 2003 for recent reviews ) . in extreme cases , the discrepancies exceed a factor of 10 . given that cel abundances are sensitive to electron temperature and density , knowledge of temperature and density variations is an essential ingredient in understanding the abundance discrepancy problem . information on temperature and density variations can be obtained by comparing results derived from plasma diagnostics which behave differently as temperature and density vary . traditionally , nebular electron temperatures and densities are derived from cel ratios . however , for pne including high - density regions ( @xmath12e@xmath13@xmath14 ) , such as ic 4997 @xcite , mz 3 @xcite and m 2 - 24 @xcite , cels are strongly suppressed by collisional de - excitation , and so depending on their critical densities , they may become unusable in probing such high - density regions . on the other hand , hydrogen recombination lines have very high critical densities ( @xmath15@xmath14 ) . consequently , the hydrogen recombination spectrum provides a powerful tool to probe high - density regions . comparison of electron temperatures and densities derived from the hydrogen recombination spectrum ( lines plus continuum ) with those derived from cels should enable us to quantify temperature and density variations in nebulae . by measuring the balmer discontinuity , @xcite determined electron temperatures in 14 pne and found that they are systematically lower than those derived from the [ ] nebular - to - auroral forbidden line ratio . this can be explained by the presence of large temperature fluctuations since the [ ] @xmath16 auroral line of higher excitation energy tends to be emitted in higher temperature regions than the [ ] @xmath17 nebular lines of lower excitation energy , as suggested by @xcite . recent reviews on temperature fluctuations of gaseous nebulae are given by @xcite and references therein . however , no photoionization models yield such large temperature fluctuations suggested by observations . moreover , temperature fluctuations are unable to explain the large disparity between the orl and cel abundances observed in some pne @xcite . instead , it is found that the hypothesis that pne may contain clumps of different temperature , density and chemical composition , embedded in diffuse ` normal ' material , can explain many of the observed patterns @xcite . peimbert s concept of temperature fluctuations alone may not be a realistic description of the physical conditions in a nebula , although it has proven to be a convenient and useful way to characterize the complexities of real nebulae . it is expected that there are also variations of electron density inside a pn . @xcite pointed out that electron densities derived from various plasma diagnostics should correlate positively with the critical densities of the diagnostic lines involved if there are large density variations within a nebula . @xcite found that electron densities derived from the [ ] @xmath9 ratios are generally lower than those derived from the optical [ ] and [ ] doublet ratios , suggesting the presence of high density regions in pne , where the 52- and 88-@xmath18 lines are suppressed by collisional de - excitation due to their relatively low critical densities . @xcite showed that the discrepancies between @xmath0([]@xmath3 ) and @xmath0(bal ) can be explained by the presence of condensations in nebulae and gave an analytical method of estimating the filling factor of condensations . our detailed studies of the peculiar pne mz 3 and m 2 - 24 suggest that the presence of high - density regions in some pne can significantly affect abundance determinations @xcite . however , density inhomogeneities alone can not explain the large disparity between the orl and cel abundances either , unless the condensations are also h - deficient @xcite . @xcite found that the discrepancy between the orl and cel abundances correlates with the difference between the temperatures derived from the [ ] collisional lines and from the balmer jump of hydrogen recombination spectrum , thus lending strong support to the argument that some extremely cold h - deficient knots may co - exist with normal nebular material in many pne . however , it is difficult to explain the origin and evolution of such h - deficient knots with the current theories of element production in intermediate - mass stars . in this paper , we present determinations of electron temperatures and densities for a large sample of pne based on the analysis of nebular hydrogen recombination spectrum . section 2 describes the sample . in section 3 , we describe a new approach to deriving electron temperature and density simultaneously from the observed nebular hydrogen recombination spectrum ( lines plus continuum ) and present the results . in section 4 , we compare results derived from a variety of temperature and density indicators and discuss the significance of variations in electron temperature and density in pne . a discussion of he / h abundance determinations is also presented in this section . in section 5 , we present electron temperatures and densities derived from the paschen discontinuity and decrements . a summary then follows in section 6 . 48 galactic pne ( 37 disc pne and 11 bulge pne ) are analyzed in the current work . the spectra were obtained in a series of observing runs using telescopes at eso la silla and la palma . 14 objects were observed using the isis double long - slit spectrograph mounted on the 4.2 m william herschel telescope ( wht ) at la palma observatory in 1996 july august and in 1997 august . a tek @xmath19 @xmath20 charge coupled device ( ccd ) was used , which yielded a spatial sampling of 0.3576 arcsec per pixel projected on the sky . in 1996 , a 600gmm@xmath21 and a 316gmm@xmath21 grating were used for the blue and red arm , respectively . in 1997 , they were replaced by a 1200gmm@xmath21 grating . 23 objects were observed with the eso 1.52 m telescope using the boller & chivens ( b&c ) long - slit spectrograph . the spectra were obtained during runs in 1995 july , 1996 july , 1997 february and 2001 june . in 1995 the detector was a ford @xmath22 @xmath23 chip , which was superseded in 1996 by a thinned uv - enhanced loral @xmath22 @xmath23 chip of much improved quantum efficiency . the b&c spectrograph has a useful slit length of about 3.5arcmin . in order to reduce the ccd read - out noise , the ccd was binned by a factor of two along the slit direction , yielding a spatial sampling of 1.63 arcsec per pixel projected on the sky . a slit width of 2arcsec was used throughout the eso 1.52 m runs . 11 objects were observed using the intermediate dispersion spectrograph ( ids ) on the 2.5 m isaac newton telescope ( int ) at the observatorio del roque de los muchachos , on la palma , spain in 2001 august . a slit 4arcmin long and 1arcsec wide was used for nebular observations . the r1200b grating was used to cover wavelengths from 35005000 at a spectral resolution of 0.95 , determined from the _ fwhm _ of the calibration arc lines , while the r300v grating , together with a gg385 order sorting filter , was used to cover wavelengths from 38007700 at a resolution of 4.0 _ fwhm_. all the spectra were reduced using midas or iraf following the standard procedure . the spectra were bias - subtracted , flat - fielded and cosmic - rays removed , and then wavelength calibrated using exposures of a calibration arc lamp . the spectra were flux - calibrated by observing spectrophotometric standard stars . 1-d spectra were obtained by integrating along the slit . using the galactic reddening law of @xcite , the logarithmic extinction at h@xmath24 , @xmath25 , was derived from the observed balmer decrements , the h@xmath26/h@xmath24 , h@xmath27/h@xmath24 and h@xmath28/h@xmath24 ratios . a mean value for each nebula was used to deredden the spectra . the object names and the extinction constants are given in table [ tbl-1 ] . amongst the pne in our sample , far - infrared spectra for 21 of them have been obtained using the long wavelength spectrometer ( lws ) on board the infrared space observatory ( _ iso _ ) . fluxes of far - ir fine structure lines derived from these spectrum have been published by @xcite except for three objects , hu 1 - 2 , ngc 6210 and ngc 6818 . for the latter three pne , we measured far - ir line fluxes using spectra retrieved from the _ iso _ data archive . in order to normalize the @xmath29 line fluxes to h@xmath24 , the absolute @xmath30 fluxes tabulated in @xcite were used . in order to calculate theoretical nebular spectra , accurate determinations of the he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios are necessary ( see section 3 ) . helium ionic abundances were derived from and recombination lines , using @xmath33 which have only a weak dependence on the adopted electron temperature and density . he@xmath31/h@xmath31 abundance ratios were calculated from the strong line @xmath34 under the assumption of case a. the effective recombination coefficient @xmath35 is derived from the formulae presented by @xcite . he@xmath32/h@xmath31 abundance ratios were calculated from the @xmath36 line , using the effective recombination coefficient @xmath37 of @xcite . the results are presented in table [ tbl-1 ] . for the purpose of comparison , various forbidden line ratios are also used to determine electron temperatures and densities . for densities , results derived from the following diagnostic ratios were used , [ ] @xmath38m/@xmath39 m , [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 . electron densities from the [ ] 52-@xmath43m/88-@xmath43 m ratio are taken from @xcite except for ngc 6210 and ngc 6818 , for which we obtain the values of @xmath44 and 2.86 ( @xmath14 ) , respectively , using our own measured line fluxes . densities deduced from the optical [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 doublet ratios are taken from wang et al . ( in preparation ) . a detailed discussion of these density indicators is given in their paper . electron temperatures derived from the [ ] @xmath2 and [ ] @xmath1 ratios together with their uncertainties are presented in table [ tbl-1 ] . for the former , values were derived assuming electron densities derived from the [ ] 52-@xmath43m/88-@xmath43 m line ratio . for the latter , the average densities derived from the [ ] @xmath41 and [ ] @xmath42 doublet line ratios were adopted , given that cl@xmath31 and ar@xmath32 have ionization potentials of 23.8 ev and 40.7 ev , respectively , comparable to the value of 35.1 ev for o@xmath31 , and therefore anticipating that the [ ] lines arise from similar ionized regions as [ ] and [ ] lines . as the density increases , the resultant temperatures become increasingly sensitive to the adopted density . when the density reaches a value higher than @xmath45 @xmath14 , the [ ] @xmath2 and [ ] @xmath1 ratios are no longer useful temperature diagnostics . for example , high density regions have been detected in m 2 - 24 and ic 4997 @xcite . for both pne , temperatures deduced from the [ ] @xmath1 ratio have large uncertainties . the recombination spectrum consists of continuum and line emission . the continuous spectrum has a jump at the limit of the balmer series at 3646 . high - order balmer lines ( @xmath46 , @xmath47 ) converge at the balmer discontinuity and merge with the continuum jump at shorter wavelength . the size of the balmer jump is a function of electron temperature . on the other hand , intensities of high - order balmer lines ( @xmath46 , @xmath48 ) relative to h@xmath24 are sensitive to electron density . therefore , the hydrogen recombination spectrum near the balmer jump region provides valuable temperature and density diagnostics . our method and results of the electron temperature and density determinations based on this technique are presented below . several processes contribute to the continuum emission from a nebula : free - free and free - bound emission and the two - photon decay . apart from hydrogen , the second most abundant element , helium also contributes significantly to the first two processes . in addition , emission from the central star and local dust can in principle contaminate the observed continuum spectrum . in the following , a detailed calculation including all the processes is presented . the free - free ( or bremsstrahlung ) emission , which is emitted when an free electron interacts but does not combine with a positively charged ion such as h@xmath31 , he@xmath31 and he@xmath32 , dominates the nebular continuum in the infrared and radio region . the emission coefficient of this process is given by @xcite , @xmath49,\ ] ] where @xmath50 and @xmath51 are ion and electron densities respectively , @xmath52 is the nuclear charge , @xmath53 is the electron temperature , @xmath54 denotes the rydberg energy and @xmath55 is the free - free gaunt factor . here @xmath55 is computed using the method outlined by @xcite . in the optical region , free - bound emission is the main contributor to the continuum . according to @xcite , the emission coefficient of continuous emission due to recombinations of h@xmath31 , he@xmath31 or he@xmath32 with electrons is @xmath56,\ ] ] where @xmath57 is the ionization potential of the ( @xmath58,@xmath59 ) state , @xmath60 is the lowest state that can contribute to the free - bound emission at the given frequency @xmath61 and @xmath62 is the photoionization cross - section for the state , which is computed using the method described by @xcite . another important source of continuous emission is two - photon decay of the 2@xmath63 level of the hydrogen atom ( two - photon emission from he@xmath31 and he@xmath64 is negligible for nebulae of cosmic compositions ; brown @xmath65 mathews , 1970 ) . single photon transition from the 2@xmath63 level to the ground 1@xmath63 is forbidden . radiative decay of the 2@xmath63 level to the ground can only occur via two photon emission , in which the electron jumps to a virtual level , which can lie in anywhere between @xmath66 and @xmath67 levels , and then to the ground state , emitting two photons in the process . as a result , continuum emission longward of ly@xmath26 will occur . the emission coefficient of the two - photon process is given by @xmath68,\ ] ] where @xmath69 and @xmath70 is the probability per second for emitting a photon in the interval d@xmath71 . approximately , the function @xmath72 is given by the following expression @xmath73^{0.8}\}+0.88[y(1-y)]^{1.53}[4y(1-y)]^{0.8}\ } { \rm s}^{-1}\ ] ] @xcite . the transition probability for the two - photon decay is @xmath74 s@xmath21 . the equilibrium population of the 2@xmath63 state is given by @xmath75 @xcite , where @xmath76 is the proton density , @xmath77 is the effective recombination coefficient for populating 2@xmath63 , and @xmath78 and @xmath79 are the collisional de - excitation transition rates of 2@xmath63 to 2@xmath80 states by electrons and protons , respectively . note that the collisional transition from 2@xmath80 to 2@xmath63 has been ignored in equation ( 6 ) . under case a conditions , this is a good approximation . under case b conditions , however , the population of the 2@xmath80 state is comparable to that of the metastable 2@xmath63 level since ly@xmath26 is trapped between the 2@xmath80 and 1@xmath63 states . consequently , there is considerable collisional excitation from 2@xmath80 to 2@xmath63 , which enhances the two - photon emission . therefore , the emission coefficient of two - photon process given by equation ( 4 ) can be significantly underestimated . however , this hardly affects our resultant electron temperatures and densities since an additional continuum ( see below ) , is added to fit our observed spectrum in order to simulate direct and scattered stellar light . for a limited wavelength region , such an approach is sufficient to account for any excess emission from the two - photon process . finally , direct or scattered light from the central star can contaminate the observed continuum level . for our observations , the slit position was normally chosen to sample the brightest parts of the nebulae , sometimes passing through the central star . some spectra were derived by scanning across the nebular surface . it is quite difficult to evaluate the level and spectral energy distribution of the contaminating stellar continuum . as central stars of pne are relatively hot ( @xmath8120,000k ) , in our calculation , a rayleigh - jeans approximation for a black body ( @xmath82 ) is adopted to simulate the stellar continuum . this stellar component also partially compensates for the underestimated two - photon continuum , as mentioned above . for simplicity , we shall call this term ` stellar continuum ' hereafter . continuum emission from dust grains peaks in the infrared and is negligible in the optical region . thus it has been neglected in our calculation . apart from continuum emission , recombination line emission from bound - bound transitions of h@xmath64 , he@xmath64 and he@xmath31 is also included . the emission coefficient of a recombination line is given by @xmath83,\ ] ] where @xmath84 is the effective recombination coefficient of the line . for and recombination lines , emission coefficients are determined as in @xcite . here the considered maximum principle quantum number of is @xmath85 . for recombination lines , emission coefficients are taken from @xcite and @xcite . all line profiles are assumed gaussian with a _ fwhm _ @xmath86 where @xmath87 is the _ fwhm _ caused by thermal broadening , given by @xmath88 where @xmath89 is the mass of the atom , and @xmath90 is the _ fwhm _ caused by other broadening effects , such as nebular expansion , turbulence , seeing conditions and instrument . for a typical pn observed here , @xmath91 . line emission coefficients are thus converted into emission coefficients per unit wavelength by @xmath92.\ ] ] just longward of the balmer jump , high - order balmer lines converge and merge into a continuum - like feature . the slope of the feature is sensitive to electron density , and thus provide an excellent density diagnostic . the region is contaminated by the @xmath933664 line of multiplet v 1 . however , the weak recombination line has a relatively narrow line width compared to the profile of the continuum - like feature and consequently , can not cause any difficulty in our analysis . summing up contributions from all the processes discussed above , the flux at a given wavelength is given by @xmath94~[{\rm erg~cm}^{-3}~{\rm s}^{-1}~{\rm \aa}^{-1}].\ ] ] equation ( 11 ) shows that , in order to synthesize the nebular spectrum near the balmer jump wavelength region , the required input parameters include nebular electron temperature and density , the magnitude of contaminating stellar continuum , he@xmath31/h@xmath31 and he@xmath32/h@xmath31 ionic abundance ratios and @xmath95 . in general , decreasing ( increasing ) of the nebular electron temperature will steepen ( flatten ) the general shape of the theoretical continuum and increase ( decrease ) the balmer discontinuities ; decreasing ( increasing ) of electron density will steepen ( flatten ) the continuous feature redwards of 3646 and decrease ( increase ) the intensities of high - order balmer lines relative to a low - order line , such as h@xmath24 , as shown in fig . [ trend ] . accordingly , matching the dereddened observed spectrum to those calculated from equation ( 11 ) by varying the electron temperature and density , we can determine the electron temperature and density simultaneously provided other parameters , such as he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios and @xmath95 are known . electron temperatures and densities for 48 pne are derived by fitting the synthetic to the observed spectra . for synthesis , the required he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios have been observationally determined , and @xmath95 was derived by fitting the line profiles of strong recombination lines deconvolved with the thermal broadening , as listed in table [ tbl-1 ] . three free parameters , nebular electron temperature and density and the fraction of the scattered rayleigh - jeans stellar continuum , are optimized simultaneously , using least chi square method to match the observed spectra . for the optimization , the fitted data points were specified explicitly by eye , within the wavelength range from 35503850 , but excluding regions of strong ( forbidden ) lines . the results are partially shown in fig . [ all ] together with the residuals of the fits . in table [ tbl-2 ] , we report the derived electron temperatures and densities along with their estimated errors . the balmer jump and decrement of the recombination spectrum have previously been used to determine nebular electron temperatures and densities in four pne , m 2 - 24 @xcite , ngc 6153 @xcite , m 1 - 42 and m 2 - 36 @xcite . for the four pne , the results derived here are in good agreement with the earlier results within the errors . the errors caused by uncertainties in the input parameters he@xmath31/h@xmath31 and he@xmath32/h@xmath31 abundance ratios are negligible due to the small measurement errors of strong and lines and their small contribution to the continuum emission compared to hydrogen ( @xmath96 ) . errors in @xmath95 can also be ignored as this quantity is well determined by fitting strong hydrogen lines . the complicated velocity fields within a nebula can cause different lines to have different deviations from their laboratory wavelengths , and thus can affect the quality of line - fitting made under the simple assumption of a uniform velocity for all ions and lines . line - fitting can also be affected by blends of unknown lines and deviations of line profiles from gaussian . inspection of fig . [ all ] shows that the observed intensities of the 2s@xmath97s@xmath98p@xmath97p@xmath99 @xmath100 lines are generally lower than its predicted values . @xcite attributed this to the destruction of lyman photons by photoionization of h@xmath64 or by absorption of dust grains . the wavelength region near the balmer jump also contains a few strong bowen fluorescence lines which are not considered in our simulations , thus giving large residuals at the corresponding wavelengths . however , these effects do not contribute to errors in the derived electron temperatures and densities since these wavelengths have been excluded in the fitting . the observed spectra plotted in fig . [ all ] , were de - reddened using extinction constants listed in table [ tbl-1 ] . all spectra were normalized to h 11 @xmath101 in order to minimize errors in derived temperatures and densities caused by uncertainties in the reddening correction and flux calibration , and taking the advantage of the small wavelength difference between the balmer jump and h 11 . the stellar continuum may have a spectral energy distribution that deviates from the rayleigh - jeans approximation adopted here . however , this continuum is expected to be smooth and vary only slightly over such a short wavelength range of fitting . therefore , we have also ignored any possible errors caused by the fitting of the stellar continuum . errors in the derived temperatures and densities are dominated by two effects . noise is an important source of error . for most pne , we found that uncertainties due to this factor are about 5@xmath102 in the derived temperatures and 0.1 dex in the densities . the other major error source originates from the extensive wings of the strong [ ] @xmath103 3726 , 3729 lines , which can affect the shape of the continuum near the balmer jump . the effect is larger for low - excitation pne , such as ngc 40 and ngc 6720 . the extensive wings of the [ ] lines can cause the continuum redwards of the balmer jump to be overestimated . as a result , the temperatures can be overestimated . the errors can be reduced by including data points at longer wavelengths in the fitting , where the extensive wings of the [ ] lines have less impact . however , increasing the wavelength range of the fitting would increase uncertainties in the reddening correction and in fitting the stellar continuum . to evaluate errors due to this effect , each spectrum was fitted twice for two wavelength ranges with one extending further into the red than the other . the extensive wings of the [ ] lines can also decrease the shape of the continuum - like feature redwards of the balmer jump , leading to overestimated densities . for most pne , this effect is negligible given the large wavelength difference ( about 70 ) between the balmer jump and the moderate strength of [ ] lines . however , the effect can be significant for those with extremely strong [ ] lines . on the other hand , the extensive wings of the [ ] lines have much less effect on the integrated intensities of high - order balmer lines ( @xmath46 , @xmath104 ) which can be determined using a line profile fitting technique . comparison of densities derived from the continuum - like feature redwards of the balmer jump with those deduced from the integrated intensities of high - order balmer lines ( e.g. liu et al . 2001b ) allows us to estimate the errors in the derived densities caused by the extensive wings of the [ ] lines . the final errors given in table [ tbl-2 ] were based on estimates of contributions from the two dominating sources of uncertainty discussed above . electron densities deduced from the balmer decrements , and from the [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 and [ ] @xmath42 doublet ratios are compared in fig . [ necorr ] . [ necorr ] shows that for most pne , the hydrogen recombination spectrum yields higher densities compared to forbidden lines diagnostics . there is a trend such that the discrepancy becomes larger at higher densities . two extreme cases are ic 4997 and m 2 - 24 . for ic 3568 , [ ] @xmath41 yields an abnormally low density . the cause of it is unclear but is probably due to measurement error . in fig . [ necorrir ] we compare electron densities derived from the [ ] far - infrared fine structure line ratio and from the hydrogen recombination spectrum . evidently , densities derived from the [ ] @xmath105 ratio are significantly lower than those derived from the balmer decrements . in fact , it has been found that the far - ir lines yield systematic lower densities , lower than even those deduced from optical forbidden line density - diagnostics @xcite . a linear fit shows that the density discrepancy increases with the electron density . o@xmath64 , s@xmath64 , cl@xmath106 , o@xmath106 and ar@xmath32 have ionization potentials of 13.6 , 10.4 , 23.8 , 35.1 and 40.7 ev , respectively , and therefore embrace a wide range of regions of differing states of ionization . on the other hand , the hydrogen recombination spectrum arises from the entire ionized nebula . thus the discrepancies between densities derived from recombination spectrum and those deduced from forbidden line ratios are not likely to be caused by ionization stratifications . we notice that the [ ] , [ ] , [ ] , [ ] and [ ] density - diagnostic lines have critical densities of @xmath107 @xmath14 . therefore , if the nebulae contain regions of higher densities , these forbidden lines will be suppressed in these regions by collisional de - excitation . in contrast , the hydrogen recombination spectrum is practically unaffected by collisional de - excitation . as a result , in such cases the hydrogen recombination spectrum will yield higher average densities compared to the [ ] , [ ] , [ ] , [ ] and [ ] lines ratios . the same argument has been developed by @xcite to explain the systematically lower densities derived from the [ ] fine - structure line ratio compare to those deduced from optical forbidden line ratios . the [ ] lines have higher critical densities than the [ ] , [ ] , [ ] lines . thus , the effects are weaker for [ ] doublet ratio than for the other cases . given that the [ ] far - infrared fine - structure lines have the lowest critical densities amongst those density - diagnostic forbidden lines ( @xmath108([]@xmath109 @xmath110 ) , the effects are largest for the [ ] @xmath105 ratio . this can be seen clearly in fig . [ necorr ] and fig . [ necorrir ] . a nebular model including these high - density condensations and their effects on temperature determinations will be discussed in section 4.3 . [ tecorr ] and fig . [ tecorrir ] show that electron temperatures derived from the [ ] @xmath1 and @xmath111 ratios [ hereafter , @xmath0([]@xmath112 and @xmath0([]@xmath113 are systematically higher than those deduced from the hydrogen recombination spectrum [ hereafter @xmath0(bal ) ] , although there are a few exceptions where @xmath0(bal ) is higher than @xmath0([]@xmath3 ) . the most extreme case is he 2 - 118 , for which @xmath0(bal ) is 5000 k higher than @xmath0([]@xmath3 ) . in our sample , the four pne with the largest temperatures discrepancies , hf 2 - 2 , m 1 - 42 , m 2 - 36 and ngc 6153 , were previously studied in detail by @xcite , @xcite and @xcite . the discrepancy between @xmath0([]@xmath3 ) and @xmath0(bal ) was first discovered by @xcite and was attributed to temperature fluctuations within the nebula @xcite . he characterized the temperature structure of a nebula by the average temperature @xmath114 and a mean square temperature fluctuation @xmath10 as follows : @xmath115 and @xmath116 where @xmath50 is the ion density for an observed emission line . assuming that all the oxygen is twice ionized and that there are no density and composition variations in the nebula , @xmath114 and @xmath10 can be derived from two of the three expressions , @xmath117_{\rm na})=t_0[1+\frac{1}{2}(\frac{9.13\times10 ^ 4}{t_0}-3)t^2],\ ] ] @xmath117_{\rm fn})=t_0[1+\frac{1}{2}(\frac{2.92\times10 ^ 4}{t_0}-3)t^2],\ ] ] and @xmath118 @xcite . in fig . [ tecorr ] and fig . [ tecorrir ] , lines showing @xmath0([]@xmath3 ) as a function of @xmath119 for the case of @xmath120 , 0.02 , 0.06 and 1.00 are over plotted . fig . [ tecorr ] shows that most nebulae have @xmath121 . excluding a few extreme nebulae having @xmath122 , we obtain a mean value of @xmath123 . the value is in excellent agreement with previous result @xmath124 derived by @xcite although their analysis contained larger measurement errors . however , even excluding the extreme nebulae , a similar comparison between @xmath0([]@xmath4 ) and @xmath119 yields an average @xmath125 . as we shall show in the following subsection , it is likely that @xmath0([]@xmath4 ) have been overestimated because of density variations in the nebula , resulting in higher @xmath126 when comparing @xmath0([]@xmath4 ) and @xmath119 . another interesting point revealed by fig . [ tecorr ] is that @xmath119 covers a wider range of values than @xmath0([]@xmath3 ) . except a few extreme cases , @xmath0([]@xmath3 ) falls between 8000 and 14000 k , whereas @xmath119 varies from 900 to 19000 k. @xcite has argued that temperature fluctuations of gaseous nebulae can cause lower @xmath119 compared to @xmath0([]@xmath3 ) . however , even @xmath123 is beyond any photoionization model prediction . another explanation of higher @xmath119 compared to @xmath0([]@xmath3 ) is that there are some condensations with @xmath127 @xmath14 in pne , which then suppress the @xmath0 ( [ ] ) @xmath1034959,5007 relative to the @xmath934363 , leading to an artificially higher @xmath0([]@xmath3 ) @xcite . but no such high density condensations have been found in pne having extremely large temperature discrepancies @xcite . recently , @xcite presented an empirical nebular model containing two components , each with its own temperature , density and chemical composition to account for the temperature discrepancy . this seems to be the most plausible explanation at present . for pne having @xmath128([]@xmath3 ) , as in he 2 - 118 , the electron temperature might conceivably increase in low - ionization regions owing to heating by shock waves in the outer regions . as a result , the average temperature of the entire nebula [ @xmath0(bal ) ] appears higher than that of high - ionization regions [ @xmath0([]@xmath3 ) ] . in 4.1 , we found that electron densities derived from the recombination spectrum are systematically higher than those derived from optical forbidden lines ratios , and confirm the earlier result of @xcite that densities derived from the [ ] far - ir fine - structure lines are the lowest of all . these results show that nebular electron densities derived from various diagnostics are correlated with the critical density of the diagnostic lines , and suggest that dense clumps are generally present in pne . in such cases , the [ ] @xmath129 ratio is no longer a good temperature indicator given the relatively low critical densities of the [ ] @xmath130- and 88-@xmath131 lines ( 3500 and 1500 @xmath14 , respectively ) . when nebulae contain condensations with densities in the range between 10@xmath132 and 10@xmath133 @xmath14 , the [ ] @xmath130- and 88-@xmath131 far - infrared lines will be heavily suppressed by collisional de - excitation , leading to overestimates of @xmath0([]@xmath4 ) . on the other hand , condensations of densities in this range will not affect @xmath0([]@xmath3 ) given that the [ ] @xmath134 and @xmath16 lines have higher critical densities of @xmath135 @xmath14 and @xmath136 @xmath14 , respectively ( cf . fig [ tecorrir ] ) . another piece of evidence pointing to the presence of dense clumps comes from fig . [ oiiite ] where we compare @xmath0([]@xmath3 ) and @xmath0([]@xmath4 ) . if nebulae have temperature fluctuations but are homogeneous in density , then @xmath0([]@xmath4 ) is expected to be always lower than @xmath0([]@xmath3 ) @xcite . however , fig . [ oiiite ] shows about two - third of the objects have higher @xmath0([]@xmath4 ) compared to @xmath0([]@xmath3 ) , suggesting that density variations play an important role in @xmath0([]@xmath4 ) determinations . @xcite showed that electron temperatures derived from the [ ] @xmath137 ratio may be overestimated if nebulae contain ionized condensations of electron densities higher than 10@xmath133 @xmath14 . however , except for a few extreme pne , such as ic 4997 @xcite , mz 3 @xcite and m 2 - 24 @xcite there is no evidence that nebulae contain a substantial amount of ionized gas in condensations with densities in excess of 10@xmath133 @xmath14 , as the presence of such condensations should have been revealed by the recombination spectrum ( liu et al . 2000 , 2001a , 2001b ; fig . [ necorr ] ) . therefore , we consider @xmath138 @xmath14 as a safe upper limit for the density of these condensations . in order to estimate the filling factor of the condensations , we use an analytical method originally developed by @xcite for explaining the discrepancy of @xmath0([]@xmath3 ) and @xmath119 . assuming that some high - density regions ( @xmath127 @xmath14 ) are present , they deduced nebular - to - auroral [ ] line ratios ( see their eq . following the same deduction of their eq . 2.3 but assuming that the electron density of these condensations is about @xmath139 @xmath14 instead of @xmath45 @xmath14 , we obtain the infrared - to - auroral line ratio @xmath140 where @xmath141 is line emission coefficient , @xmath142 , f and n refer to ir fine - structure line and nebular line , l and h refer to the low- and high - density regions , @xmath143 is the density contrast and @xmath144 is the filling factor . in the following calculation , we assume that : 1 ) density variations are the only cause of the lower @xmath0([]@xmath4 ) compared to @xmath0([]@xmath3 ) and @xmath145 ; 2 ) the nebular gas has a homogeneous electron temperature , i.e. , @xmath0([]@xmath3 ) or @xmath145 ; 3 ) the condensations have densities lower than the critical density of the [ ] auroral line but higher than the critical densities of the [ ] infrared lines , whereas low - density regions have a density that is lower than the critical densities of the [ ] auroral and infrared lines . setting @xmath146_{\rm fn}})=i_{52\micron+88\micron}/i_{4959}$ ] , we have @xmath147 . for a simple assumption @xmath148 @xmath14 and @xmath149 @xmath14 , the ratio @xmath150 is about 0.04 , whereas @xmath151 is practically independent of the density when the densities are lower than the critical density of the nebular line and thus @xmath152 . therefore , @xmath153 is given by @xmath154_{\rm fn}})/f(t_{\rm bal})-1}{0.04-f(t_{[{\rm o}~{\rm iii}]_{\rm fn}})/f(t_{\rm bal})}.\ ] ] in fig . [ tecorrir ] , the lines showing the variation of @xmath0([]@xmath4 ) as a function of @xmath0(bal ) for @xmath155 , 0.1 , 0.5 , 1.0 and 2.0 are plotted . we obtain an average value of @xmath156 . however , it should be borne in mind that the @xmath153 derived by eq . ( 18 ) is a upper limit since temperature variations are actually present and contribute partly to the measured temperature discrepancies . @xmath153 can also be estimated by comparing @xmath0([]@xmath3 ) with @xmath0([]@xmath4 ) , as shown in fig . [ oiiite ] . assuming that @xmath0([]@xmath3 ) is constant across the entire nebula , @xmath153 is likewise given by @xmath154_{\rm fn}})/f(t_{[{\rm o}~{\rm iii}]_{\rm na}})-1}{0.04-f(t_{[{\rm o}~{\rm iii}]_{\rm fn}})/f(t_{[{\rm o}~{\rm iii}]_{\rm na}})}.\ ] ] in fig . [ oiiite ] , the lines showing the variation of @xmath0([]@xmath4 ) as a function of @xmath0(bal ) for @xmath155 , 0.1 , 0.5 , 1.0 and 2.0 are also plotted . an average value of @xmath157 is obtained . it is noteworthy that the presence of temperature variations will increase @xmath0([]@xmath158 ) with respect to @xmath0([]@xmath4 ) , in contrast to density variations . therefore , @xmath153 derived from eq . ( 19 ) is a lower limit . the upper- and lower - limit of @xmath153 are listed in table . [ tbl-3b ] . for a few pne having a lower limit @xmath159 , the effect of temperature variations on the temperature determination is larger than that of density variations , causing higher @xmath0([]@xmath112 compared to @xmath0([]@xmath4 ) . for all the pne , the mean value of @xmath153 is in the range from 0.4 to 1.0 . thus , we suggest that @xmath160 may be a representative value that can be used for a typical nebula . adopting the density contrast @xmath161 , a representative value of the condensations filling factor @xmath162 is about @xmath163 . the value is similar to that found by @xcite . adopting @xmath160 for all the pne , we obtain new @xmath0([]@xmath4 ) , for which the contribution of the clumps has been deduced using equation ( 17 ) . comparing the new @xmath0([]@xmath4 ) and @xmath0(bal ) , we re - evaluate the average value @xmath164 , which is in good agreement with the value yielded by the comparison of @xmath0([]@xmath3 ) and @xmath0(bal ) ( see section 4.2 ) . in fig . [ crene ] , the differences between temperatures derived from the [ ] forbidden lines and from the balmer jump are plotted against electron densities derived from the hydrogen recombination spectrum , the [ ] @xmath5 , [ ] @xmath40 , [ ] @xmath41 [ ] @xmath42 and the [ ] @xmath165m/@xmath166 m ratios , respectively . in all cases , the differences between the two temperature are anti - correlated with nebular densities derived from various density indicators . a least - squares fit yields the following relation : @xmath167 and a linear correlation coefficient of 0.51 , where @xmath168([])@xmath169 and @xmath170 is that derived from hydrogen recombination spectrum , which represents the average density of the entire nebula . @xcite found that the discrepancies of o@xmath32/h@xmath106 abundance ratios derived from orls and from cels are greater for larger , lower - surface brightness pne than for compact , dense ones . on the other hand , @xcite found that the abundance discrepancies are greater for those with larger temperature discrepancies . our results are therefore consistent with these earlier findings . @xcite found that , @xmath171 where @xmath172 . from the two relations given in equations ( 20 ) and ( 21 ) , we find that the ratio of abundances derived from orls and from cels has a power - law dependence on density , @xmath173 where @xmath174 . in general , electron densities of pne are expected to decline monotonically as nebulae expand . @xcite found an approximate relation between nebular mean electron density and nebular size , @xmath175 , where r is the nebular radius . consequently , it is suggested that the temperature discrepancy is related to the evolutionary state of pne ; more expended pne show larger temperature discrepancies . on the other hand , the actual evolution in a given nebula may depend on many factors . furthermore , the electron density measured for a given nebula may depend on the diagnostic used . for example , electron densities of low - ionization regions , as probed by the [ ] @xmath5 and [ ] @xmath40 line ratios , can be enhanced by interactions of fast stellar winds with earlier slow winds . therefore , there are slight difference in deriving the relation of @xmath176 and @xmath170 from different density diagnostics , as shown in fig . [ crene ] . we also use electron density derived from these forbidden line ratios to estimate the @xmath26-value in equation ( 22 ) . the results show that the value is within the range of 0.3 and 0.9 . to explain the relation shown in equation ( 20 ) and ( 22 ) , we apply a chemically inhomogeneous two - component model presented by @xcite . in their hypothesis , the nebula consists of two components , h - deficient clumps and diffuse material with ` normal ' abundances ( @xmath177solar ) . these h - deficient clumps are expected to have a low electron temperature due to enhanced cooling caused by high metallicity and pressure equilibrium between the two components , so that only recombination lines are emitted . as a result , orls yield overestimated heavy element abundances . in a given pn , the contribution of the two components to the balmer temperature and abundance determinations depends on their densities , i.e. , the contribution becomes weaker as density declines . if the clumps have neutral cores , as found in abell 30 @xcite and ngc 7293 @xcite , they are expected to be quite stable during the evolution of the pne . as a result , the relative contribution of the low - density and high - temperature material to the average temperature and abundance determinations decrease as nebulae expand . consequently , the discrepancies in temperature and abundance determinations become larger as the nebulae expend . a long - standing problem in abundance determinations for pne has been that the heavy element abundances derived from orls are systematically higher than those derived from cels . the ratio of abundances derived from the two types of emission lines is found to vary from target to target . @xcite analysed a pn sample and found the orl / cel discrepancy factors for these heavy element abundances span a range from about 2 to as large as 20 . for the most extreme known pne hf 2 - 2 , the orl / cel abundance ratio reaches a record value of 84 @xcite . another problem in nebular astrophysics is that @xmath0([]@xmath3 ) are systematically higher than @xmath119 , as we have discussed in 4.2 . the two problems are found to be related . a positive correlation between the ratio of o@xmath32/h@xmath31 derived from orls and from [ ] cels and the temperature discrepancy [ @xmath176=@xmath0([]@xmath3)-@xmath119 ] has been shown by @xcite . the explanations for the dichotomy in abundance and temperature determinations have focussed on temperature and density variations and chemical inhomogeneities . however , it has been shown that the temperature and density variations alone can be ruled out by current observations @xcite . at present , the most plausible explanation is that there exist some cold h - deficient clumps in nebula @xcite where only recombination lines are emitted . to fit the observed pattern of ngc 6153 , @xcite have presented several empirical models incorporating h - deficient clumps . @xcite also constructed photoionization models of ngc 6153 and m 1 - 42 , incorporating h - deficient clumps , which reproduce the observed integrated line ratio satisfactorily . in all the models , the h - deficient clumps contain only about @xmath179 of the total mass , suggesting that the h - deficient components contain only a tiny amount of nebular gas . nevertheless , they can seriously enhance the intensities of orls , causing overestimated orl abundances . therefore , the he / h abundance ratio could be overestimated owing to the existence of the h - deficient clumps can lead to since all the observable strong and lines are recombination lines . some h - deficient components have been found in some pne , such as abell 30 and abell 78 @xcite . @xcite analysed optical spectra of the h - deficient knots in abell 30 and found the he / h abundance ratios of the knots are more than ten . these pne with extremely large abundance and temperature discrepancies , such as ngc 6153 , m 1 - 42 and hf 2 - 2 , have been found to have high he / h abundances . @xcite defined the pne having high he / h abundance ( @xmath180 ) as type - i pne . according to our suggestion , these pne with high he / h abundances might result from contamination from h - deficient clumps . quantitative analysis of overestimated he / h abundances in pne is quite difficult . for this purpose , as mentioned by @xcite , one has to rely on the measurement of @xmath9310830 line which is strongly enhanced by collisional excitation from the 2s@xmath181s metastable level . if the he / h abundance is enhanced by h - deficient clumps , one expects that there is a positive correlation between the he / h abundance and the orl / cel abundance ratio of heavy element . for most pne , however , heavy element orls are quite weak and thus yield large uncertainties in the resultant abundance . instead , we can study the relation between the he / h abundance and @xmath176 since the orl / cel abundance ratio has been found to be positively correlated with @xmath176 by @xcite . fig . [ hete ] plots he / h versus @xmath176 . the measurement error of the he / h abundance ratio is negligible . in the figure , filled and open circles represent the galactic disc and bulge pne respectively and the solid line is a linear fit represented by the relation : @xmath182 which yields a linear correlation coefficient of 0.50 . although there is considerable scatter , there is a trend of increasing he / h abundance with increasing difference between the temperature from [ ] forbidden lines and balmer discontinuity for both the galactic disc and bulge pne . therefore , it seems plausible that the he / h abundances in pne have been generally overestimated due to the existence of cold h - deficient clumps . [ hete ] also shows that the most of pne have enhancement of helium with respect to the sun , which is often ascribed to the second and third dredge - up . however , the normal model of dredge - up can not explain some extreme cases , such as he 2 - 111 which has a he / h abundance of 0.219 @xcite . for these cases , the more likely reason is that pollution of the h - deficient clumps causes an overestimate of the he / h abundance ratio . on the other hand , we can not completely reject the possibility that helium in nebulae has been partly enhanced by the products of nucleosynthesis of the central star . that may be the main cause of the scatter in the derived he / h abundances . the presence of hydrogen - deficient clumps may be further proof . the orl / cel abundance discrepancies are also found in regions @xcite . and the existence of some high density clumps in regions has been indicated by previous studies ( e.g. odell et al . therefore , the h - deficient material might also be present in regions , leading to overestimated orl abundances . if this is the case , the primordial helium abundance , which is based on observations of some metal poor extragalactic regions , could have been overestimated . the hydrogen recombination spectrum near the paschen jump can also provides a diagnostic of electron temperature and density [ hereafter , @xmath183 and @xmath184 . unfortunately , in our sample , the spectra covering the paschen jump are available for only four pne , ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 . using the same method described in section 3 , we match the theoretical with the observed spectra at 8250 . given that the observed spectra near the paschen discontinuity may include a contribution from a dust emission tail , we added a constant continuum for the match . the results are shown in fig . all the spectra are normalized to p 20 in order to decrease the uncertainty of reddening correction . table [ tbl-4 ] lists the derived temperatures and densities . one advantage of using the spectrum near the paschen jump is that the resultant temperature and density are not affected by the continuum of the hot central star and two - photon emission . however , the emission from local dust may be stronger in this band . in addition , the estimation of the continuum bluewards the paschen jump is affected by the telluric absorption ( h@xmath185o and o@xmath185 ) . and the absorption feature might also cause a error in fitting the continuum longward of the paschen jump of some pne , such as m 1 - 42 . as a result , the temperatures and densities derived from the hydrogen recombination spectra at the paschen jump are less reliable than those derived from those at the balmer jump . table [ tbl-4 ] shows that the @xmath186 are consistent with the @xmath187 within the errors . however , the @xmath183 are found to be 4000 , 1000 , 500 and 1200k lower than the @xmath119 for ngc 7027 , ngc 6153 , m 1 - 42 and ngc 7009 , respectively . this may be due to systematic errors in the reddening correction or flux calibration . as an alternative , it remains possible that the discrepancy is caused by temperature variations . further investigations including a larger data sample of pne is needed to clarify the point . in this work , we presented a method to derive nebular electron temperature and density simultaneously by matching the theoretical with the observed hydrogen recombination spectra . using the method , we derived electron temperatures and densities of 48 galactic pne . the temperatures and densities derived by this method are compared with those derived from other indicators , enabling us to study electron temperature and density variations in pne . our main findings are as follows . the temperatures derived from the nebular - auroral line ratios of [ ] are systematically higher than those derived from hydrogen recombination spectra , suggesting that temperature variations are generally present in nebulae . we obtained the rms temperature fluctuation parameter @xmath188 as a representative value . the larger discrepancy in temperatures derived from [ ] @xmath129 line ratios and from hydrogen recombination spectra , however , can not be attributed simply to the temperature variations . the presence of clumps with @xmath138 @xmath14 can account well for the discrepancy . a comparison between electron temperatures derived from [ ] @xmath129 line ratios and those derived from hydrogen recombination spectra and [ ] @xmath1 line ratios suggests that the filling factor of the condensations has a typical value of @xmath163 . the discrepancies in temperatures derived from [ ] @xmath1 line ratios and from hydrogen recombination spectra are related to the evolution of pne . the orl / cel abundance ratio is found to have a power - law dependence on electron density ( @xmath189 , @xmath174 ) . we attribute this to different evolutionary scenario of two components in pne , i.e. expanding diffuse nebular material and stable cold h - deficient clumps . the contribution of the clumps to the temperature and abundance determinations increases as the nebulae expand . he / h abundance has a positive correlation with the temperature discrepancy , suggesting that he / h abundances might be overestimated due to the existence of h - deficient clumps . if this is the case , traditional up - dredge theory should be modified . the primordial helium abundance determination will be affected if there are significant condensations in regions . further investigations are needed . the densities derived from the spectrum near the paschen jump are in good agreement with those derived form that near the balmer jump . however , the former yield generally lower temperatures than the latter . the reason remain unknown . we conclude that the existence of h - deficient clumps in pne is the most plausible explanation for the origin of temperature and density variations and the orl / cel abundance discrepancies . however , it is a challenge to clarify the origin and evolution of these clumps . in further work , we will study the statistical relationships between the temperature and abundance discrepancies and other physical conditions of pne such as radio flux , morphology , chemical abundances and the nature of the central star . the authors are grateful to w. wang for his help with part of the data reduction . we also thank j .- s chen and s .- g . luo for their help with the preparation of this paper . constructive suggestions and comments by the referee s. m. viegas were appreciated . this work was supported in part by beijing astrophysics center ( bac ) . lcccccc hu 1 - 2 & 0.512 & 0.046 & 0.079 & @xmath190 & & 144 + ic 3568 & 0.261 & 0.095 & 0.001 & @xmath191 & @xmath192 & 132 + ngc 7027 & 1.380 & 0.058 & 0.041 & @xmath193 & @xmath194&82 + ngc 7009 & 0.130 & 0.093 & 0.011 & @xmath195 & @xmath196 & 132 + ngc 40 & 0.700 & 0.062 & 0.000 & @xmath197 & @xmath198 & 132 + ngc 6210 & 0.055 & 0.099 & 0.001 & @xmath199 & @xmath200 & 148 + ngc 6826 & 0.064 & 0.099 & 0.000 & @xmath201 & @xmath202 & 132 + ngc 6572 & 0.400 & 0.110 & 0.000 & @xmath197 & @xmath203 & 132 + ngc 6741 & 1.100 & 0.081 & 0.031 & @xmath204 & @xmath205 & 132 + ngc 6790 & 1.100 & 0.081 & 0.003 & @xmath206 & & 132 + ngc 6884 & 1.050 & 0.081 & 0.016 & @xmath207 & @xmath208 & 132 + ngc 7662 & 0.180 & 0.061 & 0.038 & @xmath209 & @xmath210 & 156 + ngc 6543 & 0.100 & 0.117 & 0.000 & @xmath211 & @xmath212 & 140 + m 2 - 24 & 0.800 & 0.099 & 0.005 & @xmath213 & & 156 + cn 2 - 1 & 1.063 & 0.104 & 0.004 & @xmath214 & & 132 + ngc 6720 & 0.200 & 0.094 & 0.017 & @xmath215 & @xmath216 & 132 + ngc 6620 & 0.760 & 0.099 & 0.021 & @xmath217 & & 132 + ngc 6567 & 0.670 & 0.105 & 0.007 & @xmath218 & @xmath219 & 132 + h 1 - 35 & 1.360 & 0.115 & 0.001 & @xmath220 & & 132 + h 1 - 50 & 0.700 & 0.100 & 0.011 & @xmath221 & & 132 + he 2 - 118 & 0.158 & 0.065 & 0.000 & @xmath222 & & 148 + m 1 - 20 & 1.200 & 0.094 & 0.000 & @xmath223 & & 123 + m 3 - 21 & 0.570 & 0.100 & 0.007 & @xmath224 & & 140 + m 3 - 32 & 0.579 & 0.120 & 0.013 & @xmath225 & & 140 + ic 1297 & 0.220 & 0.079 & 0.042 & @xmath226 & & 132 + ic 4634 & 0.454 & 0.097 & 0.000 & @xmath227 & @xmath228 & 140 + ic 4776 & 0.265 & 0.088 & 0.000 & @xmath229 & & 136 + ic 4997 & 0.438 & 0.139 & 0.000 & @xmath230 & & 127 + ngc 5873 & 0.004 & 0.063 & 0.040 & @xmath231 & & 127 + ngc 5882 & 0.380 & 0.113 & 0.003 & & & 127 + ngc 6302 & 1.460 & 0.105 & 0.067 & @xmath232 & @xmath233 & 132 + ngc 6818 & 0.400 & 0.050 & 0.051 & @xmath204 & @xmath234 & 148 + ngc 6153 & 1.300 & 0.123 & 0.011 & @xmath235 & @xmath236 & 148 + m 2 - 36 & 0.270 & 0.133 & 0.003 & @xmath237 & @xmath238 & 148 + m 1 - 42 & 0.700 & 0.139 & 0.009 & @xmath239 & @xmath240 & 140 + ngc 6778 & 0.890 & 0.155 & 0.008 & @xmath241 & & 140 + hf 2 - 2 & 0.200 & 0.131 & 0.002 & @xmath242 & & 132 + dddm 1 & 0.137 & 0.060 & 0.000 & @xmath243 & & 99 + hu 2 - 1 & 0.777 & 0.097 & 0.000 & @xmath244 & & 99 + ic 2003 & 0.347 & 0.054 & 0.049 & @xmath245 & & 99 + ic 5217 & 0.501 & 0.088 & 0.008 & @xmath246 & & 115 + me 2 - 2 & 0.343 & 0.138 & 0.000 & @xmath247 & & 90 + ngc 6803 & 0.869 & 0.112 & 0.004 & @xmath248 & & 107 + ngc 6807 & 0.642 & 0.084 & 0.000 & @xmath249 & & 107 + ngc 6833 & 0.000 & 0.078 & 0.000 & @xmath250 & & 99 + ngc 6879 & 0.401 & 0.091 & 0.003 & @xmath251 & & 99 + ngc 6891 & 0.287 & 0.094 & 0.000 & @xmath252 & & 115 + vy 1 - 1 & 0.421 & 0.094 & 0.000 & @xmath253 & & 99 + lcc hu 1 - 2 & @xmath254 & @xmath255 + ic 3568 & @xmath256 & @xmath257 + ngc 7027 & @xmath258 & @xmath259 + ngc 7009 & @xmath260 & @xmath261 + ngc 40 & @xmath262 & @xmath263 + ngc 6210 & @xmath264 & @xmath261 + ngc 6826 & @xmath265 & @xmath266 + ngc 6572 & @xmath267 & @xmath268 + ngc 6741 & @xmath269 & @xmath270 + ngc 6790 & @xmath271 & @xmath272 + ngc 6884 & @xmath273 & @xmath261 + ngc 7662 & @xmath274 & @xmath275 + ngc 6543 & @xmath276 & @xmath257 + m 2 - 24 & @xmath277 & @xmath278 + cn 2 - 1 & @xmath279 & @xmath280 + ngc 6720 & @xmath281 & @xmath282 + ngc 6620 & @xmath283 & @xmath284 + ngc 6567 & @xmath285 & @xmath286 + h 1 - 35 & @xmath287 & @xmath280 + h 1 - 50 & @xmath288 & @xmath289 + he 2 - 118 & @xmath290 & @xmath291 + m 1 - 20 & @xmath292 & @xmath293 + m 3 - 21 & @xmath294 & @xmath295 + m 3 - 32 & @xmath296 & @xmath270 + ic 1297 & @xmath297 & @xmath298 + ic 4634 & @xmath299 & @xmath300 + ic 4776 & @xmath301 & @xmath280 + ic 4997 & @xmath302 & @xmath303 + ngc 5873 & @xmath304 & @xmath305 + ngc 5882 & @xmath306 & @xmath298 + ngc 6302 & @xmath307 & @xmath308 + ngc 6818 & @xmath309 & @xmath310 + ngc 6153 & @xmath311 & @xmath312 + m 2 - 36 & @xmath311 & @xmath313 + m 1 - 42 & @xmath314 & @xmath315 + ngc 6778 & @xmath316 & @xmath317 + hf 2 - 2 & @xmath318 & @xmath319 + dddm 1 & @xmath320 & @xmath321 + hu 2 - 1 & @xmath322 & @xmath323 + ic 2003 & @xmath324 & @xmath325 + ic 5217 & @xmath326 & @xmath325 + me 2 - 2 & @xmath327 & @xmath328 + ngc 6803 & @xmath329 & @xmath330 + ngc 6807 & @xmath331 & @xmath332 + ngc 6833 & @xmath333 & @xmath334 + ngc 6879 & @xmath335 & @xmath336 + ngc 6891 & @xmath337 & @xmath338 + vy 1 - 1 & @xmath337 & @xmath325 + lccccc ic 3568 & 0.527 & 0.254 & 0.315 & 1.174 & 0.253 + ngc 7027 & 0.281 & 0.169 & 0.254 & 0.708 & 0.532 + ngc 7009 & 1.210 & 0.720 & 0.433 & 0.730 & 0 . + ngc 40 & 1.345 & 0.433 & 0.374 & 2.408 & 0 . + ngc 6210 & 0.641 & 0.392 & 0.469 & 0.681 & 0.208 + ngc 6826 & 0.641 & 0.374 & 0.520 & 0.768 & 0.415 + ngc 6572 & 0.401 & 0.151 & 0.374 & 1.857 & 1.639 + ngc 7662 & 0.271 & 0.238 & 0.228 & 0.145 & 0 . + ngc 6543 & 1.505 & 0.688 & 0.872 & 1.300 & 0.282 + ngc 6720 & 0.836 & 0.303 & 0.374 & 1.981 & 0.246 + ic 4634 & 0.688 & 0.258 & 0.411 & 1.867 & 0.634 + ngc 6818 & 0.258 & 0.171 & 0.254 & 0.543 & 0.516 + ngc 6153 & 2.543 & 0.714 & 0.254 & 2.987 & 0 . + m 2 - 36 & 2.543 & 2.056 & 0.714 & 0.249 & 0 . + m 1 - 42 & 2.543 & 1.422 & 0.543 & 0.850 & 0 . + average & & & & 1.0 & 0.4 +
by embedding four - rod resonators inside double - split ring resonators superlattice , a planar composite metamaterial possessing tripod plasmonic resonances is fabricated . double fano resonances are observed where a common subradiant driven oscillator is coupled with two superradiant oscillators . as a classical analogue of four - level tripod atomic system , the extinction spectrum of the composite metamaterial exhibits a coherent effect based on double fano resonances . transfer of the absorbed power between two orthogonal superradiant oscillators is shown to be mediated by the common subradiant oscillator .
1309.6219
control of fano resonance has been a focus of research in application of artificially structured nano - sized materials including quantum dots , nano - wires , photonic - crystal slabs , plasmonic nanostructures , as well as metamaterials . @xcite metamaterials can be designed to possess a broad bright mode and a non - resonant dark mode , which couples each other to exhibit a destructive interference effect such as classical analogue of electromagnetically induced transparency ( eit ) phenomenon , corresponding to fano resonance with the asymmetry parameter equal to zero @xcite . as a further development to engineer metamaterials possessing a higher sensitivity as well as a broad spectral response for sensor applications , multiple fano resonances are investigated in plasmonic metamaterials . multiple fano resonances can be achieved in two different schemes . the first scheme is to employ one bright mode and two or more dark modes . examples are abundant where a bright dipole couples with two dark dipoles or quadrupoles : one bright dipole resonator and a pair of split ring resonators ( srrs ) as the dark elements@xcite , double symmetrical u - shaped srrs with a nanorod between the two srrs@xcite , plasmonic clusters of heptamer@xcite and pentamer@xcite , and 3d metamaterials@xcite . the second scheme is to employ one dark mode and two or more bright modes , which already has been utilized in a four - level tripod atomic system to achieve quantum coherent effects such as polarization photon gate@xcite , enhanced cross - phase modulation@xcite , and two slow - pulse matching@xcite . in the four - level tripod atomic system , two different low - lying energy levels @xmath0 and @xmath1 , bright states , are coupled to a common excited state @xmath2 by weak probe and trigger lasers , while the third low - lying energy level @xmath3 , dark state , is coupled to the excited state @xmath2 by a strong pumping laser to provide pathways to both probe and trigger lasers necessary for double eits . while a quantum coherence of @xmath0 and @xmath1 is maintained , double eit - based coherent effects are observed . as a classical analogue , a theoretical analysis of fano correlation effect is reported for a plasmon - exciton - plasmon interaction in an artificial hybrid molecular system , one quantum dot placed in the gap of two identical metal nanoparticles@xcite . a classical model has already been introduced , where two bright oscillators with different resonance frequencies are coupled to one dark oscillator@xcite . ( a ) schematics of a composite metamaterial is shown with the dipoles in four - rod resonator and double - split ring resonator superlattice for a normally incident wave with horizontal polarization . red and blue arrows correspond to two superradiant oscillators , and green arrow correspond to one subradiant oscillator . ( b ) four - level tripod is shown . @xmath0 and @xmath1 are two bright states coupled to a common excited state @xmath2 by weak probe and trigger lasers , while the third low - lying energy level @xmath3 , dark state , is coupled to the common excited state @xmath2 by a strong pump laser . ( c ) classical model of one subradiant oscillator , green circle , coupled with two superradiant oscillators , red and blue squares , is shown with coupling strengths @xmath4 and @xmath5 . , width=415 ] in this work , we experimentally demonstrate a classical analogue of double fano resonances in a planar composite metamaterial possessing tripod plasmonic resonances , where a common subradiant dipole corresponding to the dark mode is coupled with two superradient dipoles corresponding to two bright modes . the composite metamaterial is structured such that four - rod resonators ( frr ) are embedded inside double - split ring resonators ( dsrr ) superlattice . see fig . [ fig : coupled - fano - model ] . two dipole resonances of dsrr are superradiant modes corresponding to the excitations of @xmath6 and @xmath7 , and one dipole resonance of frr is subradiant mode corresponding to the excitation of @xmath8 . proximity of the inner diameter of dsrr and the rod - length of frr permits near - field coherent couplings , and the composite metamaterial is a tripod metamaterial system possessing two superradiant dipole resonances of dsrr coupled coherently with one common subradiant dipole resonance of frr in near - field , similar to a tripod atomic system with four atomic levels coupled coherently by coherent optical fields . important finding is that double fano resonances in the composite metamaterial are correlated , showing up as a transfer of the absorbed power from one superradiant dipole to the other superradient dipole in dsrr . in sec.2 , we introduce a planar composite metamaterial possessing tripod plasmonic resonances , and thz extinction spectrum of the fabricated sample is presented along with the finite - difference - time - domain ( fdtd ) simulated spectrum . in sec.3 , double fano resonances are analyzed analytically by a set of coupled equations of motion of two superradiant oscillators and one subradiant oscillators to identify coherent processes of the energy transfer between two superradiant oscillators . in particular , the phase evolutions of superradiant and subradiant oscillators are closely examined . in sec.4 , the multi - polar nature of near - field coherent interaction is examined by electric field distribution obtained from fdtd theoretical simulation . sec.5 summarizes the major findings with conclusion . in appendix the general feature of plasmonic fano resonance is examined where both superradiant and subradiant oscillators are externally driven . analysis based on two coupled oscillators model leads to the fact that the characteristic asymmetric fano resonance formula of plasmonic structure is kept the same with a modification in the asymmetry parameter @xmath9 . polarization - dependent optical density plots of measured thz extinction spectra of ( a ) symmetric superlattice array of double - split ring resonators and ( b ) asymmetric superlattice array of double - split ring resonators for @xmath10-(black curve ) and @xmath11-(red curve ) polarizations are shown . , width=415 ] double split ring resonator ( dsrr ) is composed of two split rings , and we can construct two different kinds of superlattice arrays of dsrr by arraying them in square lattice either symmetrically or asymmetrically . we compare extinction spectra of symmetric and asymmetric superlattice arrays in fig . [ fig : s1 ] , as discussed in @xcite . when the gap orientation of dsrr is along @xmath10-axis corresponding to the symmetric superlattice array , two distinct resonances of a low energy @xmath12 and a high - energy @xmath13 take place in fig . [ fig : s1](a ) . while the high - energy @xmath13 resonance is excited for @xmath11-polarization , the low - energy @xmath12 resonance is excited for @xmath10-polarization . from the extinction spectra of the symmetric superlattice array we find that dsrr exhibits two orthogonal dipolar resonances , a low - energy @xmath12 and a high - energy @xmath13 , which is equivalent to the notion that one dsrr accommodates two independent dipole oscillators with resonances at @xmath12 and @xmath13@xcite . in the asymmetric superlattice array of fig . [ fig : s1](b ) , dsrrs are arrayed in a square lattice such that the gap orientation of dsrr is rotated by angles @xmath14 alternately with respect to @xmath10-axis , which allows simultaneous excitations of both low - energy @xmath12 and high - energy @xmath13 resonances of dsrr upon incidence of a linearly polarized electromagnetic wave . the amount of relative excitations of @xmath12 and @xmath13 dipole oscillators depends on the angle between the polarization direction and the gap orientation of dsrr@xcite . we pay a particular attention to the extinction spectrum for @xmath10-polarization , i.e. , the black curve in fig . [ fig : s1](b ) . we observe that the high - energy @xmath13 is excited as well as the low - energy @xmath12 in the asymmetric superlattice array . refer to fig . [ fig : coupled - fano](a ) for a discussion of the high - energy @xmath13 excitation . in the study of double fano resonances associated with two bright modes and one dark mode , it is necessary to have simultaneous excitations of two bright modes . in this respect , the asymmetric superlattice array provides an appropriate meta - structure where both high - energy @xmath13 and low - energy @xmath12 are excited as bright modes for @xmath10-polarization . on the other hand , frr possesses a polarization - independent broad resonance , @xmath15 , which will be examined more carefully later in connection with double fano resonances . optical microscopy images and structural dimensions of ( a ) double - split ring resonator ( dsrr ) metamaterial , mm@xmath16 , ( b ) four - rod resonator ( frr ) metamaterial , mm@xmath17 , and ( c ) composite metamaterial , mm@xmath18 are shown . @xmath20 m , @xmath21=14 @xmath20 m , @xmath22= 22.5@xmath23 , and @xmath24=26 @xmath20 m , @xmath25=4 @xmath20 m , @xmath26=45@xmath23 . , width=377 ] metamaterials are fabricated by a standard photo - lithography and subsequent lift - off process of au film with micrometer resolution , which is prepared from e - beam evaporation of 10 nm thick titanium adhesion layer and 200 nm thick gold layers on 500@xmath20 m thick @xmath27-doped si wafer . we call dsrr metamaterial as mm@xmath16 , frr metamaterial as mm@xmath17 , and the composite metamaterial as mm@xmath18 , respectively , and optical microscopy images are shown in fig . [ fig : sample - microscope ] ( a ) , ( b ) , and ( c ) , with structural dimensions . mm@xmath16 is an asymmetric superlattice array of dsrrs with the inner radius @xmath21=14 @xmath20 m , the outer radius @xmath19=18 @xmath20 m , the width of circle line 4@xmath20 m , which are arranged with the gap orientation @xmath14 and the lattice constant 50 @xmath20 m . previous studies showed that mm@xmath16 exhibits two polarization - dependent dipole resonances with different q - values@xcite . mm@xmath17 is an array of frrs with @xmath28 , @xmath24=26 @xmath20 m , @xmath25=4 @xmath20 m , and the lattice constant 50 @xmath20 m , possessing a 4-fold rotational symmetry . the relative orientation of rods , i.e. , @xmath29 , is set such that the electromagnetic response of mm@xmath17 is the same regardless of the polarization angle of a normally incident electromagnetic wave in an effective medium approximation . that is , mm@xmath17 is isotropic with respect to the polarization direction . in mm@xmath18 , the relative orientation of split - rings in dsrr and rods in frr results in non - symmetric dipolar interactions between rod of frr and split - ring of dsrr for the upper and lower arc split - rings . it is important to distinguish the electromagnetic characteristics of au metal in the spectral range of thz and visible regimes . as well - known , in thz frequency regime the real part of dielectric constant of au has a very large negative value with a small positive imaginary part , which leads that au behaves as a perfect electric conductor and au can not support surface plasmons , electromagnetic surface excitations localized near the surface . however , there exist thz meta - resonances in metamaterials prepared in au film , originating from an electromagnetic resonance in subwavelength structures , which is called plasmonic resonance@xcite . time - domain polarized terahertz transmission measurements are carried out by a teraview tps spectra 3000 spectrometer with a resolution of 1.2 @xmath30 at room temperature in vacuum@xcite . sample size is @xmath31 @xmath32 and thz beam diameter is about 5@xmath33 . fdtd simulations of electromagnetic responses in mm@xmath16 , mm@xmath17 , and mm@xmath18 are carried out by commercial software fdtd solution lumerical@xcite . thickness of au film in metamaterial samples is 200 @xmath34 . however , in the fdtd simulation we adopted the thickness of au film as @xmath35 = 4 @xmath36 , which satisfies the relation @xmath37 , with the setting of an override maximum mesh step size of dz = 1 @xmath36 and minimum mesh step size 0.00025 @xmath36 . as far as the relation @xmath37 holds , the thickness of au is not relevant , since au film itself does not support any resonances and thz resonances originate from the subwavelength structure of metamaterial@xcite . a unit cell of metamaterial has been set as the unit cell of fdtd simulation and the periodic boundary conditions are adopted in @xmath10- and @xmath11-directions and the perfectly matched layer boundary condition is adopted in @xmath38-direction . in order to identify whether the broad high - energy @xmath13 resonance of mm@xmath16 stems from a coupling between neighboring dsrrs , we examined fdtd simulated extinction spectrum by varying the horizontal distance between dsrrs . and we found that there always exists the broad high - energy @xmath13 resonance with the identical spectral shape except for a decrease in the magnitude of extinction simply owing to a reduced number of meta - particles per unit area . for the dielectric permittivity of @xmath27-doped silicon substrate we employed the values obtained from thz spectrum measurement , and the dielectric permittivity of au is adopted from crc handbook @xcite , which gives a typical value of @xmath39 at 2.0 thz . ( a ) polarization - dependent optical density plots of measured thz extinction spectra of mm@xmath16 , mm@xmath17 , and mm@xmath18 are shown at @xmath10- ( @xmath40 ) and @xmath11- ( @xmath41 ) polarizations . ( b ) polarization - dependent optical density plots of fdtd simulated spectra are shown for mm@xmath16 , mm@xmath17 , and mm@xmath18 . , width=453 ] figure [ fig : s2 ] shows the polarization - dependent optical density plots of ( a ) measured and ( b ) simulated thz extinction spectra of mm@xmath18 for a normally incident wave with @xmath10-polarization , @xmath42 , ( red curve ) and @xmath11-polarization , @xmath43 , ( blue curve ) . also shown are the extinction spectra of mm@xmath16 with @xmath10-polarization , @xmath42 , ( solid black curve ) and @xmath11-polarization , @xmath43 , ( dashed black curve ) and the extinction spectrum of mm@xmath17 ( gray curve ) . mm@xmath17 exhibits polarization - independent broad resonance ( gray curve ) , and the composite metamaterial mm@xmath18 shows polarization - dependent resonances ( red and blue curve ) , which is associated with polarization - dependent meta - resonances of mm@xmath16 . comparison of fig . [ fig : s2](a ) and ( b ) shows that there is an excellent agreement between measured and simulated extinction spectra . ( a ) optical density plots of measured thz extinction spectra of mm@xmath16 ( black ) , mm@xmath17 ( gray ) , and mm@xmath18 ( red ) are shown for @xmath10-polarization . ( b ) optical density plots of fdtd simulated spectra are shown for mm@xmath16 ( black ) , mm@xmath17 ( gray ) , and mm@xmath18 ( red ) . , width=453 ] in the study of double fano resonances taking place in mm@xmath18 , we closely examine the extinction spectra with @xmath10-polarization , @xmath42 , which are re - plotted in fig . [ fig : measurement fano ] . as a digression we deconvolute the extinction spectra of mm@xmath17 to 2 lorentzian resonances . figure [ fig : s3 ] shows how the extinction spectrum of mm@xmath17 is composed of lorentzian resonances . we find that the broad extinction spectrum of mm@xmath17 is a linear sum of one resonance @xmath44 with a width @xmath45 corresponding to @xmath46 ( green curve ) and the other resonance @xmath47 with a width @xmath48 corresponding to @xmath49 ( cyan curve ) . comparison of the sum of 2 lorentzian resonances ( solid gray curve ) and the measured extinction spectrum ( dashed gray curve ) shows a good agreement . as will be discussed later , the lorentzian resonance @xmath44 takes part in giving rise to double fano resonances , while the lorentzian resonance @xmath47 stays as a background absorption since the spectral location of resonance is far away from bright mode resonances of mm@xmath16 . refer to fig . [ fig : coupled - fano ] and its discussion as well as tab . [ table - isolate ] . let s examine the characteristics of mm@xmath16 and mm@xmath17 spectra in fig . [ fig : measurement fano](a ) . mm@xmath16 ( black ) possesses one low - energy resonance with a large oscillator strength at @xmath12 = 1.4thz and the other high - energy resonance with a small oscillator strength stretching with @xmath13 spanning from 1.8thz to 3.0thz . on the other hand , mm@xmath17 ( gray ) possesses a broad resonance which is a linear sum of @xmath50 = 1.8thz and @xmath51 = 2.3thz . mm@xmath17 extinction spectrum is deconvoluted to 2 lorentzian resonances . linear sum of one resonance @xmath44 and @xmath45 with @xmath46 ( green curve ) and the other resonance @xmath47 and @xmath48 with @xmath49 ( blue curve ) is shown in the solid gray curve along with the measured extinction spectrum of mm@xmath17 ( dashed gray curve ) . , width=340 ] comparison of the linear absorption spectra of mm@xmath16 and mm@xmath17 shows two notable features . first , the total absorption cross section of mm@xmath16 is much larger than that of mm@xmath17 , indicating that the dipolar interaction with an incident electric field is much stronger in mm@xmath16 than in mm@xmath17 . therefore , mm@xmath18 is composed of a strongly driven mm@xmath16 oscillator and a weakly driven mm@xmath17 oscillator with the two interacting by a near - field coupling . second , the entire spectrum of mm@xmath17 is contained under the envelope of mm@xmath16 spectrum , with @xmath50 resonance in mm@xmath17 positioned between @xmath12 resonance and @xmath13 resonance in mm@xmath16 . next , we examine the difference in extinction spectrum between mm@xmath16 and mm@xmath18 qualitatively by paying attention to the low - frequency and high - frequency peaks in fig . [ fig : measurement fano](a ) . at the low - frequency peak , mm@xmath18 ( red ) spectrum shows a dip at @xmath52 1.6thz , the spectral position being on the blue side of mm@xmath16 ( black ) low - energy resonance @xmath12 = 1.4thz . on the other hand , at the high - frequency peak , the extinction is enhanced almost two - fold at 2.4thz in mm@xmath18 ( red ) spectrum when compared with mm@xmath16 ( black ) spectrum . these features at low- and high - frequency peaks show that there exists a correlation between changes in the low - frequency and high - frequency peaks . we note that the spectral positions of dip at @xmath52 1.6thz and enhancement in the extinction at 2.4thz are located on the red and blue sides of mm@xmath17 ( gray ) resonance @xmath50=1.8thz , respectively . now we examine the correlation between changes in the low - frequency and high - frequency peaks in terms of double fano resonances analytically . let s call the superradiant dipole with @xmath12 resonance and the superradiant dipole with @xmath13 resonance in dsrr as @xmath53 and @xmath54 , respectively , and the subradiant dipole with @xmath50 resonance in frr as @xmath55 . in the composite metamaterial mm@xmath18 , dsrrs are arrayed with the orientation angle @xmath14 , and upon normal incidence of @xmath10-polarization wave , both @xmath53 and @xmath54 are excited . noting that a pair of superradiant dipoles at the upper and lower arcs of dsrr interact with subradiant dipole of frr symmetrically , dipolar interactions between the superradiant dipoles @xmath53 & @xmath54 and the subradiant dipole @xmath55 are expressed as @xmath56 where @xmath4 and @xmath5 are the coupling strengths of @xmath53 and @xmath54 with @xmath55 , respectively . @xmath57 is the frequency associated with the characteristic coherent coupling between superradiant dipoles and subradiant dipole . as shown in appendix , an asymmetric fano lineshape takes place even when the subradiant oscillator is externally driven . in the composite metamaterial mm@xmath18 , frr resonator @xmath55 corresponds to the externally driven subradiant oscillator . in double fano resonances we have two superradiant oscillators , and from eq . ( [ eq : b ] ) and ( [ eq : d ] ) in appendix the coupled equations of motion are straightforward with superradiant @xmath53 and superradiant @xmath54 and subradiant @xmath55 externally driven by a normally incident @xmath10-polarization wave . @xmath58 @xmath59 @xmath60 note that the subradiant dipole @xmath55 is externally driven with a driving amplitude of @xmath61 . in the absence of near - field couplings with the subradiant oscillator @xmath55 , i.e. , @xmath62 , two superradiant oscillators @xmath53 and @xmath54 are excited independently with the driving amplitudes of @xmath63 and @xmath64 , respectively . when near - field couplings of @xmath53 and @xmath54 with @xmath55 are turned on , i.e. , @xmath65 and @xmath66 , there occurs an interplay between @xmath53 and @xmath54 through coupling @xmath67 . that is , @xmath12 oscillator @xmath53 and @xmath13 oscillator @xmath54 reside in one and the same dsrr , and fano resonances in @xmath53 and @xmath54 are correlated . this is similar to the control of eits in a tripod atomic system where the absorption spectrum of the hyperfine field shows a dependence on the powers of the zeeman eit signal@xcite . a classical example is to control the absorbed power spectrum of one metal nanoparticle by manipulating the distance between quantum dot and the other metal nanoparticle in a system composed of one quantum dot placed in the gap of two identical metal nanoparticles@xcite . analytical expression of double fano resonances by the coupled equations of motion eq . ( [ eq : x])-([eq : dark ] ) allows to identify how the two independent orthogonal superradiant oscillators interact indirectly via near - field couplings with the common subradiant oscillator . the extinction spectra of absorbed powers are plotted . in ( a ) , red , blue , and green curves correspond to the absorbed powers @xmath68 of isolated dipoles @xmath69 , @xmath70 $ ] obtained from the coupled eq . ( [ eq : x])-([eq : dark ] ) with @xmath62 . in ( b ) , gray , pink , and dark brown curves correspond to the absorbed powers @xmath68 of coupled dipoles @xmath69 with @xmath65 and @xmath66 , respectively . black curve in ( b ) is the sum of gray , pink , and dark brown curves . in ( c ) and ( d ) , purple , green , and black curves are the measured extinction spectra of mm@xmath16 , mm@xmath17 , and mm@xmath18 re - plotted from fig . [ fig : measurement fano](a).,width=453 ] in fig . [ fig : coupled - fano ] we compare the spectra of absorbed power obtained from eq . ( [ eq : x])-([eq : dark ] ) with the measured extinction spectra . the spectra of absorbed power @xmath68 of both isolated and coupled dipoles @xmath70 $ ] are calculated from the relation @xmath71 , @xmath72 . when near - field couplings are absent , i.e. , @xmath62 , the red , blue , and green curves in fig . [ fig : coupled - fano](a ) correspond to the absorbed powers in the isolated lorentzian oscillators with resonances at @xmath12 and @xmath13 of dsrr in mm@xmath16 and @xmath15 of frr in mm@xmath17 , respectively , we set the parameters as shown in table [ table - isolate ] for resonance frequency in sec@xmath73 , damping @xmath74 in sec@xmath73 , and quality factor q of isolated lorentzian oscillators @xmath53 , @xmath54 , and @xmath55 of fig . [ fig : coupled - fano](a ) . note that we adopt the lorentzian resonance @xmath44 and @xmath45 with @xmath46 in fig . [ fig : s3 ] as the subradiant mode @xmath55 . the quality factor q of @xmath15 oscillator of frr , the isolated @xmath55 dipole , is the highest , asserting that @xmath15 oscillator is subradiant . > p0.3inp0.3inp0.25in|p0.3inp0.3inp0.25in|p0.3inp0.3inp0.25 in , & red & & _ v(t ) _ , & blue & & @xmath75 , & green & + @xmath76&@xmath77 & @xmath78 & @xmath79&@xmath80 & @xmath81 & @xmath82&@xmath83 & @xmath84 + 47&15&3.1 & 70&25&2.8 & 60 & 14 & 4.3 + [ table - isolate ] > m1.2 in |>m1.2in|>m1.2 in & _ g _ & @xmath61 + 1.0&1.0&0.1 + [ table - coupled ] in the presence of near - field couplings , i.e. , @xmath65 and @xmath66 , two independent superradiant oscillators @xmath53 and @xmath54 residing in dsrr couple with a common subradiant oscillator @xmath55 of frr via near - field couplings . with the parameters listed in table [ table - coupled ] , we obtain the absorbed powers of each coupled oscillator as shown in fig . [ fig : coupled - fano](b ) . gray , pink , and dark brown curves correspond to the absorbed powers of @xmath12 and @xmath13 oscillators of dsrr and @xmath15 oscillator of frr in mm@xmath18 , respectively , in the presence of near - field couplings . black curve is the sum of gray , pink , and dark brown curves . near - field couplings of @xmath12 oscillator ( red ) and @xmath13 oscillator ( blue ) with the subradiant @xmath15 oscillator ( green ) give rise to the characteristic asymmetric fano resonances of the coupled dipoles in fig . [ fig : coupled - fano](b ) . gray curve of fig . [ fig : coupled - fano](b ) shows that there occurs a fano dip in the high energy side of @xmath12 resonance of isolated @xmath53 ( red ) of fig . [ fig : coupled - fano](a ) . pink curve of fig . [ fig : coupled - fano](b ) shows that there occurs a fano dip in the low energy side of @xmath13 resonance of isolated @xmath54 ( blue ) of fig . [ fig : coupled - fano](a ) . while the dipolar excitations in @xmath12 oscillator and @xmath13 oscillator are independent , the resulting double fano resonances in the black curve show that the decrease and increase in the absorbed powers of dsrr are correlated by the common @xmath15 oscillator . for comparison , in fig . [ fig : coupled - fano](c ) and ( d ) , we plotted the measured spectra of mm@xmath16 and mm@xmath17 in purple and green curves , respectively , along with mm@xmath18 spectrum in black curve . we find that the relation of mm@xmath16 @xmath85 mm@xmath17 spectra and mm@xmath18 spectrum is well described by the coupled equations of motion eq . ( [ eq : x])-([eq : dark ] ) . in other words , in the composite mm@xmath18 possessing tripod plasmonic resonances , the correlation between fano dip and extinction enhancement stems from an interplay of @xmath12 and @xmath13 oscillators in double fano resonances , mediated by coupling @xmath67 coherently coupled to the common narrow oscillator . phase @xmath86 , amplitude @xmath87 , and absorbed powers @xmath88 are plotted for @xmath89 ( left panel ) , @xmath90 ( center panel ) , and @xmath91 ( right panel ) . in all the figures , as a reference , phase @xmath86 , amplitude @xmath87 , and absorbed power @xmath88 of isolated dipoles are plotted in the absence of near field couplings , @xmath92 , in red , blue , and green curves . see text for a detailed description . , width=529 ] in fano resonance , a destructive interference between superradiant and subradiant modes can be understood by examining how phases of superradiant and subradiant modes evolve upon increasing the driving frequency@xcite . in the case that the resonance frequency of subradiant mode is higher than that of superradiant mode , the superradiant mode experiences a phase of @xmath29 when going through its own superradiant resonance and the phase keeps on increasing , and when passing through the subradiant resonance the phase experiences a rapid decrease to the minimum phase of a value less than @xmath29 and recovers back to its original value , eventually accumulating to a total value of @xmath93 . the phase evolution of subradiant mode , however , is very different . the subradiant mode does not experience a decrease in the phase but keeps on accumulating the phase to a total value of @xmath94 by going through the superradiant resonance and its own subradiant resonance . hence near the subradiant resonance there occurs a spectral region where superradiant and subradiant modes oscillate out - of - phase , which gives rise to a destructive interference for fano resonance . when double fano resonances take place in tripod plasmonic resonances , on the other hand , two superradiant modes are coupled to the common subradiant mode , and the phase evolutions of two superradiant modes are not independent of each other but are correlated in a way determined by the relative spectral positions of the superradiant resonances with respect to the subradiant resonance . in fig . [ fig : phase ] we closely examine phase @xmath86 and amplitude @xmath87 of dipole oscillators comparing with the absorbed powers @xmath88 , where @xmath86=@xmath95 $ ] , @xmath87=@xmath96 , and @xmath88=@xmath97 , where @xmath98 is an external force driving @xmath69 dipole oscillator , @xmath70 $ ] . figure [ fig : phase](a-1 ) , ( a-2 ) , and ( a-3 ) show phase @xmath86 , and fig . [ fig : phase](b-1 ) , ( b-2 ) , and ( b-3 ) show amplitude @xmath87 , and fig . [ fig : phase](c-1 ) , ( c-2 ) , and ( c-3 ) show absorbed power @xmath88 of @xmath69 dipole oscillator , @xmath70 $ ] . in all the figures , as a reference , we show phase @xmath86 , amplitude @xmath87 , and absorbed power @xmath88 of isolated dipoles in the absence of near field couplings , @xmath92 , in red , blue , and green curves . red and blue curves correspond to the superradiant isolated dipoles @xmath53 and @xmath54 , respectively , while green curve corresponds to the subradiant isolated dipole @xmath55 with the parameters of lorentzian oscillators shown in table [ table - isolate ] . we consider 3 different cases of near - field couplings , i.e. , @xmath99 ( left panel ) , @xmath100 ( center panel ) , and @xmath101 ( right panel ) . in the presence of coupling between superradiant modes and subradiant mode red , blue , and green curves in the plots correspond to the isolated dipoles @xmath69 from a comparison of red and gray curves in fig . [ fig : phase ] ( a-1 ) , we find that the phase evolution of @xmath53 ( gray ) exhibits a dip near the resonance of subradiant @xmath55 ( dark brown ) , a feature characteristic of a single fano resonance . comparison of green and dark brown curves shows that the phase of subradiant @xmath55 ( dark brown ) increases to @xmath93 when passing the resonance of superradiant @xmath53 ( red ) which takes place before its own resonance of @xmath55 ( green ) . in fig . [ fig : phase ] ( a-2 ) we observe a similar behavior in the phase evolution of @xmath54 ( pink ) , but the phase of subradiant @xmath55 ( dark brown ) increases to @xmath93 by passing its own resonance of @xmath55 ( green ) before it comes across the resonance of superradiant @xmath54 ( blue ) . we note that the spectral positions of @xmath93 phase of the subradiant dipole @xmath55 are located below and above its resonance when coupled to @xmath53 and @xmath54 , respectively . it is evident that the phase dips in superradiant @xmath53 and superradiant @xmath54 of fig . [ fig : phase ] ( a-1 ) and ( a-2 ) are responsible for single fano resonance as seen in the extinction spectra of fig . [ fig : phase ] ( c-1 ) ( gray ) and ( c-2 ) ( pink ) . black curves are the extinction spectra of mm@xmath18 when a single fano resonance takes place . that is , black curve is the sum of gray , blue , and dark brown curves in fig . [ fig : phase ] ( c-1 ) , and is the sum of red , pink , and dark brown curves in fig . [ fig : phase ] ( c-2 ) . a numerical integration of absorbed powers shows that the sum of areas under red , blue , and green curves is identical to the sum of areas under gray , pink , and dark brown curves . negative values of the absorbed powers by the coupled dipoles @xmath54 ( pink ) and @xmath55 ( dark brown ) which has the additional phase shifts stand for a transfer of powers to the other dipole , which is observed in fano resonance in plasmonic structures@xcite . however , we note that the amplitudes of the coupled dipoles @xmath54 and @xmath55 are always positive - definite . in double fano resonances , however , the phases of @xmath53 and @xmath54 are not independent of each other since a common subradiant dipole @xmath55 is coupled to both @xmath53 and @xmath54 . in particular , the phase of @xmath54 ( pink ) is strongly affected by the presence of @xmath53 , while the phase of @xmath53 ( gray ) is not much different from that in single fano resonance , as can be seen in fig . [ fig : phase ] ( a-1 ) , ( a-2 ) , and ( a-3 ) . most importantly , a phase increase to @xmath93 of subradiant @xmath55 ( dark brown ) takes place when passing the resonance of superradiant @xmath53 ( red ) , even though @xmath55 is also coupled to superradiant @xmath54 at the same time . as a result , the phase of @xmath54 ( pink ) is pulled down sharply near the resonance of @xmath55 ( green ) , which leads to a delay in recovering back to its original value , when compared with what happens in a single fano resonance . see blue and pink curves at the phase @xmath102 in fig . [ fig : phase](a-2 ) and ( a-3 ) . we also examine the amplitude evolution of the superradiant dipole @xmath54 ( pink ) in fig . [ fig : phase ] ( b-2 ) and ( b-3 ) . we find that the spectral position of amplitude maximum at the low frequency is red - shifted to a frequency below the resonance of superradiant dipole @xmath53 ( red ) . these two changes in phase and amplitude of @xmath54 , namely , a delay in phase recovery and a red - shift of amplitude maximum , are mainly responsible for the difference in extinction spectra ( black curves ) between single fano resonances of fig . [ fig : phase ] ( c-1 ) & ( c-2 ) and double fano resonances of fig . [ fig : phase ] ( c-3 ) . in fig . [ fig : phase ] ( c-3 ) we can identify how double fano resonances take place by examining the extinction spectra of isolated @xmath53 ( red ) , isolated @xmath54 ( blue ) , and mm@xmath18 ( black ) where @xmath53 and @xmath54 are coupled to @xmath55 ( green ) . in region , both enhancement and red - shift of extinction in @xmath53 as seen in the change from red curve to black curve originate from the red - shift of amplitude maximum of dipole @xmath54 . in region , on the other hand , enhancement and blue - shift of extinction in @xmath54 as seen in the change from blue curve to black curve is due to the delay in phase recovery of dipole @xmath54 . these two changes give rise to a fano dip wider than that in a single fano resonance in region . these features are examples of coherent effects in double fano resonances resulting from near - field couplings of two superradiant oscillators @xmath53 and @xmath54 to a common subradiant oscillator @xmath55 in a tripod metamaterial system . in the extinction spectra of fig . [ fig : phase ] ( c-3 ) the coherent effects show up as an energy transfer from low - energy @xmath12 oscillator to high - energy @xmath13 oscillator in dsrr , which was mentioned in the discussion of fig . [ fig : coupled - fano ] . @xmath38-components of near - field electric field , @xmath103 , of mm@xmath18 are plotted for a normally incident wave with horizontal polarization ( @xmath10-axis ) . ( a ) , ( b ) , and ( c ) correspond to regions , , and of fig . [ fig : phase ] , respectively . in particular , ( b ) corresponds to the frequency 1.8thz which is away from the fano dip.,width=415 ] @xmath38-components of near - field electric field , @xmath103 , of ( a ) the isolated dsrr , ( b ) the isolated frr , and ( c ) the composite mm@xmath18 are plotted at the frequency 1.6thz , where the fano dip takes place . , width=415 ] one advantage of the coupled equations of motion eq . ( [ eq : x])-([eq : dark ] ) in describing double fano resonances is that two orthogonal dipolar oscillations @xmath53 and @xmath54 residing in dsrr of mm@xmath18 can be identified . on the other hand , an fdtd simulation provides the information on the electric field distribution of mm@xmath18 , which allows us to identify the electric multipolar nature of coherent coupling between dsrr and frr in mm@xmath18 . all the fdtd simulation is conducted with a commercial software fdtd solution lumerical @xcite . we plot distributions of @xmath38-component of electric fields , which are shown in fig . [ fig : e_z - plot ] for a normally incident wave with horizontal polarization ( @xmath10-axis ) . in a planar metamaterial , it is known that @xmath38-components of the electric field , normal to the planar metamaterial surface ( @xmath104 plane ) , provide the information on planar charge distributions of the plamonic excitations@xcite . first of all , we note that excitations in dsrr and frr of the composite mm@xmath18 are dipolar , similar to dipolar double fano resonances reported in ref . @xcite , where a single dipolar resonance is coupled two dipoles placed side - by - side , and double fano resonances are observed in opposite ends of the spectrum with respect to a single dipolar resonance . in region , dipolar excitation of @xmath12 oscillator is dominant in dsrr , and the dipoles in dsrr and frr are anti - parallel forming a bonding hybridization . in region , on the other hand , dipolar excitation of @xmath13 oscillator is dominant in dsrr , and the dipoles in dsrr and frr are parallel forming an anti - bonding hybridization . in region , a fano dip takes place at the frequency 1.6thz . in order to show explicitly the excitation of subradiant mode , distributions of z - component of electric fields in the isolated dsrr , the isolated frr , and the composite mm@xmath18 are shown in fig . [ fig : three - distributions ] . comparison of figs . [ fig : three - distributions](a ) and ( c ) shows that the induced dipole of dsrr is significantly suppressed by a coherent coupling with frr , giving rise to the fano dip in the spectrum of dsrr in mm@xmath18 . most importantly , the induced dipole of frr in fig . [ fig : three - distributions](c ) is opposite to that of frr in fig . [ fig : three - distributions](b ) . while the fdtd simulation provides the electric field distribution , thereby , clarifies the multipolar nature of excitations of double fano resonances , it can not distinguish two orthogonal superradiant dipolar excitations in dsrr of mm@xmath16 or dsrr of mm@xmath18 . schematics of asymmetric fano spectral shape are drawn for ( a ) dipole - quadrupole and ( b ) dsrr - frr structures at the fano dip . , width=377 ] by comparing fano resonances dipole - quadrupole and dsrr - frr structures , we clarify the fano mode of dipole excitations in frr . figure [ fig : schematics - quadrupole - frr ] shows schematics of fano mode excitations . the left panel represents the uncoupled ( a ) dipole - quadrupole structure and ( c ) dsrr - frr structure . while a quadrupole is excited in two parallel rods in dipole - quadrupole structure , a weak dipole is excited in frr in dsrr - frr structure . a fano dip in dipole - quadrupole structure comes from a suppression of dipole excitation owing to a coherent coupling with the quadrupole excitation in two parallel rods as shown in fig . [ fig : schematics - quadrupole - frr](b ) . in dsrr - frr structure , on the other hand , a dipole excitation in subradiant frr induced by a near - field coupling with dsrr suppresses a superradiant mode dipolar excitation in dsrr of mm@xmath16 as shown in fig . [ fig : schematics - quadrupole - frr](d ) . we briefly examine how double fano resonances behave as the relative orientation of dsrr and frr changes by fdtd simulation . as shown in fig . [ fig : s4 ] , mm@xmath18 with @xmath105 exhibits a deeper fano dip at the spectral location of a lower energy than mm@xmath18 with @xmath106 , while mm@xmath18 with @xmath107 possesses double fano resonances similar to mm@xmath18 with @xmath106 . we suppose that the highest electric field confinement takes place near edges of dsrr and frr for mm@xmath18 with @xmath105 . , @xmath106 , and @xmath105 , extinction spectra of the composite mm@xmath18 are shown for @xmath10-polarization . ( a ) show schematics of the composite mm@xmath18 and ( b ) shows the fdtd simulated extinction spectra for @xmath107 ( pink curve ) , @xmath106 ( red curve ) , and @xmath105 ( blue curve ) along with that of mm@xmath16.,width=340 ] [ fig : s4 ] by embedding four - rod resonators inside double - split ring resonators superlattice , a composite metamaterial system is introduced , where a common subradiant oscillator in four - rod resonator is coupled with two superradiant oscillators in double - split ring resonator to give rise to double fano resonances . as a classical analogue of four - level tripod atomic system , the extinction spectrum of the composite metamaterial exhibits a double fano - based coherent effect . double fano resonances are correlated owing to the conjugation of two coupling strengths @xmath4 and @xmath5 , which shows up as a transfer of absorbed power from one superradiant dipole to the other superradiant dipole in double - split ring resonators . double fano resonances in a tripod metamaterial system can be utilized in designing plasmonic resonances of metamaterial to engineer coherent effects in metamaterials . this work is supported by quantum metamaterial research center program and global frontier program of center for advanced meta - materials ( ministry of science , ict and future planning , national research foundation , republic of korea ) . 10 niels verellen , yannick sonnefraud , heidar sobhani , feng hao , victor v moshchalkov , pol van dorpe , peter nordlander , and stefan a maier . fano resonances in individual coherent plasmonic nanocavities . , 9(4):16631667 , 2009 . andrey e. miroshnichenko , sergej flach , and yuri s. kivshar . fano resonances in nanoscale structures . , 82(3):22572298 , aug 2010 . boris lukyanchuk , nikolay i zheludev , stefan a maier , naomi j halas , peter nordlander , harald giessen , and chong tow chong . the fano resonance in plasmonic nanostructures and metamaterials . , 9(9):707715 , 2010 . mohsen rahmani , boris lukyanchuk , and minghui hong . fano resonance in novel plasmonic nanostructures . , 7(3):329349 , 2013 . shuang zhang , dentcho a. genov , yuan wang , ming liu , and xiang zhang . plasmon - induced transparency in metamaterials . , 101(4):047401 , jul 2008 . na liu , lutz langguth , thomas weiss , jurgen kastel , michael fleischhauer , tilman pfau , and harald giessen . plasmonic analogue of electromagnetically induced transparency at the drude damping limit . , 8(9):758762 , 2009 . n. liu , t. weiss , m. mesch , l. langguth , u. eigenthaler , m. hirscher , c. sonnichsen , and h. giessen . planar metamaterial analogue of electromagnetically induced transparency for plasmonic sensing . , 10(4):11031107 , 2010 . xiaojun liu , jianqiang gu , ranjan singh , yingfang ma , jun zhu , zhen tian , mingxia he , jiaguang han , and weili zhang . electromagnetically induced transparency in terahertz plasmonic metamaterials via dual excitation pathways of the dark mode . , 100(13):131101131101 , 2012 . junqiao wang , chunzhen fan , jinna he , pei ding , erjun liang , and qianzhong xue . double fano resonances due to interplay of electric and magnetic plasmon modes in planar plasmonic structure with high sensing sensitivity . , 21(2):22362244 , 2013 . shao - ding liu , zhi yang , rui - ping liu , and xiu - yan li . multiple fano resonances in plasmonic heptamer clusters composed of split nanorings . , 6(7):62606271 , 2012 . shao - ding liu , ming - jiang zhang , wen - jie wang , and yun - cai wang . tuning multiple fano resonances in plasmonic pentamer clusters . , 102(13):133105133105 , 2013 . na liu , mario hentschel , thomas weiss , a paul alivisatos , and harald giessen . three - dimensional plasmon rulers . , 332(6036):14071410 , 2011 . timothy j davis , mario hentschel , na liu , and harald giessen . analytical model of the three - dimensional plasmonic ruler . , 6(2):12911298 , 2012 . alp artar , ahmet a yanik , and hatice altug . multispectral plasmon induced transparency in coupled meta - atoms . , 11(4):16851689 , 2011 . s rebi , d vitali , c ottaviani , p tombesi , m artoni , f cataliotti , and r corbalan . polarization phase gate with a tripod atomic system . , 70(3):032317 , 2004 . shujing li , xudong yang , xuemin cao , chunhong zhang , changde xie , and hai wang . enhanced cross - phase modulation based on a double electromagnetically induced transparency in a four - level tripod atomic system . , 101(7):073602 , 2008 . andrew macrae , geoff campbell , and ai lvovsky . matched slow pulses using double electromagnetically induced transparency . , 33(22):26592661 , 2008 . yong he and ka - di zhu . fano correlation effect of optical response due to plasmon exciton plasmon interaction in an artificial hybrid molecule system . , 30(4):868873 , 2013 . zhengyang bai , chao hang , and guoxiang huang . classical analogs of double electromagnetically induced transparency . , 291:253258 , 2013 . boyoung kang , jh woo , e choi , hyun - hee lee , es kim , j kim , tae - jong hwang , young - soon park , dh kim , and jw wu . optical switching of near infrared light transmission in metamaterial - liquid crystal cell structure . , 18(16):1649216498 , 2010 . woo , e.s . kim , e. choi , boyoung kang , hyun - hee lee , j. kim , y.u . lee , tae y. hong , jae h. kim , and j.w . cryogenic temperature measurement of thz meta - resonance in symmetric metamaterial superlattice . , 19:43844392 , 2011 . yu lee , ey choi , jh woo , es kim , and jw wu . reflection resonance switching in metamaterial twisted nematics cell . , 21(15):1749217497 , 2013 . jb pendry , aj holden , dj robbins , and wj stewart . low frequency plasmons in thin - wire structures . , 10(22):4785 , 1998 . john b pendry , a j _ holden , dj robbins , and wj stewart . magnetism from conductors and enhanced nonlinear phenomena . , 47(11):20752084 , 1999 . jb pendry , l martin - moreno , and fj garcia - vidal . mimicking surface plasmons with structured surfaces . , 305(5685):847848 , 2004 . yingfang ma , zhongyang li , yuanmu yang , ran huang , ranjan singh , shuang zhang , jianqiang gu , zhen tian , jiaguang han , and weili zhang . plasmon - induced transparency in twisted fano terahertz metamaterials . , 1(3):391399 , 2011 . j wallauer and m walther . fano line shape and phase reversal in a split - ring resonator based metamaterial . , 88(19):195118 , 2013 . see http://www.teraview.com . see http://www.lumerical.com . hou - tong chen , willie j padilla , joshua mo zide , arthur c gossard , antoinette j taylor , and richard d averitt . active terahertz metamaterial devices . , 444(7119):597600 , 2006 . see http://docs.lumerical.com/en/metamaterials@[email protected] . robert c weast , melvin j astle , and william h beyer . , volume 69 . crc press boca raton , fl , 1988 . yong s joe , arkady m satanin , and chang sub kim . classical analogy of fano resonances . , 74:259266 , 2006 . martin frimmer , toon coenen , and a. femius koenderink . signature of a fano resonance in a plasmonic metamolecule s local density of optical states . , 108:077404 , feb 2012 . jianqiang gu , ranjan singh , xiaojun liu , xueqian zhang , yingfang ma , shuang zhang , stefan a maier , zhen tian , abul k azad , hou - tong chen , et al . active control of electromagnetically induced transparency analogue in terahertz metamaterials . , 3:1151 , 2012 . xiaopeng shen and tie jun cui . photoexcited broadband redshift switch and strength modulation of terahertz metamaterial absorber . , 14(11):114012 , 2012 . xun jun he , jian min wang , xiao hua tian , jiu xing jiang , and zhao xin geng . dual - spectral plasmon electromagnetically induced transparency in planar metamaterials based on bright - dark coupling . , 291(0):371 375 , 2013 . andreas bitzer , alex ortner , hannes merbold , thomas feurer , and markus walther . terahertz near - field microscopy of complementary planar metamaterials : babinet s principle . , 19(3):25372545 , jan 2011 . alp artar , ahmet ali yanik , and hatice altug . directional double fano resonances in plasmonic hetero - oligomers . , 11(9):36943700 , 2011 . u. fano . effects of configuration interaction on intensities and phase shifts . , 124:18661878 , dec 1961 . benjamin gallinet and olivier j. f. martin . ab initio theory of fano resonances in plasmonic nanostructures and metamaterials . , 83:235427 , 2011 . benjamin gallinet and olivier j. f. martin . relation between near field and far field properties of plasmonic fano resonances . , 19(22):2216722175 , oct 2011 . cl garrido alzar , mag martinez , and p nussenzveig . classical analog of electromagnetically induced transparency . , 70:37 , 2002 . b gallinet , t siegfried , h sigg , p nordlander , oj martin , et al . plasmonic radiance : probing structure at the angstrom scale with visible light . , 13(2):497503 , 2013 . f lpez - tejeira , r paniagua - domnguez , r rodrguez - oliveros , and ja snchez - gil . fano - like interference of plasmon resonances at a single rod - shaped nanoantenna . , 14(2):023035 , 2012 . feng hao , yannick sonnefraud , pol van dorpe , stefan a maier , naomi j halas , and peter nordlander . symmetry breaking in plasmonic nanocavities : subradiant lspr sensing and a tunable fano resonance . , 8(11):39833988 , 2008 . feng hao , elin m larsson , tamer a ali , duncan s sutherland , and peter nordlander . shedding light on dark plasmons in gold nanorings . , 458(4):262266 , 2008 . benjamin gallinet and olivier jf martin . influence of electromagnetic interactions on the line shape of plasmonic fano resonances . , 5(11):89999008 , 2011 . andrea lovera , benjamin gallinet , and peter nordlander . mechanisms of fano resonances in coupled plasmonic systems . , 10 , 2013 . we show that the characteristic asymmetry feature of fano resonance is retained in plasmonic structure when both superradiant and subradiant oscillators are externally driven . originally fano resonance is introduced to explain an asymmetric peak in excitation spectra observed near autoionization discrete energy level overlapping with continuum energy level of helium.@xcite while the discrete state ( dark mode ) is not directly excited , a rapid @xmath93 phase change takes place in the phase of a slowly varying - phase background continuum ( bright mode ) near the discrete resonance . hence in - phase and out - of - phase interferences take place between continuum and discrete resonances , giving rise to an asymmetric line shape in the continuum spectrum . the sequence of constructive and destructive interferences depends on whether the discrete resonance falls on the red or blue side of the background continuum resonance . the asymmetric fano line shape is characterized by 3 parameters , the resonance shift @xmath109 , the resonance width @xmath110 , and the asymmetry parameter @xmath9 . differently from a dark mode in atomic spectroscopy , there exits an intrinsic non - radiative joule loss @xmath111 in a plamonic dark mode when electromagnetically driven , and gallinet _ et _ _ al . _ @xcite obtained a general formula of fano - like asymmetric line shape in metallic plasmonic structures , where a modulation damping parameter @xmath112 is introduced , a measure of the decrease in fano resonance contrast @xcite . it is also shown that a classical two coupled oscillators model @xcite with a lossy dark oscillator provides the same general formula when the complex amplitude of the bright oscillator is expanded near the dark resonance @xcite . furthermore , even without resorting to a plasmonic dark mode , a far - field fano - like asymmetric line shape is observed in plasmonic structures . the occurrence of a fano - like interference in a single nano - rod plasmonic structures is attributed to a spatial overlap of two non - orthogonal adjacent plasmonic modes @xcite . and in a nonconcentric ring / disk cavity plasmonic structure , fano line shape still persists when dark multipolar resonances are optically excited by a retardation effect at a non - normal incidence angle @xcite . in a heptamer plasmonic nanostructure composed of one center disc with surrounding hexamer ring , an examination of local density of optical states by a selective e - beam excitation of center and ring showed that a far - field interference between the dressed superradiant and subradiant eigenmodes fully describes fano resonance @xcite . in fact , asymmetric fano line shapes observed in a variety of plasmonic structures are fitted to the general fano formula with an excellent agreement @xcite . these studies lead to the notion that in plasmonic structures an interference between superradiant and subradient eigenmodes overlapping both spectrally and spatially gives rise to an asymmetric fano line shape resonance @xcite . now we raise the question whether asymmetric fano line shape is retained even when both superradiant and subradiant oscillators are driven simultaneously in the presence of a coherent coupling . we examine how the broad lorentzian spectrum of a superradiant oscillator is modified when coupled with a driven subradiant narrow oscillator by adopting a simple mechanical two coupled oscillators model . two coupled oscillators are described by a set of equations of motion . @xmath113 narrow spectrum of the subradiant oscillator @xmath25 is located within broad spectrum of the superradiant oscillator @xmath112 . putting @xmath114 , we obtain the amplitude @xmath115 of the superradiant oscillator @xmath112 near resonance @xmath116 of narrow subradiant oscillator @xmath25 . @xmath117 where @xmath118 and @xmath119 . an external driving @xmath120 of the subradiant oscillator contributes to the superradiant amplitude via a coupling @xmath5 . following procedures in ref . @xcite , the superradiant oscillator spectrum has the following relation . @xmath121 where @xmath122 formula of asymmetric fano resonance line shape reported in ref . @xcite is kept the same as a closed form with a modification in the asymmetry parameter from @xmath9 to @xmath123 . we observe that a dark mode is not necessarily prerequisite for an asymmetric fano resonance to take place in plasmonic structures . the dark mode is simply an example of non - driven narrow resonance . the characteristic asymmetry feature of fano resonance in plasmonic structures is retained even in the presence of an external driving of narrow oscillator as far as a dispersive coherent coupling between two oscillators ensures a @xmath93-phase shift in the broad superradiant oscillator near the narrow subradiant oscillator resonance .
we present a simplified model to study the orbital evolution of a young hot jupiter inside the magnetospheric cavity of a proto - planetary disk . the model takes into account the disk locking of stellar spin as well as the tidal and magnetic interactions between the star and the planet . we focus on the orbital evolution starting from the orbit in the 2:1 resonance with the inner edge of the disk , followed by the inward and then outward orbital migration driven by the tidal and magnetic torques as well as the roche - lobe overflow of the tidally inflated planet . the goal in this paper is to study how the orbital evolution inside the magnetospheric cavity depends on the cavity size , planet mass , and orbital eccentricity . in the present work , we only target the mass range from 0.7 to 2 jupiter masses . in the case of the large cavity corresponding to the rotational period days , the planet of mass jupiter mass with moderate initial eccentricities ( ) can move to the region au from its central star in years , while the planet of mass jupiter mass can not . we estimate the critical eccentricity beyond which the planet of a given mass will overflow its roche radius and finally lose all of its gas onto the star due to runaway mass loss . in the case of the small cavity corresponding to the rotational period days , all of the simulated planets lose all of their gas even in circular orbits . our results for the orbital evolution of young hot jupiters may have the potential to explain the absence of low - mass giant planets inside au from their dwarf stars revealed by transit surveys .
0911.4532
the number of discovered extrasolar planets has been increasing quickly since the first extrasolar planet , 51 pegasi b , was detected around a solar - type star @xcite . 51 pegasi b is the prototype of a class of exoplanets known as hot jupiters " because they are generally jupiter - mass planets located @xmath8 au from their parent stars and therefore their equilibrium temperatures are much higher than that of jupiter . the high temperature of the protoplanetary disk at the orbital radii where they are now provides an environment extremely different from where giant planets are expected to form via either the core - accretion model @xcite or the gravitational instability model @xcite . one of the most commonly adopted solutions is that hot jupiters have formed at larger orbital radii in the protoplanetary disk and then moved inward to their current locations via the disk - planet interactions ( e.g. * ? ? ? the magnetic fields of classical t tauri stars ( hereafter ctts ) are typically strong enough to truncate the inner regions of protoplanetary disks to the corotation radius , where the keplerian angular velocity of the disk is the same as the stellar - spin angular velocity , and create an inner magnetospheric cavity . besides , the migration due to the planet - disk interaction is expected to slow dramatically once a young hot jupiter passes the inner disk edge @xcite . the pile - up population of hot jupiters at @xmath9 au from their solar - type parent stars revealed from radial - velocity searches @xcite may be attributed to the existence of the cavity ( * ? ? ? * and references therein ) . on one hand , an attempt has been made by @xcite to search for a hot jupiter around a ctts , although @xcite substituted an explanation of a cool stellar spot for an orbiting hot jupiter . future observations will provide more evidence for planets of early ages . on the other hand , the simulated orbital evolution of giant planets as they migrate into the cavity was carried out by @xcite . their results suggest that high - mass hot jupiters would move closer to their central stars and even be destroyed because planet s entry into the magnetospheric cavity results in a growth in orbital eccentricity @xmath10 . although the eccentricity excitation is subject to the uncertain properties such as disk mass and viscosity , this model may provide a mechanism to pump up the eccentricity of a hot jupiter inside the cavity and in turn affect the orbital evolution of the planet via the tidal interactions between the star and planet . in addition , the magnetic interactions between the star and planet in the magnetospheric cavity would be important as well , which may be an upscale analogy to the magnetic interactions between jupiter and the galilean satellites ( * ? ? ? * and references therein ) . the coupled tidal evolution of the radius and orbit of hot jupiters in the absence of a disk has been investigated @xcite . to obtain noticeable influence on @xmath11 , the tidal heating is assumed to be deposited deep in the planet in these models @xcite . the purpose of @xcite and @xcite is to explain the large radii of some transiting hot jupiters using tidal heating . on the other hand , @xcite showed that for modest eccentricities , young hot jupiters of radius @xmath12 jupiter radii at @xmath13 au swell beyond two jupiter radii and their internal degeneracy is partially lifted on the timescale of @xmath14 gyrs . ] . thereafter , their thermal equilibria become unstable and they undergo runaway inflation until their radii exceed the roche radius . then roche lobe overflow ensues . under the assumption that the overflowing gas immediately loses its orbital angular momentum to the planet and plunges into the central star , these mass - losing planets migrate outwards , such that their semi - major axes and roche radii increase while their mass , eccentricity , and tidal dissipation rate decrease until the mass loss is quenched . of course , including a disk would change the orbital evolutions of roche - lobe - overflowing planets , as has been calculated by @xcite who do not consider the tidal expansion of @xmath11 and the magnetospheric cavity though . as more and more hot jupiters located @xmath15 au have been discovered by transit surveys for recent years , it is noteworthy that almost no giant planets of mass @xmath5 jupiter mass ( @xmath16 ) have been detected within @xmath17 au from their solar - type parent stars ( see figure [ correl ] ; also see * ? ? ? * and references therein ) . this puzzle may be attributed to the subsequent evolutions of hot jupiters after they enter the magnetospheric cavity of the protoplanetary disk . since thermal evaporation ( e.g. * ? ? ? * ) has not been undoubtedly identified to be the major cause ( * ? ? ? * and references therein ) , we shall discuss the potential implication of our model to explain the observation in the end of the paper . for the above reasons , in this work we concentrate on the orbital evolution of a young hot jupiter inside the magnetospheric cavity of a protoplanetary disk by taking into account the tidal and magnetic interactions between the star and planet . to restrict to the problem without significantly involving the planet - disk interaction for simplicity , we consider the evolution after the hot jupiter enters the 2:1 orbital resonance with the inner edge of a disk . the initial eccentricity , which is expected to be excited in the cavity before the planet migrates to the resonance location , is treated as one of the free parameters of the problem . the equations for the orbital and interior evolution of a hot jupiter inside a magnetospheric cavity are described in 2 . the results are presented in 3 , and are summarized and discussed in 4 . figure [ model ] illustrates our toy model that shows a hot jupiter orbiting a ctts in a magnetospheric cavity of a protoplanetary disk . the magnetic dipole fields of a protostar truncate the inner part of the protoplanetary disk at the disk radius @xmath18 , which is the radius of the cavity . the initial orbit of the planet is set to lie at the 2:1 orbital resonance with the inner edge of the disk , which allows us to ignore the tidal interactions between the disk and the planet @xcite . the semi - major axis @xmath19 of this resonant orbit is denoted as @xmath20 . however , when the inner edge of the disk extends inwards for any reasons in our model such that @xmath21 , the planet is artificially moved inwards as well to maintain the 2:1 orbital resonance with the inner edge of the disk . moreover , we assume that there is no magnetic linkage between the disk and the planet as illustrated in figure [ model ] and therefore neglect magnetic interactions between them . in other words , the planet s orbit evolves due to tidal and magnetic interactions with the central star , whereas the star interacts with both the planet and the disk . therefore , this section consists of two parts , namely the magnetic and tidal effects , to describe the model in detail . 2.1 and 2.2 describe the star - disk and star - planet magnetic interactions , which leads to disk locking and contributes to the orbital evolution of the planet , respectively . 2.3 and 2.4 then describe the tidal interactions between the ctts and the planet . the circularization of the planet s orbit is due to tides on both the planet and the star . the thermal inflation of the planet is due to tides on the planet excited by the star . on the other hand , the tides on the star raised by the planet affect the evolution of the planet s orbit and the stellar spin . finally , we summarize the key formulae for our model in 2.5 . there have been a number of models for the star - disk interactions to account for the rotation of cttss as well as star and jet formation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition to the closed field lines connecting from the protostar to the disk to allow for funnel accretion , the open field lines may diverge primarily from the inner edge of the disk ( i.e. x - wind model , see * ? ? ? * ) or emanate largely from the star and the disk ( i.e. stellar and disk wind model , see * ? ? ? * ) . furthermore , the stellar field topology may be in the complex form of multipole @xcite or tilted dipole relative to the stellar spin @xcite . in this work , we ignore these complications and adopt the simple non - tilted dipole configuration of stellar fields that thread through the protoplanetary disk as has been illustrated in figure [ model ] . to calculate the magnetic torque @xmath22 on the star due to the magnetic linkage to the disk , we adopt the approach by @xcite in which the magnetic torque acting on an annulus of the disk is due to the azimuthal twisted field lines @xmath23 caused by the difference between the stellar spin and the orbital motion of the annulus @xcite . hence , the torques integrated from the disk inner edge at @xmath18 to infinity contribute to the net magnetic torque between the star and the disk : @xmath24 where @xmath25 is the stellar dipole moment , @xmath26 is the magnetic field strength at the stellar surface , @xmath27 is the stellar radius , @xmath28 is the corotation radius , @xmath29 is the stellar spin rate , and @xmath30 is the stellar mass . we employ the same scaling law as used in @xcite to specify the strength of stellar magnetic field : @xmath31 where the scaling factor @xmath32 is a constant indicative of the surface magnetic strength for the stellar spin period of 4 days . the magnetospheric radius @xmath18 is determined by the balance between the stellar magnetic stress and the disk reynolds stress ( starczewski et al . 2007 , and references therein ) : @xmath33 where @xmath34 is the disk accretion rate and @xmath35 is a dimensionless factor of order unity . for a star with an aligned dipole field , @xmath36 . in addition , the mass transfer rate throughout the disk is assumed to be comparable to an observational inferred accretion rate from protostellar disks onto cttss , which decays with time @xmath37 over the timescale @xmath38 years and can be fitted by @xmath39 where @xmath40 is the earth mass ( ida & lin 2004 ) . the equation for the spin evolution of the protostar is then given by @xmath41 in the above equation , @xmath42 is the moment inertia of the ccts and @xmath43 is the net torque arising from the presence of the disk , where @xmath44 is the torque spinning up the star due to the disk accretion : @xmath45 with @xmath46 being the keplerian angular speed at the radius @xmath18 @xcite . @xmath27 and @xmath42 are obtained from the interior structure of the contracting ctts modeled by @xcite , who modified the eggleton code @xcite for the pre - main sequence evolution . when the star rotates faster than the inner edge of the disk ( i.e. @xmath47 ) , the mass accreting onto the star should be ceased by the centrifugal force and @xmath44 is therefore set to be zero in our model . meanwhile , the star experiences immediate magnetic braking as a result of its strong fields , leading to disk locking of the stellar spin . disk locking makes the star and the inner disk edge corotate again and resumes the accretion . therefore precisely speaking , the disk density near the inner edge of the disk , @xmath18 , and @xmath48 should fluctuate with time . in practice , we have simply employed equations ( [ eq : r_mdisk ] ) and ( [ eq : mdot ] ) to implement the disk properties rather than modelling the detailed disk structure near @xmath18 . this is certainly a limitation of our model . with regard to the overall disk accretion rate , as we shall see in 3 , the fluctuation of the balance between the mass accretion and the magnetic braking ( i.e. disk locking ) is rather fast compared to the timescales of interest , meaning that equation ( [ eq : mdot ] ) is a proper approximation to describe the average mass accretion rate in our problem . the torque on the planet due to magnetic linkage to the star is modelled as the magnetic stress ( @xmath49 ) multiplied by the cross - sectional area of the planet s magnetosphere and the distance @xmath50 between the planet and its parent star . it is assumed that the planet has a magnetosphere in which the gas behaves like plasma . in addition , the magnetosphere is assumed to be as large as the planet s size , so the area of the planet interacting with the stellar field is @xmath51 . applying the star - disk magnetic interaction described in 2.1 to the star - planet case , we obtain the magnetic torque on the planet @xmath52 where @xmath53 is the orbital angular velocity . the magnetic torque changes with the distance between the planet and the central star while the planet moves in an eccentric orbit , so we employ the mean torque over one orbit period @xmath54 . that is , @xmath55 to carry out the above integral , we make the approximation that the orbit does not vary significantly during one orbital period ; i.e. , the orbital angular velocity @xmath56 and @xmath57 , where @xmath19 is the semi - major axis , @xmath10 is the eccentricity , and @xmath58 is the mean motion of the mutual orbit of the planet and the star . then after integrating and being divided by @xmath54 , equation ( [ equation3 ] ) gives an expression for the mean magnetic torque on the planet : @xmath59 . \label{eq : t_planet}\ ] ] here , we have introduced a coefficient @xmath60 , where @xmath61 , to regulate the magnetic torque on the planet , a formulism similar to that applied to the magnetic torques between the galilean satellites and jupiter ( zarka 2007 ) . in this regard , the energy dissipation associated with the magnetic torque can be generally considered as the poynting flux at the magnetopause scaled by @xmath60 . hence , @xmath62 corresponds to the upper limit of the magnetic torque . the planet is assumed to have zero obliquity and zero orbital inclination . the gravitational tides on the planet affect the evolution of the eccentricity and the spin of the planet . for the sake of simplicity , we do not consider the influence of the planet - star magnetic interactions on the spin and eccentricity of the planet . therefore , the rate of the eccentricity change of the orbit is due solely to gravitational tides on the planet and the star . the rate contributed from the tidal dissipation in the planet alone is given by @xmath63 , which can be obtained from ( mardling & lin 2002 ) @xmath64 , \label{eq : tidalfrict}\ ] ] where @xmath65 for the tidal dissipation in the star is also shown and will be considered in 2.4 . in the above equation , @xmath66 and @xmath67 . moreover , @xmath68 and @xmath69 are the tidal quality factors of the star and the planet , respectively . all the uncertainties associated with the physical processes of tidal heating are contained in the @xmath70 values . for a fiducial value , it has been suggested that the @xmath71 values inferred for jupiter from io s orbital evolution is @xmath72 ( yoder & peale 1981 ) . with this @xmath71 value , orbits of planets with @xmath73 and @xmath11 comparable to those of jupiter and with a period less than a week , are circularized within the main - sequence life span of solar - type stars . @xmath68 has been inferred to be @xmath74 for the very young binary stars since the synchronous timescale of the planet is much shorter than the circularization timescale @xcite , we simply assume that the planet s spin equals the orbital rotation throughout the evolution . this assumption may be invalid during the phase of the roche - lobe overflow if the orbit changes dramatically due to quick mass loss ( see eq . ( [ equation18 ] ) below ) , causing a large discrepancy between the planet spin and orbital motion . this means that the tidal heating due to synchronization is unrealistically ignored during this special phase in our model . moreover , the tidal process of orbital circularization shrinks the planet s orbit ; i.e. , @xmath75 during the orbital circularization , there is substantial tidal heating within the planet . the tidal heating of the planet may have contributed significantly to the thermal budget that governs the planet s physical properties , including its radius @xcite . ignoring the rotation energy of the planet , we can write the first law of the thermodynamics for a tidally heated planet as follows : @xmath76 where @xmath77 is the tidal heating rate ( see mardling & lin 2002 for the expression ) , @xmath78 is the intrinsic luminosity from the photosphere of the planet , @xmath79 is the internal energy , and @xmath80 is the gravitational potential energy . suppose that @xmath81 and @xmath82 with the coefficients @xmath83 and @xmath84 depending on the interior structure of the planet . then in the absence of mass loss ( i.e. @xmath85 ) , the above energy equation gives rise to the expansion / contraction rate of the planet ( cf . * ) : @xmath86 even though the expanding / contracting planet is not in the thermal equilibrium , @xmath83 , @xmath84 , and @xmath78 can be simply expressed in terms of functions of @xmath11 and @xmath73 as long as the timescales under consideration are @xmath87 the convective turnover time for an almost convective planet with the same solid - core mass and a thin radiative envelope of the same opacity ( i.e. an almost polytropic interior , see gu et al . 2003 , 2004 ) . in our approach , the core mass of the planet is assumed to be @xmath88 , which is motivated by the core accretion model for the formation of giant planets ( pollack et al . however , the thermal properties of the core are not modelled in our work . @xmath89 , @xmath90 , and @xmath91 are obtained only from certain different @xmath73 by fitting to the numerical data for the interior structures of tidally inflated jupiters described in @xcite . subsequently , their values and derivatives with respect to @xmath11 for any desired @xmath73 can be interpolated from these available fitted functions . once the planet s radius exceeds its roche radius @xmath92 ( see the appendix ) , the mass loss ensues . during the planet overflow phase , the @xmath11 adjustment after losing mass plays an important role in determining the subsequent evolution . @xmath11 changes with @xmath73 as well according to the equation of state of the planet even though the entropy of the planet does not change . since the interpolated data derived from our interior structure code is unable to deal with the processes that are as quick as the adiabatic mass loss , we adopt the following simple methodology . assuming a polytropic structure of a planet , the adiabatic mass - radius exponent @xmath93 can be estimated from the polytropic index of the planet @xcite . @xmath94 can then be related to the gravitational potential energy . it can be easily shown that @xmath95 . for simplicity we shall use this relation to estimate @xmath94 from @xmath84 for a roche - lobe filled planet , albeit these polytropic relations are derived for an isolated gravitationally bound body . therefore , the complete expression of @xmath96 in the presence of the mass loss through the lagrangian 1 ( l1 ) point should read @xmath97 where we shall adopt the formalism in @xcite to estimate @xmath98 ( see 2.5 for the detail ) . in the case of a giant planet with large radius such as a young hot jupiter , @xmath94 is always negative . in addition , @xmath99 increases with @xmath11 , because the planet degeneracy is being lifted when its density goes down @xcite . similar to the physical condition for the formation of the magnetospheric cavity of the disk , planet s overflow may either form a ring / disk or funnel along the stellar field lines onto the ctts depending on whether @xmath98 is strong enough to dominate over the stellar fields . therefore , we modify equation ( [ eq : r_mdisk ] ) by replacing @xmath48 with @xmath98 to define the overflow cavity radius @xmath100 " where the reynolds stress of the overflowing gas is comparable to the magnetic stress of the stellar fields . thus , if this radius is larger than planet s semi - major axis , where the overflow occurs , the overflowing gas will flow through the magnetic field to the central star . the stellar spin will speed up under the action of the torque from planet s overflow , @xmath101 , which is an analogy to the spin - up torque due to the disk accretion described in equation ( [ equation17 ] ) . this leads to the evolution of the stellar spin caused by the planet s overflow : @xmath102 on the other hand , if the overflow cavity radius is smaller than the semi - major axis , the overflowing gas will form a ring / disk around the star instead of directly funnelling along the field lines onto the star . in this stage , we assume that the total angular momentum of the overflowing gas is immediately transferred to the orbit of the planet , leading to the outward migration of the mass - losing planet : @xmath103 and therefore no spin - up torque is exerted on the star . the above equation for outward migration suggests that the mass - losing planet can be detached from its roche lobe as long as @xmath11 does not expand as fast as @xmath92 . therefore , it is worthwhile to examine the linear stability of a roche - lobe filled planet against mass loss . the definition of @xmath92 from the appendix and equation ( [ eq : a_outward ] ) give @xmath104 . equation ( [ eq : rdot ] ) then indicates that the expansion rate of @xmath11 is @xmath105 provided that the thermal adjustment of @xmath11 is less important than the adjustment of @xmath11 due to the adiabatic mass loss . once the planet s photosphere reaches the roche radius ( i.e. @xmath106 ) , the above analysis infers that the mass loss of a roche - lobe filled planet is linearly unstable ( i.e. undergoes exponential growth ) if @xmath107 . in addition to the tides on the planet induced by the star , the planet also raises the tides on the star to circularize the orbit and to lock the rotation of the star . the circularization rate described in the preceding subsection has to be complemented by the tidal dissipation in the star ; namely , @xmath108 where @xmath65 has been displayed in equation ( [ eq : tidalfrict ] ) . the process of the tidal locking of the ctts leads to the exchange of the angular momentum between the planet s orbit and stellar spin . the stellar spin is assumed to be normal to the orbital plane of the planet and the disk . owing to this tidal effect , the angular momentum of the stellar spin changes at the rate @xcite @xmath109 where @xmath110^{1/2}$ ] is the total angular momentum of the planet s orbit , @xmath111 , and @xmath112 . the evolution of stellar spin associated with equation ( [ eq : j*dot ] ) is then given by @xmath113 . \label{eq : omega_*}\ ] ] after describing the various interactions separately , we list and recapitulate the final equations incorporating these interactions for solving the orbital evolution of a young hot jupiter in a magnetospheric cavity . the rate of orbital circularization ( i.e. eq . ( [ eq : circularization ] ) ) is given by @xmath114 the evolution of planet s size ( i.e. eq . ( [ eq : rdot ] ) ) has the form @xmath115 the first term on the right side is due to thermal adjustment , whereas the second term is due to roche - lobe overflow driven by an adiabatic process that occurs much faster than the first term . when @xmath116 , the mass loss occurs in our simulation . this implicitly implies that we neglect the mass loss from the optically thin atmosphere before the planet s photosphere reaches its roche radius . we also ignore the possible mass loss through the l2 point @xcite . the mass - loss rate of the planet is estimated from the 1-d interior structure of the planet along with the formula in kolb & ritter 1990 ( and see the definitions of the notations therein ) : @xmath117 the orbital decaying rate of the planet is given by @xmath118 the first term on the right side represents the orbital decay due to circularization ( i.e. eq . ( [ equation19 ] ) ) . the second term is due to the magnetic torque on the planet ( i.e. eq . ( [ eq : t_planet ] ) ) and the third term arises from tides on the star ( i.e. eq . ( [ eq : j*dot ] ) ) . the last term is included only when the overflow occurs ( i.e. eq . ( [ eq : a_outward ] ) ) . moreover , when @xmath18 shrinks such that the 2:1 resonance location lies in the disk , the planet is expected to slowly migrate inwards @xcite . however , when this happens , we assume just for the sake of simplicity that @xmath19 is decreased instantaneously to maintain the 2:1 resonance with the inner edge of the disk . the stellar spin evolution is determined by a number of factors : @xmath119^{1/2}\}+\dot{\omega}_\ast . \label{equation21}\ ] ] the first term in the curly bracket arises from the evolution of the stellar moment of inertia during the pre - main - sequence phase and the second term represents the torque exerted by the disk on the star due to disk accretion and the magnetic linkage ( i.e. eq . ( [ equation15 ] ) ) . the third and the fourth terms are due to the magnetic linkages with the planet ( i.e. eq . ( [ eq : t_planet ] ) ) and the overflowing gas from the overfilled roche lobe ( i.e. eq . ( [ equation18 ] ) ) , respectively . the last term @xmath120 results from the tides on the star ( i.e. eq . ( [ eq : omega _ * ] ) ) . to solve the odes in the preceding section ( eq . ( [ eq : circularization ] ) , ( [ eq : rdot ] ) , ( [ eq : m_dot ] ) , ( [ equation20 ] ) , ( [ equation21 ] ) ) , we employ the 4th order runge - kutta method to study the inward migration of a young hot jupiter in the magnetospheric cavity due to tidal and magnetic interactions with its ctts . the simulations start from @xmath121 yr to @xmath4 yr . the planet is not introduced to the orbit with @xmath122 until @xmath123 years , a timescale comparable to the type ii migration time of a giant planet in a proto - planetary disk @xcite . in our model , @xmath32 , @xmath10 , and @xmath73 are parameterized to investigate the various evolutions of the planetary migration in the magnetospheric cavity . besides , @xmath124 , @xmath71 and @xmath35 are fixed as @xmath125 , @xmath126 , and @xmath127 throughout the simulation . the initial period of the central star is set to be 15 days at @xmath121 years ( see the left panel of figure [ locking ] ) . the figure shows the stellar spin will be locked by the disk in a few @xmath128 years and therefore the initial spin is insensitive to the tidal and magnetic interactions , which proceed on the timescales of @xmath129 years . the final disk - lock period ( i.e. the keplerian period of the disk inner edge ) depends on the magnitude of the magnetic field of the central star @xcite . we discuss two cases of cavity sizes corresponding to @xmath130 g ( 3.1 ) and @xmath131 g ( 3.2 ) separately . the disk with a big cavity will lock the central star at the rotational period of @xmath132 days and that with a small cavity will lock it at @xmath133 days . @xcite revealed the peculiar bimodal distribution of periods for young stars in the orion nebula cluster . the distribution of periods at an age of about 1 myr exhibit two peaks , one located at a period of about 8 days and the other at a period of about 2 - 3 days @xcite . the disk - lock periods in our model approximately match the observation . we consider the planets of four different planetary masses @xmath134 , @xmath135 , @xmath136 , and @xmath137 to study their different evolutions of migration . from the observations , the close - in transiting giant planets have radii ranging from @xmath140.8 @xmath138 to @xmath141.7@xmath139 . thus 1.7@xmath138 could be considered as a lower limit of the radius of a young jupiter although some of the large planet s sizes have been postulated to be attributed to tidal heating @xcite or evaporation @xcite . also , 1.7 @xmath138 corresponds to the theoretical radius of a young jupiter ( marley et al . thus we fix the initial @xmath140 for all the cases . when @xmath123 yr and the planet is introduced to the initial location at @xmath122 , the star has been locked by the disk . the right panel of figure [ locking ] shows the decrease in the periods of stellar spin and planet s orbit before @xmath141 years . this is caused primarily by the contraction of the cavity as @xmath27 decreases ( see eq . ( [ eq : r_mdisk ] ) ) . orbiting interior to the co - rotation radius , the planet then migrates inwards from the 2:1 resonance due to faster revolution of the planet than the spin of the central star . this is illustrated by the black curve during the time @xmath142 years in the right panel of figure [ locking ] where the planet with @xmath136 and @xmath143 is used as an example . meanwhile , the star should spin up via the tidal and magnetic interactions with the planet . however , the disk locking dominates the evolution of the stellar spin as indicated by the overlap between the red solid and blue dashed curves in the right panel of figure [ locking ] . in other words , the inner edge of the disk acts as the sink of total angular momentum of the system in our model in fact , the rotational period of the disk inner edge fluctuates slightly from the spin period of the star on short timescales . nevertheless , these two periods are almost the same over long timescales , justifying the argument given in 2 that equation ( [ eq : mdot ] ) is a proper approximation to describe the average mass accretion rate and determine the cavity size in our problem . as mentioned in 2.3 , the planet will lose its mass if the planet s radius exceeds the roche radius . however , if the planet s mass decreases to 20 earth masses , the simulation will be forced to stop . this is because the core mass of a giant planet is 19.4 earth masses in our model . a large cavity is opened up by the strong stellar magnetic fields with the scaling factor @xmath130 g. in this case , the planet starts to migrate from @xmath144 au , which corresponds to @xmath20 at @xmath123 years . the migration evolutions of the planets with same masses will differ as a result of different initial eccentricities . the distance , over which the planet can migrate inward from the initial position , is shortest when @xmath145 and increases as @xmath146 increases . this is because firstly , the circularization of a eccentric orbit makes the orbit shrink . secondly , the inflation of a planet due to tidal heating enhances the star - planet magnetic interaction . also , the tidal torque on the star caused by the planet is more significant as @xmath146 is larger . all of the above factors accelerate the migration rate of the planet . when @xmath146 @xmath147 the critical eccentricity @xmath148 , the planet s radius will exceed its roche lobe and start to lose mass onto the central star . we perform the calculations for three cases of @xmath146 : ( 1 ) @xmath145 ; ( 2 ) @xmath149 ; ( 3 ) @xmath150 for each planet mass . the orbital evolutions are shown in figure [ bigcavity ] and the explanations are presented below . in the case of @xmath145 ( solid lines in the figure ) , the results for different masses are not significantly difference . without an eccentric orbit to bring the planet closer to the central star ( i.e. the perihelion ) , the tidal and magnetic effects are excited less effectively in a large distance between the planet and the central star . hence , planets with different masses initially in circular orbits migrate to almost the same final positions , @xmath151 au at @xmath152 yr , as the cavities contract and drive the planets to the 2:1 resonances . when @xmath149 ( dotted lines in the figure ) , the planets start to overflow their l1 points and lose mass . as shown in the figure , their evolutions are terminated before the end of the simulations ( @xmath152 years ) , suggesting that all of the planets are disrupted . the termination of the simulations occurs when the discrepancy between @xmath11 and @xmath92 is so large during the overflow phase that @xmath153 is too large to be resolved in the simulation with the time step @xmath154 years . this time step of 100 years is already shorter than the time step used for the interpolated data derived from our interior structure code . when the runaway solutions happen , @xmath99 is always larger than 5/3 , which is in agreement with the linear instability of the mass loss . for the @xmath155 and @xmath136 cases , the mass loss rate is high . the overflow gas is able to form a ring / disk and hence the planet migrates outward according to equation ( [ eq : a_outward ] ) in our model . however , once their overflows occur , @xmath156 at the same time owing to their large @xmath11 . the radius adjustment is predominated by the adiabatic mass loss instead of tidal inflation , leading to the runaway solutions to the odes . on the other hand for the @xmath135 and @xmath134 cases , they migrate fast to the locations closer to the central stars owing to the stronger tides raised by these massive planets on their central stars . in addition , these more massive planets are more difficult to inflate by tidal heating due to their higher gravitational binding energies and cooling rates @xcite . hence , when they overflow their roche lobes , @xmath11 is not too large and therefore @xmath157 , leading to the stable overflow . although the overflowing gas initially funnels to the central star , the overflow rate becomes substantially high shortly such that @xmath158 . consequently the massive planets have undergone outward migration almost since the overflow began . the outward migration is primarily driven by inward migration due to the tidal and magnetic interactions rather than tidal inflation . the overflow finally turns to be unstable when @xmath99 increases to 5/3 as @xmath19 and therefore @xmath11 increases . the evolution then stops and the planet loses all its gas to the central star . during the phase of the fast outward migration , the planet s spin may become unlocked with its orbital motion . despite not being included in our model , the tidal heating associated with the asynchronism would speed up the planet inflation and lead to the same catastrophic fate of the planet . comparing these cases for different initial @xmath73 , massive planets require higher @xmath148 thus more tidal heating to be inflated against their greater gravitational binding energies and cooling rates . @xmath159 requires @xmath160 and destructs at @xmath161 au , while @xmath162 needs @xmath163 and destructs at @xmath164 au . now we turn to the case for @xmath150 ( i.e. @xmath146 is slightly smaller than @xmath148 ) to study the conditions in which the planet of a given mass can migrate inward the farthest and still survive in the end of the simulation . as can be expected from the previous results for @xmath149 , the dashed curves in the figure show that a more massive planet undergoes faster migration due to the stronger tidal torque . thus , the planet with @xmath159 and @xmath165 can arrive at @xmath166 au , while the planet of @xmath167 and 1 @xmath16 can not migrate to the region @xmath168 au . furthermore , a low mass planet is more easily inflated by tidal heating , giving rise to the relatively stronger magnetic interaction ( see eq . ( [ eq : t_planet ] ) ) . this explains why the planet of @xmath169 migrates slightly farther in than the planet of @xmath136 in the end of the simulation . in equaiton ( [ eq : t_planet ] ) , the coefficient @xmath60 regulates the magnetic torque on the planet , and the upper limit @xmath62 has been applied to obtain the above results . here , we employ @xmath170 to study the orbital evolutions in the absence of the magnetic torque . in comparison to the previous orbital evolutions shown in figure [ bigcavity ] , figure [ eps0 ] shows the lower migration rates because of the lack of the star - planet magnetic interaction for the @xmath1710.7 , 1 , and 1.5@xmath16 cases . however , the stellar tides induced by the @xmath134 planet are far dominant over the magnetic interaction , so the massive planet still migrates to @xmath172 au . that position is even farther inward than @xmath166 au for the @xmath62 case as the planet with even higher @xmath173 can still survive in the absence of the magnetic torque . moreover , figure [ eps0 ] confirms the previous subtle result about the dependence of the magnetic torque on @xmath11 : the slightly more inward migration of the planet of @xmath137 than @xmath136 does not reappear because these exists no magnetic torque to speed up inward migration for a lighter planet of a larger inflated radius . as to the cases for @xmath149 , the simulated planets overflow their roche radii and perish in the same way as in the @xmath62 cases ; namely , @xmath99 of various cases finally is @xmath174 and the mass loss becomes runaway . for the cases of @xmath137 , @xmath136 , and @xmath135 , @xmath156 once their overflow occurs . on the other hand , after reaching the roche lobe , the planet of @xmath134 migrates outward for a period of time due to stable overflow and inward migration driven by the tidal torque . it is finally destroyed due to the large @xmath99 that increases with @xmath11 as the mass - losing planet moves outwards . the weaker stellar fields due to the smaller scaling factor @xmath131 g form a smaller magnetospheric cavity , so the initial position of the planet is much closer to the central star ; namely , @xmath175 au . the planet and the star exert remarkable tidal forces on each other in such short distances , and the magnetic torque on the planet is not negligible as well . @xmath92 is getting smaller while the planet migrates inwards . as soon as @xmath176 , the overflow occurs , leading to the destruction . consequently , even @xmath145 ( i.e. , no tidal heating in the planet ) , all the planets destruct irrespective of their mass . if @xmath73 is bigger , the migration is faster due to the stronger tidal torque . therefore , the more massive the planet is , the quicker the planet destructs , as shown in figure [ eps0sc ] for @xmath62 . the orbital evolutions of the the @xmath136 and @xmath134 cases are characterized by the alternating out / inward orbital migrations . once the overflow occurs , the large expansion of @xmath92 due to the high mass loss rate dominates over the adiabatic inflation , hence leading to the fast outward migration . as a result , @xmath11 temporarily detaches from @xmath92 for a while until the planet moves in and the next roche overflow occurs again . finally , the planet will lose all its mass when the overflow occurs accompanied with high @xmath99 ( i.e. @xmath174 ) . in the absence of the magnetic torque ( i.e. @xmath170 ) , the tides raised by massive planets on the stars still drive significant inward migration of the planets . figure [ eps0sc ] shows that shutting down the magnetic torque can not change the outcome of planet destruction for @xmath177 . however , the planet of @xmath169 can survive in @xmath4 years because the inward migration driven by the weak tidal torque from the low mass planet is slow . therefore , we estimate @xmath178 for the planet to perish and show the evolution in the leftmost panel of figure [ eps0sc ] . we construct a simple model to study the orbital evolution of a young hot jupiter in an eccentric orbit inside a magnetospheric cavity of a proto - planetary disk around a ctts . for the sake of simplicity , we assume that the magnetospheric cavity of the protoplanetary disk is truncated by a stellar dipole field . through the magnetic linkage between the central star and the disk , the rotational period of the star is quickly locked by the inner edge of disk . we then introduce a young hot jupiter at @xmath20 and restrict ourselves to the orbital evolution of a hot jupiter inside the cavity . assuming that the planet behaves like a plasma , we apply the same formulism for the star - disk magnetic interaction to that for the star - planet interactions . we adopt the equilibrium - tide equations with @xmath179 and @xmath180 to model the star - planet tidal interactions . in so doing , our model focuses on the planet migration due to tidal and magnetic interactions between the planet and the star , and does not consider any interactions between the planet and the disk . however , as the size of the magnetospheric cavity evolves , the planet is artificially pushed to @xmath20 if it lies beyond @xmath20 . we vary three parameters @xmath32 ( @xmath181 and 1500 g ) , initial @xmath73 ( @xmath182 , 1 , 1.5 , 2 @xmath16 ) , and initial @xmath10 ( @xmath145 , @xmath183 , and @xmath148 ) in our simulations to investigate the fate of the planet migration under the influence of the cavity size , planet mass , and orbital eccentricity . two sizes of magnetospheric cavities are considered to approximately match the bimodal distribution of spin periods of young stars in the orion nebula cluster . the initial @xmath20 corresponding to the large ( small ) cavity opened up by the stronger ( weaker ) stellar fields is @xmath184 au ( @xmath185 au ) . in the case of the large cavity ( @xmath130 g ) , the planets require nonzero @xmath146 to enhance their tidal and magnetic interactions with their central stars and make significant inward migration . the migration rate increases with @xmath146 for a given planet mass . when @xmath146 is as large as the critical value @xmath148 , @xmath11 exceeds @xmath92 and the roche overflow occurs . for planets with different masses , massive planets ( @xmath186 ) have higher @xmath148 ( i.e. more intense tidal heating ) than less massive ones ( @xmath187 ) because massive planets are more difficult to thermally inflate due to their greater gravitational binding energies and cooling rates . as a result , stronger stellar tides raised by the planet with higher @xmath73 and @xmath146 allow for faster migration . overall , massive planets can therefore migrate further in than less massive ones , as shown in the simulations for @xmath150 . when @xmath149 , low - mass planets ( @xmath167 and 1 @xmath16 ) inflate by tidal heating faster than inward migration . hence they overflow their large roche lobes at large @xmath19 , resulting in the low density and therefore the runaway mass loss . on the other hand , the high - mass planets ( @xmath188 and 2 @xmath16 ) migrate inwards fast without significant tidal inflation . in contrast to the low - mass planets , once these planets fill their small roche lobe at small @xmath19 , their density is high enough that they can undergo stable l1 overflow . during this phase , tidal and magnetic interactions instead of tidal inflation drive them to lose mass , expand , and migrate outward . this stable mass - loss phase proceeds until these outward migrating planets are large enough to become unstable against mass loss . finally , they lose all gas to their central stars . in the case of the small cavity ( @xmath131 g ) , the simulated planets in circular orbits all quickly migrate in due to the fierce tidal and magnetic interactions until they stably overflow their roche radii . after that , they can move outwards due to the mass loss and move inwards again by the tidal and magnetic torques . as their mass goes down and their degeneracy is lifted , all of these planets are finally destroyed , suffering from the runaway mass loss as in the large - cavity case . to study the significance of the magnetic interactions between the star and the planet in our model , we also simulate the cases by shutting down the interaction ( i.e. @xmath170 ) . we found that the migration of less massive planets is more sensitive to the magnetic interaction , which is enhanced by their easily inflated radii . setting @xmath170 makes the migration rate of low - mass planets slower , as shown in the case for @xmath189 with @xmath190 in the large - cavity case . by contrast , the inward migration of massive planets is less sensitive to the magnetic interaction but more to the tidal interaction . as has been summarized in the preceding paragraph , all planets in the small cavity destruct under both the tidal and magnetic torques at such close - in initial positions from the stars even if @xmath145 . when @xmath170 , only the planet of @xmath137 can survive unless the low - mass planet starts with an eccentric orbit with @xmath178 . the model presented here for the orbital evolution of young hot jupiters is different from the works by @xcite and @xcite . in this work , we employ simple models including the star - planet interactions and disk locking , which are not considered by @xcite . furthermore , we take into account the planet expansion due to the adiabatic mass loss and model the mass loss rate using the approach in @xcite , whereas @xcite ignore the radius adjustment to the mass loss and use a free parameter to control the mass loss rate . ignoring the mass - radius relation during the roche - lobe overflow phase as done in @xcite , a roche - lobe filled giant planet with @xmath149 continuously loses its mass via tidal inflation and migrates outwards until the tidal heating rate is weaker than the cooling rate , leading to a survived planet of lower mass . on the other hand , @xcite consider young hot jupiters moving inwards via the type ii migration without a magnetospheric cavity and tidal inflation . in their work including the mass - radius relation , once the planets of mass @xmath191 get close enough to their central stars and therefore overflow their roche lobes , they migrate outwards and finally lose all their mass onto the star . in this regard , our results are compatible with theirs . one of the major differences is that the mass loss is driven by the type ii migration in the model by @xcite , while in our work the mass loss is driven by the loss of orbital angular momentum via the tidal and magnetic interactions with the parent star whose spin is almost locked by the disk . moreover , the planets of mass @xmath192 with @xmath193 survive in the cavity during the ctts phase in our model . for the last few years , transit surveys have revealed the peculiar @xmath194 correlation for hot jupiters : less massive hot jupiters ( @xmath195 jupiter mass ) are almost absent within @xmath196 au from their parent dwarf stars of the ages @xmath14 gyrs . our results show that during the ctts phase , planets of @xmath197 with modest initial eccentricities ( @xmath198 ) and sufficient magnetic interactions in a large cavity can migrate to @xmath199 au , while planets of @xmath200 can not . more specifically , given @xmath193 , the planet of @xmath134 can safely arrive at @xmath201 au and the planet of @xmath135 can get to @xmath202 au in the end of our simulations . whether or not these planets can further migrate toward their spun - down parent stars during the main - sequence phase depends on @xmath68 , which is not yet well understood . @xcite suggested that @xmath68 induced by hot jupiters may become large ( i.e. @xmath203 ) as cttss evolve to main - sequence stars based on their dynamical - tide model . as they and others @xcite pointed out , we would be extremely fortunate to observe these very close - in hot jupiters if the current @xmath204 , implying that @xmath68 of the solar - type main - sequence host stars may be quite large . however , it should be noted that these very close - in planets were all discovered by transit surveys ( see figure [ correl ] ) . although transit surveys overall prefer to detect close - in planets @xcite , the detectability of a hot jupiter as a function of the semi - major axis of interest here ( i.e. @xmath205 au ) is extremely ill - defined as the various transit surveys are subject to different observational limitations such as the transit depth , sampling rate , red noise , and detection threshold @xcite . if @xmath68 indeed gradually becomes @xmath203 as cttss evolve to main - sequence stars , therefore allowing the planets to move a little bit inward during the short transition after @xmath152 years while halting the further planet migration for most of the main - sequence phase , our model may have the potential to explain the absence of low - mass hot jupiters within @xmath206 au . on the other hand , a large body of studies have suggested that @xmath207 associated with the tidal dissipation induced by the hot jupiters in main - sequence dwarfs could be as small as @xmath129 . as a result of the efficient tidal dissipation , the orbits of the hot jupiters can decay so significantly that the planets may plunge into their host stars during the main - sequence phase @xcite . this accretion " model may account for the paucity of the extremely close - in hot jupiters concluded from radial - velocity surveys ( * ? ? ? * also see figure [ correl ] ) . in addition , the relatively young ages of the extremely close - in hot jupiters revealed by transit surveys may also lend support to the model @xcite . although the planet - metallicity correlation independent of the spectral type does not seem to favor the accretion model @xcite , excessive heavy elements in the shallow convection zone of high - mass dwarfs may be able to sink down to the radiative zone via the double diffusive effect , thereby eliminating the metallicity enhancement and resolving the problem for the model @xcite . furthermore , @xcite found an excess spin of the host stars of hot jupiters , which may arise from tidal spin - up , although the uncertainty of the stellar ages and the limitation of a small number of samples demand more detections in the future to confirm the conclusion . the study by @xcite for the accretion model is carried out based only on one - mass case ( i.e. @xmath136 ) . if the accretion model is the main cause responsible for the observed @xmath194 relation , it would be expected that the extremely close - in giant planets of mass @xmath208 of younger ages should also be detected by transit surveys . the orbital decay timescale during the main - sequence phase due only to the tidal dissipation is given by @xmath209 ( see equation([eq : j*dot ] ) ) . in the case of a planet of 0.7@xmath210 , the planet can survive in @xmath211 yrs from the distance of @xmath196au at @xmath212 myrs . in contrast , the orbit of a planet of @xmath213 will decay on the timescale of @xmath214 yrs from a distance of @xmath215au at @xmath212 myrs . in other words , during the main - sequence phase , less massive planets excite weaker tides on their parent stars and therefore still stay outside @xmath216 au , while the massive planets keep falling into the stars in the way as interpreted by @xcite and @xcite . the recently discovered hot jupiter wasp-18b with an orbital period of 0.94 days and a mass of @xmath217 may provide a contraint on @xmath218 of a main - sequence star in the next few years @xcite . in this paper , we have attempted to bring several physical components to the context of the evolution of young hot jupiters . nonetheless , a large number of assumptions have been made to simplify the physical processes included in our model . the structure of the truncated disk is not modelled . how a planet migrates to @xmath20 and how the eccentricity evolves during its entry to the cavity ( e.g. * ? ? ? * ) are not addressed . the disk accretion rate is assumed to be quasi - steady to determine the cavity sizes , while observations show otherwise ( e.g. * ? ? ? the dipole configuration of the stellar fields connecting to the disk ignores any types of winds even though the effect of disk locking is introduced in a simplified manner to imitate the sink of the total angular momentum in our model . the star - planet magnetic interactions are parameterized to scale with the poynting flux at the magnetopause of the planet rather than appealing to any specific models , such as the dissipation due to the alfvn waves and unipolar inductor ( e.g. * ? ? ? * ) or induced by tilted stellar dipole fields ( e.g. laine et al . 2008 ; cf . papaloizou 2007 ) . the values of @xmath68 and @xmath69 chosen for the calculations are quite arbitrary ; we have not explored a wide range of the values . the interior structure of an inflated planet and the mass - radius relation of a hot jupiter are inferred from the 1-d numerical simulation @xcite even for a roche - lobe filled planet . also , the code ignores thermal properties of the solid core of a giant planet , leaving a question as to what would happen to the core under the intense tidal heating . in a hot jupiter - star system , the mass loss may occur via the lagrangian 2 point as well @xcite , which is not considered in the present work . besides , the mass loss from a planet in an eccentric orbit may be intermittent and nonconservative , while the mass loss rate in our calculation is simply estimated from the l1 overflow at @xmath219 . to use a more appropriate expression for the roche radius and to consider the more realistic mass loss in asynchronous eccentric planetary systems , a more careful treatment ( e.g. * ? ? ? * ) is desired to refine the results . in light of these limitations and possible uncertainties , the simulations covering a wide range of the parameter space along with more realistic modelling on individual issues will be explored in a future work . we are grateful to g. i. ogilvie , f. pont , and r. e. taam for useful discussions . this work has been supported by the nsc grant in taiwan through nsc 98 - 2112-m-001 - 011-my2 . the roche radius @xmath92 is defined as the distance between l1 and the planet s center of mass . the conventional expression of @xmath92 is derived from the circular restricted three - body problem ( e.g. * ? ? ? @xcite assumed that the roche lobe overflow occurs around perihelion and made an approximation that @xmath220 . here we investigate the validity of the approximation by employing the expression for the l1 point in the elliptical restricted three - body problem @xcite . the exact values of the roche radius in units of @xmath19 are calculated using the equations in @xcite and are shown in table [ tbl-1 ] for various mass ratios and eccentricities relevant to our problem . the table shows that the roche radius at perihelion is larger than that at aphelion in some of the cases . it is because the planet orbits faster ( slower ) than the fictitious " circular motion performed at perihelion ( aphelion ) . as a result , the centrifugal force due to the orbital motion is larger ( smaller ) at perihelion ( aphelion ) than the centrifugal force of the fictitious circular motion at the point , therefore enlarging ( reducing ) the roche radius relative to the roche radius " associated with the circular motion . furthermore , the above analysis is based on the assumption that the spin of the planet is instantaneously synchronized with the orbit at any moment @xcite . while it is true for a synchronized circular orbit , in an eccentric orbit the planet s spin is normally synchronized " with its orbital motion close to perihelion where the tidal forcing is expected to be near its maximum @xcite . following the same line of the above argument , a synchronized " planet spins faster than its orbit at aphelion , implying that the roche radius is smaller than the value associated with the circular motion at the point . it should be noted that equation ( [ eq : m_dot ] ) for the mass loss rate is also derived for a circular orbit . the equation should be modified as the l1 point changes its location along an eccentric orbit . besides , when the planet s photosphere is close to its time - varying roche lobe in an eccentric orbit , the outer part of the planet should expand or contract with the roche lobe on the local dynamical timescale , which introduces additional complexity to estimate the mass loss rate . crrrrrrrrrrr @xmath222 & @xmath10 & roche radius at perihelion @xmath223 & roche radius at aphelion @xmath223 + @xmath224 & 0.1 & 0.0851 & 0.0842 + @xmath224 & 0.3 & 0.0831 & 0.0833 + @xmath225 & 0.1 & 0.0411 & 0.0265 + @xmath225 & 0.3 & 0.061 & 0.0221 + anderson et al . 2009 , submitted to armitage , p. j. , & clarke , c. j. 1996 , .280 , 458 baraffe , i. , selsis , f. , chabrier , g. , barman , t. s. , allard , f. , hauschildt , p. h. , & lammer , h. 2004 , , 419 , l13 baraffe , i. , chabrier , g. , & gallardo , j. 2009 , , 702 , 27 bodenheimer , p. lin , d. n. c. , & mardling , r. a. 2001 , , 548 , 466 boss , a. 2004 , , 610 , 456 bouvier , j. 2007 , iaus , 243 , 231 carr , j. s. 2007 , iaus , 243 , 135 choi , p. i. , & herbst , w. 1996 , , 111 , 283 davis , t. a. , & wheatley , p. j. 2009 , , accepted fischer , d. a. , & valenti , j. , , 622,1102 ford , e. b. , & rasio , f. a. 2008 , , 686 , 621 gammie , c. f. , 2001 , , 553 , 174 gaudi , b. s. , seager , s. , & mallen - ornelas , g. , 2005 , , 623 , 472 gilliland r. l. 1986 , , 300 , 339 gu , p. g. , lin , d. n. c. , & bodenheimer , p. 2003 , , 588,509 gu , p. g. , bodenheimer , p. , & lin , d. n. c. 2004 , , 608,1076 hellier , c , anderson , d. r. , cameron , a. c. , gillon , m. , hebb , l. , maxted , p. f. l. , queloz , d. , smalley , b. , triaud , a. h. m. j. , west , r. g. , wilson , d. m. , bentley , s. j. , enoch , b. , horne , k. , irwin , j. , lister , t. a. , mayor , m. , parley , n. , pepe , f. , pollacco , d. l. , segransan , d. , udry , s. , wheatley , p. j. 2009 , nature , 460 , 1098 herbst , w. , & mundt , r 2005 , , 633.967 hulamo , n. , figueira , p. , bonfils , x. , santos , n. c. , pepe , f. , gillon , m. , azevedo , r. , barman , t. , fernndez , m. , di folco , e. , guenther , e. w. , lovis , c. , melo , c. h. f. , queloz , d. , & udry , s. 2008 , , 489 , 9 hut , p. 1981 , , 99 , 126 ibgui , l. , & burrows , a. 2009 , arxiv:0902.3998v1 ida , s. & lin , d. n. c. 2004 , , 604 , 388 ivanov , p. b. , & papaloizou , j. c. b. 2007 , , 376 , 682 jackson , b. , greenberg , r. , & barnes , r. 2008 , iaus , 249 , 119 jackson , b. , barnes , r. , & greenberg , r. 2009 , , 698,1357 jiang , i. g. , ip , w. h. , & yeh , l. c. 2003 , , 582 , 449 kokubo , e. , & ida , s. 2002 , , 581 , 666 kolb , u. , & ritter , h. 1990 , , 236 , 385 laine , r. o. , lin , d. n. c. , & dong , s .- f . 2008 , , 685 , 521 langton , j. , & laughlin , g. 2008 , , 483 , l25 lin , d. n. c. , bodenheimer , p. , & richardson , d. c. 1996 , nature , 380 , 606 lin , d. n. c. , & papaloizou , j. c. b. 1985 , in protostars and planet ii , ed . d. c. black & m. s. matthew ( tuscon : univ . arizona press ) , 981 livio , m. , & pringle , j. e. 1992 , , 259 , 23 long m. , romanova m. m. , & lovelace r. v. e. 2005 , , 634 , 1214 lovelace , r. v. e. , romanova , m. m. , & barnard , a. w. 2008 , , 389 , 1233 mardling , r. a. 2007 , , 382 , 1768 mardling , r. a. & lin , d. n. c. 2002 , , 573 , 829 marley , m. s. , fortney , j. j. , hubickyj , o. , bodenheimer , p. , & lissauer , j. j. 2007 , , 655 , 541 matt , s. , & pudritz , r. e. 2005 , 632 , 135 , mayor , m. , & queloz , d. 1995 , nature , 378 355 miller , n. , fortney , j. j. , jackson , b. 2009 , , 702 , 1413 murray , c. d. & dermott , s. f. 1999 , solar system dynamics , cambridge univ . press murray - clay , r. , chiang , e. , & murray , n. 2009 , , accepted nagasawa , m. , ida , s. , & bessho , t. 2008 , , 678 , 498 narita , n. , sato , b. , hirano , t. & tamura , m. 2009 arxiv 0908.1673n ogilvie , g. i. , & lin , d. n. c. 2007 , , 661 , 1180 papaloizou , j. c. b. 2007 , , 463 , 775 ptzold , m. & rauer , h. 2002 , , 568 , l117 pollack , j. b. , hubickyj , o. , bodenheimer , p. , lissauer , j. j. , podolak , m. , & greenzweig , y. 1996 , , 124 , 62 pols , o. r. , tout , c. a. , eggleton , p. p. , han , z. 1995 , , 274 , 964 pont , f. , zucker , s. , & queloz , d. 2006 , , 373,231 pont , f. , 396,1789 pratt , j. p. , & strittmatter , p. a. , 1976 , , 204 , 29 rafikov , r. r. 2005 , , 621 , 69 rice , w. k. m. , armitage , p.j . , & hogg , d. f. 2008 , , 384 , 1242 romanova , m. m. & lovelace , r. v. e. 2006 , , 645 , 73 sepinsky , j. f. , willems , b. , & kalogera , v. 2007 , , 660 , 1624 setiawan , j. , henning , th . , launhardt , r. , muller , a. , weise , p. , & kurster , m. 2008 , nature , 451 , 38 shu , f. , najita , j. , ostriker , e. , wilkin , f. , ruden , s. , & lizano , s , 1994 , , 429 , 781 starczewski , s. , gawryszczak , a. j. , wunsch , r. , & rozyczka , m. 2007 , aca , 57,123 symington , n.h . harries , t.j . , kurosawa , r. , & naylor , t. 2005 , , 358 , 977 todoran , i. 1993 , , 199,257 trilling , d. e. , benz , w. , guillot , t. , lunine , j. i. , hubbard , w. b. , & burrows , a. 1998 , , 500 , 428 udry , s. , & santos , n. c. 2007 , annu . rev . astron . astrophys . , 45 , 397 vauclair , s. 2004 , , 605 , 874 zarka , p. 2007 , planetary and space science , 55(5 ) , 598
epidemic dynamics in a stochastic network of interacting epidemic centers is considered . the epidemic and migration processes are modelled by markov s chains . explicit formulas for probability distribution of the migration process are derived . dependence of outbreak parameters on initial parameters , population , coupling parameters is examined analytically and numerically . the mean field approximation for a general migration process is derived . an approximate method allowing computation of statistical moments for networks with highly populated centres is proposed and tested numerically . * igor sazonov and mark kelbert * college of engineering , swansea university , singleton park , sa2 8pp , u.k . national research university higher school of economics , moscow , r.f . department of mathematics , swansea university , singleton park , sa2 8pp , u.k . * key words : * spatial epidemic models ; migration dynamics ; ourbreak time * ams subject classification : * 92d30 , 91b70 , 91b72
1505.00569
epidemic outbreak in a populated center or in a network of populated centra develops stochastically due to random interaction between individuals inside a center and due to a random migration between centers of the network . conventionally , an epidemic outbreak in a highly populated center is described by deterministic processes in according with law of large numbers ( lln ) @xcite . the mean field approximation ( hydrodynamic limits ) of the appropriate statistical models establishes the basic relation of stochastic description to the dynamical equations , say the sir model ( susceptible / infected/ removed ) and its more sophisticated modifications ( seir , sis , msir , etc . ) however , there are two important cases when stochastic effects are essential . firstly , it is obviously important when the populations in centers are not large . the second less obvious scenario can occur at the initial stage of outbreak when the number of infectives is small , then the discreteness of population can essentially affect the dynamics of the outbreak making it stochastic . for an isolated center , this case is thoroughly studied in @xcite . analysis of a network of interacting epidemic centers requires an account of migration fluxes between them . so , if the initial number of infectives triggering the outbreak in a particular populated center is small ( that is typical for many outbreaks ) than the lln fails at least at initial stage until the number of infectives is large enough . for this reason the observed number of infectives can be significantly different from the prediction of a deterministic model , i.e. the standard deviation of the number of infectives and the peak outbreak time can be wide even in highly populated center or a network of such centers . in principle , the probability density function ( pdf ) , its standard deviation , and other important characteristics for the outbreak forecast could be determined by a direct numerical simulation . however , this simulation may be very costly and require too much cpu time even for modern computers . our goal is to develop a technique for a rough estimation of the outbreak statistical characteristics skipping such huge computation by applying some perturbation methods . our toolkit is the so - called small initial contagion ( sic ) approximation for the case of a large populated center when initial number of infectives is small , cf . @xcite in the case of a single populated center . for a network of highly populated sir centers in the framework of deterministic model such technique is described in @xcite . in this paper we develop a stochastic version of sic approach based on the assumption that in the real epidemic centra the number of infectives triggering an outbreak is still small . in a sense , the sic approach looks like a key to solving cumbersome epidemic networks . in @xcite , a randomized analogue of the standard sir model is considered . in the sic approximation , one distinguishes two linked stages of epidemic evolution . at the stage 1 of initial contamination the number of infectives is small and a discrete formulation is vital . at this stage the system is randomized , and governed by stochastic equations . at the stage 2 of developed outbreak when the numbers of individuals in all the components are large the lln works and the standard deterministic sir model can describe the outbreak process accurately enough . therefore it is natural to consider a deterministic system with randomized initial conditions linked to the stochastic stage 1 . the statistical characteristics of the complete model are obtained by applying deterministic equations with random initial conditions using the matching asymptotic expansions technique ( cf . in contrast to the traditional technique , the asymptotic approximation of a randomized evolution ( at a brief initial period ) is matched with a deterministic evolution with randomized initial conditions ( for all other times ) . nevertheless , as in the traditional approach , we match the approximations at some intermediate time @xmath4 in the interval where the both approximations are valid and which drops out from the final results ( cf . @xcite ) . the markov chain ( mc ) describing the randomized sir model has been studied previously , e.g. , in @xcite where a partial differential equation ( pde ) for the moment generating function was derived . we develop a similar approach for a pair of linked centers and obtain approximate formulae for their main statistical characteristics . the results of large - scale numerical simulation are in a good agreement with the appropriate models . the paper is organized as follows . in section [ sec : mc ] we introduce a mc model describing randomized epidemic outbreak in two populated centers coupled by a random migration of all types of species . here we consider convergence of this model to a deterministic mean - field model proposed in @xcite . in section [ sec : migration ] we describe a model of random migration between two interacting sir centers taking place before the outbreak started to determine all the initial conditions for the mc model . migration between centers is also described as a markov chain . here we derive the master / kolmogorov s equations for the probability generating functions ( pgf ) and solve them analytically . this analysis confirms the diffusion - like model of migration heuristically proposed in @xcite . in section [ sec : numerics ] the numerical algorithm for direct solving the mc model is described , the dependence of outbreak parameters on the population size , initial number of infectives and migration parameters are presented and discussed . in section [ sec : sic ] the generalization of the mc model on an arbitrary network of the epidemic centers is studied , and difficulties of direct numerical modelling are estimated . here the simplified model which looks relevant for a network of a highly populated centers is proposed . in discussions we make a comparison with some previously considered models and outline the perspective of the future development . consider two populated nodes , 1 and 2 , with populations @xmath5 and @xmath6 , respectively . let @xmath7 , @xmath8 , @xmath9 be the numbers of host susceptibles , infectives and removed , respectively , in node @xmath10 at time @xmath4 . let @xmath11 , @xmath12 , @xmath13 be numbers of guest susceptibles , infectives and removed , respectively , in node @xmath10 migrated from node @xmath14 at time @xmath4 . note that in the standard sir model , removed individuals do not interact with other species , do not affect dynamics of susceptibles and infectives , and can be omitted from consideration @xcite . assume that the populations in every node is completely mixed , the contamination rate @xmath15 of a susceptible individual in node @xmath10 at time interval @xmath16 $ ] is proportional to the number of all infectives in node @xmath10 : host infectives @xmath17 at time @xmath4 plus guest infectives @xmath18 immigrated from node @xmath14 . next , every infective in node @xmath19 can be removed with probability rate @xmath20 ( cf . @xcite ) . the model is described by the migration rate @xmath21 from node @xmath10 to node @xmath14 and return rate @xmath22 for a guest individual to return to his host node , they may be different for different species , i.e. we specify the migration process for susceptibles by parameters @xmath23,@xmath24 and for infectives@xmath25,@xmath26 ( cf . obviously , return rate of the host node should be higher than the migration rate to a neighbour node , i.e. , @xmath27 . taking into account all above , we model a network of two sir centers interacting due to migration of individuals between them a markov chain ( mc ) which full description is given in table 1 . if @xmath28 infectives appear in center 1 at time @xmath29 than the initial conditions for this process are@xmath30 here the initial numbers of guest susceptibles @xmath31 and @xmath32 are random and determined by migration processes between centers taking place before appearing of a single infective . in section [ sec : migration ] we determine this distribution ( which turns out to be binomial ) by considering pure migration processes taking place before the outbreak : see eq . ( [ eq : s0-distribution ] ) below . mean values for @xmath31 and @xmath33 are given by ( [ eq : s12:mean ] ) . numerical simulation based on this model is presented and discussed in section [ sec : numerics ] . analytical approach via master / kolmogorov equations ( see @xcite ) is too cumbersome , their analysis and solution in the general case is rather complicated , so we will develop reasonable approximations . lclcl @xmath34 & process # @xmath34 & rate , @xmath35 & restriction & description + 1 & @xmath36 & @xmath37 & @xmath38 & contamination in 1 ( host ) + 2 & @xmath39 & @xmath40 & @xmath41 & contamination in 1 ( guest ) + 3 & @xmath42 & @xmath43 & & recovering in 1 ( host ) ' '' '' + 4 & @xmath44 & @xmath45 & & recovering in 1 ( guest ) ' '' '' + 5 & @xmath46 & @xmath47 & @xmath48 & migration from 1 to 2 + 6 & @xmath49 & @xmath50 & @xmath51 & migration from 1 to 2 + 7 & @xmath52 & @xmath53 & @xmath54 & return from 2 to 1 + 8 & @xmath55 & @xmath56 & @xmath38 & return from 2 to 1 + 9 & @xmath57 & @xmath58 & @xmath59 & contamination in 2 ( host ) + 10 & @xmath60 & @xmath61 & @xmath51 & contamination in 2 ( guest ) + 11 & @xmath62 & @xmath63 & & recovering in 2 ( host ) ' '' '' + 12 & @xmath64 & @xmath65 & & recovering in 2 ( guest ) ' '' '' + 13 & @xmath66 & @xmath67 & @xmath68 & migration from 2 to 1 + 14 & @xmath69 & @xmath70 & @xmath71 & migration from 2 to 1 + 15 & @xmath72 & @xmath73 & @xmath74 & return from 1 to 2 + 16 & @xmath75 & @xmath76 & @xmath59 & return from 1 to 2 + * proposition 1 . * the scaled markov chain ( mc ) @xmath77 ( @xmath78 , @xmath79)@xmath80 in a populations of sizes @xmath81 related with process @xmath82 @xmath83 defined in table 1 and subjected to initial conditions ( [ eq : ini ] ) by scaling the transition rates @xmath84 , and scaling of initial conditions as@xmath85where the pdfs of independent random variables @xmath31 and @xmath32 have binomial distributions ( [ eq : s0-distribution ] ) , converges in distribution as @xmath86 to the deterministic functions @xmath87 satisfying odes introduced in @xcite @xmath88subjected to the initial conditions@xmath89where@xmath90 in fact , equations ( [ ode : sn])([ode : imn ] ) can be derived phenomenologically : if the number of species is large enough , its change by one or by few can be considered as infinitesimally small . for example , the number of infectives @xmath91 can increase due to process # 1 and # 8 with rates @xmath37 and @xmath92 , respectively , or decrease due to process # 3 and # 6 with rates @xmath43 and @xmath50 , respectively . therefore the rate of @xmath93 in time interval @xmath94 can be estimated as @xmath95 \!\mathrm{d}t-\left [ \alpha _ { 1}i_{1}+\gamma _ { 12}^{i}i_{1}\right ] \!\mathrm{d}t$ ] , that gives eq . ( [ ode : in ] ) for @xmath96 . the same holds for all other species . this argument can be made rigorous with the help of law of large numbers ( lln ) . then it indicates that the mean values of the variables converge to the mean - field ( hydrodynamic ) limit . the more delicate question is to establish the convergence in probability . sketch of the proof is given in appendix . in order to elaborate distributions for @xmath97 taking place at the very beginning of the outbreak we study the pure migration process setting @xmath98 . in this case the mc described in table 1 can be split into two independent processes : @xmath99 , and @xmath100 . for each of them we have the following mc in terms of a single random variable @xmath101 , @xmath78:@xmath102 let @xmath103 be the probability distribution in node @xmath14 at instant @xmath4 . then masters / kolmogorov s equations take the form @xmath104 % & & \hspace{1em}{}+\delta \ \left ( \left\lceil k+1\right\rceil % _ { 0}^{n}\right ) p_{k+1}-\delta kp_{k}% % \end{array } \label{eq : master - migration}\]]where @xmath105 ; for the sake of simplicity we temporary set @xmath106 , @xmath107 , @xmath108 . here the following notation is introduced to write equations for @xmath109 and @xmath110 in the same form as others @xmath111equations ( [ eq : master - migration ] ) implies the following pde @xmath112 \label{eq : pde - migr}\]]for the probability generating function ( pgf ) @xmath113the initial condition @xmath114 implies @xmath115 the solution to problem ( [ eq : pde - migr])([eq : g : ini ] ) can be found explicitly @xmath116 ^{n}. \label{eq : pde - sol}\]]now one can easily calculate all the moments of distribution @xmath117 , say @xmath118% % \begin{array}{r@{\;}c@{\;}l } % \mathbb{e}s(t ) & \equiv & \mu _ { 1}(t)=g_{z}(1,t)=n\varepsilon \left [ % 1-e^{-t/\tau } \right]% % \end{array } & & \label{eq : mu1:migration } \\ \mathrm{var}(s(t ) ) & \equiv & \mu _ { 2}(t)=g_{zz}(1,t)+\mu _ { 1}-\mu _ { 1}^{2 } = \mu _ { 1}(t)\left [ \varepsilon e^{-t/\tau } + ( 1-\varepsilon ) \right]% % \begin{array}{r@{\;}c@{\;}l } % \mathrm{var}(s(t ) ) & \equiv & \mu _ { 2}(t)=g_{zz}(1,t)+\mu _ { 1}-\mu _ { 1}^{2 } % \\[0.2527pc ] % & & \hspace{1em}{}=\mu _ { 1}(t)\left [ \varepsilon e^{-t/\tau % } + ( 1-\varepsilon ) \right]% % \end{array } % & & \label{eq : mu2:migration}\end{aligned}\]]where @xmath119 , @xmath120 . if the migration process lasts long enough before the outbreak starts then the pgf takes its limiting form for @xmath121@xmath122which is the mgf for a binomial distribution:@xmath123from here on we return to the indexed notation@xmath124 this distribution has the following first two moments@xmath125the relative standard deviation ( i.e. for the process @xmath126 decays as @xmath127 : @xmath128hence , when @xmath129 , the migration process tends in probability to the deterministic limit described in @xcite . thus , in the mc model defined by table [ tab : mc ] the initial conditions @xmath130 and @xmath32 can be selected randomly with the binomial distribution ( [ eq : s0-distribution ] ) or approximated by a gaussian function if @xmath131 is large enough . parameter @xmath132 indicates the mean share of individuals from node @xmath10 migrated to node @xmath14 . obviously this share in average should be small for highly populated centers : say , half population of a city hardly can be on a visit into another city for an essential time . parameter @xmath133 can be treated as a coupling parameter characterizing how intensive are migration fluxes between populated centers . analogous fluxes of infectives hardly are more intensive , therefore @xmath134 . so , for highly populated centers , the following inequality should be kept @xmath135 the second important characteristic of migration process is the characteristic migration time @xmath136 . from ( [ eq : mu1:migration ] ) one can see that it indicates how soon the dynamic equilibrium established after the migration process started or the population changes suddenly . both pairs of parameters : @xmath137 and @xmath138 are uniquely related . in the numerical simulation of the randomized sir model the time interval was divided into small steps @xmath139 such that the sum of all rates from table 1 multiplied by @xmath139 is essentially less than 1 : @xmath140where @xmath141 is the admitted threshold , say , @xmath142 . probability that at least one event occurs in unit of time is bounded by the sum of rates of all the processes @xmath143 : @xmath144 in this relation we majorize @xmath145 , @xmath146 , @xmath147 , @xmath148 , then @xmath149actually this value overestimates the really occurred total rate significantly as it is very improbable that numbers of guest susceptibles and infectives in a highly populated center essentially exceeds values @xmath150 and @xmath151 , respectively , where @xmath152 is defined in ( [ eq : s0-distribution ] ) , in virtue of ( [ eq : e<<1 ] ) . for this reason we can account that @xmath153 and @xmath154 ( and also use the rigorous inequalities @xmath155 ) . then we obtain the more realistic estimation:@xmath156 the following numerical scheme is used : 1 . assign the initial values to 8 variables@xmath157 where @xmath158 are random numbers distributed in accordance with ( [ eq : s0-distribution ] ) . 2 . calculate the current rates @xmath159 indicated in table 1 . calculate the current probability of at least one event occurrence in accordance with eq . ( [ eq : rate : sum]):@xmath160 4 . generate uniformly distributed random number@xmath161 $ ] . 5 . if @xmath162 then no events occur . in this case : 1 . advance at one time step : @xmath163 without changing variables @xmath164 , also @xmath165 and @xmath166 . if @xmath167 terminate the process , otherwise go to step 4 . + if @xmath168 then at least one event occurs . in this case calculate the intervals @xmath169 $ ] , @xmath170 2 . generate the second random number @xmath171 uniformly distributed in @xmath172 $ ] 3 . find in which interval @xmath171 falls . 4 . perform the process described in table 1 with the correspondent rate 5 . advance at one time step : @xmath163 6 . if @xmath167 terminate the process , otherwise go to step 2 . for numerical computation a basic model with two identical centers has parameters : @xmath173 , @xmath174 , @xmath175 , @xmath176 , @xmath177 where @xmath178 are reproduction numbers . the initial number of infectives in the first node @xmath179 where @xmath180 is taken for the basic model . a few realizations of numerical computation are depicted in figure fig : realisation . here the time variations of the total number of infectives @xmath181 in every node are shown and compared with the curves based on integration of the deterministic initial value problem ( [ ode : sn])([eq : s12:mean ] ) ) by thin grey lines and in node 2 ( @xmath182 ) by thin black lines . bold dashed lines indicates the hydrodynamic limit . ( @xmath183k , @xmath184 , @xmath185 , @xmath186),width=302 ] in the first set of numerical experiments the total population size varies from @xmath187 up to 10@xmath188 . number of realization @xmath189 was taken @xmath190 mainly but the number of realization is taken greater for small population @xmath191 and @xmath192 and smaller for extremely high populations : 250k and 1000k . the current mean number of total infectives @xmath193 and standard deviation are computed . the results of the first set are shown in figure [ fig : i - vs - n ] . observe that the mean value of the random process ( thin lines ) converges to the solution of correspondent deterministic problem ( bold lines ) . but the convergence is much slower for node 2 : for @xmath194k the mean trajectory practically coincides with the deterministic limit , on the other side the same effect in node 2 requires @xmath195k . @xmath196 ( left ) and their standard deviations ( right ) . grey curves for node 1 , black lines for node 2 . dashed lines indicate the hydrodynamic limit described by eqs . ( [ ode : sn])([ode : imn ] ) . the node population is indicated near the top of the correspondent curve.,title="fig:",width=302 ] @xmath196 ( left ) and their standard deviations ( right ) . grey curves for node 1 , black lines for node 2 . dashed lines indicate the hydrodynamic limit described by eqs . ( [ ode : sn])([ode : imn ] ) . the node population is indicated near the top of the correspondent curve.,title="fig:",width=302 ] the convergence rate is examined in figure [ fig : stdimax - vs - n ] . one can see that for node 1 the convergence rate almost coincides with @xmath197 , as for node 2 the decay rate tends to @xmath198 only for sufficiently large population : @xmath199 . thus for the secondary contaminated node the account of randomness is essential even if its population is large provided that the migration parameters @xmath200 are small ( 0.01 in this case ) . say , if @xmath201 the standard deviation exceeds the mean value up to the outbreak time . and @xmath202 vs population @xmath203 . black curves for node 1 , grey lines for node 2 . the slope of dashed line corresponds to the decay law @xmath204.,width=302 ] in the second set of numerical experiments we study dependence of mean number of infectives and their standard deviation from the initial number of infectives @xmath28 in node 1 . it varies from @xmath205 to @xmath206 ( the share @xmath207 varies from @xmath208 to @xmath209 ) . the results are plotted in figure [ fig : i1-vs - i0 ] . ( black ) and its standard deviation @xmath210 ( grey ) for different @xmath211 . the initial number of infectives in node 1 is indicated near top of the corresponding curve . dashed line indicates the deterministic limiting solution . right : the same for @xmath212.,title="fig:",width=302 ] ( black ) and its standard deviation @xmath210 ( grey ) for different @xmath211 . the initial number of infectives in node 1 is indicated near top of the corresponding curve . dashed line indicates the deterministic limiting solution . right : the same for @xmath212.,title="fig:",width=302 ] because the outbreak time depends on initial number of infectives , the mean - field curves become essentially different . therefore it is appropriate to shift the time so that the peak outbreak for different initial condition is at the same instant , say , @xmath29 . then all the curve are very close to each other and practically coincide with curve for the limiting solution introduces in @xcite . observe that the smaller the number of initial infectives the greater is the standard deviation ( std ) and the larger is the difference between the mean curve for randomized process and the mean - field curve . also observe that for node 1 , the discrepancy of mean number of infectives from the hydrodynamic limit as well as the standard deviation monotonically decay with the growth of @xmath28 . in node 2 the analogous discrepancy and the std slightly change when the number of initial infective varies from 10 to 100 . and @xmath202 process vs population @xmath213 grey curve is for node 1 , black curve is for node 2 . the dashed line has the slope corresponding to the decay law @xmath214.,width=302 ] in the third set of numerical experiments we study dependence of the mean number of infectives on the coupling coefficient @xmath215 ( the same for all species ) . it was varied in the range @xmath216 . the results are plotted in figures [ fig : i1-vs - eps ] and [ fig : i2-vs - eps ] . observe that @xmath215 practically does not affect the standard deviation of total number of infectives in node 1 . discrepancy of the mean curve from the mean - field curve becomes noticeable only for small @xmath215 : @xmath217 . ( left ) and its standard deviations @xmath210 ( right ) for different migration coefficient @xmath218 . dashed lines indicate the mean - field limit . the initial number of infectives in node 1 is indicated near top of the corresponding curve.,title="fig:",width=302 ] ( left ) and its standard deviations @xmath210 ( right ) for different migration coefficient @xmath218 . dashed lines indicate the mean - field limit . the initial number of infectives in node 1 is indicated near top of the corresponding curve.,title="fig:",width=302 ] ( left ) and its standard deviations @xmath219 ( right ) for different migration coefficient @xmath218 . dashed lines indicate the mean - field limit . the initial number of infectives in node 1 is indicated near top of the corresponding curve.,title="fig:",width=302 ] ( left ) and its standard deviations @xmath219 ( right ) for different migration coefficient @xmath218 . dashed lines indicate the mean - field limit . the initial number of infectives in node 1 is indicated near top of the corresponding curve.,title="fig:",width=302 ] as for the second node , the standard deviation grows monotonically with the decrease of the coupling . that indicates the importance to account for randomness of the epidemic process in the case of weak coupling ( i.e. in the case of relatively slow migration fluxes ) . and @xmath220 vs population @xmath221 . grey curve is for node 1 , black curve is for node 2 . the dashed line has the slope corresponding to the decay law @xmath204 . right : outbreak value for @xmath222 and @xmath220 processes vs population @xmath221 grey curve is for node 1 , black curve is for node 2 . the dashed line are for the mean - field values.,title="fig:",width=302 ] and @xmath220 vs population @xmath221 . grey curve is for node 1 , black curve is for node 2 . the dashed line has the slope corresponding to the decay law @xmath204 . right : outbreak value for @xmath222 and @xmath220 processes vs population @xmath221 grey curve is for node 1 , black curve is for node 2 . the dashed line are for the mean - field values.,title="fig:",width=302 ] the mc model for two coupled sir centra can be readily generalized for an arbitrary network of @xmath223 mutually interacting sir centers : @xmath224where @xmath225 . generically , to study migration fluxes ( @xmath226 ) between every pair of centers ( although of different rates ) we have to consider @xmath227 host and @xmath228 guest species . the correspondent mc model will contain @xmath229 fluxes . the total rate can be evaluated via @xmath230 . therefore for a network containing significant number of highly populated centra ( say , main cities in the uk ) , the time interval @xmath139 will have to be taken extremely small , and , hence , the gpu time for a single realization will be too long to get the necessary statistics . thus the mc should be simplified for numerical modelling . the sic approximation is based on the assumptions which look relevant for a network of highly populated centers : * population in every center is high : @xmath231 . * migration fluxes between the centers are small : @xmath232 . * initial number of infectives in the first contaminated center ( say , @xmath233 ) , is small : @xmath234 . * reproduction number exceeds unity and is not close to it in all the nodes : @xmath235 where @xmath236 . in these assumptions , the outbreak process in every center can be split into the following main stages : 1 . * contaminating stage * : number of infectives is small @xmath237 and the fluxes of infectives caused by migration are essential for the outbreak process ( except the first node ) . * developed outbreak * : @xmath238 , when the contribution of migration fluxes is negligible , ( also the mean - field description for every individual realization looks adequately ) . 3 . * recovering stage * : the node is not affected by infective immigrants and slightly affects contamination of other nodes . from these assumptions follows that the outbreak dynamics in the first node can be considered independently , and can be described by the following mc @xmath239with the initial condition @xmath240 , @xmath241 studied in @xcite . at the contamination stage we have @xmath242 and the mc can be further simplified @xmath243 epidemic dynamics in node 2 at the contamination stage ( @xmath244 ) can be described by analogous mc with additional flux @xmath245 of infectives migrated from node 1:@xmath246with @xmath247 where @xmath248 at the contamination stage processes @xmath249 and @xmath250 are practically independent of process @xmath251 . thus we have to consider mc ( [ mc : i2 ] ) with a random flux @xmath245 which statistics will be specified later . first we consider a single realization of the flux @xmath245 and treat it as a deterministic function . later we will use averaging on @xmath245 to calculate moments of distribution for @xmath251 . let @xmath253 be the probability of @xmath254 infectives @xmath255 at instant @xmath4 . initial condition is @xmath256 kolmogorov s equations for mc ( [ mc : i2 ] ) are@xmath257here @xmath258 . for the pgf @xmath252 tending @xmath259 ( with @xmath260 , @xmath261 ) we obtain the following pde @xmath262 \label{eq : pde - nu}\]]with the initial condition @xmath263 . its solution can be written in the integral form @xmath264where @xmath265 is the initial growth rate of infectives in the deterministic sir model in the limit @xmath266 ( limiting solution introduced in @xcite ) . the first conditional moment @xmath268 for fixed @xmath245 is @xmath269 averaging over all realizations for @xmath245 ( with a time varying pdf @xmath270 ) : @xmath271 @xmath272\mathrm{d}% \nu = \int_{0}^{t}\bar{\nu}\left ( t^{\prime } \right ) e^{\lambda _ { 2}(t - t^{\prime } ) } \mathrm{d}t^{\prime } \label{eq : mu1-nu}\]]where @xmath273 . thus the average number of infectives in node 2 in the contamination stage relates with the flux @xmath245 via the convolution@xmath274 \equiv \mu _ { 1}\left ( t\right ) = \bar{\nu}\left ( t\right ) \ast e^{\lambda _ { 2}t}. \label{sic : ei2}\ ] ] to calculate it we apply the law of total variation ( e.g. @xcite ) : @xmath276 = \mathbb{e}\left [ \mathrm{var}% ( i_{2}(t)\,|\,\nu ) \right ] + \mathrm{var}\left [ \mathbb{e}(i_{2}(t)\,|\,\nu ) % \right ] . \label{eq : var(i2)+}\ ] ] * 1 . * the first addend in ( [ eq : var(i2)+ ] ) can be found through the pgf @xmath252 : @xmath277 \mathrm{d}t^{\prime } + \mu _ { 1}(t\,|\,\nu ) .% \end{array}%\]]after the averaging through @xmath278 we obtain @xmath279 = \displaystyle\mu _ { 1}(t)+\frac{2\beta _ { 2}^{\prime } } { \lambda _ { 2}}\int_{0}^{t}\bar{\nu}% ( t^{\prime } ) \left [ e^{2\lambda _ { 2}(t - t^{\prime } ) } -e^{\lambda _ { 2}(t - t^{\prime } ) } \right ] \mathrm{d}t^{\prime } .\]]thus a first addend in ( [ eq : var(i2)+ ] ) can written as a sum of two convolutions @xmath279 = \frac{2\beta _ { 2}^{\prime } } { \lambda _ { 2}}\bar{\nu}(t)\ast e^{2\lambda _ { 2}t}+(1-\frac{% 2\beta _ { 2}^{\prime } } { \lambda _ { 2}})\bar{\nu}(t)\ast e^{\lambda _ { 2}t } \label{eq : var(i2):1}\]]where @xmath280 . * 2 . * to calculate the second addend in ( [ eq : var(i2)+ ] ) , we temporary add @xmath281 to it . now it can be expressed via the covariance of flux @xmath245:@xmath282 + \mu _ { 1}^{2}(t)=\mathbb{e}\left ( \int_{0}^{t}\nu \left ( t^{\prime } \right ) e^{\lambda _ { 2}(t - t^{\prime } ) } \mathrm{d}t^{\prime } \right ) ^{2 } \\[0.5pc ] \displaystyle=\int_{0}^{t}\int_{0}^{t}\mathbb{e}\left [ \nu \left ( t^{\prime } \right ) \nu \left ( t^{\prime \prime } \right ) \right ] e^{\lambda _ { 2}(t - t^{\prime } ) } \mathrm{d}t^{\prime } e^{\lambda _ { 2}(t - t^{\prime \prime } ) } \mathrm{d}t^{\prime \prime } .% \end{array } \label{eq : var(i2)a}\]]the function in the integrand can be represented as a sum @xmath283 = \mathrm{cov}\left [ \nu \left ( t^{\prime } \right ) , \nu \left ( t^{\prime \prime } \right ) \right ] + \bar{\nu}(t^{\prime } ) \bar{\nu}% ( t^{\prime \prime } ) . \label{eq : e[nu(t')*nu(t")]}\]]integration of the second addend in ( [ eq : e[nu(t)*nu(t " ) ] ] ) gives just the temporary added term@xmath284thus the second addend in ( [ eq : var(i2)+ ] ) can be written through the following integral@xmath285 = \int_{0}^{t}\int_{0}^{t}\mathrm{cov}\left [ \nu \left ( t^{\prime } \right ) , \nu \left ( t^{\prime \prime } \right ) \right ] e^{\lambda _ { 2}(t - t^{\prime } ) } % \mathrm{d}t^{\prime } e^{\lambda _ { 2}(t - t^{\prime \prime } ) } \mathrm{d}% t^{\prime \prime } . \label{eq : var(i2)b}\]]in which we have to calculate the covariance of flux @xmath245 . if flux @xmath245 is a random process controlled by a mc , calculation of its covariance is a complicated task , and consideration of this needs a separate work . remind that the flux is a linear combination of two mc processes ( [ eq : nu ] ) : @xmath286 . here we approximate @xmath249 and @xmath250 by two mutually independent poisson processes with variable rates @xmath287 and @xmath288 , respectively , where @xmath289 and @xmath290 are calculated below . in this approximation , using the independence of increments of the inhomogeneous poisson flow ( e.g. @xcite ) we can write@xmath291 & \approx & \bar{i}_{12}(\min \left\ { t^{\prime } , t^{\prime \prime } \right\ } ) , \\ \mathrm{cov}\left [ i_{21}\left ( t^{\prime } \right ) , i_{21}\left ( t^{\prime \prime } \right ) \right ] & \approx & \bar{i}_{21}(\min \left\ { t^{\prime } , t^{\prime \prime } \right\ } ) , \\ \mathrm{cov}\left [ i_{12}\left ( t^{\prime } \right ) , i_{21}\left ( t^{\prime \prime } \right ) \right ] & \approx & 0.\end{aligned}\ ] ] we justify this approximation numerically below . thus , for the covariance of the flux we have@xmath292 = \varpi ( \min \left\ { t^{\prime } , t^{\prime \prime } \right\ } ) , \quad \varpi ( t)\equiv \left ( \beta _ { 2}^{\prime } \right ) ^{2}% \bar{i}_{12}(t)+\left ( \delta _ { 12}^{i}\right ) ^{2}\bar{i}_{21}(t ) . \label{eq : cov : po}\]]equation ( [ eq : cov : po ] ) holds true because the second central moment of the poisson process coincides with the first moment . in this approximation , it is sufficient to compute the first moment of flux @xmath245 in order to compute the second moment for @xmath293 . with the account for ( [ eq : cov : po ] ) , we split integral ( [ eq : var(i2)b ] ) into two parts:@xmath294 & \approx & j_{1}+j_{2 } \\ j_{1 } & = & \int_{0}^{t}\int_{0}^{t^{\prime } } \varpi ( t^{\prime \prime } ) e^{\lambda _ { 2}(t - t^{\prime \prime } ) } \mathrm{d}t^{\prime \prime } e^{\lambda _ { 2}(t - t^{\prime } ) } \mathrm{d}t^{\prime } \\ j_{2 } & = & \int_{0}^{t}\int_{t^{\prime } } ^{t}\varpi ( t^{\prime } ) e^{\lambda _ { 2}(t - t^{\prime \prime } ) } \mathrm{d}t^{\prime \prime } e^{\lambda _ { 2}(t - t^{\prime } ) } \mathrm{d}t^{\prime } .\end{aligned}\ ] ] integrating @xmath295 by parts and @xmath296 directly we obtain that they both give the same answer@xmath297 finally combining the above results we have@xmath298 \bar{i}_{21}\ast e^{2\lambda _ { 2}t } \\[0.7pc ] \displaystyle\;{}-\big [ \frac{4\left ( \beta _ { 2}^{\prime } \right ) ^{2}}{\lambda _ { 2}}-\beta _ { 2}^{\prime } \big ] \bar{i}_{12}\ast e^{\lambda _ { 2}t}-\big [ \frac{2\beta _ { 2}^{\prime } \delta _ { 12}^{i}}{\lambda _ { 2}}+\frac{2\left ( \delta _ { 12}^{i}\right ) ^{2}}{\lambda _ { 2}}-\delta _ { 12}^{i}\big ] \bar{i}% _ { 21}\ast e^{\lambda _ { 2}t}.% \end{array } \label{eq : var(i2)}\ ] ] it is natural to split the total flux into two parts @xmath300 . flux process @xmath301 described by the mc @xmath302(with @xmath303 ) coincides with that described by ( [ mc : i2 ] ) if we set @xmath304 , @xmath305 , @xmath306 . then we can immediately write down a solution for the pgf @xmath307and the first moment@xmath308 flux process @xmath309 is more complicated and can be described by the following mc@xmath310with the initial conditions @xmath311 , @xmath312 . if we split the third event into two independent events @xmath313 and @xmath314 with the same rate , we can split mc ( [ mc : nu21 ] ) into two mcs . the first mc describes migration of host susceptives from node 2 to node 1 and their possible removal due to contamination:@xmath315it is independent of the second mc with the rates@xmath316which describes migration of susceptibles to a neighbor node , their contamination there and return to the host node as infected species . first , we study the first mc : ( [ mc : s21 ] ) . in accordance with ( eq : s0-distribution ) it has the binomial initial distribution:@xmath317the probabilities @xmath318 of @xmath254 guest susceptibles @xmath319 at instant @xmath4 satisfy kolmogorov s equations for mc ( [ mc : i2 ] ) @xmath320 . \label{eq : s21:pk}\]]where @xmath321 , @xmath322 . system ( [ eq : s21:pk ] ) implies the following pde for mgf @xmath323 @xmath324 . \label{eq : s21:pde}\]]the initial condition is @xmath325the initial value problem ( [ eq : s21:pde])([sic : g0 ] ) admits the explicit solution @xmath326 ^{n}\exp \left\ { \nu ( z-1)\phi ( t)\int_{0}^{t}\frac{\mathrm{d}t^{\prime } } { \phi ( t^{\prime } ) } % \right\}. \label{sic : g}\]]where @xmath327 . from here we have@xmath328 \phi ( t ) . \label{eq : es21}\ ] ] now for process ( [ mc : i21 ] ) in analogy with processes ( [ mc : i2 ] ) and ( [ mc : i12 ] ) we can immediately write@xmath329 \mathrm{d}t^{\prime } . \label{eq : ei21}\]]neglecting the mutual dependence of processes @xmath330 and @xmath331 we approximate@xmath332@xmath333 \approx \mathbb{e}s_{21}(t)\bar{i}% _ { 1}(t ) . \label{eq : e(s21*i1)}\]]below we show numerically that it is satisfactory approximation for our applications . remind that equations ( [ eq : mu1-nu ] ) , ( [ eq : var(i2)b ] ) , ( eq : cov : po ) , ( [ eq : var(i2 ) ] ) , ( [ eq : ei12 ] ) , ( [ eq : ei21 ] ) , ( eq : e(s21*i1 ) ) are valid at the contamination stage only ( @xmath334 ) . they allow us to calculate the first and second moments for the number of infectives without modelling the random process directly . to evaluate the moments at the developed outbreak we use the same approach as in @xcite : by approximating the outbreak via the mean field solution for a single sir node with random initial conditions . for this aim we define the intermediate time @xmath335 such that the number of infective is large enough to use the mean field solution but the number of infective still slightly deviate from @xmath6 : @xmath336 and @xmath337 . then we generate @xmath189 times a random number @xmath338 lognormally distributed ( to guarantee the positiveness ) with mean @xmath339 and variance @xmath340 and integrate the classical sir equations : @xmath341 , @xmath342 with initial condition @xmath343,@xmath344 /\beta _ { 2}^{\prime } \]]where @xmath345 $ ] is the @xmath254th branch of the lambert function lambert . let us emphasize that in the classical sir model , the numbers of susceptives and infectives are related as @xmath346 /\beta ^{\prime } $ ] where @xmath347 ( cf . resolving this relation with respect to @xmath348 we can write @xmath349 /\beta ^{\prime } \]]where @xmath350 for the growing part and @xmath109 for the decaying part of the outbreak . if the outbreak is triggered by a infinitesimal number of infectives we can set @xmath351 /\beta ^{\prime } $ ] . also note that it is natural to approximate the solution to a standard sir model by the limiting solution ( @xmath352 ) introduced in skg08 . the limiting solution is independent of the initial condition , therefore it is not necessity to integrate the odes @xmath189 times but only once . thus the proposed two - stage model of a coupled randomized epidemic centers allows us to calculate its first moments much faster than the direct simulation summarized in table [ tab : mc ] . to show the accuracy of the proposed model we compare the solution obtained by different approaches . again the basic model of section [ sec : numerics ] is used : @xmath353k , @xmath354 , @xmath174 , @xmath355 , @xmath356 , @xmath357 but also model with the smaller population @xmath358k . we compare ( i ) the full randomized model described in table [ tab : mc ] which we regard as a benchmark ; ( ii ) the sic approximate randomized model where only flux of infectives from node 1 to node 2 is accounted , it is described in table [ tab : mc : sic ] ; ( iii ) the two - stage semi - randomized model proposed in this section above . in the two - stage model we take time @xmath359 for transition from a contamination randomized stage to the mean - field stage with random initial condition . the expected number of infectives in node 2 at time @xmath360 is @xmath361 which is large enough and at the same time much smaller than the node population . we take @xmath362 for number of realization in the second stage to evaluate the moments . in the sic approximation the randomized model presented in table tab : mc : sic comprises four consequently independent mcs . processes 1 and 2 represent independent outbreak in node 1 ( [ mc : i1s1 ] ) . processes 3 and 4 represent migration of its infectives to node 2 ( [ mc : i12 ] ) . processes 5 and 6 represent migration of host susceptives from node 2 to node 1 and their possible removal due to contamination ; they are analogous to mc ( mc : s21 ) with @xmath6 substituted by @xmath363 to be valid for all the stages . analogously processes 7 and 8 represent contamination of @xmath319 in the first node and migration of appeared infectives @xmath364 to their host node ( [ mc : i21 ] ) . finally processes 910 represent the outbreak in node 2 ; they are analogous to mc with an additional flux ( [ mc : i2 ] ) with the accurate account of the number of succeptives @xmath363 at all the stages . lll # & process & rate + 1 & @xmath365 & @xmath366 + 2 & @xmath367 & @xmath43 ' '' '' + 3 & @xmath368 & @xmath50 + 4 & @xmath369 & @xmath56 + 5 & @xmath370 & @xmath67 + 6 & @xmath371 & @xmath372 + 7 & @xmath373 & @xmath374 + 8 & @xmath375 & @xmath45 + 9 & @xmath376 & @xmath377 + 10 & @xmath378 & @xmath63 ' '' '' + the results of computation are presented in figures [ fig : sic10k ] and fig : sic2k . here the bold lines indicate the full randomized model , the thin solid lines indicate the sic approximated model , the line with dots indicate the two - stage semi - randomized model . also the mean - field solution is presented as well and indicated by dashed lines . k. bold line full randomized model ( table 1 ) , thin line approximate randomized model ( table 2 ) , line with dots the proposed two - stage semi - randomized model , dashed line the mean field solution.,title="fig:",width=245 ] k. bold line full randomized model ( table 1 ) , thin line approximate randomized model ( table 2 ) , line with dots the proposed two - stage semi - randomized model , dashed line the mean field solution.,title="fig:",width=245 ] but for @xmath183k.,title="fig:",width=226 ] but for @xmath183k.,title="fig:",width=226 ] evidently , the proposed two - stage semi - randomized model gives a quite satisfactory approximation for the first two moments of total number of infectives in node 2 if the population is @xmath379k but only qualitative similarity for the smaller population . this justifies the used simplifications for rather moderate populated sites , but for the sites with population 2k and smaller a more sophisticated models should be developed . this can be a material of the consequent works . the randomized network sir model coupled by randomized migration fluxes is described in terms of a markov chain ( mc ) . in the absence of infectives the pure migration model give reasonable modelling of migration of individuals is described as a simple mc : being disturbed the system returns to a dynamic equilibrium exponentially fast that resembles a diffusion process in physics . in the mean - field ( hydrodynamic ) limit the mc converges to a non - standard network sir model : the host and guest species are treated separately in the corresponding odes ( [ ode : sn])([ode : imn ] ) . note that traditionally the coupling with other nodes is described by adding some transport terms ( cf . @xcite ) @xmath380 simple analysis shows that pure migration in the such equation results in exponentially growing solutions @xcite . this instability is ignored as it can be hidden in the background of the outbreak and not be observable in certain epidemic model scenarios . the model proposed in @xcite @xmath381gives more stable pure migration but the simple analysis shows that in this model we obtain the fully mixed population in all the nodes @xcite ( @xmath382 in our terms ) . thus the dynamics of this model seems to be more realistic but nevertheless does not satisfy an intuitive interpretation of the equilibrium of the migration process . the both above models are characterized by only one parameter describing migration of a given species . this makes impossible to tune the model to obtain the realistic migration in the absence of the outbreak . in our earlier work @xcite a migration model is introduced without splitting species into host and guest with two migration parameters : @xmath383 and @xmath384 describing migration process of a given species . but model proposed in @xcite also does not give a completely satisfactory solution for pure migration in the case of different migration times : @xmath385 and @xmath386 as shown in @xcite . note that for some combination of the network parameters ( epidemic and migration ) effect of the more correct accounting of migration can be very small but it can become essential when parameters of the model vary . it is interesting to compare three different techniques for the model under consideration : a mc describing the number of individuals from all categories in both centers , its hydrodynamical limit in the form of a system of dynamic equations and a simple description of contamination stage at the node 2 as an isolated center with an inflow of infectives neglecting the backward migration . we developed an approach when the random evolution on the contamination stage either in its full or simplified form is coupled with the dynamical description on the stage of saturation . this makes the problem computationally feasible . our intention is to apply this technique to a network of interacting population centers in the next paper . note that the direct simulation of the network is far too expensive in terms of computational time compared with the integration of systems of dynamical equations . the computational time may be considerably reduced if the simulation is required during a relatively short contamination periods only . the spatial dynamics of multi - species epidemic models is widely discussed in the literature ( see @xcite and the reference therein ) . for a purely deterministic model the account of different dynamics of host and guest species on the epidemic speed was studied in @xcite . the simplifying assumptions make the analysis tractable by may not adequately reflect reality . it seems that a network of stochasticly interacting centres of the type discussed above may provide more realistic but still tractable setting . in the next paper we intend to derive the traveling wave characteristic equation ( cf . @xcite,@xcite,@xcite ) and explore analytically an numerically the dependence of the mean epidemic speed and its standard deviation on the network parameters . the financial support within the framework of a subsidy granted to the national research university higher school of economics for the implementation of the global competitiveness is acknowledged . 99 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 s.m . ross , _ first course of probability _ , nj : prentice hall , 2008 i. sazonov , m. kelbert , m. b. gravenor , a two - stage model for the sir outbreak : accounting for the discrete and stochastic nature of the epidemic at the initial contamination stage , _ mathematical biosciences _ 234 ( 2011 ) 108117 . i. sazonov , m. kelbert , m. b. gravenor , the speed of epidemic waves in a one - dimensional lattice of sir models , _ mathematical modelling of natural phenomena _ 3 ( 4 ) ( 2008 ) 2847 . i. sazonov , m. kelbert , m. b. gravenor , travelling waves in a network of sir epidemic nodes with an approximation of weak coupling , _ mathematical medicine and biology _ 28 ( 2 ) ( 2011 ) 165183 . a.n . nayfeh , _ introduction to perturbation technique _ , n.y . : j. wiley @xmath387 sons , 1973 . d. daley , j. gani , _ epidemic modeling _ , cambridge university press , cambridge , 1999 . i. sazonov , m. kelbert , a new view on migration processes between sir centra : an account of the different dynamics of host and guest . _ arxiv _ : 1401.6830v1 [ q - bio . pe ] , 2013 mollison , d. , ed . . cambridge university press , cambridge , 1995 . murray , j. . springer , london , 1993 . y. suhov , m. kelbert , _ probability and statistics by example . vol.ii , markov chains : a primer in random processes and their applications_. cambridge : cambridge university press , 2008 r.m . corless , g.h . gonnet , d.e.g . hare , d.j . jeffrey , and d.e . knuth , on the lambert _ w _ function . _ advances in computational math . _ , 5 ( 1996 ) , 329359 . w. wang , and x. zhao , . , ( 2004 ) , 97112 . w. wang , and g. mulone , . , 1 ( 2003 ) , 321335 . j. arino , j. davis , d. hartley , r. jordan , j. miller and p. van den driessche , a multi - species epidemic model with spatial dynamics , _ mathematical medicine and biology - a journal of the i m a _ 22(2 ) ( 2005 ) , 129142 . d. mollison , dependence of epidemic and population velocities on basic parameters . ( 1991 ) , 255287 . b. bolker , and b. grenfell , space , persistence and dynamics of measles epidemics . , 1325 ( 1995 ) , 309320 . grenfell , a. kleczkowski , c.a . gilligan , and b.m . bolker , spatial heterogeneity , nonlinear dynamics and chaos in infectious diseases . , 2 ( 1995 ) , 160183 . grenfell , b.t . bjrnstad , and j. kappey , travelling waves and spatial hierarchies in measles epidemics . , 6865 ( 2001 ) , 716723 . j. arino , j. jordan and p. van den driessche , quarantine in a multi - species epidemic model with spatial dynamics . _ mathematical biosciences 206 _ ( 2007 ) , 4660 . s. bansal , b.t . grenfell and l.a . meyers large - scale spatial - transmission models of infectious disease . _ science _ 316 , 5829 ( 2007 ) , 12981301 . v. colizza , and a. vespignani , invasion threshold in heterogeneous metapopulation networks . ( 2007 ) , 148701 . v. colizza , and a. vespignani , epidemic modeling in metapopulation systems with heterogeneous coupling pattern : theory and simulations . , 3 ( 2008 ) , 450467 . j. burton , l. billings , d.a . cummings , and i.b . schwartz , disease persistence in epidemiological models : the interplay between vaccination and migration . , 1 ( 2012 ) , 9196 . a. shwartz , a. weiss , _ large deviation for performance analisys_. london : chapman and hall , 1995 m. kelbert , i. manolopoulou , i. sazonov and y.m . suhov , large deviations for a model of excess of loss re - insurance . _ markov processes and related fields_. * 13*(1 ) ( 2007 ) , 137158 a. muller , d. stoyan , _ comparison methods for stochastic models and risks _ , n.y . : wiley , 2002 first , the mean values of the markov chain ( mc ) converges to the solution of initial value problem ( [ ode : sn])([eq : s12:mean ] ) by the lln . phenomenological sketch is given section [ sec : mc ] , and the rigorous proof is analogous to that presented in @xcite . we also have to establish the convergence in probability . for definiteness consider @xmath251 and apply chebyshev s inequality for any @xmath388 @xmath389 } { \lambda ^{2}\epsilon ^{2}}.\]]recall that @xmath390 is the population scaling parameter ( see section sec : mc ) . so , it is enough to check that @xmath276 = o(\lambda ) , \qquad \lambda \rightarrow \infty . \label{app : cov}\]]it is demonstrated numerically in section [ sec : numerics ] ( see figure fig : stdimax - vs - n ) . actually we see that the normalized standard deviation decays as @xmath204 , or equivalently , the non - normalized standard deviation grows as @xmath391 ( i.e. , as @xmath392 ) , that implies ( [ app : cov ] ) . the rigorous argument runs as follows . consider the processes in table 1 which cause the change in number of infectives in node 2 and outline the fluxes of infectives . these processes are @xmath393{0pc}{1.5056pc } \\ \hline\hline \mbox{9,16 } & i_{2}\rightarrow i_{2}+1 & \beta _ { 2}i_{2}s_{2}+\underline{% \beta _ { 2}i_{12}s_{2}+\delta _ { 12}^{i}i_{21}}\rule[-0.5019pc]{0pc}{1.5056pc } \\ \hline \mbox{11,14 } & i_{2}\rightarrow i_{2}-1 & \alpha _ { 1}i_{2}+\underline{\gamma _ { 21}^{i}i_{2}}\rule[-0.5019pc]{0pc}{1.5056pc } \\ \hline\hline \end{array } \label{app : process-1}\]]here the flux terms are underlined , the remaining terms describe the mc based stochastic sir model @xcite . so the real flux can be defined as@xmath394 we construct the process @xmath395@xmath396{0pc}{1.5pc } \\ \hline\hline \tilde{i}_{2}\rightarrow \tilde{i}_{2}+1 & \beta _ { 2}^{\prime } \tilde{i}_{2}+% \tilde{\nu}\rule[-0.5pc]{0pc}{1.5pc } \\ \hline \tilde{i}_{2}\rightarrow \tilde{i}_{2}-1 & \alpha _ { 1}\tilde{i}_{2}% \rule[-0.5pc]{0pc}{1.5pc } \\ \hline\hline \end{array } \label{app : process-2}\]]with the majorized constant flux@xmath397remind that @xmath258 is a constant when @xmath86 . for this process we have a randomized si model ( considered in @xcite ) with the constant poisson flux @xmath398 . this problem is solved in section [ sec : sic ] and it is shown that its variance grows as @xmath399 . the second order stochastic domination means that for any convex function @xmath403 we have the inequality for all @xmath404 @xmath405\leq \mathbb{e}[\psi ( { \tilde{i}}_{2}(t))],\]]@xmath406 is the number of susceptible in the tractable model described by ( [ app : process-2 ] ) . in our case this implies the inequality @xmath408 \mathrm{var}\left [ \tilde{i}_{2}(t)\right ] $ ] .
the quantum electrodynamical vacuum polarization effects arising in the collision of a high - energy proton beam and a strong , linearly polarized laser field are investigated . the probability that laser photons merge into one photon by interacting with the protons electromagnetic field is calculated taking into account the laser field exactly . asymptotics of the probability are then derived according to different experimental setups suitable for detecting perturbative and nonperturbative vacuum polarization effects . the experimentally most feasible setup involves the use of a strong optical laser field . it is shown that in this case measurements of the polarization of the outgoing photon and and of its angular distribution provide promising tools to detect these effects for the first time .
0906.5576
with the advent of quantum field theory our idea of vacuum has been strongly modified with respect to that of classical physics . vacuum is no more intended as a literal nothing but as a physical entity with a complex structure . within the framework of quantum electrodynamics ( qed ) , the interest in investigating the properties of quantum vacuum has been growing considerably in the last years , mostly due to the advancement of the strong field laser technique . in particular , many experimental schemes have been put forward to measure these properties ( see the recent reviews @xcite and the references therein ) . in fact , qed predicts that in the presence of very intense electromagnetic fields quantum vacuum manifests dielectric properties due to the polarization of the virtual electron - positron background and even breaks down in stronger fields when virtual electron - positron pairs , acquiring sufficient energy from the field , become real @xcite . the strength scale of the electromagnetic field where the qed vacuum effects become apparent is determined by the so - called critical fields : @xmath0 and @xmath1 ( here @xmath2 and @xmath3 are the electron charge and mass , respectively and natural units with @xmath4 are used ) . the investigation of the properties of qed vacuum is closely connected with the possibility of testing qed in the presence of strong background electromagnetic fields . however , the values of the critical fields are far beyond the strength of electromagnetic fields that can be produced in terrestrial laboratories . a remarkable exception to this fact are highly charged nuclei . in the context of qed , `` highly charged nuclei '' are nuclei with a charge number @xmath5 such that @xmath6 , with @xmath7 the fine structure constant . in fact , the electric field of highly charged nuclei at the typical qed length @xmath8 ( compton length ) is comparable with the critical field @xmath9 . this explains the recent experimental successes in measuring the delbrck scattering @xcite , i. e. the scattering of a high - energy photon by the field of a highly charged ion , and the photon splitting in a coulomb field @xcite confirming the theoretical predictions @xcite . the development of laser technology is opening up the possibility of probing the properties of qed vacuum and of testing qed in the presence of intense wave fields . although the next generation of petawatt laser systems is expected to reach an intensity of `` only '' @xmath10 @xcite , different theoretical schemes have been put forward to approach the so - called schwinger intensity of @xmath11 @xcite . a significant step towards the schwinger intensity will be realized in the near future at the extreme light infrastructure ( eli ) where unprecedented intensities of the order of @xmath12 will be attained @xcite . the ever - increasing available laser intensities have stimulated numerous theoretical proposals to observe `` refractive '' vacuum polarization effects ( vpes ) induced by strong laser fields @xcite , while the observation of `` absorptive '' vpes in colliding laser beams , namely electron - positron pair creation , requires field intensities of the order of @xmath13 @xcite . however , laser - assisted pair production was observed about ten years ago at the stanford linear accelerator center ( slac ) in the collision of a high - energy electron beam ( @xmath14 ) and an optical laser beam with an intensity of the order of @xmath15 @xcite . due to the large electron energy , the effective laser field amplitude and frequency in the rest frame of the electron are much larger than their value in the laboratory frame and this has allowed electron - positron pair creation . a qed process in a strong laser field , that is characterized by a typical four - momentum of a particle @xmath16 and by a laser field with four - vector potential amplitude @xmath17 , field tensor amplitude @xmath18 , electric field amplitude @xmath19 and frequency @xmath20 , depends on the two invariant parameters @xmath21 and @xmath22 ( the metric @xmath23 is employed ) @xcite . the classical field parameter @xmath24 can be interpreted as the work done by the laser field on the electron in the compton length @xmath8 in units of the laser photon energy and , therefore , determines the role of multiphoton processes ( at @xmath25 the process is multiphoton ) . the same parameter @xmath24 also determines the adiabaticity of the process ( the limit @xmath26 and @xmath27 fixed corresponds to the constant crossed field case ) @xcite . the quantum field parameter @xmath27 determines the magnitude of the quantum nonlinear effects . when @xmath28 multiphoton effects play a minor role and the probability of the process goes to its perturbative weak field limit described by the feynman diagram(s ) of lowest order in the laser field . in the opposite case of @xmath26 the process is adiabatic , the laser field is almost constant and the photon picture does not hold anymore . in this limit another perturbative weak field regime exists when the parameter @xmath27 is much smaller than unity . in the following , by `` perturbative regime '' we mean that either @xmath29 or @xmath30 when the process probability can be represented as a power expansion in the corresponding small parameter . in the case of two equal , counterpropagating laser beams , the parameter @xmath27 results from the @xmath18 of one laser beam and from the @xmath31 of a photon of the other laser beam : @xmath32 . the schemes proposed in @xcite to observe vpes in colliding laser beams are limited to _ perturbative _ processes . this is because for usual parameters of strong optical lasers , though the condition @xmath33 is easily fulfilled , nevertheless , @xmath34 since @xmath35 and @xmath36 , and , consequently , the nonperturbative effects are suppressed @xcite . however , the vpes can be increased if a strong laser beam collides with a high - energy proton beam . analogously to the slac experiment @xcite , this is due to the lorentz boost of the laser field amplitude and frequency in the rest frame of the relativistically fast moving proton . the perturbative regime ( @xmath34 and @xmath37 ) of scattering of a strong laser field with arbitrary polarization by a coulomb field has been considered in @xcite . the first attempt to go beyond the perturbative regime of vpes in this process has been done in the early work of ref . @xcite , where the laser photon merging was considered in the scattering of a strong _ circularly _ polarized laser field by a coulomb field . due to the symmetry of the system , however , only two laser photons can merge if the laser field is circularly polarized , i. e. the possibility of multiphoton vpes is in principle ruled out in this case . moreover , the author explicitly considers only regions of parameters where nonperturbative effects either are completely negligible or cancel out each other , in such a way that the final results are , in fact , perturbative . already in ref . @xcite it was pointed out that the large lorentz boost of a high - energy proton beam enhances the cross - section of the photon merging process in the perturbative case . moreover , the enhancement of the laser field amplitude and frequency may open a way to nonperturbative regimes : @xmath33 and @xmath38 . in @xcite we have done the first calculation of probabilities of nonperturbative refractive vpes in the collision of a high - energy proton beam with a linearly polarized , strong laser field and proposed a setup that will allow an observation of the merging of multiple pairs of laser photons into a single high - energy photon due to vpe induced by the proton field and the laser field . the proton features fit very well the requirements of this setup . in fact , on the one hand it is light enough to be accelerated to very high energies like up to @xmath39 at the tevatron or even up to @xmath40 at the forthcoming large hadron collider ( lhc ) @xcite , such that the laser field amplitude and frequency in the proton rest frame are significantly larger than their values in the laboratory frame . on the other hand , the proton is heavy enough that multiphoton thomson scattering of the laser photons by the proton itself is strongly suppressed . this is of great advantage with respect to employing , for example , an electron beam because , as we will see , multiphoton thomson scattering represents a background of our process of laser photon merging . the nonperturbative vpes during a proton and a laser beam collision manifest themselves in two ways . first , the scaling of the photon merging probability is very sensitive to the laser field parameters . thus , in the limit @xmath33 the probability @xmath41 of the merging of @xmath42 photons scales as @xmath43 if @xmath44 , while in the perturbative case ( @xmath30 ) the scaling is @xmath45 . second , the multiphoton processes ( merging of @xmath42 laser photons with @xmath46 ) , while being negligibly small in the perturbative regime , become significant and observable in the nonperturbative regime . in our previous letter @xcite we have investigated the most favorable case for observation of multiphoton vpes by employing proton accelerators already available and the next generation of petawatt optical laser systems . accordingly , we have calculated the probabilities of the photon merging in the domain of @xmath26 and @xmath27 fixed . however , there are other interesting regimes of parameters which show peculiarities in the amplitudes of the process and which are connected with the possibility of the experimental observation of laser photon merging with the next generation of x - ray - free electron lasers ( x - fels ) like those available in the near future at desy @xcite and at slac @xcite . the investigation of these regimes is carried out in the present paper . in the present paper we thoroughly investigate the laser photon merging process during the proton and laser beam collision in nonperturbative regimes , taking into account exactly the influence of the laser field . starting from the general expressions of the amplitudes of the process , we obtain analytical asymptotics valid in different parameters regions and analyze ranges of parameters not considered yet . in particular , additionally to the previous results , we obtain new analytical expressions for the amplitudes of the process in the domain of @xmath28 and for any @xmath47 , as well in the regime of @xmath48 and @xmath25 that are relevant for x - fel . we also analyze the experimental feasibility of the process with different possible experimental devices . attention is devoted to the question of how advantageous for the photon merging process could be the complementary virtues of an x - fel with respect to an optical laser , namely , the high photon energy , the high repetition rate and the relatively large space - time volume of the laser beam . as we will see , the use of optical strong laser fields is much more favorable from an experimental point of view . for this reason , in addition to the results in @xcite , we also investigate in this case the polarization and the complete angular distribution of the emitted photons . the paper is organized as follows : the theoretical model is considered in sec . the asymptotics of the probability of the process are analyzed in sec . iii . in the same section , in each asymptotic limit the corresponding physical situations are considered and the process feasibility is discussed . the summary of sec . iv concludes the paper . for the sake of completeness , we have reported in an appendix the expression of the polarization operator found in @xcite that will be our starting point here . below , we consider the vpes arising in the head - on collision of a high - energy proton with a strong laser beam . we assume that the laser beam propagates in the positive @xmath49 direction and that it is linearly polarized along the @xmath50 direction . the amplitude and the intensity of the laser field will be indicated as @xmath19 and @xmath51 , respectively , whereas @xmath20 is the laser frequency . concerning the proton , we have modeled it as an external field due to its large energy and we have ensured that the recoil on the proton of the emitted high - energy photon was negligible in the numerical examples . we assume that the proton moves with velocity @xmath52 along the negative @xmath49 direction . the proton mass is indicated as @xmath53 and its relativistic lorentz factor as @xmath54 . as the proton is considered as an external field , the feynman diagram representing the process of photon merging is that shown in fig . [ feynman_diagram ] . the crossed photon leg corresponds to the proton electromagnetic field which is represented by the four - vector potential @xmath55 with @xmath56 being the four - velocity of the proton . since the proton charge number is @xmath57 , the proton field can be safely taken into account perturbatively at lowest order . the thick fermion lines in fig . [ feynman_diagram ] are dressed fermion propagators which exactly include the laser field . this is achieved by employing the volkov states to build the fermion propagator @xcite . finally , the photon line on the right in fig . [ feynman_diagram ] indicates the emitted photon resulting from the merging . we indicate its four - momentum as @xmath58 , with @xmath59 being the outgoing photon energy , and its two polarization four - vectors as @xmath60 , with @xmath61 . the presence of the laser field can in principle induce a mass different from zero of the outgoing photon through vacuum polarization and the appearance of a third , longitudinal photon mode @xcite . in order to estimate these effects we assume that the emitted photon counterpropagates with the laser field . from the expression of the polarization operator in a laser field one sees that at fixed @xmath62 then these effects are @xmath63 times smaller than those considered here , apart from the asymptotic case @xmath64 in which they are @xmath65 times smaller @xcite . in both cases , then , they can be safely neglected for realistic parameters . instead , in the constant crossed field limit @xmath33 and @xmath66 fixed , these effects scale as @xmath67 if @xmath68 and @xmath69 if @xmath70 @xcite . in this situation they become important at @xmath71 but we will not consider this extreme case . the two amplitudes of the laser photon merging process depending on the final photon polarization state are given by ( see fig . [ feynman_diagram ] ) @xmath72 where @xmath73 is the vacuum polarization tensor in a laser field @xcite and where @xmath74 is the four - dimensional fourier transform of the four - vector potential in eq . ( [ a ] ) , with @xmath75 and @xmath76 being the dirac @xmath77 function . the vacuum polarization tensor @xmath73 in a laser field has been calculated in @xcite and we report in the appendix the final expression valid for an elliptically polarized laser field with the main axes directed along the @xmath78 and @xmath50 direction _ and for initial and final four - momenta @xmath79 and @xmath80 such that , as in the situation at hand , @xmath81 and @xmath82_. by starting from that expression of the vacuum polarization tensor in the case of a linearly polarized laser field along @xmath50 ( one has to set @xmath83 and @xmath84 in eq . ( [ a_mu ] ) ) , it can be shown that the amplitudes @xmath85 can be written as @xmath86 to obtain this equation we have exploited the four - dimensional @xmath77 function present in eq . ( [ c_r ] ) that gives @xmath87 and then @xmath88 . also , we have employed the auxiliary four - vectors @xmath89 given in eq . ( [ lambda_j_app ] ) : @xmath90 where @xmath91 , @xmath92 and where @xmath93 is the four - momentum of the laser photons , and we have introduced the coefficients @xmath94\delta_{j,1}\right.\\ & \qquad\qquad\left.+\xi^2\frac{\sin^2\rho}{1-v^2}\mathcal{j}_n(z_{2n})+\frac{\eta_{2n}}{2}\left(n - i\frac{1-v^2}{4\rho}\right)j_n(z_{2n})\right\ } , \end{split}\ ] ] where @xmath95 is the kronecker @xmath77 function , @xmath96/2\}/\eta_{2n}(1-v^2)$ ] , @xmath97/\eta_{2n}(1-v^2)$ ] , @xmath98 and @xmath99 , with @xmath100 being the ordinary bessel function of order @xmath101 and @xmath102 its derivative . as it is clear from the expressions of the coefficients @xmath103 , the invariant amplitudes of the laser photon merging process essentially depend on the laser field only through the two gauge- and lorentz - invariant parameters @xmath104 in the expression of the parameter @xmath105 , we have introduced the four - momentum @xmath106 of the outgoing photon when @xmath42 ( @xmath107 ) laser photons merge , which is defined as @xmath108 here , @xmath109 is the energy of the outgoing photon when @xmath42 laser photons merge : @xmath110 while @xmath111 and @xmath112 are the polar and the azimuthal angle in the momentum space , being the laser propagation direction ( @xmath49 axis ) the polar axis . a few observations are in order . first , as it can also be inferred from the furry theorem , only an even number of laser photons can merge in the process at hand @xcite . second , we note that due to gauge and lorentz invariance the scalar coefficients @xmath103 can not depend on the scalar products @xmath113 and this is why they do not depend as well on the azimuthal angle @xmath112 . third , the parameter @xmath27 defined in the introduction depends here on the number of laser photons merged and it is equal to @xmath114 this is because the momentum @xmath16 entering the definition of @xmath27 is given here by @xmath106 or , which is equivalent , by the momentum transfer from the proton . it is worth mentioning here that , in general , the series in eq . ( [ m ] ) also contains terms with @xmath115 . these terms vanish in the situation at hand because , _ as we have shown in the appendix , _ the four - dimensional @xmath77 functions in eq . ( [ c_r ] ) can not be fulfilled . however , on the one hand , the term with @xmath116 contributes to the modification to the coulomb potential of the proton induced by vacuum polarization @xcite . on the other hand , it also gives rise to a radiative correction to thomson scattering without net absorption of laser photons . both these corrections are @xmath63 times smaller than the effect considered here and will be neglected . the coefficients @xmath103 show a highly nonperturbative dependence on the laser field parameters through @xmath24 and @xmath105 . the parameter @xmath24 , as already mentioned in the introduction , is responsible for the magnitude of high - order vacuum effects in the laser intensity . the quantity @xmath117 is the total energy of a laser photon and the outgoing photon in their center - of - momentum frame when @xmath42 laser photons merge in unit of the electron mass . at this point , in order to calculate the total rate @xmath118 of emitted photons one has to employ the amplitudes @xmath85 in eq . ( [ m ] ) and apply the fermi golden rule @xcite : @xmath119 here , we have assumed a space - time integration volume equal to unity and we have exploited the dirac @xmath77 function to perform the integral on the energy of the outgoing photon . @xmath120 indicates the differential solid angle relative to the outgoing photon direction . by employing the expressions of the four - vectors @xmath121 in eq . ( [ lambda_j ] ) , one can easily show that @xmath122 also , since the four - vectors @xmath89 are both perpendicular to @xmath80 ( see eq . ( [ lambda_j ] ) ) , when we sum in eq . ( [ r ] ) over the outgoing photon polarizations we can employ the substitution @xcite @xmath123 note that , in general , the sum should include the four polarization states of the photon ( in this case the above equation becomes an equality ) . however , due to gauge invariance only the two transverse photon polarization states contribute to the photon merging rate . this can also be verified through a direct calculation by observing that @xmath124 and by reminding that @xmath82 . in this way , the integration over the angle @xmath112 in eq . ( [ r ] ) can be easily taken and , by introducing the rate @xmath125 of the laser photon merging with an absorption of @xmath42 laser photons , we obtain @xmath126 from this expression it is clear that the evaluation of the laser photon merging rate reduces to the evaluation of the two coefficients @xmath103 . their general expressions in eq . ( [ c_j_n_ex ] ) are rather complex and , as we will see in the next section , they can be simplified in different asymptotical regions of the parametric space @xmath127 . we conclude this section with the following observation that we have mentioned in the introduction . it can easily be seen that the photons emitted via 2@xmath101-multiphoton thomson scattering of laser photons by the proton have the same energy as the photons emitted via the 2@xmath101-laser photon merging process because the kinematics of the two processes is the same ( see the review @xcite for a detailed discussion of multiphoton thomson scattering in the case of electrons ) . both processes occur in the collision of a proton and a laser beam and they interfere . as it is clear from the feynman diagram in fig . [ feynman_diagram ] , the photon merging process is qualitatively different from the multiphoton thomson scattering ( see fig . [ multiphot_thomson ] ) . in the latter , the photons of the laser interact _ directly _ with the proton through its electric charge . in the former , instead , the laser photons interact with the _ electromagnetic field _ of the proton through a virtual electron - positron pair . an immediate consequence of this difference is that the process of multiphoton thomson scattering , though being a process of lower order in @xmath63 than the process of @xmath42-laser photon merging , is nevertheless suppressed due to the large proton mass @xmath128 and in the limit @xmath129 the amplitude of the process vanishes . this is not the case of the laser photon merging process which has a finite leading - order amplitude in the limit @xmath129 that exactly corresponds to the feynman diagram in fig . [ feynman_diagram ] . the expressions ( [ c_j_n_ex ] ) of the coefficients @xmath103 hold for any value of the two physical parameters @xmath24 and @xmath105 . however , these expressions are rather complex and difficult to use for quantitative estimations . simpler analytical expressions can be found in different limits of the parameters @xmath24 and @xmath105 . as we will see below , this corresponds to various physical situations when different laser systems could be employed to observe the process of laser photon merging like , for example , x - fels or strong optical lasers . to the sake of completeness , we report here also the results obtained in @xcite concerning the case of an optical laser field ( see par . the expressions of the amplitudes for the case in which the strong laser field is an x - fel are obtained here for the first time and are reported in par . iii.a and par . iii.c . we consider in this paragraph the asymptotic expressions of the coefficients @xmath103 when the parameter @xmath24 can be considered much smaller than unity , while the parameter @xmath105 is arbitrary but fixed . this case is particularly suitable for treating the practical situation in which the strong laser field is that of an x - fel . in fact , due to the large laser frequency and the low intensity , the parameter @xmath24 for an x - fel is much smaller than unity . as a check , we consider the following laser parameters that will be available with the tesla project at desy @xcite : a laser frequency @xmath130 , a laser peak power of @xmath131 and a spot radius of @xmath132 corresponding to a peak intensity @xmath133 , and we merely obtain @xmath134 . on the other hand , being the laser frequency only two - three orders of magnitude smaller than the electron mass , it would then be limiting ( as we will see in a numerical example below ) to restrict anyway the values of the parameter @xmath105 ( see eq . ( [ eta ] ) ) mostly in the case of a high - energy proton beam . as we have mentioned , in the present case with @xmath29 , the laser field can be treated perturbatively . by expanding the exponential and the bessel functions in the coefficients @xmath103 with respect to the small parameter @xmath24 , one sees that the leading term of the series is proportional to @xmath135 . this term corresponds to the process of laser photon merging in which @xmath42 laser photons merge . the next terms in the expansion result proportional to @xmath136 , with @xmath137 being a positive integer , and correspond to the process of @xmath42-laser photon merging accompanied by the exchange of @xmath137 laser photons without _ net _ absorption . due to the dependence on @xmath135 , higher - order processes are suppressed and the leading contribution to the laser photon merging process comes from the term with @xmath138 and @xmath139 , i. e. from the merging of two laser photons without any additional exchange . one of the feynman box - diagrams representing this process is shown in fig . [ box_diagram ] . _ in the remaining two , the laser photon lines are both either above or below the proton field and the outgoing photon lines . the use of the volkov states as electron states automatically takes also into account the contribution of all these diagrams . it is also worth observing here that another diagram is present with four external photon lines , namely the one with one laser photon and two proton field lines . this diagram is the lowest - order diagram contributing to the delbrck scattering of a laser photon by the proton . the probability of this process is roughly @xmath140 times the probability of 2-laser photon merging and then it can be dominating in the present case . however , the two processes are experimentally distinguishable because the energy of the photon emitted via delbrck scattering is half that of the photon emitted via 2-laser photon merging . _ after expanding the exponential and the bessel functions in eq . ( [ c_j_n_ex ] ) , the integral over @xmath141 can be performed ( we call to mind that the convergence of this integral as @xmath142 is ensured by the usual @xmath143 prescription ) and we obtain : @xmath144\text{arctanh}\left(\frac{2}{g}\right)+(-1)^j4(1-v^2)\right.\\ & \qquad\qquad\left.-[8(v^{2(2-j)}-g)+3^{2-j}g^2(1-v^2)]\log\left(1-\frac{4}{g^2}\right)\right\ } , \end{split}\ ] ] where we have introduced the quantity @xmath145 . the remaining integration in eq . ( [ c_j_n_xi_small ] ) can be performed with mathematica and after a number of simplifications ( we have employed , in particular , the properties of the euler s dilogarithm function described in @xcite ) the final result is @xmath146\right\ } , \end{split}\\ \label{c_2_2 } \begin{split } c_{2,2}&=\frac{\xi^2}{8\eta_2 ^ 2}\left\{\eta_2 ^ 2 + 4(1 - 3\eta_2)\sqrt{\eta_2}\left(\sqrt{1+\eta_2}\text{arcsinh}\sqrt{\eta_2}-\sqrt{2+\eta_2}\text{arcsinh}\sqrt{\frac{\eta_2}{2}}\right)\right.\\ & \qquad\qquad\left.-2(1-\eta_2)\left[(1 + 2\eta_2)\text{arcsinh}^2\sqrt{\eta_2}-2(1+\eta_2)\text{arcsinh}^2\sqrt{\frac{\eta_2}{2}}\right]\right\}. \end{split}\end{aligned}\ ] ] it is remarkable that these expressions are exact in the parameter @xmath147 and they are explicitly expressed only through elementary functions . moreover , the analytical structure of the amplitudes in the present case is completely different from that in the case of a strong optical laser field ( see eq . ( [ c_n ] ) below ) . in particular , altough the present regime is perturbative in the sense that @xmath29 , a complex dependence of the amplitudes is observed on the laser frequency @xmath20 through the parameter @xmath147 . in the limit of @xmath147 much smaller than unity ( small laser frequencies ) , we obtain from eqs . ( [ c_1_2])-([c_2_2 ] ) @xmath148 in this case the total rate @xmath149 can be easily calculated and is given by @xmath150 this result does not agree with the corresponding one calculated in @xcite and presented there in eq . ( 61 ) where a factor four is missing . we point out , however , that all the conclusions of the paper @xcite are unaffected by this misprint . the result in eq . ( [ r_2 ] ) coincides with the rate calculated by starting from the lowest - order euler - heisenberg lagrangian density @xcite @xmath151,\ ] ] in which the total electromagnetic field @xmath152 is replaced by the sum of the proton field , the laser field and the photon field and it is expanded up to linear terms in the proton and the photon fields . this result which may appear obvious when also looking at fig . [ box_diagram ] deserves investigation . for example , the same would not be true if a laser photon line were substituted in fig . [ box_diagram ] by a proton line ( delbrck scattering ) @xcite . in fact , the use of the euler - heisenberg lagrangian density is allowed when the momenta flowing in the electron propagators ( see fig . [ box_diagram ] ) are much smaller than the electron mass @xmath3 . in the case of delbrck scattering this is not true for the momentum flowing between the two photon legs representing the proton electromagnetic field . instead , in our case the momentum @xmath79 absorbed by the proton is fixed by the energy - momentum conservation relation @xmath153 . in this way , @xmath154 and @xmath155 , all the momenta flowing in the box diagram are then much smaller than @xmath3 . in the opposite limit @xmath156 , one obtains for the coefficients in eqs . ( [ c_1_2 ] ) and ( [ c_2_2 ] ) the following expressions : @xmath157.\ ] ] the behavior of the amplitudes in this limit depends on the polarization of the laser field : in @xcite it was found that in the same limit the amplitudes of the laser photon merging process in a circularly polarized laser field become independent of the parameter @xmath147 . although interesting from a theoretical point of view , this case is of less practical interest because , in general , it would require too high laser frequencies . in fact , if the emission angle @xmath111 is much smaller than unity , the condition @xmath156 means that @xmath158 . instead , at intermediate angles such that @xmath159 , the condition @xmath156 implies that @xmath160 . a less restrictive condition is obtained if @xmath161 ( outgoing photon almost collinear with the proton ) . in fact , in this case the condition @xmath156 is fulfilled if @xmath162 . this strong inequality can in turn be fulfilled by combining an x - fel with @xmath163 @xcite with a large accelerator like tevatron or lhc for which @xmath164 @xcite . however , one has to bear in mind that at @xmath111 exactly equal to @xmath165 the laser photon merging rate vanishes . we conclude this paragraph by showing in fig . [ small_xi ] the rate of 2-photon merging by employing the aforementioned laser parameters that will be available at the tesla facility at desy : @xmath130 and @xmath133 . the figure shows the behavior of the rate , calculated by employing eqs . ( [ c_1_2 ] ) and ( [ c_2_2 ] ) , as a function of the emission angle @xmath111 at a typical value @xmath166 for the lorentz relativistic factor of the proton . the rate is almost peaked for back - emitted photons where the parameter @xmath147 is larger . at those values of @xmath111 where the spectrum has a peak , one can see that the value of @xmath147 can be larger than unity . finally , the figure also shows the advantage of using strong optical laser fields in order to observe the laser photon merging process experimentally : in the present case , the photon merging rate is several orders of magnitude smaller than that obtained in the next section for the optical laser case ( cf . also the numerical simulation relative to fig . 2 in @xcite ) . it can be easily checked that the high effective repetition rate of the x - fel ( 30000 pulses in one second @xcite ) and its relatively large space - time volume ( @xmath167 waist size and @xmath168 pulse duration @xcite ) can not compensate for the negligibly small rate . available optical lasers easily exceed the electron relativistic threshold @xmath171 . in fact , for a typical optical photon energy of @xmath172 , the condition @xmath173 is fulfilled at a laser intensity of about @xmath174 . on the other hand , as the laser photon energy is much smaller than the electron mass @xmath175 , one can also assume that @xmath169 , even in the case when the gamma factor of the proton is as large as about @xmath176 like at the lhc . now , in the asymptotic limit @xmath177 and @xmath178 in such a way that the parameter @xmath179 remains finite ( see eq . ( [ chi_n ] ) ) , it can be shown that the leading - order contributions to the coefficients @xmath103 are @xmath180+\frac{i_n(x_{2n})}{\lambda}\right\}\ ] ] where @xmath181 and where @xmath182 is the modified bessel function of order @xmath101 and @xmath183 its derivative . in this limit the coefficients @xmath103 depend only on the parameter @xmath184 . the above expressions are valid for any value of @xmath184 and we have also calculated the two limiting expressions of @xmath103 when the parameter @xmath184 is much smaller than unity and much larger than unity ( see eqs . ( 6 ) and ( 7 ) in @xcite ) . in the former case , the rate of photon merging with absorption of more than two laser photons is negligible and the total photon merging rate with absorption of two laser photons is given again by eq . ( [ r_2 ] ) . this is an interesting point that deserves examination . in fact , as we have seen in the previous paragraph , the perturbative approach holds when @xmath29 while we are working in the opposite limit @xmath33 here . however , one can show that for the process at hand the limit @xmath33 and @xmath185 gives the same result of the limit @xmath29 and @xmath169 . this is a peculiarity of the photon merging process and it was also pointed out in @xcite in the case of a circularly polarized laser field : it was noted by affleck and kruglyak in @xcite that this is not true in the case of photon splitting in a laser field ( see also @xcite ) where the results of the two limits are different . in the case of photon splitting @xmath186 , where @xmath187 , with @xmath80 being the four - momentum of the incoming photon . in the limit @xmath33 and @xmath30 the photon splitting amplitude corresponds to the hexagon feynman diagram with three laser photon legs and not to the box diagram as in the limit @xmath29 and @xmath188 . the regime of the photon merging @xmath26 and @xmath189 is the most interesting nonperturbative regime . in the limit @xmath190 the scaling of the probability of photon merging is nonperturbative ( see eq . ( 7 ) in @xcite ) : @xmath191 . in this regime merging of many laser photon pairs is not negligible as in the perturbative regime ( @xmath185 ) where it is damped by a small factor @xmath192 . nevertheless , if @xmath193 , then the rate of multiphoton processes decreases slowly with increasing @xmath101 as @xmath194 . in @xcite we have analyzed the feasibility of observing experimentally nonperturbative vpes in a strong laser field in the regime @xmath26 and @xmath189 . as we have mentioned in the introduction , various proposals have already been put forward to observe perturbative vpes in a strong laser field @xcite . we provide below a numerical example showing that with our scheme the requirements for the laser field to observe perturbative vpes are easily fulfilled nowadays . as a strong laser beam we consider , in fact , a typical multiterawatt laser beam with the following parameters : a laser photon energy of @xmath195 , a pulse energy of @xmath196 , a pulse duration of @xmath197 at @xmath198 repetition rate and a spot radius of @xmath199 ( the resulting laser intensity is @xmath200 ) @xcite . due to the low photon energy we have to use a high - energy proton beam so that multiphoton thomson scattering does not conceal the laser photon merging process . to show this , we calculate the ratio @xmath201 between the total rate @xmath202 of 2-photon thomson scattering @xcite @xmath203 and the total rate @xmath149 in eq . ( [ r_2 ] ) , and we obtain @xmath204 ^ 4\approx 5\times 10 ^ 7\left[\frac{1}{\gamma(1+\beta)}\frac{m^2}{\omega_0 m}\right]^4.\ ] ] if we use the above laser photon energy and we assume that @xmath205 , we find @xmath206 . in this way , if we require that the background process of 2-photon thomson scattering does not conceal our effect , i. e. that at most @xmath207 , then proton relativistic factors larger than @xmath208 are required that will be available at lhc . the main proton beam parameters at lhc are @xcite : a proton energy of @xmath209 , a number of protons per bunch of @xmath210 , a bunch transversal radius of @xmath211 , a bunch length of @xmath212 . as we have done in @xcite , in fig . [ optical_weak ] we compare the differential rate @xmath213 of photons emitted only by 2-photon thomson scattering ( dashed line , result via @xcite ) and the differential rate @xmath214 of photons emitted also by including the 2-photon merging process ( continuous line ) . we have already observed that in general these two processes interfere because the photons emitted have the same frequency in the two cases . the contribution of the vpes is clear from the figure and it becomes more evident if we calculate the total number of photons emitted in one hour only via 2-photon thomson scattering and only via 2-photon merging . to do this we first integrate numerically the corresponding differential rates over the angle @xmath111 . then , we need the effective number of protons that pass through the laser beam in one shot , the laser pulse duration ( the proton bunch is temporally much longer than the laser pulse ) and then the laser repetition rate ( the repetition rates of conventional proton accelerators are orders of magnitude larger than the laser repetition rate ) . in turn , we can estimate the effective number @xmath215 of protons passing through the laser beam as @xmath216 with @xmath217 being the laser rayleigh length . by using these quantities we can estimate the total number of photons emitted in one hour via 2-photon merging as @xmath218\tau_l[\text{s}]\nu_l[\text{hz}]$ ] and we obtain 180 events . analogously , we obtain that the total number of photons emitted in one hour only via 2-photon thomson scattering is again 180 . if one includes the two processes by summing their amplitudes , one obtains about 360 events as the total number of photons produced . this shows that there is no interference between the two processes in the present regime . in fact , at small @xmath219 the coefficients @xmath220 are real and the amplitude of the 2-photon merging process is purely imaginary ( see eq . ( [ m ] ) ) . on the contrary , it can be shown starting from the results in @xcite , that the amplitude of the 2-photon thomson scattering is purely real . the values of the parameter @xmath219 shown in the upper horizontal axis in fig . [ optical_weak ] indicate that the results are perturbative in the laser field because @xmath221 . also , as it is expected in the perturbative regime , the merging of 4-laser photons is completely negligible : only @xmath222 events per hour . this is a clear indication of the qualitative and quantitative difference between the perturbative and the nonperturbative regime in the process of laser photon merging . it is worth observing that here the number of events per hour is not much smaller than that obtained in the numerical example shown in @xcite where in the nonperturbative regime we obtained about 390 2-photon merging events per hour . this is due to the fact that here the spacetime overlapping of the laser beam and the proton beam is much larger than in @xcite where a tight focused ( @xmath223 ) , short ( @xmath224 ) laser pulse was employed . in fact , the 2-photon merging rate here is about 4 - 5 orders of magnitude smaller than that in @xcite . in the above example a proton beam with the parameters available at the lhc is required . however , the perturbative regime of the photon merging process can be realized even in an all - optical setup using laser parameters envisaged at eli @xcite . the proton beam could be produced in the so - called laser - piston regime by shooting an ultra - strong laser beam onto a plasma slab @xcite . according to the simulations of the latter , proton energies of about @xmath225 can be envisaged . the other relevant proton beam parameters are : number of protons in the bunch @xmath226 , bunch transversal radius @xmath227 and bunch length @xmath228 . the relatively low proton energy can be compensated by employing a strong attosecond pulse of extreme ultraviolet ( xuv ) radiation ( see @xcite ) : intensity of @xmath229 , photon energy of @xmath230 and pulse duration of @xmath231 . this xuv pulse , according to @xcite , can be produced by the reflection from a plasma surface of an ultra - strong laser pulse of @xmath232 duration , intensity @xmath233 , focused onto a spot - radius of @xmath234 which assumes conversion efficiency of @xmath235 ( we have extrapolated the values of the conversion efficiency that in @xcite are available for initial laser intensities up to about @xmath236 ; however , the conversion efficiency weakly depends on the laser intensity at high relativistic intensities ) . the dependence of the 2-photon merging rates @xmath213 and @xmath214 is similar to that in fig . 5 and we do nt report it . for the total rates we obtain 6.9 photons per shot only due to 2-photon thomson scattering , 5.2 photons per shot only due to 2-laser photon merging and 12.1 photons per shot due to the two processes together . we also stress that the simulations performed in @xcite have been carried out for laser intensities much smaller than those available at eli . if the results of @xcite can be scaled to intensities of the order of @xmath12 like those available at eli , much larger rates can be expected . moreover , the values of the parameters @xmath184 can reach the order of unity and non - perturbative , multiphoton vpes ( merging of more than two laser photons ) could become observable . since the present case of a strong optical laser field is the most favorable from an experimental point of view , we also investigate here for this case the polarization properties of the emitted photons and their complete angular distribution . starting from eq . ( [ r ] ) we obtain the rate per unit of solid angle and for each polarization @xmath237 as @xmath238 the two polarization unit vectors of the final photons have been chosen as @xmath239 with @xmath240 by employing these polarization four - vectors and the four - vectors @xmath241 given in eq . ( [ lambda_j ] ) , we obtain the following expressions for the four - dimensional scalar products occurring in eq . ( [ r_a ] ) @xmath242 in the above paragraph we have seen that the contribution of the laser photon merging to the total rate of photons produced together with multiphoton thomson scattering is in principle measurable . in the following we show that the contribution of the laser photon merging process is more clearly detectable by measuring the polarization and the angular distribution along the angle @xmath112 of the final photons . to do this we reconsider the first numerical example already discussed in the above paragraph which is based on already existing laser and proton acceleration technique . in figs . 6 and 7 we show the photon rate emitted per unit of solid angle and with polarization 1 and 2 , respectively . in the part a ) of the figures we have plotted only the contribution @xmath243 due to 2-photon thomson scattering ( see the review @xcite for a detailed discussion about multiphoton thomson scattering of electrons ) while in the part b ) we have plotted the total contribution @xmath244 including also the 2-laser photon merging process . in the case of polarization 1 the two plots are very similar , implying that the contribution of the laser photon merging is smaller than that of 2-photon thomson scattering . however , in the case of polarization 2 the opposite situation occurs : the contribution of the 2-laser photon merging process is dominating and the angular distribution along @xmath112 is clearly different in the two cases . in particular , the modulation depth of the total photon rate at varying @xmath112 and at fixed @xmath245 is at least 4 times larger than that for the thomson scattering case only . this property will allow to distinguish experimentally the contribution of the 2-laser photon merging process without a measurement of an absolute photon number . it is also worth observing from eqs . ( [ e_1_lambda_1])-([e_2_lambda_2 ] ) that the dependence of the two differential rates @xmath246 on the azimuthal angle @xmath112 is rather complex and , in particular , it can not be concluded that @xmath247 , with @xmath248 being an arbitrary fixed angle . this is a consequence of the fact that the linear polarization of the laser field along the @xmath50 direction breaks the cylindrical symmetry around the @xmath49 direction . to the sake of clarity we also present in figs . 8 and 9 the above photon rates but integrated with respect to @xmath112 ( and multiplied times @xmath249 ) . the two curves in fig . 8 corresponding to the case of polarization 1 are again very similar . however , the two curves in fig . 9 corresponding to the case of polarization 2 are different not only because the contribution of the 2-photon thomson scattering is very small but also because the angular distributions of the two rates show a peak at different angles @xmath111 . in conclusion , measurements of the polarization of the final photons and of their angular distribution can be a useful experimental tool to detect the process of 2-laser photon merging in the presence of a strong optical laser field . finally , we observe that the differential rates @xmath250 and @xmath251 integrated with respect to @xmath112 can be obtained from figs . 8 and 9 by subtracting the dotted curves from the continuous ones because the interference between 2-photon thomson scattering and 2-photon merging is in this case negligible . as we have already mentioned , the difference between these two rates is a consequence of the symmetry breaking induced by the linear polarization of the laser field . as we have mentioned in par . [ small_xi ] , this asymptotic limit is of more theoretical than practical interest due to the large laser frequencies required , in general , to achieve this regime . moreover , in the present case we assume that the parameter @xmath24 is not much smaller than unity and this circumstance is verified only by employing the rather speculative `` goal '' parameters for the future x - fel at desy as given in @xcite . however , this regime is interesting because it can indicate modifications of the behavior of qed at high energies due to the presence of a strong laser field . this circumstance has already been noted , for example , in @xcite , in investigating the pair - creation process in a strong laser field and a nuclear field . whereas , the opposite situation was found in @xcite where at high energies of the incoming photon the photon splitting amplitude resulted the same as that of photon - photon scattering in vacuum . in order to find the leading term of the coefficients @xmath103 it is more convenient to take first the integral on @xmath253 in eq . ( [ c_j_n_ex ] ) . this integral can be performed from @xmath254 to @xmath255 due to the symmetry of the integrand and then it is convenient to change the variable according to : @xmath256 . now , in the asymptotic limit @xmath252 it is possible for each @xmath101 to find a constant @xmath257 such that @xmath258 . after that , one divides the integral over @xmath259 into two integrals , one from @xmath254 to @xmath257 and one from @xmath257 to @xmath255 . in this way the coefficients @xmath103 receive two contributions and they can be written as @xmath260 for which we have pointed out that the two single contributions @xmath261 and @xmath262 ( but not their sum ) depend , in general , on the arbitrary quantity @xmath257 . the calculation of the leading terms in @xmath263 to @xmath261 and @xmath262 is lengthy but straightforward . below , we give the final results : @xmath264\right.\\ & + i\frac{b^3}{16a^2}{_3f_2}\left(1,1,\frac{3}{2};2,3;\frac{b^2}{a^2}\right)+\frac{\xi^2}{2}\left(1+\frac{\sin^2\rho}{\rho^2}-\frac{\sin 2\rho}{\rho}\right)(2-j)+\\ & -i\frac{b}{2}\left(c+i\frac{\pi}{2}-\log\left(\frac{2\eta_2}{a}\right)+\frac{i}{2\rho}\right)\bigg\ } \end{split}\ ] ] if @xmath138 , and @xmath265\\ & \left.+\frac{b}{2(n^2 - 1)}\left(\frac{b}{ia}\right)^{n-1}\left(1+n\sqrt{1-\frac{b^2}{a^2}}\right)\left(1+\sqrt{1-\frac{b^2}{a^2}}\right)^{-n}\right\ } \end{split}\ ] ] if @xmath46 . in eq . ( [ c_j_2_large_eta ] ) we have introduced the euler constant @xmath266 and the hypergeometric function @xmath267 , with @xmath268 and @xmath269 being arbitrary real parameters and with @xmath50 being in general a complex variable @xcite . we have also used the following notation @xmath270,\\ b&=\rho\xi^2\left(\frac{\sin^2\rho}{\rho^2}-\frac{\sin 2\rho}{2\rho}\right).\end{aligned}\ ] ] it is worth noting that in the present case the amplitudes have a complex dependence on the two parameters @xmath24 and @xmath147 _ independently _ , while in the case of an optical laser field they depend only on the combination @xmath179 . at large @xmath147 the dominant contribution with logarithmic accuracy comes from eq . ( [ c_j_2_large_eta ] ) and it is given by @xmath271 by employing the numerical parameters indicated as `` goal '' in [ 28 ] , i. e. @xmath272 and @xmath273 , we obtain @xmath274 and @xmath275 at @xmath161 and @xmath276 and then @xmath277 . however , the final number of photon merging events per hour turns out to be negligibly small due to the fact that the laser field is supposed to be focused to one wavelength of @xmath278 to obtain the quoted intensity . then the overlapping volume between laser and proton beams is very small . we have checked that if @xmath29 then the largest coefficients are those with @xmath138 and their expressions coincide with those already found in par . iii a ( see eq . ( [ c_j_n_xi_small_large_eta ] ) ) . in the strong - field limit @xmath33 we observe that both @xmath279 and @xmath280 become proportional to @xmath281 . then , independently of the number of laser photons that merge the amplitudes result proportional to @xmath281 . moreover , the coefficients @xmath220 also maintain a nontrivial logarithmic dependence on @xmath24 through the function @xmath282 . it is interesting to note that in this limit the amplitudes @xmath220 depend effectively on the parameter @xmath283 , with @xmath284 being the electron mass dressed in the laser field . in conclusion , in this paper we have continued the study of the process of laser photon merging in laser - proton collisions that we started in @xcite . we confirm that the most interesting regime from an experimental point of view is that investigated in @xcite , which offers the possibility of observing at least in principle nonperturbative refractive vacuum polarization effects ( vpes ) in a laser field . in fact , we have seen that the higher intensity of strong available optical laser fields compensates for the high photon energy , the high repetition rate of an x - fel and for the large overlapping region between the x - fel and the proton beams . we have shown here with a numerical example that , in order to observe perturbative vpes with our setup , optical laser intensities of the order of @xmath285 are already sufficient . due to the experimental relevance of the case of a strong optical laser field , we have also investigated for this case the polarization and the complete angular distribution of the emitted photons . interestingly , we have seen that both polarization measurements and angular distribution measurement can be performed to clearly discriminate between the process of laser photon merging and the background process of multiphoton thomson scattering . the other regimes of parameters studied in the present paper have shown interesting features as well . first , we have found the leading - order contribution to the coefficients @xmath103 for small @xmath24 and arbitrary @xmath105 . we have seen that this regime is suitable when the laser beam used to polarize the vacuum is an x - fel . the leading - order contributions in this regime are those corresponding to 2-laser photon merging and they are expressed only through elementary functions of @xmath147 without any further integration . moreover , we have studied the limit of large @xmath105 and fixed @xmath24 and we have seen that the coefficients @xmath103 show a complex dependence on the relativistic parameter @xmath24 . in the strong field limit @xmath33 the amplitudes show the same proportionality to @xmath281 , independently of the number of laser photons that merge . finally , in the same limit the coefficients @xmath220 ( corresponding to the 2-photon merging process ) also show a nontrivial logarithmic dependence on @xmath24 . we report here the final expression found in @xcite for the polarization operator in the plane wave with four - vector potential @xmath286 where @xmath91 , @xmath92 and where @xmath93 is the four - momentum of the laser photons . we consider only the case where the initial photon with four - momentum @xmath79 is virtual _ and @xmath81 _ , while the final one with four - momentum @xmath80 is real and @xmath82 . the polarization operator @xmath73 can be expressed as ( we note that in the more general case with @xmath287 a fifth term would be present in the following sum ) @xmath288 in this expression the four - vectors @xmath89 with @xmath289 are given by @xmath290 and the coefficients @xmath291 with @xmath292 can be written as @xmath293\right\}\\ & \times\sum_{l=1}^{\infty}\delta(q+2lk_0-k)g_{r , l } , \end{split}\ ] ] where @xmath294 is the four - dimensional dirac @xmath77 function and where @xmath295i^{-l}j_{-l}(z ) , \end{split}\\ g_{4,l}&=(-1)^lg_{3,l}(\xi_1 ^ 2\longleftrightarrow\xi_2 ^ 2),\end{aligned}\ ] ] with @xmath296 , @xmath100 being the ordinary bessel function of order @xmath101 and @xmath102 its derivative and with @xmath297 _ we mention that , in general , the series in eq . ( [ c_r ] ) also contains terms with @xmath298 . however , their contributions vanish in our case because the corresponding @xmath77 functions can not be fulfilled . in fact , from the relation @xmath299 it follows that @xmath300 . then , since @xmath301 , it must be @xmath302 . _ . the parameters of the laser field are @xmath130 and @xmath133 and the proton s relativistic lorentz factor is @xmath166 . the upper horizontal axis shows the values of the parameter @xmath147 as a function of @xmath111 . ] of photons emitted only via 2-photon thomson scattering ( dashed line ) and via both 2-photon thomson scattering and 2-photon merging ( continuous line ) . the upper horizontal axis shows the values of the parameter @xmath219 as a function of @xmath111 . the parameters of the laser field are @xmath195 and @xmath200 and the proton s relativistic lorentz factor is @xmath303 . ] per unit of solid angle @xmath304 of photons emitted per second with polarization 1 only via 2-photon thomson scattering ( part a ) ) and via both 2-photon thomson scattering and 2-laser photon merging ( part b ) ) . the parameters of the laser field and of the proton are the same as in fig . 5 . ] per unit of solid angle @xmath304 of photons emitted with polarization 2 only via 2-photon thomson scattering ( part a ) ) and via both 2-photon thomson scattering and 2-laser photon merging ( part b ) ) . the parameters of the laser field and of the proton are the same as in fig . 5 . ] of photons emitted with polarization 1 only via 2-photon thomson scattering ( dashed line ) and via both 2-photon thomson scattering and 2-photon merging ( continuous line ) . the upper horizontal axis shows the values of the parameter @xmath219 as a function of @xmath111 . the parameters of the laser field and of the proton are the same as in fig . 5 . ] of photons emitted with polarization 2 only via 2-photon thomson scattering ( dashed line ) and via both 2-photon thomson scattering and 2-photon merging ( continuous line ) . the upper horizontal axis shows the values of the parameter @xmath219 as a function of @xmath111 . the parameters of the laser field and of the proton are the same as in fig . 99 y. i. salamin _ 427 * , 41 ( 2006 ) ; m. marklund and p. k. shukla , rev . 78 * , 591 ( 2006 ) ; g. mourou , t. tajima , and s. v. bulanov , _ ibid . _ , 309 ( 2006 ) . w. dittrich and m. reuter , _ effective lagrangians in quantum electrodynamics _ ( springer - verlag , berlin , 1985 ) . w. greiner , b. mller , and j. rafelski , _ quantum electrodynamics of strong fields : with an introduction into modern relativistic quantum mechanics _ ( springer , berlin , 1985 ) w. dittrich and h. gies , _ probing the quantum vacuum _ ( springer - verlag , berlin , 2000 ) . akhmadaliev , et al . c * 58 * , 2844 ( 1998 ) . sh . akhmadaliev _ et al . lett . * 89 * , 061802 ( 2002 ) . r. n. lee , a. i. milstein , v. m. strakhovenko , zh . 112 * , 1921 ( 1997 ) [ jetp * 85 * , 1049 ( 1997 ) ] ; phys . a * 57 * , 2325 ( 1998 ) ; phys . a * 58 * , 1757 ( 1998 ) . r. n. lee _ et al . _ , phys . rep . * 373 * , 213 ( 2003 ) . j. norby , laser focus world , january 1 ( 2005 ) . t. tajima and g. mourou , phys . rev . beams * 5 * , 031301 ( 2002 ) ; s. v. bulanov , t. esirkepov and t. tajima , phys . lett . * 91 * , 085001 ( 2003 ) ; s. gordienko , a. pukhov , o. shorokhov , and t. baeva , phys . rev . lett . * 94 * , 103903 ( 2005 ) . see the eli proposal at http://www.extreme - light - infrastructure.eu/. a. a. varfolomeev , sov . jetp * 23 * , 681 ( 1966 ) ; z. bialyncka - birula and i. bialyncki - birula , phys . d * 2 * , 2341 ( 1970 ) ; e. brezin and c. itzikson , phys . rev . d * 3 * , 618 ( 1971 ) ; i. affleck and l. kruglyak , phys . 59 * , 1065 ( 1987 ) ; n. n. rozanov , jetp * 76 * , 991 ( 1993 ) ; d. bernard _ et al . _ , j. d * 10 * , 141 ( 2000 ) ; g. brodin , m. marklund , and l. stenflo , phys . lett . * 87 * , 171801 ( 2001 ) ; a. di piazza , k. z. hatsagortsyan , and c. h. keitel , phys . d * 72 * , 085005 ( 2005 ) ; e. lundstrm _ et al . lett . * 96 * , 083602 ( 2006 ) ; j. lundin _ a * 74 * , 043821 ( 2006 ) ; t. heinzl _ et al . _ , opt . commun . * 267 * , 318 ( 2006 ) ; a. di piazza , k. z. hatsagortsyan , and c. h. keitel , phys . lett . * 97 * , 083603 ( 2006 ) ; a. m. fedotov and n. b. narozhny , phys . lett . a * 362 * , 1 ( 2007 ) . a. i. milstein _ et al . _ , a * 72 * , 052104 ( 2005 ) . a. di piazza , a. i. milstein , and c. h. keitel , phys . a * 76 * , 032103 ( 2007 ) . d. l. burke _ et al . _ , * 79 * , 1626 ( 1997 ) . we point out that in the slac experiment the electron - positron pairs were not produced only by laser fields . either the coulomb field of the electron or the intermediate @xmath305-photon produced via compton scattering assists the laser field in producing pairs . the situation considered in @xcite is slightly different . we found that the photon splitting rate shows nonperturbative dependence on the laser parameters . however , we have not found an experimental scheme suitable to observe the process of photon splitting even with forthcoming laser facilities like eli . a. di piazza , k. z. hatsagortsyan , and c. h. keitel , phys . * 100 * , 010403 ( 2008 ) . see also arxiv:0708.0475 [ hep - ph ] . particle data group : w .- m . _ , j. phys . g * 33 * , 1 ( 2006 ) . v. p. yakovlev , zh . . fiz . * 51 * , 619 ( 1966 ) [ sov . jetp * 24 * , 411 ( 1967 ) ] . tesla project homepage : http://xfel.desy.de/. linear coherent light source ( lcls ) project homepage : http://www-ssrl.slac.stanford.edu/ lcls/. w. becker and h. mitter , j. phys . a * 8 * , 1638 ( 1975 ) ; h. mitter , acta phys . austr . suppl . xiv , 397 ( 1975 ) ; v. n. baier , a. i. milstein , and v. m. strakhovenko , zh . fiz . * 69 * , 1893 ( 1975 ) [ sov . jetp * 42 * , 961 ( 1976 ) ] . v. b. berestetskii , e. m. lifshitz , and l. p. pitaevskii , _ quantum electrodynamics - 2nd ed . _ ( elsevier , oxford , 1982 ) . f. mandl and g. shaw , _ quantum field theory _ ( wiley interscience , chichester , 1984 ) . h. bateman and a. erdlyi , _ higher trascendental functions , vol . 1 _ , mcgraw - hill , malabar florida , 1953 . p. papatzacos and k. mork , phys . rep . * 21 * , 81 ( 1975 ) . j. c. kieffer , proceedings of the 19th annual meeting of the ieee , 536 ( 2006 ) . t. esirkepov , _ et al . lett . * 92 * , 175003 ( 2004 ) . g. tsakiris , _ et al . _ , new j. phys . * 8 * , 19 ( 2006 ) . a. ringwald , phys . b * 510 * , 107 ( 2001 ) . a. i. milstein _ a * 73 * , 062106 ( 2006 ) . i. s. gradshteyn and i. m. ryzhik , _ tables of integrals , series and products _ ( academic press , san diego , 2000 ) .
we prove a formula that expresses the viterbo maslov index of a smooth strip in an oriented-manifold with boundary curves contained in-dimensional submanifolds in terms the degree function on the complement of the union of the two submanifolds .
1205.0535
we assume throughout this paper that @xmath2 is a connected oriented 2-manifold without boundary and @xmath3 are connected smooth one dimensional oriented submanifolds without boundary which are closed as subsets of @xmath2 and intersect transversally . we do not assume that @xmath2 is compact , but when it is , @xmath4 and @xmath5 are embedded circles . denote the standard half disc by @xmath6 let @xmath7 denote the space of all smooth maps @xmath8 satisfying the boundary conditions @xmath9 and @xmath10 . for @xmath11 let @xmath12 denote the subset of all @xmath13 satisfying the endpoint conditions @xmath14 and @xmath15 . each @xmath13 determines a locally constant function @xmath16 defined as the degree @xmath17 when @xmath18 is a regular value of @xmath19 this is the algebraic number of points in the preimage @xmath20 . the function @xmath21 depends only on the homotopy class of @xmath19 . we prove that the homotopy class of @xmath19 is uniquely determined by its endpoints @xmath22 and its degree function @xmath21 ( theorem [ thm : trace ] ) . the main theorem of this paper asserts that the viterbo maslov index of an element @xmath23 is given by the formula @xmath24 where @xmath25 denotes the sum of the four values of @xmath21 encountered when walking along a small circle surrounding @xmath26 , and similarly for @xmath27 ( theorem [ thm : maslov ] ) . the formula plays a central role in our combinatorial approach @xcite to floer homology @xcite . an appendix contains a proof that the space of paths connecting @xmath4 to @xmath5 is simply connected under suitable assumptions . * acknowledgement . * we thank david epstein for explaining to us the proof of proposition [ prop : dbae ] . define a cell complex structure on @xmath2 by taking the set of zero - cells to be the set @xmath28 , the set of one - cells to be the set of connected components of @xmath29 with compact closure , and the set of two - cells to be the set of connected components of @xmath30 with compact closure . ( there is an abuse of language here as the `` two - cells '' need not be homeomorphs of the open unit disc if the genus of @xmath2 is positive and the one - cells " need not be arcs if @xmath31 . ) define a boundary operator @xmath32 as follows . for each two - cell @xmath33 let @xmath34 where the sum is over the one - cells @xmath35 which abut @xmath33 and the plus sign is chosen iff the orientation of @xmath35 ( determined from the given orientations of @xmath4 and @xmath5 ) agrees with the boundary orientation of @xmath33 as a connected open subset of the oriented manifold @xmath2 . for each one - cell @xmath35 let @xmath36 where @xmath37 and @xmath38 are the endpoints of the arc @xmath35 and the orientation of @xmath35 goes from @xmath37 to @xmath38 . ( the one - cell @xmath35 is either a subarc of @xmath4 or a subarc of @xmath5 and both @xmath4 and @xmath5 are oriented one - manifolds . ) for @xmath39 a * @xmath40-chain * is defined to be a formal linear combination ( with integer coefficients ) of @xmath40-cells , i.e. a two - chain is a locally constant map @xmath41 ( whose support has compact closure in @xmath2 ) and a one - chain is a locally constant map @xmath42 ( whose support has compact closure in @xmath43 ) . it follows directly from the definitions that @xmath44 for each two - cell @xmath33 . each @xmath13 determines a two - chain @xmath21 via @xmath45 and a one - chain @xmath46 via @xmath47 here we orient the one - manifolds @xmath48 and @xmath49 from @xmath50 to @xmath51 . for any one - chain @xmath52 denote @xmath53 conversely , given locally constant functions @xmath54 and @xmath55 , denote by @xmath56 the one - chain that agrees with @xmath57 on @xmath58 and agrees with @xmath59 on @xmath60 . [ def : trace]fix two ( not necessarily distinct ) intersection points @xmath11 . * ( i ) * let @xmath61 be a two - chain . the triple @xmath62 is called an * @xmath63-trace * if there exists an element @xmath64 such that @xmath21 is given by . in this case @xmath65 is also called the * @xmath63-trace of @xmath19 * and we sometimes write @xmath66 . * ( ii ) * let @xmath67 be an @xmath63-trace . the triple @xmath68 is called the * boundary of @xmath69 . * * ( iii ) * a one - chain @xmath52 is called an * @xmath70-trace * if there exist smooth curves @xmath71\to\alpha$ ] and @xmath72\to\beta$ ] such that @xmath73 , @xmath74 , @xmath75 and @xmath76 are homotopic in @xmath2 with fixed endpoints , and @xmath77 [ rmk : nu]assume @xmath2 is simply connected . then the condition on @xmath75 and @xmath76 to be homotopic with fixed endpoints is redundant . moreover , if @xmath78 then a one - chain @xmath46 is an @xmath70-trace if and only if the restrictions @xmath79 and @xmath80 are constant . if @xmath81 and @xmath82 are embedded circles and @xmath83 denote the positively oriented arcs from @xmath26 to @xmath27 in @xmath82 , then a one - chain @xmath46 is an @xmath70-trace if and only if @xmath84 and @xmath85 in particular , when walking along @xmath4 or @xmath5 , the function @xmath46 only changes its value at @xmath26 and @xmath27 . [ le : boundary ] let @xmath11 and @xmath64 . then the boundary of the @xmath63-trace @xmath86 of @xmath19 is the triple @xmath87 where @xmath46 is given by . in other words , if @xmath21 is given by and @xmath46 is given by then @xmath88 . choose an embedding @xmath89\to\sigma$ ] such that @xmath19 is transverse to @xmath90 , @xmath91 for @xmath92 , @xmath93 , @xmath94 are regular values of @xmath19 , @xmath95 is a regular value of @xmath96 , and @xmath90 intersects @xmath4 transversally at @xmath97 such that orientations match in @xmath98 denote @xmath99)$ ] . then @xmath100 is a @xmath1-dimensional submanifold with boundary @xmath101 if @xmath102 then @xmath103 we orient @xmath104 such that the orientations match in @xmath105 in other words , if @xmath102 and @xmath106 , then a nonzero tangent vector @xmath107 is positive if and only if the pair @xmath108 is a positive basis of @xmath109 . then the boundary orientation of @xmath104 at the elements of @xmath110 agrees with the algebraic count in the definition of @xmath111 , at the elements of @xmath112 is opposite to the algebraic count in the definition of @xmath113 , and at the elements of @xmath114 is opposite to the algebraic count in the definition of @xmath115 . hence @xmath116 in other words the value of @xmath46 at a point in @xmath58 is equal to the value of @xmath21 slightly to the left of @xmath4 minus the value of @xmath21 slightly to the right of @xmath4 . likewise , the value of @xmath46 at a point in @xmath60 is equal to the value of @xmath21 slightly to the right of @xmath5 minus the value of @xmath21 slightly to the left of @xmath5 . this proves lemma [ le : boundary ] . [ thm : trace ] * ( i ) * two elements of @xmath7 belong to the same connected component of @xmath7 if and only if they have the same @xmath63-trace . * ( ii ) * assume @xmath2 is diffeomorphic to the two - sphere . then @xmath67 is an @xmath63-trace if and only if @xmath117 is an @xmath70-trace . * ( iii ) * assume @xmath2 is not diffeomorphic to the two - sphere and let @xmath11 . if @xmath46 is an @xmath70-trace , then there is a unique two - chain @xmath21 such that @xmath118 is an @xmath63-trace and @xmath119 . we prove ( i ) . `` only if '' follows from the standard arguments in degree theory as in milnor @xcite . to prove `` if '' , fix two intersection points @xmath120 and , for @xmath121 , denote by @xmath122 the space of all smooth curves @xmath123\to x}$ ] satisfying @xmath124 and @xmath125 . every @xmath64 determines smooth paths @xmath126 and @xmath127 via @xmath128 these paths are homotopic in @xmath2 with fixed endpoints . an explicit homotopy is the map @xmath129 ^ 2\to\sigma\ ] ] where @xmath130 ^ 2\to{{\mathbb d}}$ ] is the map @xmath131 by lemma [ le : boundary ] , he homotopy class of @xmath132 in @xmath133 is uniquely determined by @xmath134 and that of @xmath135 in @xmath136 is uniquely determined by @xmath137 . hence they are both uniquely determined by the @xmath63-trace of @xmath19 . if @xmath2 is not diffeomorphic to the @xmath0-sphere the assertion follows from the fact that each component of @xmath138 is contractible ( because the universal cover of @xmath2 is diffeomorphic to the complex plane ) . now assume @xmath2 is diffeomorphic to the @xmath0-sphere . then @xmath139 acts on @xmath140 because the correspondence @xmath141 identifies @xmath140 with a space of homotopy classes of paths in @xmath138 connecting @xmath133 to @xmath136 . the induced action on the space of two - chains @xmath142 is given by adding a global constant . hence the map @xmath143 induces an injective map @xmath144 this proves ( i ) . we prove ( ii ) and ( iii ) . let @xmath21 be a two - chain , suppose that @xmath145 is an @xmath70-trace , and denote @xmath146 let @xmath147\to\alpha}$ ] and @xmath148\to\beta}$ ] be as in definition [ def : trace ] . then there is a @xmath149 such that the map @xmath150 is homotopic to @xmath75 and @xmath151 is homotopic to @xmath76 . by definition the @xmath63-trace of @xmath152 is @xmath153 for some two - chain @xmath154 . by lemma [ le : boundary ] , we have @xmath155 and hence @xmath156 is constant . if @xmath2 is not diffeomorphic to the two - sphere and @xmath69 is the @xmath63-trace of some element @xmath13 , then @xmath19 is homotopic to @xmath152 ( as @xmath138 is simply connected ) and hence @xmath157 and @xmath158 . if @xmath2 is diffeomorphic to the @xmath0-sphere choose a smooth map @xmath159 of degree @xmath160 and replace @xmath152 by the connected sum @xmath161 . then @xmath69 is the @xmath63-trace of @xmath19 . this proves theorem [ thm : trace ] . [ rmk : trace]let @xmath67 be an @xmath63-trace and define @xmath162 * ( i ) * the two - chain @xmath21 is uniquely determined by the condition @xmath163 and its value at one point . to see this , think of the embedded circles @xmath4 and @xmath5 as traintracks . crossing @xmath4 at a point @xmath164 increases @xmath21 by @xmath165 if the train comes from the left , and decreases it by @xmath165 if the train comes from the right . crossing @xmath5 at a point @xmath166 decreases @xmath21 by @xmath167 if the train comes from the left and increases it by @xmath167 if the train comes from the right . moreover , @xmath57 extends continuously to @xmath168 and @xmath169 extends continuously to @xmath170 . at each intersection point @xmath171 with intersection index @xmath51 ( respectively @xmath50 ) the function @xmath21 takes the values @xmath172 as we march counterclockwise ( respectively clockwise ) along a small circle surrounding the intersection point . * ( ii ) * if @xmath2 is not diffeomorphic to the @xmath0-sphere then , by theorem [ thm : trace ] ( iii ) , the @xmath63-trace @xmath69 is uniquely determined by its boundary @xmath173 . * ( iii ) * assume @xmath2 is not diffeomorphic to the @xmath0-sphere and choose a universal covering @xmath174 . choose a point @xmath175 and lifts @xmath176 and @xmath177 of @xmath4 and @xmath5 such that @xmath178 then @xmath69 lifts to an @xmath179-trace @xmath180 more precisely , the one chain @xmath181 is an @xmath70-trace , by lemma [ le : boundary ] . the paths @xmath71\to\alpha$ ] and @xmath72\to\beta$ ] in definition [ def : trace ] lift to unique paths @xmath182\to{{\tilde{\alpha}}}$ ] and @xmath183\to{{\tilde{\beta}}}$ ] connecting @xmath184 to @xmath185 . for @xmath186 the number @xmath187 is the winding number of the loop @xmath188 about @xmath189 ( by rouch s theorem ) . the two - chain @xmath21 is then given by @xmath190 to see this , lift an element @xmath64 with @xmath63-trace @xmath69 to the universal cover to obtain an element @xmath191 with @xmath192 and consider the degree . [ def : catenation]let @xmath193 . the * catenation of two @xmath63-traces @xmath67 and @xmath194 * is defined by @xmath195 let @xmath64 and @xmath196 and suppose that @xmath19 and @xmath152 are constant near the ends @xmath197 . for @xmath198 sufficiently close to one the * @xmath199-catenation of @xmath19 and @xmath152 * is the map @xmath200 defined by @xmath201 [ le : catenation ] if @xmath64 and @xmath196 are as in definition [ def : catenation ] then @xmath202 thus the catenation of two @xmath63-traces is again an @xmath63-trace . this follows directly from the definitions . [ def : maslov]let @xmath11 and @xmath64 . choose an orientation preserving trivialization @xmath203 and consider the lagrangian paths @xmath204\to{{\mathbb{r}\mathrm{p}}}^1\ ] ] given by @xmath205 the * viterbo maslov index of @xmath19 * is defined as the relative maslov index of the pair of lagrangian paths @xmath206 and will be denoted by @xmath207 by the naturality and homotopy axioms for the relative maslov index ( see for example @xcite ) , the number @xmath208 is independent of the choice of the trivialization and depends only on the homotopy class of @xmath19 ; hence it depends only on the @xmath63-trace of @xmath19 , by theorem [ thm : trace ] . the relative maslov index @xmath209 is the degree of the loop in @xmath210 obtained by traversing @xmath211 , followed by a counterclockwise turn from @xmath212 to @xmath213 , followed by traversing @xmath214 in reverse time , followed by a clockwise turn from @xmath215 to @xmath216 . this index was first defined by viterbo @xcite ( in all dimensions ) . another exposition is contained in @xcite . [ rmk : mascat]the viterbo maslov index is additive under catenation , i.e. if @xmath217 are @xmath63-traces then @xmath218 for a proof of this formula see @xcite . [ def : arc]let @xmath67 be an @xmath63-trace and @xmath162 @xmath69 is said to satisfy the * arc condition * if @xmath219 when @xmath69 satisfies the arc condition there are arcs @xmath220 and @xmath221 from @xmath26 to @xmath27 such that @xmath222 here the plus sign is chosen iff the orientation of @xmath223 from @xmath26 to @xmath27 agrees with that of @xmath4 , respectively the orientation of @xmath224 from @xmath26 to @xmath27 agrees with that of @xmath5 . in this situation the quadruple @xmath225 and the triple @xmath226 determine one another and we also write @xmath227 for the boundary of @xmath69 . when @xmath13 and @xmath228 satisfies the arc condition and @xmath229 then @xmath230 is homotopic in @xmath4 to a path traversing @xmath223 and the path @xmath231 is homotopic in @xmath5 to a path traversing @xmath224 . [ thm : maslov ] let @xmath67 be an @xmath63-trace . for @xmath232 denote by @xmath233 the sum of the four values of @xmath21 encountered when walking along a small circle surrounding @xmath18 . then the viterbo maslov index of @xmath69 is given by @xmath234 we first prove the result for the @xmath0-plane and the @xmath0-sphere ( section [ sec : plane ] ) . when @xmath2 is not simply connected we reduce the result to the case of the @xmath0-plane ( section [ sec : lift ] ) . the key is the identity @xmath235 for every lift @xmath236 to the universal cover and every deck transformation @xmath237 . a connected oriented @xmath0-manifold @xmath2 is called * planar * if it admits an ( orientation preserving ) embedding into the complex plane . [ prop : maslovc ] equation holds when @xmath2 is planar . assume first that @xmath238 and @xmath62 satisfies the arc condition . thus the boundary of @xmath69 has the form @xmath239 where @xmath220 and @xmath221 are arcs from @xmath26 to @xmath27 and @xmath240 is the winding number of the loop @xmath241 about the point @xmath242 ( see remark [ rmk : trace ] ) . hence the formula can be written in the form @xmath243 here @xmath244 denotes the intersection index of @xmath223 and @xmath224 at a point @xmath245 , @xmath246 denotes the value of the winding number @xmath21 at a point in @xmath247 close to @xmath26 , and @xmath248 denotes the value of @xmath21 at a point in @xmath247 close to @xmath27 . we now prove under the assumption that @xmath69 satisfies the arc condition . the proof is by induction on the number of intersection points of @xmath224 and @xmath4 and has seven steps . * step 1 . * _ we may assume without loss of generality that @xmath249,\qquad x < y,\ ] ] and @xmath250 is an embedded arc from @xmath26 to @xmath27 that is transverse to @xmath251 . _ choose a diffeomorphism from @xmath2 to @xmath252 that maps @xmath223 to a bounded closed interval and maps @xmath26 to the left endpoint of @xmath223 . if @xmath4 is not compact the diffeomorphism can be chosen such that it also maps @xmath4 to @xmath251 . if @xmath4 is an embedded circle the diffeomorphism can be chosen such that its restriction to @xmath224 is transverse to @xmath251 ; now replace the image of @xmath4 by @xmath251 . this proves step 1 . * step 2 . * _ assume and let @xmath253 be the @xmath254-trace obtained from @xmath69 by complex conjugation . then @xmath69 satisfies if and only if @xmath255 satisfies . _ step 2 follows from the fact that the numbers @xmath256 change sign under complex conjugation . * step 3 . * _ assume . if @xmath257 then @xmath69 satisfies . _ in this case @xmath224 is contained in the upper or lower closed half plane and the loop @xmath258 bounds a disc contained in the same half plane . by step 1 we may assume that @xmath224 is contained in the upper half space . then @xmath259 , @xmath260 , and @xmath261 . moreover , the winding number @xmath21 is one in the disc encircled by @xmath223 and @xmath224 and is zero in the complement of its closure . since the intervals @xmath262 and @xmath263 are contained in this complement , we have @xmath264 . this proves step 3 . * step 4 . * _ assume and @xmath265 , follow the arc of @xmath224 , starting at @xmath26 , and let @xmath266 be the next intersection point with @xmath251 . assume @xmath267 , denote by @xmath268 the arc in @xmath224 from @xmath266 to @xmath27 , and let @xmath269 $ ] ( see figure [ fig : maslov4 ] ) . if the @xmath63-trace @xmath270 with boundary @xmath271 satisfies so does @xmath69 . _ . ] by step 2 we may assume @xmath272 . orient @xmath224 from @xmath26 to @xmath27 . the viterbo maslov index of @xmath69 is minus the maslov index of the path @xmath273 relative to the lagrangian subspace @xmath274 . since the maslov index of the arc in @xmath224 from @xmath26 to @xmath266 is @xmath51 we have @xmath275 since the orientations of @xmath276 and @xmath268 agree with those of @xmath223 and @xmath224 we have @xmath277 now let @xmath278 be the intersection points of @xmath251 and @xmath224 in the interval @xmath279 and let @xmath280 be the intersection index of @xmath251 and @xmath224 at @xmath281 . then there is an integer @xmath282 such that @xmath283 and @xmath284 . moreover , the winding number @xmath21 slightly to the left of @xmath26 is @xmath285 it agrees with the value of @xmath21 slightly to the right of @xmath286 . hence @xmath287 it follows from equation for @xmath270 and equations , , and that @xmath288 this proves step 4 . * step 5 . * _ assume and @xmath265 , follow the arc of @xmath224 , starting at @xmath26 , and let @xmath266 be the next intersection point with @xmath251 . assume @xmath289 , denote by @xmath268 the arc in @xmath224 from @xmath266 to @xmath27 , and let @xmath269 $ ] ( see figure [ fig : maslov5 ] ) . if the @xmath63-trace @xmath270 with boundary @xmath271 satisfies so does @xmath69 . ] by step 2 we may assume @xmath272 . since the maslov index of the arc in @xmath224 from @xmath26 to @xmath266 is @xmath50 , we have @xmath290 since the orientations of @xmath276 and @xmath268 agree with those of @xmath223 and @xmath224 we have @xmath291 now let @xmath292 be the intersection points of @xmath251 and @xmath224 in the interval @xmath293 and let @xmath280 be the intersection index of @xmath251 and @xmath224 at @xmath281 . since the value of @xmath21 slightly to the left of @xmath266 agrees with the value of @xmath21 slightly to the right of @xmath26 we have @xmath294 since @xmath295 is the sum of the intersection indices of @xmath251 and @xmath268 at all points to the left of @xmath266 we obtain @xmath296 it follows from equation for @xmath270 and equations , , and that @xmath297 this proves step 5 . * step 6 . * _ assume and @xmath265 , follow the arc of @xmath224 , starting at @xmath26 , and let @xmath298 be the next intersection point with @xmath251 . assume @xmath299 . denote by @xmath268 the arc in @xmath224 from @xmath27 to @xmath298 , and let @xmath300 $ ] ( see figure [ fig : maslov6 ] ) . if the @xmath63-trace @xmath270 with boundary @xmath301 satisfies so does @xmath69 . ] by step 2 we may assume @xmath272 . since the orientation of @xmath268 from @xmath27 to @xmath298 is opposite to the orientation of @xmath224 and the maslov index of the arc in @xmath224 from @xmath26 to @xmath298 is @xmath50 , we have @xmath302 using again the fact that the orientation of @xmath268 is opposite to the orientation of @xmath224 we have @xmath303 now let @xmath304 be all intersection points of @xmath251 and @xmath224 and let @xmath280 be the intersection index of @xmath251 and @xmath224 at @xmath281 . choose @xmath305 such that @xmath306 then @xmath307 and @xmath308 for @xmath309 the intersection index of @xmath251 and @xmath268 at @xmath281 is @xmath310 . moreover , @xmath311 is the sum of the intersection indices of @xmath251 and @xmath268 at all points to the left of @xmath27 and @xmath312 is minus the sum of the intersection indices of @xmath251 and @xmath268 at all points to the right of @xmath298 . hence @xmath313 we claim that @xmath314 to see this , note that the value of the winding number @xmath21 slightly to the left of @xmath26 agrees with the value of @xmath21 slightly to the right of @xmath298 , and hence @xmath315 this proves the first equation in . to prove the second equation in we observe that @xmath316 and hence @xmath317 this proves the second equation in . it follows from equation for @xmath270 and equations , , and that @xmath318 here the first equality follows from , the second equality follows from for @xmath270 , the third equality follows from , and the fourth equality follows from . this proves step 6 . * step 7 . * _ equation holds when @xmath238 and @xmath69 satisfies the arc condition . _ it follows from steps 3 - 6 by induction that equation holds for every @xmath63-trace @xmath67 whose boundary @xmath319 satisfies . hence step 7 follows from step 1 . next we drop the assumption that @xmath69 satisfies the arc condition and extend the result to planar surfaces . this requires a further three steps . * step 8 . * _ equation holds when @xmath238 and @xmath78 . _ under these assumptions @xmath320 and @xmath321 are constant . there are four cases . * case 1 . * _ @xmath4 is an embedded circle and @xmath5 is not an embedded circle . _ in this case we have @xmath322 and @xmath323 . moroeover , @xmath4 is the boundary of a unique disc @xmath324 and we assume that @xmath4 is oriented as the boundary of @xmath324 . then the path @xmath71\to\sigma$ ] in definition [ def : trace ] satisfies @xmath325 and is homotopic to @xmath326 . hence @xmath327 here the last equation follows from the fact that @xmath69 can be obtained as the catenation of @xmath57 copies of the disc @xmath324 . * case 2 . * _ @xmath4 is not an embedded circle and @xmath5 is an embedded circle . _ this follows from case 1 by interchanging @xmath4 and @xmath5 . * case 3 . * _ @xmath4 and @xmath5 are embedded circles . _ in this case there is a unique pair of embedded discs @xmath324 and @xmath328 with boundaries @xmath4 and @xmath5 , respectively . orient @xmath4 and @xmath5 as the boundaries of these discs . then , for every @xmath329 , we have @xmath330 hence @xmath331 here the last equation follows from the fact @xmath69 can be obtained as the catenation of @xmath57 copies of the disc @xmath324 ( with the orientation inherited from @xmath2 ) and @xmath169 copies of @xmath332 ( with the opposite orientation ) . * case 4 . * _ neither @xmath4 nor @xmath5 is an embedded circle . _ under this assumption we have @xmath333 . hence it follows from theorem [ thm : trace ] that @xmath334 and @xmath335 for the constant map @xmath336 . thus @xmath337 this proves step 8 . * step 9 . * _ equation holds when @xmath238 . _ by step 8 , it suffices to assume @xmath81 . it follows from theorem [ thm : trace ] that every @xmath64 is homotopic to a catentation @xmath338 , where @xmath339 satisfies the arc condition and @xmath340 . hence it follows from steps 7 and 8 that @xmath341 here the last equation follows from the fact that @xmath342 and hence @xmath343 for every @xmath232 . this proves step 9 . * step 10 . * _ equation holds when @xmath2 is planar . _ choose an element @xmath64 such that @xmath344 . modifying @xmath4 and @xmath5 on the complement of @xmath345 , if necessary , we may assume without loss of generality that @xmath4 and @xmath5 are mebedded circles . let @xmath346 be an orientation preserving embedding . then @xmath347 is an @xmath348-trace in @xmath252 and hence satisfies by step 9 . since @xmath349 , @xmath350 , and @xmath351 it follows that @xmath69 also satisfies . this proves step 10 and proposition [ prop : maslovc ] [ rmk : maslov]let @xmath352 be an @xmath63-trace in @xmath252 as in step 1 in the proof of theorem [ thm : maslov ] . thus @xmath353 are real numbers , @xmath223 is the interval @xmath354 $ ] , and @xmath224 is an embedded arc with endpoints @xmath22 which is oriented from @xmath26 to @xmath27 and is transverse to @xmath251 . thus @xmath355 is a finite set . define a map @xmath356 as follows . given @xmath357 walk along @xmath224 towards @xmath27 and let @xmath358 be the next intersection point with @xmath251 . this map is bijective . now let @xmath359 be any of the three open intervals @xmath279 , @xmath70 , @xmath360 . any arc in @xmath224 from @xmath18 to @xmath358 with both endpoints in the same interval @xmath359 can be removed by an isotopy of @xmath224 which does not pass through @xmath22 . call @xmath69 a * reduced @xmath63-trace * if @xmath361 implies @xmath362 for each of the three intervals . then every @xmath63-trace is isotopic to a reduced @xmath363-trace and the isotopy does not effect the numbers @xmath256 . -traces in @xmath252 . ] let @xmath364 ( respectively @xmath365 ) denote the set of all points @xmath366 where the positive tangent vectors in @xmath367 point up ( respectively down ) . one can prove that every reduced @xmath63-trace satisfies one of the following conditions . * case 1 : * if @xmath368 then @xmath369 . @xmath370 $ ] . * case 3 : * if @xmath371 then @xmath369 . @xmath372 $ ] . ( examples with @xmath259 and @xmath260 are depicted in figure [ fig : maslov ] . ) one can then show directly that the reduced @xmath63-traces satisfy equation . this gives rise to an alternative proof of proposition [ prop : maslovc ] via case distinction . if @xmath2 is diffeomorphic to the @xmath0-plane the result has been established in proposition [ prop : maslovc ] . hence assume @xmath373 let @xmath64 . if @xmath19 is not surjective the assertion follows from the case of the complex plane ( proposition [ prop : maslovc ] ) via stereographic projection . hence assume @xmath19 is surjective and choose a regular value @xmath374 of @xmath19 . denote @xmath375 for @xmath376 let @xmath377 according to whether or not the differential @xmath378 is orientation preserving . choose an open disc @xmath379 centered at @xmath18 such that @xmath380 and @xmath381 is a union of open neighborhoods @xmath382 of @xmath383 with disjoint closures such that @xmath384 is a diffeomorphism for each @xmath385 which extends to a neighborhood of @xmath386 . now choose a continuous map @xmath387 which agrees with @xmath19 on @xmath388 and restricts to a diffeomorphism from @xmath386 to @xmath389 for each @xmath385 . then @xmath18 does not belong to the image of @xmath152 and hence equation holds for @xmath152 ( after smoothing along the boundaries @xmath390 ) . moreover , the diffeomorphism @xmath391 is orientation preserving if and only if @xmath392 . hence @xmath393 by proposition [ prop : maslovc ] equation holds for @xmath394 and hence it also holds for @xmath86 . this proves theorem [ thm : maslov ] when @xmath2 is simply connected . the key step for extending proposition [ prop : maslovc ] to non - simply connected two - manifolds is the next result about lifts to the universal cover . [ prop : gm ] suppose @xmath2 is not diffeomorphic to the @xmath0-sphere . let @xmath395 be an @xmath63-trace and @xmath396 be a universal covering . denote by @xmath397 the group of deck transformations . choose an element @xmath398 and let @xmath176 and @xmath177 be the lifts of @xmath4 and @xmath5 through @xmath184 . let @xmath399 be the lift of @xmath69 with left endpoint @xmath184 . then @xmath400 for every @xmath401 . [ le : gm ] suppose @xmath2 is not diffeomorphic to the @xmath0-sphere . let @xmath69 , @xmath402 , @xmath403 , @xmath236 be as in proposition [ prop : gm ] . if @xmath404 for every @xmath401 then equation holds for every @xmath401 . if does not hold then there is a deck transformation @xmath405 such that @xmath406 since there can only be finitely many such @xmath405 , there is an integer @xmath407 such that @xmath408 and @xmath409 for every integer @xmath410 . define @xmath411 . then @xmath412 and @xmath413 for every integer @xmath414 define @xmath415 then @xmath416 is diffeomorphic to the annulus . let @xmath417 be the obvious projection , define @xmath418 , @xmath419 , and let @xmath420 be the @xmath421-trace in @xmath416 with @xmath422 , @xmath423 , and @xmath424 then @xmath425 by proposition [ prop : maslovc ] both @xmath236 and @xmath426 satisfy equation and they have the same viterbo maslov index . hence @xmath427 here the last equation follows from . this contradicts and proves lemma [ le : gm ] . [ le : ab ] suppose @xmath2 is not diffeomorphic to the @xmath0-sphere . let @xmath69 , @xmath402 , @xmath403 , @xmath236 be as in proposition [ prop : gm ] and denote @xmath428 and @xmath429 . choose smooth paths @xmath430\to{{\tilde{\alpha}}},\qquad \gamma_{{\tilde{\beta}}}:[0,1]\to{{\tilde{\beta}}}\ ] ] from @xmath431 to @xmath432 such that @xmath433 is an immersion when @xmath434 and constant when @xmath435 , the same holds for @xmath436 , and @xmath437 define @xmath438),\qquad { { \tilde{b}}}:= \gamma_{{\tilde{\beta}}}([0,1]).\ ] ] then , for every @xmath439 , we have @xmath440 @xmath441 @xmath442 the same holds with @xmath443 replaced by @xmath444 . if @xmath4 is a contractible embedded circle or not an embedded circle at all we have @xmath445 whenever @xmath237 and this implies , and . hence assume @xmath4 is a noncontractible embedded circle . then we may also assume , without loss of generality , that @xmath446 , the map @xmath447 is a deck transformation , @xmath402 maps the interval @xmath448 bijectively onto @xmath4 , and @xmath449 with @xmath450 . thus @xmath451 $ ] and , for every @xmath452 , @xmath453 \quad\iff\quad 0\le k\le { { \tilde{y}}}-{{\tilde{x}}}\quad\iff\quad { { \tilde{y}}}-k\in[{{\tilde{x}}},{{\tilde{y}}}].\ ] ] similarly , we have @xmath454 \quad\iff\quad [ { { \tilde{x}}}+k,{{\tilde{y}}}+k]\cap[{{\tilde{x}}},{{\tilde{y}}}]=\emptyset\ ] ] and @xmath455 \quad\iff\quad [ { { \tilde{x}}}+k,{{\tilde{y}}}+k]\subset[{{\tilde{x}}},{{\tilde{y } } } ] \quad\iff\quad k=0.\ ] ] this proves , , and for the deck transformation @xmath456 . if @xmath457 is any other deck transformation , then we have @xmath458 and so , , and are trivially satisfied . this proves lemma [ le : ab ] . [ le : wg ] suppose @xmath2 is not diffeomorphic to the @xmath0-sphere . let @xmath69 , @xmath402 , @xmath403 , @xmath236 be as in proposition [ prop : gm ] , and let @xmath459 be as in lemma [ le : ab ] . then the following holds . * ( i ) * equation holds for every @xmath439 that satisfies @xmath460 . * ( ii ) * if @xmath69 satisfies the arc condition then holds for every @xmath401 . we prove ( i ) . let @xmath439 such that @xmath460 and let @xmath461 be as in lemma [ le : ab ] . then @xmath187 is the winding number of the loop @xmath462 about the point @xmath186 . moreover , the paths @xmath463\to{{\mathbb c}},\qquad g\gamma_{{\tilde{\beta}}}:[0,1]\to{{\mathbb c}}\ ] ] connect the points @xmath464 . hence @xmath465 similarly with @xmath457 replaced by @xmath466 . moreover , it follows from lemma [ le : ab ] , that @xmath467 hence @xmath468 here we have used the fact that every @xmath439 is an orientation preserving diffeomorphism of @xmath252 . thus we have proved that @xmath469 since @xmath460 , we have @xmath470 and the same identities hold with @xmath457 replaced by @xmath466 . this proves ( i ) . we prove ( ii ) . if @xmath69 satisfies the arc condition then @xmath471 and @xmath472 for every @xmath401 . in particular , for every @xmath401 , we have @xmath460 and hence holds by ( i ) . hence it follows from lemma [ le : gm ] that holds for every @xmath401 . this proves lemma [ le : wg ] . the next lemma deals with @xmath63-traces connecting a point @xmath473 to itself . an example on the annulus is depicted in figure [ fig : annulus3 ] . [ le : zero ] suppose @xmath2 is not diffeomorphic to the @xmath0-sphere . let @xmath69 , @xmath402 , @xmath403 , @xmath236 be as in proposition [ prop : gm ] . suppose that there is a deck transformation @xmath474 such that @xmath475 . then @xmath69 has viterbo maslov index zero and @xmath476 for every @xmath477 . -trace on the annulus with @xmath78 . ] by assumption , we have @xmath478 and @xmath479 . hence @xmath4 and @xmath5 are noncontractible embedded circles and some iterate of @xmath4 is homotopic to some iterate of @xmath5 . hence , by lemma [ le : dbae3 ] , @xmath4 must be homotopic to @xmath5 ( with some orientation ) . hence we may assume , without loss of generality , that @xmath446 , the map @xmath447 is a deck transformation , @xmath402 maps the interval @xmath448 bijectively onto @xmath4 , @xmath480 , @xmath481 , @xmath482 , and that @xmath483 is an integer . then @xmath484 is the translation @xmath485 let @xmath486\subset{{\tilde{\alpha}}}$ ] and let @xmath487 be the arc connecting @xmath488 to @xmath489 . then , for @xmath186 , the integer @xmath187 is the winding number of @xmath490 about @xmath189 . define the projection @xmath491 by @xmath492 denote @xmath493 and @xmath494 , and let @xmath495 be the induced @xmath421-trace in @xmath252 with @xmath496 then @xmath497 and @xmath498 are embedded circles and have the winding number @xmath489 about zero . hence it follows from step 8 , case 3 in the proof of proposition [ prop : maslovc ] that @xmath426 has viterbo maslov index zero and satisfies @xmath499 . hence @xmath236 also has viterbo maslov index zero . it remains to prove that @xmath476 for every @xmath477 . to see this we use the fact that the embedded loops @xmath4 and @xmath5 are homotopic with fixed endpoint @xmath26 hence , by a theorem of epstein , they are isotopic with fixed basepoint @xmath26 ( see ( * ? ? ? * theorem 4.1 ) ) . thus there exists a smooth map @xmath500\to\sigma$ ] such that @xmath501 for all @xmath502 and @xmath503 $ ] , and the map @xmath504 is an embedding for every @xmath505 $ ] . lift this homotopy to the universal cover to obtain a map @xmath506\to{{\mathbb c}}$ ] such that @xmath507 and @xmath508,\quad { { \tilde{f}}}(s,1)\in{{\tilde{b}}}_1,\quad { { \tilde{f}}}(0,t)={{\tilde{x}}},\quad { { \tilde{f}}}(s+1,t)={{\tilde{f}}}(s , t)+1\ ] ] for all @xmath509 and @xmath503 $ ] . here @xmath510 denotes the arc in @xmath444 from @xmath488 to @xmath1 . since the map @xmath504 is injective for every @xmath511 , we have @xmath512\times[0,1])\ ] ] for every every @xmath439 . now choose a smooth map @xmath513 with @xmath192 ( see theorem [ thm : trace ] ) . define the homotopy @xmath514\times[0,1]\to{{\mathbb c}}$ ] by @xmath515 . then , by theorem [ thm : trace ] , @xmath516 is homotopic to @xmath517\times[0,1]}$ ] subject to the boundary conditions @xmath518 , @xmath519 , @xmath520 , @xmath521 . hence , for every @xmath522 , we have @xmath523 in particular , choosing @xmath189 near @xmath524 , we find @xmath525 for every @xmath439 that is not one of the translations @xmath456 for @xmath526 . this proves the assertion in the case @xmath527 . if @xmath528 it remains to prove @xmath529 for @xmath530 . to see this , let @xmath531 $ ] , @xmath510 be the arc from @xmath488 to @xmath1 , @xmath532 be the winding number of @xmath533 about @xmath534 , and define @xmath535 then , by what we have already proved , the @xmath179-trace @xmath536 satisfies @xmath537 for every @xmath439 other than the translations by @xmath488 or @xmath1 . in particular , we have @xmath538 for every @xmath539 and also @xmath540 . since @xmath541 for @xmath186 , we obtain @xmath542 for every @xmath543 . this proves lemma [ le : zero ] . the next example shows that lemma [ le : wg ] can not be strengthened to assert the identity @xmath476 for every @xmath439 with @xmath460 . [ ex : annulus1]figure [ fig : annulus1 ] depicts an @xmath63-trace @xmath67 on the annulus @xmath544 that has viterbo maslov index one and satisfies the arc condition . the lift satisfies @xmath545 , @xmath546 , @xmath547 , and @xmath548 . thus @xmath549 . -trace on the annulus satisfying the arc condition . ] the proof has five steps . * step 1 . * _ let @xmath459 be as in lemma [ le : ab ] and let @xmath439 such that @xmath550 ( an example is depicted in figure [ fig : torus ] . ) then holds . _ -trace on the torus not satisfying the arc condition . ] the proof is a refinement of the winding number comparison argument in lemma [ le : wg ] . since @xmath551 we have @xmath237 and , since @xmath552 , it follows that @xmath4 is a noncontractible embedded circle . hence we may choose the universal covering @xmath174 and the lifts @xmath176 , @xmath177 , @xmath236 such that @xmath446 , the map @xmath447 is a deck transformation , the projection @xmath402 maps the interval @xmath448 bijectively onto @xmath4 , and @xmath553 by assumption and lemma [ le : ab ] there is an integer @xmath40 such that @xmath554 thus @xmath457 is the deck transformation @xmath456 . since @xmath551 and @xmath555 it follows from lemma [ le : ab ] that @xmath556 and @xmath557 and hence , again by lemma [ le : ab ] , we have @xmath558 with @xmath433 and @xmath436 chosen as in lemma [ le : ab ] , this implies @xmath559 since @xmath560 , there exists a constant @xmath561 such that @xmath562 the paths @xmath563 and @xmath564 both connect the point @xmath565 to @xmath566 . likewise , the paths @xmath567 and @xmath568 both connect the point @xmath569 to @xmath570 . hence @xmath571 here the last but one equation follows from . thus we have proved @xmath572 since @xmath573 step 1 follows by taking the sum of the two equations in . * step 2 . * _ let @xmath459 be as in lemma [ le : ab ] and let @xmath439 . suppose that either @xmath574 or @xmath575 . then holds . _ if @xmath460 the assertion follows from lemma [ le : wg ] . if @xmath576 and @xmath577 the assertion follows from step 1 . if @xmath578 and @xmath579 the assertion follows from step 1 by interchanging @xmath184 and @xmath185 . namely , holds for @xmath236 if and only if it holds for the @xmath179-trace @xmath580 this covers the case @xmath575 . if @xmath574 the assertion follows by interchanging @xmath443 and @xmath444 . namely , holds for @xmath236 if and only if it holds for the @xmath581-trace @xmath582 this proves step 2 . * * _ let @xmath459 be as in lemma [ le : ab ] and let @xmath439 such that @xmath583 ( an example is depicted in figure [ fig : annulus2 ] . ) then holds for @xmath457 and @xmath466 . _ -trace on the annulus with @xmath584 and @xmath585 . ] since @xmath551 ( and @xmath586 ) we have @xmath237 and , since @xmath552 and @xmath587 , it follows that @xmath588 and @xmath589 . hence @xmath4 and @xmath5 are noncontractible embedded circles and some iterate of @xmath4 is homotopic to some iterate of @xmath5 . so @xmath4 is homotopic to @xmath5 ( with some orientation ) , by lemma [ le : dbae3 ] . hence we may choose the universal covering @xmath174 and the lifts @xmath590 such that @xmath446 , the map @xmath447 is a deck transformation , @xmath402 maps the interval @xmath448 bijectively onto @xmath4 , and @xmath591 , @xmath481 , @xmath592 . thus @xmath593 $ ] is the arc in @xmath176 from @xmath488 to @xmath185 and @xmath444 is the arc in @xmath177 from @xmath488 to @xmath185 . moreover , @xmath482 and the arc in @xmath177 from @xmath488 to @xmath1 is a fundamental domain for @xmath5 . by assumption and lemma [ le : ab ] there is an integer @xmath40 such that @xmath594 and @xmath595 . hence @xmath443 does not contain any negative integers and @xmath444 does not contain any positive integers . choose @xmath596 such that @xmath597 for @xmath598 let @xmath599 and @xmath600 be the arcs from @xmath488 to @xmath601 and consider the @xmath179-trace @xmath602 where @xmath603 is the winding number of @xmath604 about @xmath605 . note that @xmath606 and @xmath607 we prove that , for each @xmath40 , the @xmath179-trace @xmath608 satisfies @xmath609 if @xmath185 is an integer , then follows from lemma [ le : zero ] . hence we may assume that @xmath185 is not an integer . we prove equation by reverse induction on @xmath40 . first let @xmath610 . then we have @xmath611 for every @xmath612 . hence it follows from step 2 that @xmath613 thus we can apply lemma [ le : gm ] to the projection of @xmath608 to the quotient @xmath614 . hence @xmath608 satisfies . now fix an integer @xmath615 and suppose , by induction , that @xmath616 satisfies . denote by @xmath617 and @xmath618 the arcs from @xmath619 to @xmath1 , and by @xmath620 and @xmath621 the arcs from @xmath1 to @xmath601 . then @xmath608 is the catenation of the @xmath179-traces @xmath622 here @xmath623 is the winding number of the loop @xmath624 about @xmath625 and simiarly for @xmath626 . note that @xmath627 is the shift of @xmath616 by @xmath1 . the catenation of @xmath616 and @xmath628 is the @xmath179-trace from @xmath488 to @xmath1 . hence it has viterbo maslov index zero , by lemma [ le : zero ] . and satisfies @xmath629 since the catenation of @xmath628 and @xmath627 is the @xmath179-trace from @xmath619 to @xmath601 , it also has viterbo maslov index zero and satisfies @xmath630 moreover , by the induction hypothesis , we have @xmath631 combining the equations , , and we find @xmath632 for @xmath539 . for @xmath633 we obtain @xmath634 here the last but one equation follows from equation and proposition [ prop : maslovc ] , and the last equation follows from lemma [ le : zero ] . hence @xmath608 satisfies . this completes the induction argument for the proof of step 3 . * step 4 . * _ let @xmath459 be as in lemma [ le : ab ] and let @xmath439 such that @xmath635 then holds for @xmath457 and @xmath466 . _ since @xmath577 we have @xmath237 . since @xmath636 we have @xmath637 and @xmath638 . hence @xmath4 and @xmath5 are noncontractible embedded circles , and they are homotopic ( with some orientation ) by lemma [ le : dbae3 ] . thus we may choose @xmath174 , @xmath176 , @xmath177 , @xmath236 as in step 3 . by assumption there is an integer @xmath639 . hence @xmath443 and @xmath444 do not contain any negative integers . choose @xmath596 such that @xmath640 assume without loss of generality that @xmath641 . for @xmath598 denote by @xmath642 and @xmath643 the arcs from @xmath488 to @xmath601 and consider the @xmath179-trace @xmath644 in this case @xmath645 as in step 3 , it follows by reverse induction on @xmath40 that @xmath608 satisfies for every @xmath40 . we assume again that @xmath185 is not an integer . ( otherwise follows from lemma [ le : zero ] ) . if @xmath610 then @xmath646 for every @xmath612 , hence it follows from step 2 that @xmath608 satisfies , and hence it follows from lemma [ le : gm ] for the projection of @xmath608 to the annulus @xmath614 that @xmath608 also satisfies . the induction step is verbatim the same as in step 3 and will be omitted . this proves step 4 . * step 5 . * _ we prove the proposition . _ if both points @xmath647 are contained in @xmath443 ( or in @xmath444 ) then @xmath648 by lemma [ le : ab ] , and in this case equation is a tautology . if both points @xmath647 are not contained in @xmath649 , equation has been established in lemma [ le : wg ] . moreover , we can interchange @xmath184 and @xmath185 or @xmath443 and @xmath444 as in the proof of step 2 . thus steps 1 and 4 cover the case where precisely one of the points @xmath647 is contained in @xmath649 while step 3 covers the case where @xmath237 and both points @xmath647 are contained in @xmath649 . this shows that equation holds for every @xmath401 . hence , by lemma [ le : gm ] , equation holds for every @xmath401 . this proves proposition [ prop : gm ] . choose a universal covering @xmath174 and let @xmath403 , @xmath176 , @xmath177 , and @xmath399 be as in proposition [ prop : gm ] . then @xmath650 here the last equation follows from proposition [ prop : gm ] . hence , by proposition [ prop : maslovc ] , we have @xmath651 this proves in the case where @xmath2 is not simply connected . [ app : a ] we assume throughout that @xmath2 is a connected oriented smooth @xmath0-manifold without boundary and @xmath652 are two embedded loops . let @xmath653,\sigma)\,|\ , x(0)\in\alpha,\,x(1)\in\beta\right\}\ ] ] denote the space of paths connecting @xmath4 to @xmath5 . by assumption , @xmath2 is oriented and has a nontrivial fundamental group . by the uniformization theorem , choose a metric of constant curvature . then the universal cover of @xmath2 is isometric to either @xmath660 with the flat metric or to the upper half space @xmath661 with the hyperbolic metric . the @xmath0-manifold @xmath659 is a quotient of the universal cover of @xmath2 by the subgroup of the group of covering transformations generated by a single element ( a translation in the case of @xmath660 and a hyperbolic element of @xmath662 in the case of @xmath661 ) . since @xmath90 is not contractible , this element is not the identity . hence @xmath659 is diffeomorphic to the cylinder . [ le : dbae3 ] let @xmath669 be noncontractible embedded loops and suppose that @xmath670 are nonzero integers such that @xmath671 is homotopic to @xmath672 . then either @xmath673 is homotopic to @xmath674 and @xmath675 or @xmath673 is homotopic to @xmath676 and @xmath677 . let @xmath667 be the covering generated by @xmath674 . then @xmath678 lifts to a closed curve in @xmath659 and is homotopic to @xmath679 . hence @xmath679 lifts to a closed immersed curve in @xmath659 . hence there exists a nonzero integer @xmath680 such that @xmath681 lifts to an embedding @xmath682 . any embedded curve in the cylinder is either contractible or is homotopic to a generator . if the lift of @xmath681 were contractible it would follow that @xmath678 is contractible , hence , by lemma [ le : dbae2 ] , @xmath683 in contradiction to our assumption . hence the lift of @xmath681 to @xmath659 is not contractible . with an appropriate sign of @xmath680 it follows that the lift of @xmath681 is homotopic to the lift of @xmath674 . interchanging the roles of @xmath674 and @xmath673 , we find that there exist nonzero integers @xmath684 such that @xmath685 in @xmath659 . hence @xmath674 is homotopic to @xmath686 in the free loop space of @xmath659 . since the homotopy lifts to the cylinder @xmath659 and the fundamental group of @xmath659 is abelian , it follows that @xmath687 if @xmath688 then @xmath673 is homotopic to @xmath674 , hence @xmath689 is homotopic to @xmath678 , hence @xmath690 is contractible , and hence @xmath691 , by lemma [ le : dbae2 ] . if @xmath692 then @xmath673 is homotopic to @xmath676 , hence @xmath693 is homotopic to @xmath678 , hence @xmath694 is contractible , and hence @xmath695 , by lemma [ le : dbae2 ] . this proves lemma [ le : dbae3 ] . orient @xmath4 and @xmath5 and and choose orientation preserving diffeomorphisms @xmath696 a closed loop in @xmath654 gives rise to a map @xmath697\to\sigma$ ] such that @xmath698 let @xmath699 denote the degree of @xmath700 and @xmath701 denote the degree of @xmath702 . since the homotopy class of a map @xmath703 or a map @xmath704 is determined by the degree we may assume , without loss of generality , that @xmath705 if one of the integers @xmath670 vanishes , so does the other , by lemma [ le : dbae2 ] . if they are both nonzero then @xmath673 is homotopic to either @xmath674 or @xmath706 , by lemma [ le : dbae3 ] . hence @xmath673 is isotopic to either @xmath674 or @xmath706 , by ( * ? ? ? * theorem 4.1 ) . hence @xmath4 is isotopic to @xmath5 , in contradiction to our assumption . this shows that @xmath707 with this established it follows that the map @xmath708\to\sigma $ ] factors through a map @xmath709 that maps the south pole to @xmath4 and the north pole to @xmath5 . since @xmath710 it follows that @xmath711 is homotopic , via maps with fixed north and south pole , to one of its meridians . this proves proposition [ prop : dbae ] .
in this paper , we investigate thermal fluctuations of the granular gas , which is driven by gaussian thermostat , on the basis of two - point kinetic theory . in particular , we consider thermal fluctuations of the inelastic variable sphere , which was proposed by yano [ j. phys . a , 46 ( 37 ) , 375502 ( 2013 ) ] . time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux are calculated on the basis of the two - point kinetic theory and compared with their numerical results , which are calculated using the direct simulation monte carlo ( dsmc ) method . finally , we compare transport coefficients , which are calculated on the basis of the kinetic theory , with those calculated using the dsmc method .
1512.08906
the first study of thermal fluctuations of the granular gas was performed by goldhirsch and van noije @xcite . in their study @xcite , green - kubo expression was considered for the viscosity coefficient ( @xmath0 ) of the granular gas on the basis of the scaled sonine polynomials in chapman - enskog method . afterwards , dufty and brey @xcite formulated green - kubo expression for the thermal conductivity ( @xmath1 ) and diffusive thermal conductivity ( @xmath2 ) together with @xmath0 . et al_. @xcite proved that their green - kubo expression for the transport coefficients of the granular gas reproduces transport coefficients , which are calculated by chapman - enskog method , with some good accuracies under the homogeneous cooling state ( hcs ) , using the direct simulation monte carlo ( dsmc ) method @xcite , whereas their numerical results also indicated that @xmath1 and @xmath2 , which are calculated on the basis of green - kubo expression using the dsmc method , deviate from @xmath1 and @xmath2 , which are calculated by the first order approximations in chapman - enskog method , in the range of @xmath3 ( @xmath4 : restitution coefficient ) and @xmath0 , which is calculated using the dsmc method , is similar to @xmath0 , which is calculated by the first order approximation in chapman - enskog method , in all the rage of @xmath4 ( @xmath5 ) . as a recent study related to thermal fluctuations of the granular gas , the thermally fluctuating hydrodynamics equation , namely , landau - lifshitz - navier - stokes - fourier ( llnsf ) equation @xcite , for the granular gas , was formulated using the inelastic boltzmann equation with the noise term brey _ et al_. @xcite . + in this paper , we aim to calculate general solutions of time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux for the granular gas driven by gaussian thermostat , which was proposed by montanero and santos @xcite . @xmath6 + to attain our aim , we extend the two - point kinetic theory for the elastic gas by tsuge and sagara @xcite to the granular gas driven by gaussian thermostat . in particular , we discuss the inelastic variable hard sphere ( ivhs ) , which was proposed by yano @xcite , as the component of the granular gas . the ivhs model is useful for understanding of the characteristics of thermal fluctuations of the granular gas driven by gaussian thermostat , because we can estimate how the characteristics of thermal fluctuations change in accordance with the change of the dependency of the collision frequency on the relative velocity between two colliding granular particles . additionally , the investigation of thermal fluctuations of the granular gas driven by gaussian thermostat is interesting , because the investigation of thermal fluctuations under the thermally nonequilibrium steady state is a universal problem in the nonequilibrium statistical mechanics . time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux for the granular gas driven by gaussian thermostat are calculated on the basis of the two - point kinetic theory . these analytical solutions of time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux are compared with dsmc results of time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux . here , we remind that three cases among the ivhs , namely , inelastic hard sphere ( ihs ) , ivhs with @xmath7 and inelastic maxwell sphere ( ims ) @xcite , are calculated , where @xmath8 will be defined in sec . finally , @xmath0 , @xmath1 and @xmath2 , which are calculated on the basis of the kinetic theory , are compared with . + this paper is organized as follows . the fourth and sixth order spherically symmetric moments are calculated for the ivhs to define the zeroth order approximation of the velocity distribution function in sec . ii . on the basis of the zeroth order approximation of the velocity distribution function , we discuss the two - point kinetic theory for the granular gas driven by gaussian thermostat in sec . analytical solutions of time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux are compared with dsmc results in sec . iv , when the granular gas under gaussian thermostat is composed of the ihs , ivhs with @xmath7 or ims . additionally , @xmath0 , @xmath1 and @xmath2 , which are calculated on the basis of the kinetic theory , are compared with . finally , we make the concluding remarks in sec . v. firstly , we calculate the fourth and sixth order spherically symmetric moments , . the inelastic boltzmann equation for the ivhs is written under the spatially homogeneous state as @xmath9 where @xmath10 is the velocity distribution function , @xmath11 and @xmath12 are velocities of two colliding ivhs , @xmath13 is a constant cross section . @xmath14 is the cooling rate and @xmath15 ( @xmath16 : flow velocity ) is the peculiar velocity , @xmath17 is the magnitude of the relative velocity of two colliding ivhs , @xmath18 $ ] is the deflection angle and @xmath19 $ ] is the scattering angle . we consider @xmath20 $ ] , where @xmath21 corresponds to the ihs and @xmath22 corresponds to the ims . of course , @xmath23 is obtained after a binary collision . + the zeroth order approximation of @xmath10 , namely , @xmath24 can be approximated using the fourth and sixth order spherically symmetric moments ( @xmath25 and @xmath26 ) as follows : @xmath27 where @xmath28 is maxwell - boltzmann distribution , and @xmath29 and @xmath30 , in which @xmath31 ( @xmath32 : gas constant ) . here , @xmath33 and @xmath34 ( @xmath35 : density ) . + substituting @xmath36 into both sides of eq . ( 1 ) , multiplying both sides of eq . ( 1 ) by @xmath37 and integrating over @xmath38 , we obtain @xmath39 where @xmath40 and @xmath41 ( @xmath42 : viscosity coefficient of the elastic vhs , @xmath43 : static pressure ) . + gaussian thermostat postulates that the total energy is always conserved . as a result , we obtain the cooling rate @xmath14 from @xmath44 in eq . ( 3 ) , when the flow velocity is always equal to zero , namely , @xmath45 , as @xmath46 + in later numerical analysis , the definition of @xmath14 in eq . ( 4 ) is not used , because @xmath47 is not satisfied owing to velocity fluctuations in the dsmc calculation . therefor , @xmath14 is calculated from the energy - loss via inelastic collisions by each time step in the dsmc calculation . + substituting @xmath36 into both sides of eq . ( 1 ) , multiplying both sides of eq . ( 1 ) by @xmath48 and @xmath49 , respectively , and integrating over @xmath38 , we obtain @xmath50 where @xmath51 and @xmath52 ( @xmath53 ) are functions of @xmath4 and @xmath8 . in eq . ( 5 ) , we neglected nonlinear terms @xmath54 , @xmath55 and @xmath56 to simplify our discussions . readers are recommended to read the paper by santos and montanero @xcite to confirm effects of those nonlinear terms . concrete forms of @xmath51 and @xmath52 ( @xmath53 ) in eq . ( 5 ) are defined by eqs . ( a1)-(a6 ) in appendix a. steady solutions of @xmath25 and @xmath26 in eq . ( 5 ) are obtained by @xmath57 in eq . finally , steady solutions of @xmath25 and @xmath26 are obtained as @xmath58 @xmath59 , @xmath60 in eq . ( 6 ) and @xmath61 , @xmath62 in eq . ( 7 ) are defined by eqs . ( a7)-(a10 ) in appendix a. @xmath25 and @xmath26 in eqs . ( 6 ) and ( 7 ) are equal to those under the hcs . on the basis of chapman - enskog method , brilliantov @xcite obtained @xmath25 and @xmath26 for the ihs under the hcs in a different form from eqs . ( 6 ) and ( 7 ) . + figures 1 , 2 and 3 shows @xmath25 ( left frame ) and @xmath26 ( right frame ) versus @xmath4 for the ihs , ivhs with @xmath7 and ims , respectively . dsmc results of @xmath25 and @xmath26 for the ihs , ivhs with @xmath7 and ims are obtained using 5000 sample particles per a cell in @xmath63 grids in the square domain @xmath64 , y \in [ 0,1]$ ] , + the left frame of fig . 1 shows that @xmath25 ( symbols ) , which is calculated using the dsmc method , namely , @xmath65 is similar to @xmath25 ( dashed - dot line ) , which is obtained as a steady solution by setting @xmath66 in eq . ( 5 ) and equal to @xmath25 , . @xmath25 in eq . ( 6 ) ( solid line ) is similar to @xmath65 in the range of @xmath67 , whereas the difference between @xmath25 in eq . ( 6 ) and @xmath65 increases , as @xmath4 decreases in the range of @xmath68 . additionally , @xmath25 in eq . ( 6 ) is similar to @xmath25 ( dashed line ) , which was obtained by brilliantov and p@xmath69schel @xcite , in the range of @xmath70 . the right frame of fig . 1 shows that @xmath26 ( symbols ) , which is calculated using the dsmc method , namely , @xmath71 is similar to @xmath26 ( solid line ) in eq . ( 7 ) in the range of @xmath72 , whereas @xmath71 is similar to @xmath26 ( dashed lime ) , which was obtained by brilliantov and p@xmath69schel @xcite , in the range of @xmath73 . therefore , @xmath26 in eq . ( 7 ) is more similar to @xmath71 than @xmath26 , which was obtained by brilliantov and p@xmath69schel @xcite . + next , we investigate @xmath25 and @xmath26 for the ivhs with @xmath7 . the left frame of fig . 2 shows that @xmath65 is similar to @xmath25 ( dashed line ) , which is obtained as a steady solution by setting @xmath66 in eq . ( 5 ) , in the range of @xmath70 . @xmath25 in eq . ( 6 ) ( solid line ) is similar to @xmath65 in the range of @xmath74 , whereas the difference between @xmath25 in eq . ( 6 ) and @xmath65 increases , as @xmath4 decreases in the range of @xmath3 . the right frame of fig . 2 shows that @xmath71 is similar to @xmath26 ( solid line ) in eq . ( 7 ) in the range of @xmath75 . + finally , we investigate @xmath25 and @xmath26 for the ims . the left frame of fig . 3 shows that @xmath65 ( symbols ) is similar to @xmath25 ( solid line ) in eq . ( 6 ) in the range of @xmath76 . here , we must remind that @xmath66 in eq . ( 5 ) is always obtained for the ims . on the other hand , @xmath26 in eq . ( 7 ) diverges at @xmath77 and becomes negative in the range of @xmath78 , as shown in the right frame of fig . 3 . @xmath26 in eq . ( 7 ) approximates @xmath71 with good accuracies in the range of @xmath67 . to avoid the divergence of @xmath26 in eq . ( 7 ) , + on the basis of @xmath25 and @xmath26 , we define the zeroth order approximation of the velocity distribution function , namely , @xmath79 , in the next section . the two - point kinetic theory was firstly discussed for the elastic gas by tsuge and sagara @xcite . in this section , we consider the two - point kinetic theory for the granular gas driven by gaussian thermostat . the stochastic inelastic boltzmann equation is the most fundamental equation in the two - point kinetic theory . in particular , the stochastic inelastic boltzmann equation under gaussian thermostat is written as @xmath80=0,\\ & & \mathscr{j}\left(\ell_1|\ell\right)\left[\varrho\left(t,\ell\right)\varrho\left(t,{\ell}_1\right)\right ] \nonumber \\ & = & \mathscr{a}\int_{\mathbb{v}^3 } d\bm{c}_1 \int_0^{2\pi } d \epsilon \int_0^\pi d\chi g^{1-\xi } \left\{\frac{1}{\alpha^{2-\xi}}\varrho\left(t,\ell^{\prime\prime}\right)\varrho\left(t,{\ell}_1^{\prime\prime}\right)-\varrho\left(t,\ell\right)\varrho\left(t,{\ell}_1\right ) \right\}\sigma \sin \chi , \nonumber \\\end{aligned}\ ] ] where @xmath81 is the microscopic distribution function defined by @xmath82,\end{aligned}\ ] ] where @xmath83 is the index of the particles , @xmath84 is the total number of particles , and @xmath85 ( @xmath86 ) . from eq . ( 8) , we assume @xmath87 . + for convenience , we set @xmath88 . in @xmath89-space , we consider the correlation between two points @xmath90 and @xmath91 . as a result , definitions @xmath92 and @xmath93 are used . at a point @xmath90 , the stochastic inelastic boltzmann equation under gaussian thermostat in eq . ( 8) is rewritten as @xmath94\varrho\left(\tilde{\alpha}\right ) \nonumber \\ & & -\mathscr{j}\left({\tilde{\alpha}}_1|\tilde{\alpha}\right)\left[\varrho\left(\tilde{\alpha}\right)\varrho\left({\tilde{\alpha}}_1\right)\right]=0.\end{aligned}\ ] ] next , we define the averaged quantity in @xmath95-space using finite volumes @xmath96 and @xmath97 as follows @xmath98 where @xmath99 is the arbitrary function defined in @xmath95-space . + on the basis of eqs . ( 12 ) and ( 13 ) , we obtain following equation by multiplying @xmath100 by both sides of eq . ( 11 ) and taking average in @xmath95-space , @xmath101\left[\overline{\varrho(\tilde{\alpha})\varrho(\tilde{\beta})}-f(\tilde{\alpha})f(\tilde{\beta})\right]\nonumber \\ & = & \mathscr{j}\left(\tilde{\alpha}|\tilde{\alpha}_1\right)\left[\overline{\varrho(\tilde{\alpha})\varrho(\tilde{\alpha}_1)\varrho(\tilde{\beta})}-\overline{\varrho(\tilde{\alpha})\varrho(\tilde{\alpha}_1)}f(\tilde{\beta } ) \right],\end{aligned}\ ] ] where @xmath102 . ( 14 ) , @xmath103 is decomposed as @xmath104 where @xmath105 is the two - point phase density of different particles and @xmath106 is the probability of finding same particles at @xmath107 and @xmath108 . @xmath109 is related to the hydrodynamic fluctuation term @xmath110 as follows @xmath111 for @xmath112 and @xmath113 , we obtain @xmath114f\left(\tilde{\alpha}\right).\end{aligned}\ ] ] assuming that hydrodynamic fluctuations are ignorable , ( i.e. , @xmath115 or @xmath116 ) , we obtain the following equation from eq . ( 14 ) @xcite @xmath117g\left(\tilde{\alpha};\tilde{\beta}\right ) \nonumber \\ & & = \mathscr{j}\left(\tilde{\alpha}|{\tilde{\alpha}}_1\right)\left(f(\tilde{\alpha})g({\tilde{\alpha}}_1;\tilde{\beta})+f({\tilde{\alpha}}_1)g(\tilde{\alpha};\tilde{\beta})\right ) , \nonumber \\\end{aligned}\ ] ] where @xmath118 . + @xmath106 is expanded using hermite polynomials @xcite as follows @xmath119 where @xmath120 , @xmath121 , @xmath122 $ ] , + @xmath123 $ ] and @xmath124 is the hermite polynomial @xcite . + from eq . ( 21 ) , we readily obtain @xmath125 we assume the initial form of @xmath126 from eq . ( 19 ) as follows @xmath127_{\epsilon=0}=\delta\left[\xi\left(\tilde{\alpha}\right)-\xi(\tilde{\beta})\right]f^{(0)}\left(\tilde{\alpha}\right),\end{aligned}\ ] ] where @xmath128 is the zeroth order approximation of the velocity distribution function at a point @xmath90 . + substituting @xmath129 , @xmath130 , @xmath131 , and @xmath132 into eq . ( 22 ) with eq . ( 23 ) , we obtain @xmath133_{\epsilon=0}&=&\rho(\tilde{\alpha})\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[{q}_{i , j}^{(1,1)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha } ) \tilde{c}^2\delta_{ij}\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[{q}_{ij , lm}^{(2,2)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha } ) \tilde{c}^4\left(1+\frac{a_4}{15}\right)\left(\delta_{il}\delta_{jm}+\delta_{jl}\delta_{im}-\frac{2}{3}\delta_{ij}\delta_{lm}\right)\delta \left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[{q}_{i , j}^{(3,3)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha } ) \tilde{c}^6 \delta_{ij}\left(10+\frac{11}{3}a_4+\frac{a_6}{3}\right)\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\end{aligned}\ ] ] where we used the assumption @xmath134 and the relation @xmath135 , which is obtained in the energy conservative system . @xmath136 corresponds to the two - point - correlation of thermal fluctuations of the density , @xmath137 corresponds to the two - point - correlation of thermal fluctuations of the momentum , @xmath138 corresponds to the two - point - correlation of thermal fluctuations of the pressure deviator , and @xmath139 corresponds to the two - point - correlation of thermal fluctuations of the two times of the heat flux . substituting @xmath140 or @xmath141 into both sides of eq . ( 20 ) and integrating in accordance with eq . ( 22 ) , we obtain time evolutions of @xmath142 and @xmath143 as follows @xmath144 where we restrict ourselves to the spatially homogeneous state . + from eqs . ( 28 ) and ( 29 ) , we obtain @xmath145_{\epsilon=0 } \exp\left[-\left(\nu_{\pi}\left(\tilde{\alpha}\right)-\zeta\left(\tilde{\alpha}\right)\right ) \epsilon\right],\\ q_{i , j}^{(3,3)}(\epsilon)&=&\left[q_{i , j}^{(3,3)}\right]_{\epsilon=0 } \exp\left[-\left(\nu_q\left(\tilde{\alpha}\right)-\frac{3}{2}\zeta\left(\tilde{\alpha}\right)\right ) \epsilon \right].\end{aligned}\ ] ] where @xmath146 and @xmath147 are dissipation rates of @xmath142 and @xmath143 , which are calculated as @xcite @xmath148 provided that @xmath149 and @xmath150 , we can approximate @xmath151 with @xmath152 , so that eqs . ( 30 ) and ( 31 ) can be rewritten as @xmath153_{\epsilon=0 } \frac{2}{\nu_{\pi}\left(\tilde{\alpha}\right)-\zeta\left(\tilde{\alpha}\right ) } \delta\left(\epsilon\right),\\ q_{i , j}^{(3,3)}\left(\epsilon\right)&=&\left[q_{i , j}^{\left(3,3\right)}\right]_{\epsilon=0}\frac{2}{\nu_q\left(\tilde{\alpha}\right)-\frac{3}{2}\zeta\left(\tilde{\alpha}\right)}\delta\left(\epsilon\right).\end{aligned}\ ] ] substituting @xmath154 into eqs . ( 34 ) and ( 35 ) , respectively , we can readily reproduce thermally fluctuating terms in llnsf equation for the elastic gas @xcite . + integrating eqs . ( 30 ) and ( 31 ) from @xmath155 to @xmath156 , we obtain @xmath157_{\epsilon=0 } \frac{1}{\nu_{\pi}\left(\tilde{\alpha}\right)-\zeta\left(\tilde{\alpha}\right)},\\ \int_0^\infty q_{i , j}^{\left(3,3\right)}\left(\epsilon\right ) d\epsilon & = & \left[q_{i , j}^{\left(3,3\right)}\right]_{\epsilon=0}\frac{1}{\nu_q\left(\tilde{\alpha}\right)-\frac{3}{2}\zeta\left(\tilde{\alpha}\right)}.\end{aligned}\ ] ] here , we remind that @xmath158 , @xmath159 and @xmath160 for the granular gas driven by gaussian thermostat are written as @xcite @xmath161 + from eqs . ( 38)-(40 ) , we obtain @xmath162_{\epsilon=0}^{-1},\\ \kappa\left(\alpha,\omega\right)&=&\left(\frac{5}{2}+\frac{a_4}{3}\right ) p\left(\tilde{\alpha}\right ) r \int_0^\infty q_{i , j}^{\left(3,3\right)}\left(\epsilon\right ) d\epsilon \left[q_{i , j}^{(3,3)}\right]_{\epsilon=0}^{-1},\\ \eta\left(\alpha,\omega\right)&=&\frac{a_4}{6 } \frac{p\left(\tilde{\alpha}\right)rt\left(\tilde{\alpha}\right)}{\rho\left(\tilde{\alpha}\right)}\int_0^\infty q_{i , j}^{(3,3)}\left(\epsilon\right ) d\epsilon \left[q_{i , j}^{(3,3)}\right]_{\epsilon=0}^{-1}.\end{aligned}\ ] ] eqs . ( 41)-(43 ) are definitions of the transport coefficients of the granular gas driven gaussian thermostat . substituting @xmath154 into eq . ( 41)-(43 ) , we can readily obtain green - kubo expression for the transport coefficients in the elastic gas . + the comparison of the viscosity coefficient of the granular gas driven by gaussian thermostat , which is calculated using the dsmc method , with that obtained by chapman - enskog method was done using the solution of the uniform shear flow by garzo and montanero @xcite . in sec . iv , we calculate following four parameters using the dsmc method @xmath163_{\epsilon=0}(\alpha)/\left[q_{xx , xx}^{(2,2)}\right]_{\epsilon=0}^{\alpha=1},\\ \psi_3\left(\alpha\right)&=&\left[q_{x , x}^{(3,3)}\right]_{\epsilon=0}(\alpha)/\left[q_{x , x}^{(3,3)}\right]_{\epsilon=0}^{\alpha=1},\\ \phi_2\left(\epsilon\right)&=&q_{xx , xx}^{(2,2)}\left(\epsilon\right)\left[q_{xx , xx}^{(2,2)}\right]^{-1}_{\epsilon=0},\\ \phi_3\left(\epsilon\right)&=&q_{x , x}^{(3,3)}\left(\epsilon\right)\left[q_{x , x}^{(3,3)}\right]^{-1}_{\epsilon=0}.\end{aligned}\ ] ] analytical solutions of @xmath164 and @xmath165 are readily calculated from eqs . ( 26 ) and ( 27 ) as @xmath166 + similarly , analytical solutions of @xmath167 and @xmath168 are readily calculated from eqs . ( 30 ) and ( 31 ) as @xmath169,\\ \phi_3\left(\epsilon\right)&=&\exp\left[-\left(\nu_q\left(\tilde{\alpha}\right)-\frac{3}{2}\zeta\left(\tilde{\alpha}\right)\right ) \epsilon \right].\end{aligned}\ ] ] in sec . iii , we considered time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux for the granular gas driven by gaussian thermostat on the basis of the two - point kinetic theory , whereas time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux for the granular gas under the hcs are discussed on the basis of the two - point kinetic theory in appendix b. the dsmc calculation is performed to confirm the validity of the two - point kinetic theory for the granular gas driven by gaussian thermostat . in the dsmc calculation , 5000 sample particles are set in a cell , whereas equally spaced @xmath63 grids are set in the square domain @xmath64 , y \in [ 0,1]$ ] . all the physical quantities are normalized as : @xmath170 , @xmath171 , @xmath172 , @xmath173 and @xmath174 . the time interval is set as @xmath175 and @xmath176 steps are iterated . @xmath177 ( @xmath178 : knudsen number ) for the ihs , @xmath179 ( @xmath178 : knudsen number ) for the ivhs with @xmath7 and @xmath180 for the ims . dsmc calculations are performed using @xmath181 and @xmath182 for the ihs , ivhs with @xmath7 and ims . the dsmc results show @xmath185_{\epsilon=0}=1.33 $ ] and @xmath186_{\epsilon=0}=10 $ ] for the ihs , ivhs with @xmath7 and ims , when @xmath154 , namely , @xmath187 in eqs . ( 26 ) and ( 27 ) . therefore , dsmc results reproduce eqs . ( 26 ) and ( 27 ) with good accuracies , when @xmath154 . + show @xmath164 versus @xmath4 ( left frame ) and @xmath165 versus @xmath4 ( right frame ) for the ihs , ivhs with @xmath7 and ims , respectively . @xmath164 in eq . ( 44 ) is calculated using @xmath25 in eq . ( 6 ) or @xmath188 , whereas @xmath165 in eq . ( 45 ) is calculated using @xmath25 in eq . ( 6 ) and @xmath26 in eq . ( 7 ) or @xmath188 and @xmath189 . the difference between @xmath190_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method , and @xmath164 in eq . ( 44 ) , which is obtained using @xmath25 in eq . ( 6 ) or @xmath188 , increases in the range of @xmath191 , as @xmath4 decreases . @xmath164 in eq . ( 44 ) , which is obtained using @xmath188 , is more similar to @xmath190_{\text{\tiny{dsmc}}}$ ] than @xmath164 in eq . ( 44 ) , which is obtained using @xmath25 in eq . ( 6 ) . in the right frame of fig . 4 , the difference between @xmath192_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method , and @xmath165 in eq . ( 45 ) , which is obtained using @xmath25 in eq . ( 6 ) and @xmath26 in eq . ( 7 ) or @xmath188 and @xmath189 , increases in the range of @xmath191 , as @xmath4 decreases . we , however , find that @xmath165 in eq . ( 45 ) , which is obtained using @xmath188 and @xmath189 , is much more similar to @xmath192_{\text{\tiny{dsmc}}}$ ] than @xmath165 in eq . ( 45 ) , which is obtained using @xmath25 in eq . ( 6 ) and @xmath26 in eq . + in the left frame of fig . 5 , the difference between @xmath190_{\text{\tiny{dsmc}}}$ ] and @xmath164 in eq . ( 44 ) , which is obtained using @xmath25 in eq . ( 6 ) or @xmath188 , increases in the range of @xmath68 , as @xmath4 decreases . @xmath164 in eq . ( 44 ) , which is obtained using @xmath188 , is more similar to @xmath190_{\text{\tiny{dsmc}}}$ ] than @xmath164 in eq . ( 44 ) , which is obtained using @xmath25 in eq . the difference between @xmath192_{\text{\tiny{dsmc}}}$ ] and @xmath165 in eq . ( 45 ) , which is obtained using @xmath25 in eq . ( 6 ) and @xmath26 in eq . ( 7 ) or @xmath188 and @xmath189 , increases in the range of @xmath193 , as @xmath4 decreases . we , however , find that @xmath165 in eq . ( 45 ) , which is obtained using @xmath188 and @xmath189 , is much more similar to @xmath192_{\text{\tiny{dsmc}}}$ ] than @xmath165 in eq . ( 45 ) , which is obtained using @xmath25 in eq . ( 6 ) and @xmath26 in eq . + in the left frame of fig . 6 , the difference between @xmath190_{\text{\tiny{dsmc}}}$ ] and @xmath164 in eq . ( 44 ) , which is obtained using @xmath25 in eq . ( 6 ) or @xmath188 , is smaller than that in the case of the ihs . there is only marked difference between @xmath190_{\text{\tiny{dsmc}}}$ ] and @xmath164 in eq . ( 44 ) , which is calculated using @xmath188 or @xmath25 in eq . ( 6 ) , when @xmath194 . in the right frame of fig . 6 , there are marked differences between @xmath192_{\text{\tiny{dsmc}}}$ ] and @xmath165 in eq . ( 45 ) , which is obtained using @xmath188 and @xmath189 , when @xmath195 and @xmath196 . additionally , there are marked differences between @xmath192_{\text{\tiny{dsmc}}}$ ] and @xmath165 in eq . ( 45 ) , which is calculated using @xmath25 in eq . ( 6 ) and @xmath26 in eq . ( 7 ) , when @xmath191 , because @xmath26 in eq . ( 7 ) diverges at @xmath197 , as shown in the right frame of fig . + we investigate @xmath198 and @xmath199 using the dsmc method and eqs . ( 46 ) and ( 47 ) . figures 7 , show @xmath198 versus @xmath200 and @xmath199 versus @xmath200 for the ihs , ivhs with @xmath7 and ims , when @xmath194 , @xmath196 , @xmath201 , @xmath202 , @xmath203 and @xmath182 , respectively . figure 7 shows that @xmath204_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method , is quite similar to @xmath198 in eq . meanwhile , @xmath205_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method , is slightly smaller than @xmath199 in eq . ( 47 ) , when @xmath154 and @xmath203 , whereas @xmath205_{\text{\tiny{dsmc}}}$ ] is slightly larger than @xmath199 in eq . ( 47 ) , when @xmath194 . the difference between @xmath205_{\text{\tiny{dsmc}}}$ ] and @xmath199 in eq . ( 47 ) , when @xmath154 , might be improved by the enhancement of the accuracy of the time integration in the dsmc method . on the other hand , the difference between @xmath205_{\text{\tiny{dsmc}}}$ ] and @xmath199 in eq . ( 47 ) , when @xmath194 , indicates that the linear response form in eq . ( 29 ) is insufficient to demonstrate @xmath199 for the ihs , when @xmath194 . + figure 8 shows that @xmath204_{\text{\tiny{dsmc}}}$ ] is quite similar to @xmath198 in eq . meanwhile , @xmath205_{\text{\tiny{dsmc}}}$ ] is slightly smaller than @xmath199 in eq . ( 47 ) , when @xmath154 , whereas @xmath205_{\text{\tiny{dsmc}}}$ ] is slightly larger than @xmath199 in eq . ( 47 ) , when @xmath194 . such tendencies of @xmath204_{\text{\tiny{dsmc}}}$ ] and @xmath205_{\text{\tiny{dsmc}}}$ ] for the ivhs with @xmath7 are quite similar to those for the ihs . + figure 9 shows that @xmath204_{\text{\tiny{dsmc}}}$ ] is quite similar to @xmath198 in eq . meanwhile , @xmath205_{\text{\tiny{dsmc}}}$ ] is larger than @xmath199 in eq . ( 47 ) , when @xmath194 and @xmath196 . the difference between @xmath205_{\text{\tiny{dsmc}}}$ ] and @xmath199 in eq . ( 47 ) increases , as @xmath4 decreases , as shown in frames of @xmath194 and @xmath196 . additionally , we can confirm that the difference between @xmath205_{\text{\tiny{dsmc}}}$ ] and @xmath199 in eq . ( 47 ) when @xmath194 increases , as @xmath8 decreases , as shown in figs . finally , we can conclude that the difference between @xmath205_{\text{\tiny{dsmc}}}$ ] and @xmath199 in eq . ( 47 ) , when @xmath194 or @xmath196 , is not caused by nonlinear terms in the collisional moments of @xmath206 @xcite for the ihs and ivhs with @xmath7 , which are not included in eq . ( 29 ) , because nonlinear terms in the collisional moments of @xmath206 never appear in the right hand side of eq . ( 29 ) for the ims , . finally , we compare transport coefficients , @xmath0 , @xmath1 and @xmath2 in eqs . ( 41)-(43 ) , which are calculated using the dsmc method , with those in eqs . ( 38)-(40 ) . as shown in eqs . ( 41)-(43 ) , we can understand that transport coefficients are related to areas of @xmath198 and @xmath199 in @xmath207 . therefore , accuracies of numerical integrations of @xmath198 and @xmath199 , which are calculated using the dsmc method , are significant . meanwhile , @xmath198 and @xmath199 fluctuate around @xmath208 . therefore , @xmath198 and @xmath199 become negative values in ranges of @xmath209 and @xmath210 . such fluctuations around 0 can be reduced by increasing the number of sample particles . we , however , find that further increase of the number of sample particles requires parallel computations . in numerical integrations of @xmath198 and @xmath199 , we integrate @xmath198 and @xmath199 in the range of @xmath211 $ ] , in which both @xmath198 and @xmath199 are always positive . in particular , we are interested in @xmath212 , @xmath213 and @xmath214 . additionally , we remind that the transport coefficients do not depend on @xmath215_{\hat{\epsilon}=0}$ ] and @xmath216_{\hat{\epsilon}=0}$ ] , as shown in eqs . ( 41)-(43 ) . consequently , differences between @xmath215_{\hat{\epsilon}=0}$ ] and @xmath216_{\hat{\epsilon}=0}$ ] , which are calculated using the dsmc method , and those in eqs . ( 26 ) and ( 27 ) do not contribute to any differences between transport coefficients , which are calculated using the dsmc method in eqs . ( 41)-(43 ) , and those in eqs . ( 38)-(40 ) . + figure 10 shows @xmath212 , @xmath213 and @xmath217 versus @xmath4 . @xmath25 in eqs . ( 39 ) and ( 40 ) are equal to a steady solution of eq . ( 5 ) with @xmath66 or eq . ( 6 ) . as mentioned above , @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , coincides with @xmath25 in eq . ( 6 ) only for the ims . figure 10 shows that @xmath218_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method in eq . ( 41 ) , is quite similar to that in eq . ( 38 ) for the ihs , ivhs with @xmath7 and ims in all the range of @xmath4 . such a similarity between @xmath219 in eq . ( 38 ) and @xmath218_{\text{\tiny{dsmc}}}$ ] was also confirmed for the ihs by garzo and his coworkers @xcite @xcite . + figure 10 shows that @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) or @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , does not fit @xmath221_{\text{\tiny{dsmc}}}$ ] , which is calculated using the dsmc method in eq . ( 42 ) , in ranges of @xmath222 and @xmath223 in the case of the ihs . meanwhile , @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) is more similar to @xmath221_{\text{\tiny{dsmc}}}$ ] than @xmath213 in eq . ( 39 ) with @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , in the range of @xmath224 in the case of the ihs . @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) or @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , does not fit @xmath221_{\text{\tiny{dsmc}}}$ ] in the range of @xmath225 in the case of the ivhs with @xmath7 . meanwhile , @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) is more similar to @xmath221_{\text{\tiny{dsmc}}}$ ] than @xmath213 in eq . ( 39 ) with @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , in the range of @xmath3 in the case of the ivhs with @xmath7 . @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) does not fit @xmath221_{\text{\tiny{dsmc}}}$ ] in the range of @xmath3 in the case of the ims . meanwhile , the difference between @xmath221_{\text{\tiny{dsmc}}}$ ] and @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) increases markedly , as @xmath4 decreases from @xmath226 to @xmath208 , in the case of the ims . as a result , the difference in the range of @xmath3 between @xmath221_{\text{\tiny{dsmc}}}$ ] and @xmath220 in eq . ( 39 ) with @xmath25 in eq . ( 6 ) increases , as @xmath8 decreases from unity ( ihs ) to zero ( ims ) . + figure 10 shows that @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) or @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , does not fit @xmath228 , which is calculated using the dsmc method in eq . ( 43 ) , in the range of @xmath68 in the case of the ihs . meanwhile , @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) is more similar to @xmath228 than @xmath227 in eq . ( 40 ) with @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , in the range of @xmath191 in the case of the ihs . @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) or @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , does not fit @xmath228 in the range of @xmath68 in the case of the ivhs with @xmath7 . meanwhile , @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) is more similar to @xmath228 than @xmath227 in eq . ( 40 ) with @xmath25 , which is a steady solution of eq . ( 5 ) with @xmath66 , in the range of @xmath191 in the case of the ivhs with @xmath7 . @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) does not fit @xmath228 in the range of @xmath229 in the case of the ims . meanwhile , the difference between @xmath228 and @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) increases markedly , as @xmath4 decreases from @xmath230 to @xmath208 , in the case of the ims . as a result , the difference in the range of @xmath191 between @xmath228 and @xmath227 in eq . ( 40 ) with @xmath25 in eq . ( 6 ) increases , as @xmath8 decreases from unity ( ihs ) to zero ( ims ) . in this paper , we investigated thermal fluctuations of the granular gas driven by gaussian thermostat on the basis of the two - point kinetic theory . in particular , we considered the inelastic variable hard sphere ( ivhs ) as the component of the granular gas . spherically symmetric moment @xmath25 , which is calculated using the dsmc method , is more similar to @xmath25 , which is analytically obtained by neglecting @xmath26 in the collisional term of @xmath25 , than @xmath25 , which is analytically obtained by including @xmath26 in the collisional term of @xmath25 in cases of the ihs and ivhs with @xmath7 . @xmath26 , which is calculated using the dsmc method , is similar to @xmath26 , which is analytically obtained , in the range of @xmath73 in the case of the ihs and in the range of @xmath75 in the case of ivhs with @xmath7 , whereas @xmath26 , which is analytically obtained for the ims , diverges at @xmath231 . correlations of thermal fluctuations of the pressure deviator and two times of the heat flux at the same time were evaluated using two parameters @xmath183 and @xmath184 . @xmath183 , which is calculated using the dsmc method , is similar to @xmath183 in eq . ( 44 ) in the range of @xmath67 in cases of the ihs and ivhs , whereas @xmath183 , which is calculated using the dsmc method , is similar to @xmath183 in eq . ( 44 ) in the range of @xmath76 in the case of the ims . meanwhile , @xmath184 in eq . ( 45 ) is similar to @xmath184 , which is calculated using the dsmc method , in the range of @xmath67 in the case of the ihs and in the range of @xmath73 in cases of the ivhs with @xmath7 and ims . the use of @xmath232 and @xmath233 in eqs . ( 44 ) and ( 45 ) improved similarities between @xmath183 , which is calculated using the dsmc method , and @xmath183 in eq . ( 44 ) with @xmath25 in eq . ( 6 ) or @xmath184 , which is calculated using the dsmc method , and @xmath184 in eq . ( 45 ) with @xmath25 and @xmath26 in eqs . ( 6 ) and ( 7 ) . time correlations of thermal fluctuations of the pressure deviator and two times of the heat flux were evaluated using two parameters , namely , @xmath198 and @xmath199 . @xmath198 , which is calculated using the dsmc method , is similar to @xmath198 in eq . ( 46 ) in cases of the ihs , ivhs with @xmath7 and ims . @xmath199 , which is calculated using the dsmc method , is slightly smaller than @xmath199 in eq . ( 47 ) , when @xmath194 for the ihs and ivhs with @xmath7 or @xmath194 and @xmath196 for the ims . the viscosity coefficient , which is calculated using the dsmc method , is quite similar to the viscosity coefficient , which is analytically obtained by the kinetic theory , in cases of the ihs , ivhs with @xmath7 and ims . the thermal conductivity , which is calculated using the dsmc method , does not fit the thermal conductivity , which is obtained by the kinetic theory , in ranges of @xmath70 and @xmath223 in the case of the ihs . the thermal conductivity , which is calculated using the dsmc method , does not fit the thermal conductivity , which is obtained by the kinetic theory , in the range of @xmath3 in cases of the ivhs with @xmath7 and ims . the diffusive thermal conductivity , which is calculated using the dsmc method , is similar to the diffusive thermal conductivity , which is obtained by the kinetic theory , in the range of @xmath74 in cases of the ihs , ivhs with @xmath7 and ims . finally , differences between the thermal conductivity and diffusive thermal conductivity at @xmath194 , which are calculated using the dsmc method and those obtained by the kinetic theory , increases , as @xmath8 decreases . @xmath51 and @xmath52 ( i=0,1,2 ) in eq . ( 5 ) are calculated as @xmath234 @xmath59 in eq . ( 6 ) are calculated as @xmath235 @xmath60 in eq . ( 6 ) are calculated as @xmath236 @xmath61 in eq . ( 7 ) are calculated as @xmath237 @xmath238 in eq . ( 7 ) are calculated as @xmath239 in this paper , we discussed the extension of the two - point kinetic theory by tsuge - sagara to the granular gas driven by gaussian thermostat . in the two - point kinetic theory by tsuge - sagara , the form of the correlation function , namely , @xmath240 , is expanded using hermite polynomials , as shown in eq . the validity of the application of the two - point kinetic theory to thermal fluctuations of the granular gas under the hcs requires further considerations . for convenience , we expand @xmath240 using modified moments @xmath241 such as @xmath242 from eq . ( b1 ) , we readily obtain @xmath243 from eqs . ( b1 ) , ( b2 ) and ( 23 ) , we obtain following relations using similar procedures to obtain eqs . ( 24)-(27 ) . @xmath244_{\epsilon=0}&=&\rho(\tilde{\alpha})\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[\tilde{q}_{i , j}^{(1,1)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha})\delta_{ij}\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[\tilde{q}_{ij , lm}^{(2,2)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha } ) \left(1+\frac{a_4}{15}\right)\left(\delta_{il}\delta_{jm}+\delta_{jl}\delta_{im}-\frac{2}{3}\delta_{ij}\delta_{lm}\right)\delta \left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right],\\ \left[\tilde{q}_{i , j}^{(3,3)}\right]_{\epsilon=0}&=&\rho(\tilde{\alpha } ) \delta_{ij}\left(10+\frac{11}{3}a_4+\frac{a_6}{3}\right)\delta\left[\bm{x}(\tilde{\alpha})-\bm{x}(\tilde{\beta})\right].\end{aligned}\ ] ] in eq . ( b2 ) , @xmath241 depends on the only density . as a result , eqs . ( b3)-(b6 ) hold true for the granular gas under the hcs . on the other hand , the calculation of @xmath245_{\epsilon=0}$ ] under the hcs is difficult , owing to the emergence of the long wave instability via the inelastic clustering . therefore , the choice of the time interval to sample @xmath246_{\epsilon=0}$ ] is significant . + finally , we must investigate whether the extension of green - kubo expression for the transport coefficients in eqs . ( 41)-(43 ) can be possible for the granular gas under the hcs . according to the two - point kinetic theory in sec . iii , time evolutions of @xmath247 and @xmath248 are calculated as @xmath249 the cooling rate @xmath14 never emerges in eqs . ( b7 ) and ( b8 ) unlike eqs . ( 28 ) and ( 29 ) , whereas @xmath250 and @xmath251 are function of time , because the temperature depends on @xmath252 . + solutions of eqs . ( b7 ) and ( b8 ) are obtained as @xmath253_{\epsilon=0},\\ \tilde{q}_{i , j}^{(3,3)}\left(\epsilon\right)&=&\exp\left(-\int_0^\epsilon \nu_{q}\left(\tilde{\alpha}\right ) ds\right)\left[\tilde{q}_{i , j}^{(3,3)}\right]_{\epsilon=0},\end{aligned}\ ] ] where @xmath254 . + from eqs . ( b9 ) and ( b10 ) , we obtain @xmath255_0^t \left[\tilde{q}_{ij , lm}^{(2,2)}\right]_{\epsilon=0},\\ \int_0^t \tilde{q}_{i , j}^{(3,3)}\left(\epsilon\right ) d\epsilon&=&-\left[\nu_{q}^{-1}\left(\tilde{\alpha},\epsilon\right)\exp\left(-\int_0^\epsilon \nu_{q}\left(\tilde{\alpha},s\right ) ds\right)\right]_0^t \left[\tilde{q}_{i , j}^{(3,3)}\right]_{\epsilon=0},\end{aligned}\ ] ] the transport coefficients of the ivhs , which are calculated by chapman - enskog method , are written as @xcite @xmath256 where @xmath257 , @xmath258 , @xmath259 , @xmath260 and @xmath261 . + @xmath158 , @xmath159 and @xmath160 at @xmath262 can not be expressed with @xmath263 , @xmath264_{\epsilon=0}$ ] , @xmath265 and @xmath266_{\epsilon=0}$ ] , explicitly , because we can not express @xmath250 with @xmath263 and @xmath264_{\epsilon=0}$ ] in eq . ( b11 ) and @xmath251 with @xmath265 and @xmath266_{\epsilon=0}$ ] in eq . ( b12 ) . on the other hand , we can obtain initial values of @xmath250 and @xmath251 by setting @xmath267 such as @xmath268_{\epsilon=0}}{\int_0^\infty \tilde{q}_{ij , lm}^{(2,2)}\left(\epsilon\right ) d\epsilon},~~\nu_q\left(\tilde{\alpha},0\right)=\frac{\left[\tilde{q}_{i , j}^{(3,3)}\right]_{\epsilon=0}}{\int_0^\infty \tilde{q}_{i , j}^{(3,3)}\left(\epsilon\right ) d\epsilon}.\end{aligned}\ ] ] substituting eq . ( b16 ) into eqs . ( b13)-(b15 ) , we obtain the modified green - kubo expression for the transport coefficients using initial values of transport coefficients . of course , such modified green - kubo expression obtained using initial values of transport coefficients are nonlinear responses . in the dsmc calculation , the transport coefficients at @xmath262 are obtained , when we calculate @xmath269 and @xmath270 by setting @xmath271 . + another approach to calculate transport coefficients for the granular gas under the hcs , we consider the map : @xmath272 , in which quantities with subscript @xmath156 correspond to representative values , in accordance with the map by dufty and brey @xcite . this map nondimensionalizes dissipation rates @xmath250 and @xmath251 , which depend on time , in time independent forms . applying this map to eqs . ( b7 ) and ( b8 ) , we obtain @xmath273 and @xmath274 as @xmath275_{\epsilon=0}\frac{1}{\tilde{\nu}_{\pi}\left(\tilde{\alpha}\right)},\\ \int_0^\infty \tilde{q}_{i , j}^{(3,3)}\left(\epsilon^\prime\right ) d\epsilon^\prime=\left[\tilde{q}_{i , j}^{(3,3)}\right]_{\epsilon=0}\frac{1}{\tilde{\nu}_q\left(\tilde{\alpha}\right)},\end{aligned}\ ] ] where @xmath276 and @xmath277 . as mentioned above , @xmath278 and @xmath279 are time independent . substituting @xmath278 and @xmath279 in eqs . ( b17 ) and ( b18 ) into eqs . ( b13)-(b15 ) , we obtain another form of the modified green - kubo expression for the transport coefficients . in the dsmc calculation , we use the time interval @xmath280 to integrate @xmath281 and @xmath282 in eqs . ( b17 ) and ( b18 ) , which are calculated using the temperature by each time step . the relation between the modified green - kubo expression for the transport coefficients , which was introduced on the basis of the two - point kinetic theory , and green - kubo expression by dufty and brey @xcite must be addressed in our future work . + + @xmath283 p_{ij},}\nonumber \\\\ & & \textcolor{black}{\frac{\partial q_i}{\partial t}+\frac{\partial q_i u_k}{\partial x_k}+\frac{5}{2}p\frac{\partial rt}{\partial x_i}+\frac{5}{2}p_{ik}\frac{\partial rt}{\partial x_k}+rt\frac{\partial p_{ik}}{\partial x_k}-rtp_{ik}\frac{\partial \ln \rho}{\partial x_k } } \nonumber \\ & & \textcolor{black}{-\frac{p_{ij}}{\rho}\frac{\partial p_{jk}}{\partial x_k}+\frac{7}{5}q_k\frac{\partial u_i}{\partial x_k}+\frac{2}{5}q_k\frac{\partial u_k}{\partial x_i}+\frac{2}{5}q_i\frac{\partial u_k}{\partial x_k}+\frac{1}{6}\frac{\partial s}{\partial x_i } } \nonumber \\ & & \textcolor{black}{=\left[\frac{3}{2}\zeta-\nu_q\left(1+\underbrace{\frac{\left(\omega-2\right)\omega\left(50+\alpha(\omega-10)+7 \omega\right)}{480\left(\alpha\left(70 + 29\omega\right)-110 - 37\omega\right)}a_4}_{\beta_q}\right ) \right]q_i,}\end{aligned}\ ] ] + + 23 i. goldhirsch and t. p. c. van noije , phys . e , 61 ( 3 ) , 3241 ( 2000 ) . j. w. dufty and j. j. brey , j. stat . phys . , 109 ( 3 - 4 ) , 433 ( 2002 ) . j. j. brey , m. j. ruiz - montero , p. maynar , and m. g. de soria , j. physics : cond . , 17 ( 24 ) , s2489 ( 2005 ) . j. j. brey and m. j. ruiz - montero , phys . e , 70 ( 5 ) , 051301 ( 2004 ) . g. a. bird , molecular gas dynamics and direct simulation of gas flows ( oxford science , oxford ) ( 1994 ) . l. d. landau and e. m. lifshitz , fluid mechanics , course of theoretical physics , 6 , ( addison - wesley , reading , 1959 ) . j. j. brey , p. maynar , and m. i. garcia de soria , phys . e , 79 , 051305 ( 2009 ) . j. m. montanero and a. santos , gran . , 2 ( 2 ) , 53 ( 2000 ) . s. tsuge and k. sagara , j. stat . phys , 5 , 5 , 403 ( 1975 ) . r. yano , j. phys . a , 46 ( 37 ) , 375502 ( 2013 ) . t. scheffler and d. e. wolf , granular matter , 4 , 3 , 103 ( 2002 ) . a. santos and j. m. montanero , gran . , 11 ( 3 ) , 157 ( 2009 ) . n. v. brilliantov and t. p@xmath69schel , europhys . lett . , 74 ( 3 ) , 424 ( 2006 ) . t. p. c. van noije and m. h. ernst , gran . matt . 1 ( 2 ) , 57 ( 1998 ) . h. grad , comm . pure and appl . , 2 , 331 ( 1949 ) . a. santos , physica a , 321 ( 3 ) , 442 ( 2003 ) . v. garzo and j. m. montanero , physica a , 313 , 336 ( 2002 ) . : + fig . 1 : @xmath25 versus @xmath4 for the ihs ( left frame ) @xmath26 versus @xmath4 for the ihs ( right frame ) . + fig . 2 : @xmath25 versus @xmath4 for the ivhs with @xmath7 ( left frame ) . @xmath26 versus @xmath4 for the ivhs with @xmath7 ( right frame ) . + fig . 3 : @xmath25 versus @xmath4 for the ims ( left frame ) . @xmath26 versus @xmath4 for the ims ( right frame ) 4 : @xmath164 versus @xmath4 ( left frame ) and @xmath165 versus @xmath4 ( right frame ) for the ihs . + fig . 5 : @xmath164 versus @xmath4 ( left frame ) and @xmath165 versus @xmath4 ( right frame ) for the ivhs with @xmath7 . + fig . 6 : @xmath164 versus @xmath4 ( left frame ) and @xmath165 versus @xmath4 ( right frame ) for the ims . : @xmath198 and @xmath199 versus @xmath200 for the ihs . + fig . 8 : @xmath198 and @xmath199 versus @xmath200 for the ivhs with @xmath7 . + fig . 9 : @xmath198 and @xmath199 versus @xmath200 for the ims . 10 : @xmath212 , @xmath213 and @xmath217 versus @xmath4 for the ihs , ivhs with @xmath7 and ims . + fig . 11 : @xmath284 versus @xmath4 in cases of the ihs , ivhs with @xmath7 and ims .
we study the resummation of large qcd radiative corrections up to the next to leading logarithmic accuracy to the process ; i.e. , we resum logarithms of the type and ( is the quark mass ) . the only source of all the logarithms to this accuracy is the off - shell sudakov form factor included into the triangle topologies of the one - loop box diagram . we prove that any other configurations of diagrams to this accuracy , either cancel in subgroups or develop a universal on - shell sudakov exponent due to the final quark anti - quark lines . we study the mechanism of cancellations between the different diagrams , which leads to the simple resummed results . we show the cancellation explicitly at three loops for the leading and at two loops for the next - to - leading logarithms . we also point out the general mechanism responsible for it , and discuss how it can be extended to higher orders . mctp-02 - 66 + * on the resummation of large qcd corrections to * + _ michigan center for theoretical physics + randall laboratory of physics , university of michigan , + ann arbor , michigan 48109 - 1120 , usa _ e - mail : [email protected] +e - mail : [email protected] +e - mail : [email protected]
hep-ph0212115
future linear colliders are expected to reveal the answers to many questions of modern particle physics . one of these concerns the the physics of the higgs particle and origin of electroweak symmetry breaking . the neutral scalar higgs boson is an important ingredient of the standard model ( sm ) and is the only sm elementary particle which has not been detected so far ( see for a review @xcite , @xcite ) . the lower limit on @xmath7 of approximately @xmath8 gev at @xmath9 c.l . has been obtained from direct searches at lep @xcite . current experiments are concentrating on the possibility of finding a higgs particle in the intermediate mass region @xmath10 in this region it decays mainly to a @xmath11 pair . the _ photon mode _ of the future linear collider ( lc ) , namely the collisions of the energetic polarized compton photons , will be used for the production and for the study of the higgs particle . in the intermediate mass range , the main production process is @xmath12 qcd as well as electroweak radiative corrections to this process have been studied very well and have been found to be small in this region @xcite . the main challenge , however , is to get under control the background process @xmath13 which gets extremely large qcd corrections . in this paper we discuss the process of the quark anti - quark production in the photon mode of the lc , @xmath0 . the amplitude for this process in the scalar channel contains large double logarithms at @xmath14 . at very high energies , the large logarithms spoil the perturbative predictions . therefore it is mandatory to develop a clear resummation procedure for this double and single logarithmic terms . let us stress , that the main interest in the process @xmath0 comes from the fact ( but is not limited to it ) that it represents the dominant background for the production of the higgs particle , @xmath15 @xcite . in fact , our motivations for this study are twofold : + ( 1 ) a detailed study of the process @xmath0 is very important due to phenomenological reasons mentioned above ; and + ( 2 ) in addition to that , the quark anti - quark production in photon collisions , being one of the simplest process in qcd , is important in its own right , e.g. for studying and understanding qcd effects . the born cross section for the polarized @xmath16 production in the scalar channel ( at @xmath17 ) , where the higgs will be studied as well , is suppressed by @xmath18 @xcite , ( here @xmath19 is the center of mass energy of the initial photons ) . however , the perturbative qcd corrections contains the large double logarithms of the form @xmath20 , which give a contribution to the cross section which is of the same order as the born contribution at high energies . the presence of the large correction was noticed by jikia in @xcite . the double logarithmic nature and the origin of these corrections were studied in @xcite . the authors studied the process to one and two loop accuracy . later , the form of the resummed results for the double logarithms has been argued in @xcite . these authors also claimed that the double logarithms have a``non - sudakov '' origin . in this paper , we first present an alternative way of understanding the resummation procedure for the double logarithms . the general idea of our approach , is that the only source of double logarithms is the off - shell sudakov form factor included in the triangle topologies of the one - loop box diagram . we have proved that the other types of the higher loop diagrams will either cancel in subgroups or develop a universal on - shell sudakov exponent due to the final quark anti - quark lines . in addition to that , ( 1 ) we extend this analysis to the next - to - leading - logarithmic ( nll ) accuracy , and ( 2 ) we study the mechanism of cancellations alluded to above , between the different diagrams which leads to very simple resummed results . we demonstrate the cancellation explicitly at three loops and point out the mechanism responsible which then allows for the generalization to higher loops , all up to the next to leading logarithmic accuracy . as an aside , we argue that all the large logarithms up to the next to leading level are related to the sudakov ones . this includes not only the leading ones of the form @xmath1 but also the next to leading ones of the form @xmath2 ( @xmath3 is the quark mass ) . it is this fact together with an understanding of the cancellation mechanism which allows us to develop an easy resummation procedure . the paper is organized as follows . in section 2.1 we discuss the one loop diagrams and explain which topologies are important . in section 2.2 we present a procedure for the resummation of the double logarithms . in section 2.3 we extend our analysis to the next - to - leading logarithmic level . in this section we consider all possible topologies and give the final result of the resummation . section 3 is devoted to the study of the different cancellations which are responsible for the simple resummed results of section 2.2 for the double logarithmic case . in section 4 we tackle the case of the form factors to the next to leading logarithmic accuracy and justify the results given in section 2.3 . the paper ends with some conclusions and discussions . we study the process of quark anti - quark pair production in photon - photon collisions @xmath21 in the scalar channel @xmath17 , at very high energies compared to the mass of the quark @xmath3 , and at large scattering angles @xmath22 the radiative corrections to this helicity amplitude contain large double and single qcd logarithms of the form @xmath23 , in the limit eq.([lim1],[lim2 ] ) . it has been shown in @xcite that the large double logarithms ( dl ) have a sudakov - like nature and can be extracted from the total result by identifying special kinematic regions . it was demonstrated that , only the box diagram contributes , and that this box diagram can be reduced to three different effective diagrams with triangle topology . we label them as topologies a , b , c following @xcite , see fig.([topologies ] ) . these effective diagrams result from the box diagram when one of the four propagators is hard , two collinear and one soft ( the soft line can be either the gluon or the quark propagator ) . hence one can shrink the hard propagator of the box diagram to a point , because it does not depend on the soft loop momenta anymore . the hard momenta would be just an ultra violet cutoff in this situation . in order to introduce notations and to get an idea about the structure of the dl , let us consider one loop calculations in the dl approximation . also , we will use these results in the next section for the construction of the dl - resummed result . as is well known , the double logarithms come from the region of soft loop momenta . hence , one can always write a factorized form for the amplitude in the dl approximation : @xmath24 the factor 2 in the last equation represents two identical contributions @xmath25 of the topology c. the amplitude then reads @xmath26 we turn now to the calculation of the form factors @xmath27 . it is convenient to use the standard sudakov technique for calculating dl contributions , as described in @xcite , and introduce sudakov parameterization @xcite . one starts with decomposing the soft momenta in terms of those along the hard external momenta , for example @xmath28 and transverse to them @xmath29 the dl contribution comes only from the region @xmath30 for different topologies it is convenient to use different decompositions of the soft momenta : @xmath31 the loop integration in terms of the new variables reads @xmath32 the integration over the transverse momenta of the soft quark is performed by taking half of the residues in the corresponding propagator , @xmath33 here @xmath34 is a small ( fictitious ) gluon mass . in this manner the one loop amplitude may be calculated in the dl approximation and topology a gives , @xmath35 topology b and c have following form at one loop in the dl approximation @xmath36 these results were derived first in @xcite and will be used in our calculation later on . the main idea behind our method is that the only origin of the double logarithms is the off - shell sudakov form factor included in the effective one loop triangle diagrams a , b and c. contributions from other types of diagrams will either cancel in subgroups or develop a simple on - shell sudakov exponent due to the final quark anti - quark pair . the cancellations amongst the diagrams will be discussed in the next section where the general mechanism is elucidated . here we study the dl resummation in the topologies a , b and c. the diagrams of topology a represent a simple case . here we do not have any `` hard '' dls , but only the `` soft '' ones , which develop the usual on - shell sudakov form factor from the gluon exchanges between the final quark antiquark pair . it is well known , that the on - shell dl sudakov form factor in qcd is the exponent of the one loop result @xmath37 with @xmath38 given in eq.([1loopa ] ) . the resummation of the dl terms coming from the diagrams of topology b is more involved . first , we have to account for the `` soft '' on - shell exponent from the final quark antiquark re - scattering , similar to the one in topology a , eq.([topa ] ) . second , we observe a new element , namely the `` hard '' double logarithms from quark anti - quark re - scattering inside the one loop diagram . the interacting particles ( quarks in this case ) are hard and slightly off - shell , @xmath39 . according to our general idea , the resummation of the `` hard '' dl can be obtained by including the off - shell sudakov form factor into the loop , and accounting only for those momenta which do not destroy the dl nature of the one loop result . we recall that the off - shell sudakov form factor , is a vertex of the production of the quark and anti - quark with small virtualities @xmath40 and @xmath41 and has a form @xcite @xmath42 assuming that @xmath43 now the momenta @xmath44 and @xmath45 become loop momenta , i.e.,they depend on @xmath46 @xmath47 because of the dl approximation , the kinematic region of interest is restricted by the inequalities @xmath48 taking into account the result eq.([sudb ] ) and including it into eq.([1loopb ] ) we get @xmath49 we next transform the exponent into a power series and find that the integral of the @xmath50th term will be of the form @xmath51 the final result at dl accuracy reads @xmath52 with @xmath53 . the index @xmath54 shows that the order of the amplitude is @xmath55 . we can clearly identify the separate contributions of the fixed orders in @xmath56 . on the other hand , if @xmath57 is large all terms in the series are important , giving altogether some analytic function @xmath58 . this function is identified with a hyper - geometric function @xmath59 , namely @xmath60 these functions are exact answers in @xmath61 in the dl approximation . for large values of the parameter @xmath62 the function @xmath63 has the following asymptotic , @xmath64 thus we see that despite the fact that perturbation theory blows up at large @xmath61 , the resummed result gives a smooth well defined function . topology c differs from b only through the color structure at the dl level . at the end of the calculations the answers for topologies b and c are similar and are related by the simple substitution : @xmath65 combining all results for the three topologies we have @xmath66 with the functions defined above . we must stress that there are other diagrams which can give dl contributions . fortunately , all of them cancel in the certain groups , as will be discussed in section 3 . this cancellation has a very simple physical interpretation ( see sec . 3 ) . our results are written in two different representations . we have checked that they are in agreement with @xcite , where the authors used a completely different method , not mentioning the off - shell sudakov form factors at all . we believe that our approach is more transparent . it shows that the new double logarithms are of sudakov type , and , therefore , the resummation procedure becomes simple . another advantage of our approach is that it opens up the possibility for resumming the single logarithms as well . it is possible to develop this approach to achieve next - to - leading - logarithmic accuracy . first , the factorization formula should be modified in order to take into account single logarithms . the amplitude reads at nll approximation @xmath67 with the function @xmath68 which can be extracted from the explicit calculations performed in @xcite by expanding the result at @xmath69 . @xmath70 is the on - shell sudakov form factor to the next - to - leading logarithmic accuracy @xcite . second , in order to calculate the form factors @xmath71 and @xmath72 at nll , we need an expression for the sudakov form factor also to this accuracy . we turn now to the calculation of the form factors @xmath71 and @xmath72 . in fact , such an analysis already exists in the literature @xcite for the case when the two external fermion lines are off - shell by the same amount , i.e. , @xmath73 : @xmath74 for our purposes , we need to extend the analysis to take into account that @xmath75 . it is easy to see that for the region , eq.([lim1 ] ) , the proof of factorization and exponentiation given in @xcite goes through with straightforward changes . the major change involves the normalization of the coupling . we have studied one and two loop diagrams for the @xmath76 production vertex with slightly off - shell quarks with the following result : @xmath77 with the normalization of the coupling constant determined to be @xmath78 . we show only the double and single ir logarithms in the exponent of eq.([slresult ] ) . in order to understand this normalization , we have to consider the diagram shown in fig.[rge ] where we keep track of the @xmath79 dependent pieces only since they are separately gauge invariant . such diagrams can be accounted for by considering the following gluon propagator @xmath80 where @xmath81 is the vacuum polarization by the gluon ; at the one loop level it is simply @xmath82 , @xmath83 being a scheme - dependent constant ( @xmath84 scheme @xmath85 ) . the diagram , fig.[rge ] , corresponds to the first term in the expansion of the gluon propagator in @xmath56 . the @xmath79part of this result is as mentioned earlier a gauge invariant subset of the complete set of two loop diagrams . because the effects of the running coupling gives only single logarithmic terms it is enough to consider the remaining integrals to dl accuracy . namely , we may trace only the terms proportional to @xmath86 from fig.[rge ] . at the dl accuracy the spin structure of the amplitude is simple , so that one needs to consider the scalar integral only , @xmath87 to evaluate this , we consider a slightly more general integral @xmath88 which after expanding in @xmath89 will give us the desired integral @xmath90 . using feynman parameters , this integral is reduced to @xmath91,\end{aligned}\ ] ] with @xmath92 the function @xmath93 has two zeros , @xmath94 : @xmath95 for very small virtualities , @xmath96 , the roots are simplified to @xmath97 . expanding the integrand of @xmath98 in @xmath89 up to second order we have @xmath99 \big).\end{aligned}\ ] ] the final integration over @xmath100 is simple , the result is @xmath101 we see , that the first term in this equation reproduces the dl result from eq.([sudb ] ) and eq.([slresult ] ) . it can be checked , that the second term suggests the normalization of the coupling constant to be , @xmath102 . indeed , returning to the integral @xmath90 , we find @xmath103.\end{aligned}\ ] ] it is clear that the last logarithm , containing the @xmath104 term , can be absorbed into the running coupling , giving @xmath105 with the normalization point @xmath106 . the exponentiation of the integral @xmath90 will give us the final off - shell sudakov exponent , eq.([slresult ] ) . in order to get single logarithms in eq.([slresult ] ) we have to include the numerator and the spin structure . we do not present the details of these calculations here . instead we note , that all logarithms we have accounted for are of infrared origin , @xmath107 . we do not show the uv logarithms which come as a result of the renormalization of the quark mass . such terms can be omitted if the quark mass in the leading order result is normalized at a large scale @xmath108 . the formula eq.([slresult ] ) reproduces the expression for the sudakov form factor at non - equal virtualities at dl accuracy derived by carrazone et . al . in @xcite , eq.([sudb ] ) , as well as at nll with equal virtualities @xmath109 derived by smilga in @xcite , eq.([smilga ] ) . in addition , the normalization point @xmath102 that we find reproduces that of the nll results with equal virtualities @xmath109 derived by smilga in @xcite , eq.([smilga ] ) . this scale , @xmath102 , has a very transparent origin . the vertex of the interaction of a soft gluon with an off - shell quark ( @xmath40 ) is described by the coupling @xmath110 . in the situation of gluon - exchange between two quarks with different virtualities , we have an effective coupling @xmath111 . using the running of the coupling @xmath112 , at one loop level , @xmath113 , we will find that the effective coupling @xmath111 is reduced to @xmath114 , which coincides with our previous results . as a next step , we include this form factor inside the one loop triangle diagram and calculate the last loop integration with the form factor which now accounts for all large logarithms to nll accuracy . the final result for the next - to - leading - logarithmic form factor reads @xmath115 with @xmath116 from eq.([dlresult ] ) and @xmath117 with @xmath118 , @xmath79 is the number of light flavors @xcite . topology c at nll gives a slightly different result . in the previous section where we have already studied the dl result , we saw that the `` hard '' sudakov form factor is developed as a result of the re - scattering of a hard quark on the hard gluon . let us start with the sudakov form factor for the quark - gluon vertex with an off - shell outgoing quark of virtuality @xmath40 and a gluon of virtuality @xmath41 . the result for the reduced form factor is @xmath119 where the coefficient @xmath120 is related to the wave function renormalization of the gluon . now we can obtain a resummed result by substituting this form factor in the known one loop integral of topology c. the dl result coincides with that of topology b ( as we discussed previously , after accounting for the substitution @xmath121 ) . the result for next - to - leading - logarithmic form factor reads @xmath122 the final result for the amplitude is eq.([ampnll ] ) together with eq.([rep1],[finalb],[finalc ] ) . that concludes our derivation of the nll form factors . we emphasize that throughout this subsection we have taken into account all logarithms of the sudakov type only . in section 4 we argue that to the nll level this is justified because of cancellations between diagrams that is similar to the dl case . in other words , logarithms of the nll type coming from other sources ( than the sudakov form factors ) cancel between contributions of subsets of diagrams . the simple expressions for the resummed results given in the previous section are a consequence of the cancellation of double logarithmic terms which are not of the sudakov type . in fact for the exponentiation to work there has to be a large number of cancellations between subgroups of diagrams . such cancellations have been discussed in the literature explicitly at the two loop @xcite and the three loop levels @xcite . we agree with the results of these papers however the new element we would like to add here is to point out the general mechanism which is responsible for these cancellations . we discuss this mechanism on various examples and write down identities which guarantee the cancellations to higher loop orders . the relevant cancellations at the double logarithmic level are due to a single mechanism - the dipole mechanism . two fermions with opposite charges form a dipole when separated by a short distance . when observing from far away , or probing by means of a gluon with long wavelength , one finds the dipole system to be neutral to leading order . to be more specific , the total amplitude for exchanging a soft gluon with two such oppositely charged fermions , as shown by the sum of the two generic feynman diagrams in fig . ( [ generic ] ) , vanishes in the infrared sensitive region ( double and single logarithms ) . this infrared sensitive region is that of a soft gluon , i.e. , a gluon with all components of the 4-momentum small , @xmath123 , @xmath124 this soft region is responsible for the double logarithms . the cancellation can be shown explicitly by employing the grammar - yennie @xcite decomposition and the ward identity.to illustrate the mechanism let us consider a special case shown in , fig . ( [ mixed ] ) . the amplitude for the first diagram is @xmath125 where we have explicitly written down only the factors relevant for the cancellation and denoted the rest of the corresponding expressions for the diagram by @xmath126 and @xmath127 . their explicit form are of no interest as long as they are common to both the diagrams , @xmath128 and @xmath129 . employing the grammar - yennie decomposition we write for the gauge boson propagator , @xmath130 and keep only the term @xmath131 , i.e. , we make the replacement @xmath132 which leaves the infrared behavior unchanged @xcite , we obtain @xmath133[(k_1-l - k)^2-m^2 ] } \gamma_2 u(p_1 ) \frac{1}{k^2(p_1 \cdot k)},\end{aligned}\ ] ] where we have applied the ward identity on the outgoing fermion line . summing the two diagrams together , we have @xmath134,\ ] ] where the relative minus sign arises from the fact that the soft gluon interacts with two oppositely charged fermions and @xmath135[(k_1-l - k)^2-m^2 ] } \gamma_2 u(p_1 ) { 1 \over k^2}.\ ] ] in order to prove the sum , @xmath136 , vanishes , we use sudakov parameterization , @xmath137 and proceed with the soft approximation , @xmath138 we arrive at @xmath139 = \frac{s}{2 } \left[\frac{1}{(\beta+\alpha)\frac{s}{2}-p_{1\bot } \cdot k_{\bot } } - \frac{1}{(\beta+\alpha)\frac{s}{2}-p_{2\bot } \cdot k_{\bot } } \right].\ ] ] again , in the soft approximation , @xmath140 and @xmath141 are negligible compared to @xmath142 due to the @xmath143-function @xmath144 arising from the pole of the gauge boson propagator . this can be readily seen in the expansion @xmath145 ^ 2 + \left[\frac{p_{1\bot } \cdot k_{\bot}}{(\beta + \alpha)\frac{s}{2}}\right]^4+\cdots\right\},\ ] ] since the odd terms in @xmath140 vanish when integrating over the angle between @xmath146 and @xmath147 . further , @xmath148^{2n } = \left\ { \begin{array}{ll } 2^n \left ( \frac{\beta}{\alpha } \right)^n \frac{p_{1\bot}^{2n}}{s^n } \cos^{2n}\theta & \alpha \gg \beta \\ \\ 2^n \left [ \frac{\alpha \beta}{(\alpha+\beta)^2 } \right]^n \frac{p_{1\bot}^{2n}}{s^n } \cos^{2n}\theta & \alpha \sim \beta \\ \\ 2^n \left ( \frac{\alpha}{\beta } \right)^n \frac{p_{1\bot}^{2n}}{s^n } \cos^{2n}\theta & \beta \gg \alpha \end{array } \right.\ ] ] and power suppression arises from the term @xmath149 with @xmath150 , @xmath151 with @xmath152 and @xmath153 for all the three cases . thus we have seen that the two diagrams indeed cancel each other in the double logarithmic approximation . the explicit calculation is applicable to the generic diagrams in fig . ( [ mixed ] ) if we recall that what really matters here is nothing but the exchange of the soft gluon with the fermion - antifermion pair . the same conclusion holds for a exchanging collinear gluon in the single logarithm approximation as we will discuss later . we will next discuss how the dipole mechanism for cancellation works in certain 3 loop examples . more specifically , we will show the cancellation of double logarithms in the diagrams @xmath154 of topology c. for simplicity of presentation we first discuss only the abelian case omitting all group theory factors and in the next section the extension to the non - abelian case will be presented . these diagrams can be grouped in twos or threes according to the cancellation as seen below . 1 . @xmath155 2 . @xmath156 3 . @xmath157 4 . @xmath158 5 . @xmath159 6 . @xmath160 these six groups cover all the abelian - like diagrams of fig . [ topc ] . * group 1 . * + we start with group 1 . the amplitudes @xmath161 and @xmath162 may be written as @xmath163 m^{(1)}_{\mu}\end{aligned}\ ] ] @xmath164 m^{(1)}_{\mu}\end{aligned}\ ] ] where @xmath165 is some one loop sub - diagram , which is identical in both @xmath161 and @xmath162 . such a representation of the diagrams is very convenient because it turns out that the cancellation of double logarithmic contributions could be traced without explicit computations of the separate diagrams . instead we observe the cancellations by comparing diagrams in some subgroups and using kinematic simplifications specific to the double logarithmic approximation . because the final quark line is hard we might write @xmath166 . the momenta @xmath167 must be softer than @xmath168 , otherwise no double logarithms can appear . then we find that integrand of the diagram @xmath161 contains @xmath169 . therefore decomposing the momenta @xmath170 into @xmath171 and @xmath45 we find that only the component parallel to @xmath171 contributes . we remind the reader that @xmath172 . introducing @xmath173 , the unit vector parallel to @xmath171 , @xmath174 we obtain for the sum of the diagrams @xmath175 m^1=0\end{aligned}\ ] ] thus the leading contribution in the integrand , responsible for the double logarithmic asymptotic , cancels . this is shown in fig . [ generic ] . as we will see below the mechanism of the cancellation is very general and is related to the dipole mechanism . indeed we may define an impact factor as a sub - diagram , where a real photon produces a nearly on - shell quark and antiquark , which interacts with the soft gluons . the remarkable fact is that in the configuration responsible for double logarithms both particles move in the same direction . because the quark momenta is harder than the gluon momenta , only the direction of its movement plays a role . @xmath176 further , because the @xmath11 system does not have a net color charge , being produced by a photon , the first term in the multi - pole expansion of interaction of this system with any number of soft gluons is the color - dipole moment . this leads to the power suppression of the sub - diagram and the loss of the double logarithms . * group 2 . * + we continue with group 2 which constitutes diagrams @xmath177 and @xmath178 . the corresponding amplitudes could be written as @xmath179 m^{(2)}_{\mu}\end{aligned}\ ] ] @xmath180 m^{(2)}_{\mu}\end{aligned}\ ] ] where as before , @xmath181 is some sub - diagram , and we may write @xmath182 . again , the momenta @xmath167 is softer than @xmath168 and only the component of @xmath168 parallel to @xmath171 contributes . we have @xmath183 m^2=0\end{aligned}\ ] ] we thus see that the leading contribution to the integrand , which gives rise to the double logarithmic asymptotic , cancels . * group 3 . * + we turn now to the amplitudes for the diagrams @xmath184 which may be written as @xmath185 m^{(5)}_{\mu,\nu}\end{aligned}\ ] ] @xmath186 m^{(6)}_{\mu,\nu}\end{aligned}\ ] ] @xmath187 m^{(7)}_{\mu,\nu}\end{aligned}\ ] ] it is clear from the diagrams that we can write to the accuracy needed for the relevant subgraphs , @xmath188 . the momenta @xmath189 are softer than @xmath168 and only the component of @xmath168 parallel to @xmath171 contributes . after simplifications , we obtain @xmath190 = 0\end{aligned}\ ] ] again , we see that the leading contribution to the integrand , which could give rise to the double logarithmic asymptotic , cancels . * group 4 . * + the amplitudes for @xmath191 could be written as @xmath192 m^{(8)}_{\mu,\nu}\end{aligned}\ ] ] @xmath193 m^{(9)}_{\mu,\nu}\end{aligned}\ ] ] @xmath194 m^{(10)}_{\mu,\nu}\end{aligned}\ ] ] with @xmath195 to the desired accuracy for the sub - diagrams . the momenta @xmath189 are softer than @xmath168 and only the component of @xmath168 parallel to @xmath171 contributes . after simplifications , we obtain @xmath196 = 0\end{aligned}\ ] ] we see that the leading contribution to the integrand cancels . the mechanism of the cancellation is identical to the one of group 3 . it is important to note specially for purposes of the next section that @xmath197 . hence one might find another 2 sets of 3 diagrams in groups 3 and 4 , which cancel in groups of three . * group 5 . * + the amplitudes for @xmath198 could be written as @xmath199 m^{(15)}_{\mu,\nu}\end{aligned}\ ] ] @xmath200 m^{(12)}_{\mu,\nu}\end{aligned}\ ] ] @xmath201 m^{(11)}_{\mu,\nu}\end{aligned}\ ] ] with @xmath202 the relevant sub - diagrams . the momenta @xmath189 have to be softer than @xmath168 and only component of @xmath168 parallel to @xmath171 contributes . after simplifications , we obtain @xmath203 = 0\end{aligned}\ ] ] we again observe the cancellation of the leading contribution to the integrand . the mechanism of cancellation is identical to the one of group 3 and is the just the dipole mechanism . * group 6 . * + finally the amplitudes for @xmath204 could be written as @xmath205 m^{(16)}_{\mu,\nu}\end{aligned}\ ] ] @xmath206 m^{(14)}_{\mu,\nu}\end{aligned}\ ] ] @xmath207 m^{(13)}_{\mu,\nu}\end{aligned}\ ] ] where as before we write to this accuracy , @xmath208 . the momenta @xmath189 have to be softer than @xmath168 and only the component of @xmath168 parallel to @xmath171 contributes . after simplifications , we obtain @xmath209 = 0\end{aligned}\ ] ] which explicitly shows the cancellation of the leading terms . the mechanism of the cancellation is identical to the one for the earlier groups and is the dipole mechanism . note that for these two groups as well , @xmath210 after interchanging @xmath167 and @xmath211 . hence one might expect to find another two sets of three diagrams in groups 5 and 6 , which cancel in groups of three . in conclusion , we have shown that all diagrams @xmath212 cancel in group of two or three due to the dipole mechanism . the inclusion of the non - abelian contributions do not pose any new difficulties . in fact , there are many simplifications because the final state quark and antiquark are produced by photons and hence carry no net color charge . when the color factors are included the cancellations take place between diagrams with the same such factors . for example , consider the cancellation between the diagrams of group 1 . here @xmath161 and @xmath162 have the same group factor since the only difference between them is an abelian vertex . the same is true for group 2 . thus the cancellation between the diagrams of groups 1 and 2 proceed as in the previous section even in the full non - abelian theory . consider next , group 3 of the previous subsection . for this case we will discuss the cancellation in two different ways . in the first method we note that only the group theory factor for the diagram @xmath213 is different from that of @xmath214 and @xmath215 . explicitly , for @xmath213 the factor is @xmath216 while for the other two it is @xmath217 . all other factors are the same as in the previous section with the rule that the group factors are overall multiplicative . in particular , the sub - diagrams @xmath218 contain no color matrices . now we can write the color factor of @xmath213 as the one for @xmath214 and @xmath215 plus a left over term which is @xmath219 . consider now the diagrams in group 4 of the previous section . here the group factor associated with @xmath220 is @xmath217 which is different from @xmath221 and @xmath222 which is @xmath223 . the difference now is @xmath224 . using the fact that @xmath225 we see that the left over pieces from groups 3 and 4 cancel each other out . it is easy to check that the same applies to the combination of groups 5 and 6 . it is clear from the above discussion , that it is much more straightforward in the non abelian case to change the assignment of the diagrams into different groups . it is easy to see that apart from the group theory factors , and in the soft approximation , the diagrammatic expressions for @xmath213 and @xmath220 of the previous subsection ( from groups 3 and 4 ) are the same ( using of course @xmath225 ) . from groups 5 and 6 of the previous subsection the same applies to the expressions for @xmath226 and @xmath227 ( in this case using @xmath228 ) . thus the following assignment of the diagrams in the non - abelian case into different groups will make sure that all diagrams in a group have the same group theoretical factors : 1 . @xmath229 2 . @xmath156 3 . @xmath230 4 . @xmath231 5 . @xmath232 6 . @xmath233 the new assignment does not change the results for the abelian case and now the cancellation in the non - abelian case proceeds within each group . the only non - trivial result one must use is the equalities @xmath225 and @xmath228 . these are always seen to be true in the soft approximation . in the present grouping , the diagrams in each group are seen to be related to each other by a cyclic permutation of the gluon lines . such a cyclic permutation leaves the color factor unchanged . the cancellation between the other non abelian diagrams shown in fig.[nonab ] also proceeds similarly : for example , @xmath234 cancels @xmath235 ; @xmath236 cancels @xmath237 and so on . this discussion now has been set up for generalization to all orders . we see that the cancellation at the 3 loop level discussed in the previous subsection relies on the following : ( 1 ) the soft approximation,(2 ) algebraic identities like @xmath238 and that diagrams in the same group are related by a cyclic permutation of the gluon lines . in the above , @xmath239 is a generic hard momentum and @xmath240 are the soft ones . the soft approximation essentially tells us that a soft gluon does not see spin and more explicitly if @xmath239 generically denotes a hard fermion momentum then we can consistently use,@xmath241 for the hard propagator and @xmath242 for the vertex factors . this will ensure the equalities like @xmath225 needed for the cancellations to occur for our process . as far as the algebraic identities are concerned , they are in fact special cases of the sterman - libby identities @xcite . these identities obtained by considering diagrams related by cyclic permutations of the gluon lines read in general : @xmath243 in the above , @xmath244 and @xmath245 . these identities are easily seen to reduce to eqs.([id_dl ] ) for the cases @xmath246 . thus we see each of the technical ingredients have a generalization to higher loop orders . the physics of the dipole mechanism and the color singlet nature of the final ( and initial ) states combined with the above guarantee the cancellations needed to all orders . the goal of this section is to identify those diagrams that either vanish or cancel some other diagrams at next - to - leading logarithmic order . the diagrams left over are then just those needed in the discussion in section 2.3 . our explicit analysis is only at the two loop level and at the end we argue that the results hold to nll accuracy . in the first subsection , we discuss the regime contributing to the next - to - leading logarithms . we then introduce a power counting technique that enables us to discover the non - vanishing diagrams and make appropriate approximations . armed with these two techniques , we are able to exclude a set of diagrams without actually carrying out the loop integrations . we remind the reader that we work throughout in the feynman gauge . in this gauge we will see that for the process under consideration , the additional logarithms at the nll level must be of collinear origin . in a two - particle scattering process , all momenta lie in the same plane . we can take two of the independent momenta , @xmath247 and @xmath248 , as + and - direction , and denote the third momentum as @xmath239 ( @xmath239 could be either @xmath44 or @xmath45 ) . the loop momentum @xmath249 can be expressed in terms of light cone variables @xmath250 we note that the following integral in the soft regime @xmath251 gives terms like @xmath252 , @xmath253 or @xmath254 . in other words , the soft regime can not give rise to a single logarithm at the one - loop level . consider an n - loop feynman integral with loop momenta @xmath255 . we can decompose @xmath256 in terms of external momenta @xmath247 and @xmath248 , @xmath257 in the soft region , @xmath258 , the feynman integral does not give the next - to - leading logarithm , @xmath259 . this is because , intuitively , each integration over the light cone variables @xmath260 and @xmath261 gives either a logarithm or a power suppression , as exemplified in the previous paragraph . we now prove this statement more rigorously . we first note the following replacement for loop momenta corresponding to a soft line , @xmath262 the rest of the propagators can be categorized into hard , collinear and soft ones . the hard propagators are irrelevant to the infrared sensitivity . the remaining possibilities are collinear to @xmath247 or @xmath248 , soft or collinear to @xmath239 . we discuss them in turn . \(1 ) collinear to @xmath247 or @xmath248 . a boson that is parallel to @xmath247 has a propagator that can be written as @xmath263 the fermionic propagator can be expressed as @xmath264 \(2 ) soft . a bosonic propagator @xmath265 using @xmath266 , and @xmath267 , it is easy to show @xmath268 hence , the expansion below , @xmath269 , \label{soft}\ ] ] converges . the fermionic propagator has a term proportional to @xmath270 , in addition to the expansion above . \(3 ) collinear to @xmath239 . again , the bosonic propagator can be expanded as @xmath271 \label{col - p}\end{aligned}\ ] ] the fermionic propagator may give rise to an additional @xmath272 in the numerator . first , we only include the first term in the series in eq.([soft ] , [ col - p ] ) and apply the following trick @xmath273 this trick reduces the number of different combinations of @xmath260 s and @xmath261 s while splitting one term into two . therefore , when the propagators in case ( 1 ) and only the first terms in the expansions of the soft and collinear - to-@xmath239 propagators are included , the integral can be reduced to a finite sum of the type @xmath274 where @xmath275 s represent @xmath260 s , @xmath261 s or the combinations thereof . it is evident that an integration over each @xmath275 gives either a logarithm or a power suppression @xmath276 . no next - to - leading logarithm can arise in the soft regime . we now include the whole series in eq.([soft ] , [ col - p ] ) , as well as their fermionic counterpart whenever appropriate . the denominators are polynomials of @xmath277 and @xmath278 , @xmath279 the numerators consists of terms proportional to @xmath280 and @xmath281 . the feynman integral is thus of the following form @xmath282 where @xmath283 and @xmath284 are both polynomial functions of their arguments and the spinor structure is not interesting . since @xmath285 is the only vector in the integrand which is not the integration variable , the integral of interest can further be reduced to @xmath286 where we have further left out the tensorial structure of @xmath287 and @xmath288 . by noting @xmath289 where the angles @xmath290 are relative to the vector @xmath291 , the integral can be cast into @xmath292 where @xmath293 and @xmath294 are polynomial functions of their arguments . we have used the @xmath143 function ( eq.([delta - func ] ) ) to perform the integration over @xmath295 and implicitly included the resulting @xmath296 functions in the polynomials @xmath293 . the summation in the second line is due to the expansions in the soft and collinear - to-@xmath239 propagators . it is a convergent series . now we examine the arguments of the polynomial @xmath293 in eq.([poly - sqrt ] ) . the @xmath297 represents various combinations of @xmath277 and @xmath298 that may appear . for such a combination , we split the integration region @xmath299 after such maniputions , we obtain a convergent series @xmath300 with p and q polynomial functions of @xmath301 and @xmath302 . in order to obtain the next - to - leading logarithm , @xmath303 of the integrations have to give logarithm while the last one gives a constant of order 1 . this is impossible for the above integral in the soft region . it follows that at least one of the loop momenta has to be taken out of soft region to correctly reproduce the next to leading logarithmic behavior in the feynman gauge . in order to identify the regime contributing to the next - to - leading logarithmic order at two - loop level , we consider inserting a gluon into the one - loop box diagram . from the previous subsection , we know the inserted gluon has to be collinear . therefore , we will consider in turn the three topologies with an additional collinear gluon inserted to each of them . throughout the following subsections , we always denote the soft loop momentum by @xmath249 and that of the collinear gluon by @xmath304 . we take two momenta @xmath239 and @xmath305 as the basis to decompose the momentum @xmath304 of the collinear gluon , @xmath306 the generic momenta @xmath239 and @xmath305 can be any two of the external momenta @xmath247 , @xmath248 , @xmath44 and @xmath45 . in the so - called collinear region , without loss of generality , we assume @xmath304 parallel to @xmath239 such that @xmath307 in general , all the propagators in a feynman diagram can be characterized as hard ( off - shell ) , soft or collinear ( to a certain direction ) . in order to get the next - to - leading logarithm at two - loop level , which is a double logarithm multiplied with a single logarithm , a feynman diagram has to contain at least four collinear and one soft propagators . specifically , the double logarithmic form factor arises from a soft virtual particle interacting with two ( nearly ) on - shell particles that are flying apart along two different directions . on the other hand , the interaction between two collinearly flying virtual particles gives rise to the single logarithmic form factor . as a result , whenever we have less than four collinear propagators , we can immediately conclude that the feynman diagram does not contribute at the next - to - leading logarithmic level . when we have exactly four collinear and a soft propagator , we only keep terms proportional to @xmath260 in the numerator . the @xmath261 term can be dropped because it cancels a @xmath261 in the denominator and effectively removes a collinear propagator . an example of such a collinear propagator is @xmath308 and when we carry out the integration over @xmath304 , we can pick up a pole from an on - shell propagator as the above one . it follows that @xmath309 and @xmath310 should also be dropped . if we get more than four collinear propagators , we will keep all the terms . however , in general the terms proportional to @xmath260 are suppressed by @xmath311 and thus leave @xmath261 and @xmath310 the leading terms . when inserting a gluon into a box diagram , we will obtain one - loop vertex subdiagrams inside four of the resulting two - loop diagrams . two of the vertex corrections each have two legs ( nearly ) on - shell and the other soft . there is no large scale of order @xmath19 , other than that from the uv cutoff , in such a sub - diagram . therefore , these two vertex functions contain no infrared logarithms . the other two vertex function subdiagrams have two ( nearly ) on - shell and one off - shell legs each . they do contribute single logarithms in all the three topologies . such vertex corrections are shown in fig.[nll_a](g , h ) , [ nll_b](g , h ) , [ nll_c](e , f ) and [ nll_cu](e , f ) . self energy corrections are understood and not explicitly drawn . hereafter , we will omit the diagrams ( and regions of diagram ) that give rise to large logarithms only of ultraviolet origin , until we are ready to run the relevant parameters using the rge . we show some of the diagrams relevant to the next - to - leading logarithms of topology a in fig.[nll_a ] . the shorter fermion lines in the box subdiagrams represent the off - shell propagators , which characterize the topology a. note fig.[nll_a](a ) and ( b ) represent different regions of the same diagram , where the soft gluons are labeled by @xmath249 and the collinear gluons by @xmath304 . the same comments apply to fig.[nll_a](c ) and ( d ) . ( we will follow the conventions that the characteristic off - shell propagator is denoted by the shorter line , @xmath249 for the soft momentum and @xmath304 for collinear momentum throughout the rest of this article . ) the reduced diagram for the first six diagrams , fig.[nll_a](a ) through ( f ) , all consist of a hard vertex with four jets attached to it , as shown in fig.[topa_jets ] . the two jets eventually emerge as quark and antiquark further interact with each other via a soft diagram ( the gluon with momentum @xmath249 ) . one of them consists of the collinear gluon and the quark or antiquark . in addition , these subset of diagrams are gauge invariant , since the reciprocal subset ( consisting of vertex correction and self - energy subdiagrams ) are gauge invariant . we can now invoke results from a general power counting analysis of infrared sensitive contributions ( both soft and collinear ) to a typical wide - angle scattering process @xcite . it was shown there that the logarithmic configuration requires that jet lines are attached to hard vertices by a single line , otherwise there is power suppression . the analysis of @xcite was made in a physical gauge , but it obviously holds for the gauge invariant set discussed above . hence , the sum of fig.[nll_a](a ) through ( f ) does not contain infrared logarithms . the remaining diagrams are fig.[nll_a](b ) and ( h ) , which contribute to the next - to - leading logarithmic order . these are just the diagrams proportional only to @xmath312 contributing to on - shell sudakov form factor to nll accuracy which is extensively discussed in @xcite now we turn to the diagrams in fig.([nll_b ] ) . in fig.([nll_b]e ) , the collinear gluon , labeled by @xmath304 , can be parallel to @xmath44 , @xmath247 or @xmath248 . we discuss them in turn . \i ) parallel to @xmath44 ( @xmath313 ) . there are only three collinear propagators left . this region can be excluded by power counting . \ii ) parallel to @xmath247 ( @xmath314 ) . the soft fermion is the one labeled by @xmath249 , while the fermion labeled by @xmath315 is collinear to @xmath247 . and the fermion with momentum @xmath316 is off - shell . there are exactly four collinear and one soft propagator left . hence we only keep the component of @xmath304 that is parallel to @xmath247 . the numerator of the diagram here is , @xmath317 which implies a vanishing contribution to the next - to - leading logarithmic order from this region . \iii ) the last possible region is @xmath318 , which vanishes due to the similar reason as in ii ) . therefore , fig.[nll_b](e ) vanishes as well as fig.[nll_b](f ) . diagrams fig.[nll_b](a ) - ( d ) can be shown to factorize . take fig.[nll_b](d ) as an example , the numerator of the amplitude is @xmath319 v(p_2 ) , & k \| k_1 \\ \bar{u}(p_1 ) \gamma^{\alpha } \left [ 4 k_1 \cdot ( -p_2 + k ) \right ] ( { /\kern-6pt k}_1 + { /\kern-4pt l } ) { \epsilon \kern-6pt/}(k_1 ) m { \epsilon \kern-6pt/}(k_2)(-{/\kern-6pt k}_2 + { /\kern-4pt l } ) \gamma_{\alpha } v(p_2 ) , & k \| p_2 \end{array } \right . \end{aligned}\ ] ] the factorization is evident now . similar results hold for the other three diagrams . however , they do not contain any large single logarithms . the remaining diagrams , fig.[nll_b](g ) and ( h ) are just the ones included in the resummation discussed in section 2.3 to this order . in topology c , the two - loop s - channel diagrams only proportional to @xmath312 are shown in fig.([nll_c ] ) . the regions contributing to the next - to - leading logarithmic approximation for fig.([nll_c]a , b , c ) are both the gluons being parallel to @xmath45 . note the incoming quark and the outgoing gluon of the hard subprocess @xmath320 are nearly on shell . in addition , the gluon labeled by @xmath304 in the three diagrams are nearly on - shell too . we can expect the contributions only proportional to @xmath312 of the three diagrams to cancel to the leading order , in the same manner as discussed earlier in section 3 . in order to show the cancellation , we decompose @xmath321 each of the three amplitude takes on the form @xmath322 with @xmath323 , @xmath324 , representing amplitudes of fig.([nll_c]a , b , c ) respectively . in this subsection , integrations over @xmath249 and @xmath304 should be understood in the amplitudes . we consistently omit common numerical factors for simplicity . to the single logarithmic approximation , @xmath325 and @xmath326 terms in the numerators can be neglected . hence , @xmath327 and we have put back the @xmath328 in the second line . the summation of @xmath329 in the three amplitudes closely follows the earlier proof using the ward identity ( see section 3.1 ) . we obtain ( for the piece proportional only to @xmath312 ) @xmath330 \\ \nonumber & \cdot { \epsilon \kern-6pt/}(k_2 ) \frac{1}{{/\kern-4pt l}-m } \gamma_{\lambda } v(p_2 ) \frac{1}{(p_2+k)^2 } \frac{1}{k^2 } \frac{1}{(p_2+k+l)^2}.\end{aligned}\ ] ] the first term is suppressed by the quark mass , while the second one vanishes in the single logarithmic approximation by simple power counting . therefore , the contributing diagram in fig.[nll_c ] are d , e , and f. similarly , the diagrams with the u - channel hard subprocess also include six diagrams . noteworthy is that fig.[nll_cu](a ) can also be drawn as a vertex correction to the s - channel diagrams , with the triangle subdiagram being the vertex correction therein . however , we notice they represent different regions . in fig.[nll_cu](a ) , the quark labeled by @xmath249 is soft , whereas in the s - channel diagram , the corresponding one is collinear to @xmath248 . a similar remark applies to fig.[nll_cu](b ) . here , it can be viewed as a vertex correction to the u - channel diagram , which in turn represents a different region . as a summary , the two diagrams shown in fig.[nll_cu](a , b ) represent the region that both the virtual gluons are parallel to @xmath45 , while in their counterparts , the two virtual gluons are parallel to @xmath248 and @xmath45 respectively . obviously , the sum of fig.[nll_cu](a , b , c ) vanishes to the next - to - leading logarithmic approximation in exactly the same way fig.[nll_c](a , b , c ) do . therefore , the contributing diagrams of the topology c at the two - loop level include fig.[nll_c](d , e , f ) and fig.[nll_cu](d , e , f ) only . we note the diagrams contributing to the `` hard '' off - shell sudakov form factor in this topology are mostly non - abelian which are of sudakov type . some of these are shown in fig.[topc_nab ] . this completes our argument justifying that the only source of logarithm at this order are of the `` sudakov '' type . in this section , we argue without detailed proof that the above results hold to the full nll accuracy . we consider only topologies b and c , since topology a , which corresponds to the on - shell sudakov form factor , has been discussed in detail in @xcite . for topologies b and c , consider the insertions of a gluon into the bare diagram . there are two cases : + i ) the gluon momentum is in the soft region . the analysis in section 3 can be applied and it can be factored out . + ii ) if the gluon momentum is in collinear region , the analysis in the previous subsections apply and again , factorization results . all subsequent insertions of gluons in case ( ii ) must be restricted to the soft region . therefore , for these the analysis of section 3 apply . in case ( i ) , we keep inserting gluons and apply the same analysis until we encounter a collinear gluon . then case ( ii ) applies . hence we can conclude that to nll accuracy , the relevant diagrams are all of the sudakov type . in this paper we have studied the resummation up to the next to leading logarithmic level of the qcd radiative corrections to @xmath331 production by photon photon collisions . apart from the phenomenological applications , this problem has inherent interest in providing a theoretical laboratory to study qcd effects . we showed that to the accuracy considered all logarithms are of the sudakov type . on a diagram by diagram basis other types of diagrams do give rise to next to leading logarithms but they cancel amongst each other by the dipole mechanism . we explicitly showed how the dipole mechanism works to 3 loops and outlined an all orders generalization . at the nll level we found that in the feynman gauge the only new logarithms are of collinear origin . the final expression for the resummed amplitude is given in section 2 , eq.([ampnll ] ) together with eq.([rep1],[finalb],[finalc ] ) . we would like to thank g. sterman for discussions . this work was supported by the us department of energy . j. f. gunion and h. e. haber , phys . d * 48 * , 5109 ( 1993 ) . k. hagiwara _ et al . _ [ particle data group collaboration ] , phys . rev . d * 66 * , 010001 ( 2002 ) . k. melnikov and o. yakovlev , phys . * b312 * , 179 ( 1993 ) [ hep - ph/9302281 ] . + a. djouadi , m. spira , j. j. van der bij and p. m. zerwas , phys . * b257 * , 187 ( 1991 ) . + j. g. korner , k. melnikov and o. i. yakovlev , phys . rev . * d53 * , 3737 ( 1996 ) hep - ph/9508334 ] . + k. melnikov , m. spira and o. yakovlev , z. phys . * c64 * , 401 ( 1994 ) i. f. ginzburg , g. l. kotkin , v. g. serbo and v. i. telnov , nucl . instrum . meth . * 205 * , 47 ( 1983 ) . v. telnov , nucl . instrum . meth . a * 355 * , 3 ( 1995 ) . v. s. fadin , v. a. khoze and a. d. martin , phys . d * 56 * , 484 ( 1997 ) [ arxiv : hep - ph/9703402 ] . d. l. borden , v. a. khoze , w. j. stirling and j. ohnemus , phys . d * 50 * , 4499 ( 1994 ) [ arxiv : hep - ph/9405401 ] . g. jikia and a. tkabladze , phys . d * 54 * , 2030 ( 1996 ) [ arxiv : hep - ph/9601384 ] . m. melles and w. j. stirling , phys . d * 59 * , 094009 ( 1999 ) [ arxiv : hep - ph/9807332 ] . v. v. sudakov , sov . jetp * 3 * , 65 ( 1956 ) [ zh . eksp . fiz . * 30 * , 87 ( 1956 ) ] . v. g. gorshkov , v. n. gribov , l. n. lipatov and g. v. frolov , sov . j. nucl . * 6 * , 95 ( 1968 ) [ yad . * 6 * , 129 ( 1967 ) ] . j. carazzone , e. c. poggio and h. r. quinn , phys . b * 57 * , 161 ( 1975 ) . j. c. collins , adv . high energy phys . * 5 * , 573 ( 1989 ) . a. v. smilga , nucl . b * 161 * , 449 ( 1979 ) . r. akhoury , h. wang and o. i. yakovlev , phys . rev . d * 64 * , 113008 ( 2001 ) [ arxiv : hep - ph/0102105 ] . g. sterman , an introduction to quantum field theory , cambridge university press , 1993 . s. b. libby and g. sterman , phys . d * 19 * , 2468 ( 1979 ) . + v. ganapathi and g. sterman , phys . rev . d * 23 * , 2408 ( 1981 ) . r. akhoury , phys . d * 19 * , 1250 ( 1979 ) . g. sterman , phys . d * 17 * , 2773 ( 1978 ) .
grb 090417b was an unusually long burst with a duration of at least 2130 s and a multi - peaked light curve at energies of 15150 kev . it was optically dark and has been associated with a bright star - forming galaxy at a redshift of 0.345 that is broadly similar to the milky way . this is one of the few cases where a host galaxy has been clearly identified for a dark gamma - ray burst and thus an ideal candidate for studying the origin of dark bursts . we find that the dark nature of grb 090417b can not be explained by high redshift , incomplete observations , or unusual physics in the production of the afterglow . assuming the standard relativistic fireball model for the afterglow we find that the optical flux is at least 2.5 mag fainter than predicted by the-ray flux . the _ swift_/xrt-ray data are consistent with the afterglow being obscured by a dense , localized sheet of dust approximately 3080 pc from the burst along the line of sight . our results suggest that this dust sheet imparts an extinction of mag , which is sufficient to explain the missing optical flux . grb 090417b is an example of a gamma - ray burst that is dark due to the localized dust structure in its host galaxy .
1005.1675
the _ swift _ observatory @xcite is a multi - instrument satellite mission that was designed to detect and rapidly localize gamma - ray bursts ( grbs ) . the observatory contains three telescopes : the burst alert telescope ( bat ; @xcite ) , the @xmath1-ray telescope ( xrt ; @xcite ) , and the ultraviolet / optical telescope ( uvot ; @xcite ) . the bat is used to identify grbs and localize them to @xmath3 in the energy range 15150 kev . once the bat has localized a burst _ swift _ slews to point the xrt and the uvot at the burst . the xrt obtains rapid @xmath1-ray localizations to @xmath4 in the energy range 0.210 kev while the uvot obtains localizations to @xmath5 then repeatedly cycles through a set of optical and ultraviolet filters . over the past five years _ swift _ has detected grbs at the rate of approximately 100 per year . almost all of these _ swift_-detected grbs have had @xmath1-ray afterglows , and optical or infrared afterglows have been detected for @xmath6% of them . the remaining @xmath7 40% of the _ swift _ bursts have no reported optical or infrared detections . in general between @xmath840% of grbs are dark grbs @xcite . the gamma - ray burst grb 090417b was detected by the bat as an image trigger at 13:17:23 ut on 2009 apr 17 @xcite . _ swift _ immediately slewed to grb 090417b and narrow - field observations began observing 387 s ( xrt ) and 378 s ( uvot ) after the bat trigger . the bat light curve showed a long period of emission starting 200 s before the trigger ( see figure [ figure : batlc ] ) . several ground - based follow - up observations were made , but no optical or infrared afterglow was detected . @xcite found a source with @xmath9 inside the xrt error circle @xcite and identified it as an sdss galaxy @xcite and the possible host galaxy of grb 090417b . @xcite measured a redshift of @xmath10 for this galaxy based on the forbidden oxygen emission lines [ ] ( 3727 ) and [ ] ( 5006 ) as well as h@xmath11 emission . dark grbs have been an enduring mystery in the lore of grb studies . in the pre-_swift _ era only about one quarter of localized gamma - ray bursts had an optical afterglow ( oa ) detected ( e.g. , * ? ? ? it was sometimes assumed that this was due to the response times of ground - based optical telescopes which allowed most afterglows to fade before they could be observed . _ swift _ however , with its ability to train the xrt and uvot on a burst within @xmath12 s of the initial detection , has been able to provide positions to within @xmath13 for most bursts that have been detected by the bat . the rapid distribution of these positions has allowed ground - based optical and infrared follow - up observations for many of these bursts , often within minutes of the bat trigger . even with such rapid follow - up observations no optical or infrared afterglows have been found for @xmath7 40% of bat - detected grbs , thus demonstrating that observational constraints were not the cause to the dark burst problem @xcite . several attempts have been made to quantify the dark burst problem by establishing operational definitions of dark bursts @xcite . most recently @xcite proposed that dark grbs be defined as those that have @xmath14 measured 11 hours after the burst . they define @xmath15 as the spectral index between the optical and @xmath1-ray regimes and @xmath16 as the spectral index in the @xmath1-ray band . the spectral index is defined by @xmath17 where @xmath18 is the flux density at a frequency @xmath19 . this definition assumes that both the @xmath1-ray and optical afterglows are due to synchrotron radiation from a relativistic fireball . the synchrotron model predicts that the spectral slope will either be the same in both regimes ( if there is no cooling break between them ) or will differ by @xmath20 ( if there is a cooling break between them ) . a difference of @xmath21 can not be explained in this model and thus requires that something be suppressing the flux at optical wavelengths , resulting in a dark burst @xcite . several possibilities have been proposed to explain dark bursts . first , several studies have suggested that dark bursts may be due to extinction along the line of sight , either in the milky way or in the host galaxy ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? @xcite suggest that this is the case for most dark grbs . further , @xmath1-ray observations show that some bursts with no optical afterglows have systematically higher column densities than those with optical afterglows @xcite . this suggests that dark bursts may be suffering from higher extinction than optically - bright bursts . however , there is significant overlap in the distribution of @xmath22 values between dark and bright bursts , so it is not clear if extinction is the sole parameter responsible for the darkness of these bursts . @xcite did panchromatic studies of several grbs that were observed with _ swift _ s xrt and uvot ( i.e. , optically - bright grbs ) in order to probe the extinction in their host galaxies . they found that the combined @xmath1-ray / ultraviolet / optical data for most of these grbs are best fit using a small magellanic cloud ( smc ) dust model @xcite . the extinctions in these galaxies were small , with @xmath23 mag . they also found a mean gas - to - dust ratio of @xmath24 , which is lower than , but similar to , the value found in the smc ( @xmath25 using equation ( 4 ) and table 2 of @xcite ) . @xcite also found that the the gas - to - dust ratio for a large sample of host galaxies of optically - bright grbs is lower than the smc value . however , the gas - to - dust ratios in the host galaxies of dark grbs tend to be more like that of the milky way than that of the smc ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . a second possible origin for dark bursts is that they are grbs located at high redshift , so the lyman break ( or the start of the lyman - alpha forest ) is observed redward of the optical band . the most distant grb observed to date is grb 090423 at @xmath26 @xcite , so some of the dark bursts could simply be at high redshift . however , @xcite have found that less than approximately 19% of grbs have @xmath27 , whereas between 25% and 42% of grbs are dark . therefore , it is unlikely that all of the _ swift _ dark burst lie at very high redshifts . the affects of relativistic beaming of grb emission at very high redshift is poorly understood due to a lack of data on high - redsift bursts . we assume here that it does not affect our discussion in this paper . third , the hypothesis that afterglow radiation is synchrotron radiation may be wrong . this could lead to a different spectrum from what is expected from the relativistic fireball model , in which case the above definition of a dark burst is invalid . long soft grbs @xcite tend to be found in host galaxies that are small , irregular , and have high specific star - formation rates @xcite . @xcite found a median infrared luminosity of @xmath28 for the host galaxies of grbs with optical afterglows and conclude that most hosts are sub - luminous at optical and near - infrared wavelengths . these grb hosts tend to be small galaxies that have intrinsic luminosities similar to those of the dwarf galaxies in the local group . however , some hosts are significantly larger , such as that of grb 990705 @xcite , which was an sc spiral with @xmath29 . to date no long soft grb has been found in an elliptical galaxy . all of the hosts of long soft grbs have morphologies that are consistent with either exponential discs , or irregular structure @xcite . grb hosts follow the size luminosity relation ( e.g. , * ? ? ? * ) and extend it towards lower luminosities @xcite . in spite of their generally small size the host galaxies of long soft grbs tend to have high specific star - formation rates ( ssfrs ) . @xcite found a median ssfr of 0.8 yr@xmath30 , similar to what is seem in lyman - break galaxies . it is important to realize , however , that these properties have been determined using grbs that have usually been localized to sub - arcsecond precisions . in practice this usually means that the sample of grb host galaxies is restricted to grbs that have had optical afterglows . the host galaxies of dark grbs may have different properties . @xcite identify host galaxies for 14 dark grbs and found that they have the same redshift distribution as the hosts of optically - bright grbs . they also found that these hosts do not appear to be significantly different in size , structure , or luminosity from the hosts of grbs with optical afterglows . they did find that a significant fraction of the hosts of dark bursts show evidence for high internal extinction ( @xmath315 mag ) . however , many of their host associations are uncertain , with more than half of the galaxies having a probability of a chance association on the sky of @xmath32 . as we will show in this paper grb 090417b is located @xmath33 from the centre of a bright galaxy with a probability of a chance association of only @xmath34 . this makes grb 090417b one of the strongest associations of a dark grb with a host galaxy . in this paper we present space- and ground - based gamma - ray , @xmath1-ray , ultraviolet , and optical observations of grb 090417b and its host galaxy . we have adopted a standard @xmath35cdm cosmology with a hubble parameter of @xmath36 km s@xmath30 mpc@xmath30 , a matter density of @xmath37 , and a cosmological constant of @xmath38 . for this cosmology a redshift of @xmath39 corresponds to a luminosity distance of 1.806 gpc and a distance modulus of 41.29 mag . one arcsecond corresponds to 6.51 comoving kpc , or 4.84 proper kpc . the look back time is 3.80 gyr . the bat scaled - map , event and dph data were reduced using the standard bat software available through heasarc . the @xmath0 duration of grb 090417b was @xmath40 s in the 15150 kev band . the light curve evolved gradually , and there were no features in the unimaged count rates that might indicate how long the burst persisted in the bat energy range past that time . even at that lower limit on @xmath0 , this was the longest grb ever seen by bat or batse @xcite , although grb 060218 approached this duration with @xmath41 @xcite . the bat light curve for grb 090417b showed four broad , overlapping peaks in the flux at approximately @xmath42 , @xmath43 , @xmath44 , and @xmath45 s , as shown in figure [ figure : batlc ] . the 1-s peak flux of @xmath46 photons @xmath47 s@xmath30 in the range 15350 kev was approximately @xmath48 s. the total fluence from @xmath49 to @xmath50 s was @xmath51 erg @xmath47 in the 15150 kev band . the spectrum was best fitted by a simple power - law function with an average photon index of @xmath52 . the index did not change significantly during the burst . all the above uncertainties are at 90% confidence . the most impressive characteristic in the prompt emission of grb 090417b was its extreme length . the burst was marginally above background ( @xmath53 ) on the 64-s interval starting immediately after a pre - planned slew maneuver ended at @xmath54 seconds , although not at a high enough level to be detectable independent of the later emission . from that time , the flux increased steadily , to be detected as a point source on - board on the interval @xmath55 to @xmath56 s , and finally as a sufficiently significant source to trigger a burst response on the interval @xmath56 to @xmath57 s. although it appeared to be declining toward background levels , the burst continued to be detected until the spacecraft slewed away due to the changing earth constraint at @xmath50 s. the xrt on board _ swift _ began observations on 2009 april 17 at 15:26:33 ut , 387 s after the bat trigger @xcite , and ended on 2009 may 5 , with a total net exposure of 1456 s in windowed timing mode ( wt ) and 83.36 ks in photon counting mode ( pc ) . the xrt observations are listed in table [ table : xrtjournal ] . lllllcr 00349450000 & 2009 - 04 - 17 15:26:40 & 2009 - 04 - 17 23:29:46 & @xmath58 & @xmath59 & wt & @xmath60 + 00349450000 & 2009 - 04 - 17 15:38:43 & 2009 - 04 - 17 23:33:26 & @xmath61 & @xmath62 & pc & @xmath63 + 00349450002 & 2009 - 04 - 18 10:29:48 & 2009 - 04 - 18 11:07:52 & @xmath64 & @xmath65 & pc & @xmath66 + 00349450003 & 2009 - 04 - 18 16:57:16 & 2009 - 04 - 18 20:20:44 & @xmath67 & @xmath68 & pc & @xmath69 + 00349450006 & 2009 - 04 - 19 01:17:02 & 2009 - 04 - 19 12:50:57 & @xmath70 & @xmath71 & pc & @xmath72 + 00349450007 & 2009 - 04 - 20 04:41:58 & 2009 - 04 - 20 22:26:57 & @xmath73 & @xmath74 & pc & @xmath75 + 00349450008 & 2009 - 04 - 21 14:13:55 & 2009 - 04 - 21 23:59:56 & @xmath76 & @xmath77 & pc & @xmath78 + 00349450009 & 2009 - 04 - 22 04:29:42 & 2009 - 04 - 22 14:20:47 & @xmath79 & @xmath80 & pc & @xmath81 + 00349450010 & 2009 - 04 - 23 00:07:52 & 2009 - 04 - 23 22:40:57 & @xmath82 & @xmath83 & pc & @xmath84 + 00349450011 & 2009 - 04 - 24 00:18:14 & 2009 - 04 - 24 22:57:56 & @xmath85 & @xmath86 & pc & @xmath87 + 00349450012 & 2009 - 04 - 25 11:39:16 & 2009 - 04 - 25 23:01:57 & @xmath88 & @xmath89 & pc & @xmath90 + 00349450013 & 2009 - 04 - 26 02:11:18 & 2009 - 04 - 27 23:12:56 & @xmath91 & @xmath92 & pc & @xmath93 + 00349450014 & 2009 - 04 - 27 00:40:17 & 2009 - 04 - 27 21:39:56 & @xmath94 & @xmath95 & pc & @xmath96 + 00349450015 & 2009 - 04 - 28 00:37:57 & 2009 - 04 - 28 23:20:56 & @xmath97 & @xmath98 & pc & @xmath99 + 00349450016 & 2009 - 04 - 29 00:41:42 & 2009 - 04 - 29 23:26:57 & @xmath100 & @xmath101 & pc & @xmath102 + 00349450017 & 2009 - 05 - 01 00:40:42 & 2009 - 05 - 02 23:37:57 & @xmath103 & @xmath104 & pc & @xmath105 + the xrt data were processed using the ftools software package distributed inside heasoft ( v6.6.3 ) . we ran the task xrtpipeline applying calibrations and standard filtering criteria . events with grades 02 and 012 were selected for wt and pc data respectively . the analysis was performed inside the 0.310 kev energy band . the best position for the afterglow was obtained using 6972 s of overlapping xrt photon counting mode data and 9 uvot @xmath106-band images in order to correct the xrt astrometry making use of the xrt uvot alignment and matching to the usno - b1 catalogue , as described by @xcite and @xcite . our best position is ra(j2000.0 ) = @xmath107 ( 13:58:46.62 ) , dec(j2000.0 ) = @xmath108 ( + 47:01:05.4 ) with an error of @xmath109 ( 90% confidence , including boresight uncertainties ) . source photons were extracted from a region with a 30 pixel radius ( 1 pixel = @xmath110 ) , with the exception of pc mode data for sequence 00349450000 where we used an annular region with radii 3 pixel and 30 pixel in order to correct for pile - up . the background was estimated from a circular region with a 50 pixel radius located away from any detected source in the field . when the count - rate dropped below the level of @xmath111 count s@xmath30 we used the sosta tool of ximage , which corrects for vignetting , exposure variations and psf losses within an optimized box , using the same background region . the 0.310 kev light curve thus obtained is shown in figure [ figure : xa_lcurve ] ( top panel ) . the light curve has been rebinned in order to achieve a minimum signal - to - noise ratio ( s / n ) of 3 for each point . the best fit to the light curve is given by a doubly broken power law with indices @xmath112 , @xmath113 , and @xmath114 , and breaks at @xmath115 s and @xmath116 s. the second break could be interpreted as a jet break . however the lack of an optical detection for this burst does not allow us to confirm or deny this hypothesis . also see [ section : dust_scattering ] for an alternate interpretation of the late - time steepening of the @xmath1-ray light curve . the first sequence of @xmath1-ray data is dominated by a strong flare . there is a rise that began before the start of xrt observations and peaked around @xmath117 s. this is followed by a double - peaked flare which started at @xmath118 s , peaked at @xmath119 s and @xmath120 s , and had a decay time of 240 s. further flaring activity is apparent up to at least @xmath121 s. the bottom panel of figure [ figure : xa_lcurve ] shows the hardness ratio for the afterglow measured using the 0.32 and the 210 kev bands . the first orbit ( up to @xmath122 ) shows a hard hardness ratio with spectral evolution that follows the flaring activity . the same is true for the remaining observations , with the hardness ratio slowly decreasing with time . in order to detect possible spectral variations we extracted spectra separately for the two flares , the early decay part from @xmath123 s to @xmath124 s , the steep decay from @xmath124 s to @xmath125 s , and the late decay from @xmath126 s to the end of observations . source and background spectra were extracted using the same regions used for the light curves , with the exception of the late decay spectrum where the count rate was too low to group the energy channels in order to have a gaussian distribution of the number of photons per grouped channel . hence , for this spectrum we replaced the @xmath127 statistic with the @xmath128 statistic @xcite , which can be used whenever the gaussian approximation is not valid , provided that the contamination from background photons is negligible . to ensure this , we extracted photons from a region with a 10 pixel radius . the ancillary response files were generated using the task xrtmkarf , and with the exception of the late decay spectrum we grouped channels in order to have at least 20 photons per bin . spectral fitting was performed using xspec ( v11.3.2 ) . for each spectrum we modelled the galactic absorption using a warm absorber with neutral hydrogen column density @xmath129 @xmath47 @xcite and included an intrinsic warm absorber at the redshift of the host galaxy ( @xmath130 ) . the best model for the spectra of the flares was a power law with a high energy cutoff , while for the remaining parts of the afterglow the best fit was obtained using a single power law . the best fit parameters are given in table [ table : xa_spectra ] . lllllll @xmath58 & @xmath61 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & 446/421 + @xmath61 & @xmath59 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & 604/523 + @xmath139 & @xmath140 & @xmath141 & @xmath142 & & @xmath143 & 39/45 + @xmath140 & @xmath144 & @xmath145 & @xmath146 & & @xmath147 & 15/29 + @xmath64 & @xmath80 & @xmath148 & @xmath149 & & @xmath150 & 13/22 + @xmath82 & @xmath104 & @xmath151 & @xmath152 & & @xmath153 & 32/24 + the fact that the first phases of the xrt observations are best fitted by a cutoff power law with the cutoff energy around 3 kev indicates that the peak energy of the flares was passing through the xrt band at the time . after the flares the spectrum is described by a power law with photon index of @xmath154 . after the first break , up to @xmath155 ks , the spectrum is described by a power law with similar parameters ( @xmath156 and an absorbing column @xmath157 @xmath47 ) . at later times the spectrum softens significantly to a photon index of @xmath158 and shows no significant variations across the break at @xmath159 s up to the end of the observations . the intrinsic absorbing column is in the range @xmath160 @xmath47 . from our analysis of the spectral energy distribution ( [ section : sed ] ) we find that the best fit to the data is given by a milky way extinction law instead of the smc - like extinction law found in many grb host galaxies . therefore , we use the relationship between @xmath161 and extinction in the milky way from @xcite to find that the observed absorbing column corresponds to an extinction @xmath162 mag . this high value for the extinction in the host along the line of sight to the burst is capable of explaining the dark optical nature of this event . swift_/uvot began observations 378 s after the bat trigger @xcite with a 9 s settling mode exposure using the @xmath106 filter . this was followed by a sequence that rotated through all seven of uvot s lenticular filters . the first of these exposures was a 147 s exposure starting at 395 s taken with the white filter . observations continued until 1325874 s (= 15.3 days ) after the trigger using primarily the @xmath163 and ultraviolet filters . no optical or ultraviolet afterglow was detected in any of the uvot data . once it was determined that there was no optical or ultraviolet afterglow for grb 090417b uvot observed primarily with its ultraviolet filters in order to obtain ultraviolet magnitudes for the nearby galaxy that can not be obtained from ground - based observatories . the sdss galaxy noted by @xcite is detected when we coadd the uvot images . we determined a centroid for the sdss galaxy of ra , dec = 13:58:46.66 , + 47:01:04.4 ( j2000.0 ) with an estimated internal uncertainty of @xmath164 and an estimated systematic uncertainty relative to the usno - b1.0 catalogue of @xmath165 @xcite . these uncertainties are the 90% confidence intervals . this galaxy is @xmath33 southeast of the centre of the uvot - enhanced xrt error circle . the field of grb 090417b is shown in figure [ figure : not_field ] . the sdss galaxy is well - isolated from other sources in the field , so there is no contamination from neighbouring sources when doing aperture photometry . we obtained the uvot data from the _ swift _ data archive on 12 may 2009 . this data has had bad pixels identified , mod-8 noise corrected , and has been transformed into fk5 coordinates . we used the standard uvot data analysis software distributed with heasoft 6.6.2 along with the standard calibration data . photometry was done on the sdss galaxy using uvotsource with a circular source aperture of radius @xmath166 for the source region and a circular aperture of radius @xmath167 centred at ra , dec = 13:58:47.57 , + 47:00:33.3 ( j2000.0 ) for a background region . the background region was selected to have similar background properties to those at the location of the galaxy , and to be free of contaminating sources . the uvot photometry of the sdss galaxy is presented in table [ table : uvot_photometry ] . we find no evidence for any change in the magnitude of the galaxy , in any filter , during the course of the uvot observations . our photometry was calibrated to the uvot photometric system described in @xcite . crcc @xmath106 & 1237 & @xmath16820.9 & + @xmath169 & 862 & @xmath16821.4 & + @xmath163 & 22554 & 23.09 & 0.38 + uvw1 & 10850 & 23.12 & 0.56 + uvm2 & 15942 & 22.49 & 0.29 + uvw2 & 30827 & 22.69 & 0.19 + white & 841 & @xmath16822.2 & + the uvot - enhanced xrt position for grb 090417b is ra = 13:58:46.62 , dec = + 47:01:05.4 ( j2000.0 ) which corresponds to galactic coordinates of @xmath170 . the line - of - sight galactic reddening in this direction is @xmath171 mag @xcite . this corresponds to extinctions in the uvot filters of of @xmath172 , @xmath173 , @xmath174 , @xmath175 , @xmath176 , and @xmath177 , and @xmath178 mag . since there is a candidate host galaxy for grb 090417b it is not possible to put a direct faint limit on the luminosity of the optical afterglow beyond the observed magnitude of the galaxy itself ( @xmath179 @xcite ) . in order to constrain the maximum possible luminosity of the afterglow we used the uvot data to look for variability in the galaxy s magnitude . to do this we divided the data into early and late epochs . the early epoch consisted of all the uvot data taken during _ swift _ s initial automated observations of grb 090417b ( obsid 00349450000 ) , which covers times up to 29604 s ( @xmath7 8 hr ) after the bat trigger . this is the period when the optical afterglow is expected to be brightest . the late epoch included all observations later than this . the uvot exposures for each epoch were coadded to produce two deep images for each uvot filter . we photometered the galaxy in each image and recorded the difference between the two magnitudes . if we assume that the first epoch magnitude is due to the host+afterglow and the second epoch magnitude is due to the host alone then the difference between these two magnitudes puts an upper limits on the luminosity of the afterglow . we find no evidence for any change in the magnitude of the host galaxy between the first and last epochs larger than @xmath180 mag . therefore , the afterglow could not have contributed more light than what corresponds to a 0.2 mag increase in luminsity . from this we obtain the following constraints on the magnitude of the optical afterglow at 4 hr (= 14800 s ) after the bat trigger : @xmath181 , uvw1@xmath182 , uvm2@xmath183 , and uvw2@xmath184 . the uncertainty in these estimates is about 1.0 mag . we obtained optical images of grb 090417b on 2009 apr 1718 with the alfosc - fasu mounted on the nordic optical telescope ( not ) . alfosc - fasu was operated in its high gain state ( 0.726 e@xmath185/adu ) . the read - out noise was 3.2 e@xmath185/pixel and the pixel scale was @xmath186/pixel . the not @xmath187-band image is shown in figure [ figure : not_field ] . data reduction was done following the standard procedure for optical ccd data and the individual images were aligned by centroiding on several stars in each exposure . photometry was performed on the combined not images using sextractor @xcite . we calibrated the data using the standard alfosc zero points . we find @xmath188 ( bessel @xmath189 ) , @xmath190 ( bessel @xmath187 ) , and @xmath191 ( interference @xmath192 ) for the sdss galaxy . the galactic extinctions for the not data are @xmath193 , @xmath194 , @xcite , and @xmath195 . we used the methodology of @xcite to estimate the probability that a galaxy would lie within @xmath33 of the centre of the xrt error circle by chance , @xmath196 . this assumes that galaxies are randomly distributed on the sky and that there is no clustering . this assumption is not formally correct as galaxies do cluster . however , an examination of the field of grb 090417b suggests that this grb did not occur in a significant over - density of background galaxies , so distortions in the computed @xmath196 value are likely to be small . @xcite use the galaxy number counts of @xcite to generate the surface density of galaxies with @xmath187 magnitudes brighter than some limiting magnitude . in this case the @xmath187 band is for the keck 10-m low resolution imaging spectrometer ( lris ) @xmath187 filter . we adopt @xmath190 ( [ section : not_data ] ) as the magnitude of the sdss galaxy . using the @xcite formula for an error circle containing a galaxy this yields @xmath197 . this is a fairly small value , which suggests that sdss j135846.65 + 470104.5 is likely to be the host galaxy of grb 090417b . a chance alignment probability of @xmath198 is large enough that one can not rule out the possibility of a misidentification . a visual examination of the not images shows that the xrt error circle includes a significant fraction of the sdss galaxy ( see figure [ figure : not_field ] ) . the observed separation between the @xmath1-ray afterglow and the centre of the galaxy is @xmath33 . at a redshift of @xmath39 this corresponds to a projected separation of 5.23 proper kpc . the galaxy appears point - like in all of our data . the not @xmath187-band data has a seeing fwhm of @xmath199 , which represents an upper limit on the effective radius of the galaxy of @xmath200 proper kpc @xcite found @xmath201 for the galaxy , which implies an absolute magnitude of @xmath202 . this puts the galaxy inside the observed range of grb host sizes and luminosities ( see @xcite , their fig . 4 ) . in order to test our hypothesis that sdss j135846.65 + 470104.5 is the host galaxy of grb 090417b we obtained the _ observations of the @xmath1-ray afterglow that were taken on 2009 may 11 , 24 days ( @xmath203 s ) after the burst ( proposal number : 10900117 , pi : andrew levan ) . an @xmath1-ray source ( consistent with the xrt position ) is clearly detected at the position of the sdss galaxy . the observed flux from this source is consistent with the @xmath1-ray flux that is predicted from the xrt light curve assuming a power - law decay with @xmath204 . this suggests that the _ chandra _ source is the @xmath1-ray afterglow of grb 090417b , and that there is minimal ( if any ) contamination from the galaxy . taken together these arguments strengthen the case that sdss j135846.65 + 470104.5 is the host galaxy . therefore , we conclude that we have identified the host of the dark burst grb 090417b to a high degree of confidence . we used the combined ultraviolet , optical , and infrared photometry of the sdss galaxy to determine its spectral energy distribution ( sed ) . photometry for this galaxy was taken from our observations , the sdss web site , and the various gcn circulars on this burst , and is listed in table [ table : host_photometry ] . we used xspec to fit a power law to the observed sed . the best fit , with no host extinction , has a power - law index of @xmath205 . however , the goodness of fit is only @xmath206 , which is very poor . therefore , we also tried including extinction due to dust in the host galaxy when fitting the sed . dust laws for the small magellanic cloud , large magellanic cloud and the milky way were tried . these led to a significant improvement in the goodness of fit with the best fit being found for milky way extinction in the host galaxy . our best fit has a spectral index of @xmath207 , and a host extinction of @xmath208 . cccl uvw2 & 22.69 & 0.19 & 1 + uvm2 & 22.49 & 0.29 & 1 + uvw1 & 23.12 & 0.56 & 1 + @xmath209 & 23.09 & 0.38 & 1 + @xmath210 & 23.23 & 0.53 & 2 + @xmath211 & @xmath16821.4 & & 1 + @xmath189 & 23.25 & 0.13 & 1 + @xmath212 & 22.02 & 0.16 & 2 + @xmath212 & 22.47 & 0.16 & 3 + @xmath213 & @xmath16820.9 & & 1 + @xmath187 & 21.34 & 0.03 & 1 + @xmath214 & 21.9 & 0.3 & 4 + @xmath214 & 21.62 & 0.09 & 2 + @xmath214 & 21.62 & 0.10 & 3 + @xmath192 & 20.82 & 0.07 & 1 + @xmath215 & 21.41 & 0.11 & 2 + @xmath215 & 21.31 & 0.12 & 3 + @xmath216 & 20.78 & 0.25 & 2 + @xmath216 & 21.49 & 0.30 & 3 + @xmath217 & 20.0 & & 5 + @xmath218 & 18.5 & & 5 + sdss j135846.65 + 470104.5 has a colour of @xmath219 , making it a fairly red galaxy . the typical absolute magnitude for a red galaxy at @xmath220 is @xmath221 @xcite , assuming a cosmology with @xmath222 . for our adopted cosmology that corresponds to @xmath223 . at @xmath10 the rest - frame @xmath189 band corresponds roughly to the observed @xmath214 band . assuming that the galaxy has a power - law spectrum ( see [ section : sed ] ) and @xmath224 , then it has a rest - frame luminosity in the @xmath189 band of @xmath225 where @xmath226 is the rest - frame @xmath189-band luminosity of a typical red galaxy at @xmath39 . this indicates that the host is approximately an @xmath227 galaxy with a total luminosity that is roughly similar to that of the milky way . however , we stress that this result is somewhat uncertain because @xmath228 is highly correlated with the slope of the faint end of the galaxy luminosity function , and with its normalization @xcite . we can estimate the star - formation rate in the host galaxy using eq . 2 of @xcite . at the redshift of the burst 1500 approximately corresponds to the uvot uvw2 filter while 2800 approximately corresponds to the sdss @xmath210 filter . correcting for galactic extinction these magnitudes for the host become uvw2@xmath229 and @xmath230 . these yield star - formation rates ( in @xmath231 yr@xmath30 ) of 0.3 ( 1500 ) and 0.7 ( 2800 ) assuming a @xcite initial mass function , and 0.6 ( 1500 ) or 1.2 ( 2800 ) assuming a @xcite initial mass function . these values assume that there is no extinction in the host galaxy . this is clearly not correct , so they actually represent a lower limit on the star - formation rate of approximately 1 @xmath231 yr@xmath30 . correcting for the mean extinction in the host ( @xmath232 mag ) gives a star - formation rate of @xmath233 @xmath231 yr@xmath30 . however , this corrected values should be treated with caution since extinction in the host is likely to be highly variable and may not be affecting all star - forming regions equally . the host galaxy appears to be roughly similar in luminosity to the milky way , but with a significantly higher star - formation rate . this puts it on the massive end of the grb host galaxy distribution . no afterglow has been detected for grb 090417b at infrared , optical , or ultraviolet wavelengths . changes in the near - ultraviolet flux from the host galaxy between early and late times are consistent with no contribution from an afterglow down to a limiting magnitude of @xmath234 . we are assuming that the bright limit on the afterglow magnitude is valid for all time after four hours after the bat trigger . there is no evidence for a brightening of the host galaxy at late times , so this is a reasonable assumption . therefore we adopt this as an upper limit to the luminosity of the afterglow at 11 hr and compare it to the @xmath1-ray flux at 11 hr . this yields @xmath235 , making it a dark burst @xcite . the @xmath1-ray spectrum at this time is evolving from @xmath236 to @xmath237 , so using the definition of @xcite we find that @xmath238 , also making grb 090417b a dark burst . our limit on @xmath239 is somewhat uncertain due to the large uncertainty in our estimate of @xmath240 . however , for grb 090417b to be an optically bright burst by the definition of @xcite then the optical afterglow would have had to have had @xmath241 , which is inconsistent with the observed host+afterglow magnitude ( @xmath242 ) . our computation of the spectral index between the optical and @xmath1-ray regimes depends on the limiting magnitude of the afterglow in the uvot @xmath163 band as determined from the statistical error in the observed magnitude of the host galaxy at two epochs . in order to test that grb 090417b really is a dark burst we compute the expected uvot @xmath163-band magnitude for the afterglow given the observed @xmath1-ray spectrum and assuming that the cooling frequency is between @xmath1-ray and optical wavelengths . this yields a prediction of @xmath241 at 11 hours , which is ruled out by the observations . therefore , we conclude that grb 090417b is a dark grb . the difference between the predicted @xmath163 magnitude and the upper limit that we derive in [ section : uvot_data ] is @xmath243 mag . if we assume a milky way extinction law this corresponds to @xmath244 mag of extinction in the host galaxy along the line of sight to the burst . if we assume the canonical @xmath1-ray light curve for grb afterglows @xcite then we can use the temporal decay indices and the spectral slopes to constrain the underlying physics . we stress that the following analysis is only valid if we assume the relativistic fireball model . alternate models for the @xmath1-ray afterglows of grbs are available , such as the cannonball model @xcite , the disc model of @xcite , the late internal emission model @xcite , the long - lived reverse shock model @xcite , and the prior outflow emission model @xcite . the time and post - break decay index of the late - time break in the @xmath1-ray light curve at @xmath245 s after the bat trigger are consistent with a jet break . however , the dust scattering interpretation of the late - time spectral softening ( see [ section : dust_scattering ] ) suggests that the steepening of the light curve after @xmath246 s is due to scattered emission due to dust dominating over the synchrotron component of the afterglow . unfortunately there is no data in other wavelength regimes to test if the late - time steepening is achromatic ( which would argue for a jet break ) , therefore it is uncertain if this is a jet break . @xcite give the relationships between the temporal and spectral slopes and the electron power - law distribution index , @xmath247 , for a blast expanding into a homogeneous external medium , and @xcite give the relationships for a stellar wind ( @xmath248 ) environment . both of these papers assume that @xmath249 . for the case were @xmath250 the relationships of @xcite are used . @xcite found that grb afterglows exhibit a range of electron distributions with values for the ten grbs they studied being between @xmath251 and @xmath252 . the @xmath1-ray spectral index during the first part of phase iii of the afterglow ( @xmath253 s ) is @xmath254 . if the cooling frequency is above the @xmath1-ray regime during this time then the predicted electron index is @xmath255 , which is well outside the range of values found by @xcite . if the cooling frequency is below the @xmath1-ray band during this time then @xmath256 , which is consistent with the @xcite range of electron indices . for @xmath257 the spectral index predicts a temporal decay of @xmath258 for either a homogeneous interstellar medium ( ism ) or a stellar wind environment . the observed @xmath1-ray decay during this time is @xmath259 , which is consistent with the predicted value . therefore , we conclude that the cooling frequency is most likely below the @xmath1-ray band at @xmath253 s. unfortunately it is not possible to tell , from the @xmath1-ray data , if the burst is expanding into a uniform external medium or a pre - existing stellar wind . the broad agreement between @xmath260 and @xmath261 with the predictions of the fireball model suggests that this is a reasonable explanation for the physics of this afterglow during this phase of the afterglow . between 68980 s and 428400 s the spectral index softens to @xmath262 . in this interval the fireball model predicts @xmath263 for a homogeneous ism and a wind if @xmath257 . if @xmath264 then the environment must be windy ( since @xmath265 is increasing ) and @xmath266 is predicted to be @xmath267 . the observed value is @xmath268 . neither case is consistent with the data in this time interval . the data before 68980 s is broadly consistent with a relativistic fireball expanding into either a homogeneous ism or a wind - stratified external medium , and a cooling frequency below the @xmath1-ray band in the period @xmath269 s. however , this model can not explain the data later than 68980 s. some other physical process is needed to explain the late - time softening of the @xmath1-ray spectrum . for grbs occurring in dusty star forming regions , the @xmath1-rays emitted during the prompt phase may be scattered by dust grains in small angles , which leads to a delayed @xmath1-ray afterglow component @xcite . placing circumburst dust at @xmath7 10100 pc away from the burst , the calculated dust echo @xmath1-ray lightcurve displays a shallow decay lightcurve followed by late steepening , as is commonly observed in many grbs @xcite . one major feature of this model is the strong spectral softening evolution in the echo emission @xcite . such a softening feature has not commonly been observed in the majority of grbs , so the dust scattering model is not favored to interpret those bursts . however , grb 090417b shows a spectral softening during the shallow decay phase . also the optically dark nature of the afterglow suggests that there is a substantial amount of dust in the circumburst environment . all these make this burst a very probable candidate for the dust scattering model . in modeling this burst , we assume a dust sheet with distance @xmath270 from the burst , with other dust grain properties the same as in figure 1 of @xcite . we find the calculated light curve can reproduce quite well the xrt light curve from @xmath271 s to @xmath272 s ( we consider the flares prior to this time as not being due to dust scattering because the @xmath273-ray emission was still on at that time ) for @xmath274 3080 pc and for a broad range of the source emission spectral index values . a strong test of the model is whether the spectral index evolution can be reproduced . in figure [ figure : dust_fit ] we plot the temporal evolution of the observed xrt spectral index . the dust scattered emission spectral index is calculated in the same way as in @xcite . the amount of softening predicted by the model is large , i.e. , @xmath275 34 , and it is independent of the source ( prompt ) emission spectral index . we tried to use the reported prompt bat spectral index , @xmath276 , and the even harder spectral indices observed in the two @xmath1-ray flares ( @xmath277 s ) , @xmath278 , as the source @xmath1-ray spectral index , but found that the model predicted emission is significantly softer than what is observed . however , if one assumes an even harder spectral index of @xmath279 for the prompt @xmath1-rays , both the @xmath1-ray light curve and spectral evolution can be properly reproduced ( see figure [ figure : dust_fit ] ) . such a hard @xmath1-ray spectrum is consistent with the self - absorbed regime of a synchrotron or inverse compton scattering spectrum . we therefore assume that the grb prompt emission spectrum has a break above the @xmath1-ray band due to the self - absorption frequency , @xmath280 , below which the spectrum has @xmath281 or @xmath282 ( e.g. , * ? ? ? a large @xmath280 is possible if the prompt emission radius is small enough ( e.g. , * ? ? ? * ) . the best model fit to the lightcurve and the spectral evolution is shown in figure [ figure : dust_fit ] . the fit to the @xmath1-ray light curve has a reduced @xmath127 of 2.7 , and the fit to the spectral index evolution has a reduced @xmath127 of 1.5 . we also estimate the amount of dust in terms of @xmath283 , the scattering optical depth to the 1 kev photons , using the same formula as in @xcite . the source @xmath1-ray fluence is extrapolated from the observed @xmath273-ray fluence using the assumed spectral form , and the scattered emission @xmath1-ray fluence is estimated from the observed xrt fluence for @xmath284 s using an intermediate value of observed spectral index @xmath285 . we find @xmath286 . this means that over half of prompt 1 kev photons would be scattered by the dust . this @xmath283 value also implies @xmath287 mag , according to some empirical relations between the two dust properties used in @xcite . this @xmath288 is much higher than the average value for the host galaxy that was derived from the spectral energy distribution in section [ section : sed ] , but is consistent with the extinction derived from the @xmath1-ray measurement of the hydrogen column along the line of sight to the burst , @xmath289 mag . this high line - of - sight extinction naturally explains the non - detection of an optical afterglow for grb 090417b . our model indicates that the dust grain column density along the line of sight is @xmath290 @xmath47 , and the density of the intervening dust sheet is @xmath291 @xmath292 . the data of grb 090417b is consistent with the dust scattering model under the condition that the prompt @xmath1-rays have a self - absorbed spectrum in the @xmath1-ray range . the large amount of dust inferred from the @xmath1-ray modeling is consistent with the dark nature of the optical afterglow . bat observations of grb 090417b were cut off at approximately 2105 s after the bat trigger when _ swift _ went into an earth limb constraint . therefore the total duration , and thus the total fluence , of the burst is not known . however , the fluence while the burst was observed was @xmath51 erg @xmath47 in the 15150 kev band . this puts a lower limit on the total fluence . we can use this lower limit to estimate a lower limit on the isotropic equivalent energy of grb 090417b . the bat spectrum is best fit by a single power law , so at a redshift of @xmath39 this corresponds to an isotropic equivalent energy of @xmath293 erg . we performed a @xmath294 correction following the prescription of @xcite and find @xmath295 ( with some uncertainty due to the arbitrary choice of cut - off energy for the power - law spectrum ) . this yields a @xmath294-corrected total isotropic energy of @xmath296 erg in the 202000 kev band . our data does not allow us to constrain the time of a jet break , so we are unable to estimate the jet opening angle or the intrinsic gamma - ray energy of grb 090417b . to date no supernova signature has been seen for any dark grb . if a supernova were detected in a dark grb that would have profound implications for origin of that dark burst . the detection of a supernova would indicate that the burst was not dark due to extinction or high redshift , but must have a fundamentally different afterglow than conventional optically - bright bursts . in light of this we examined our data to see if there is any evidence for a supernova component for grb 090417b . uvot observations were taken up to 15 days after the bat trigger , so we searched this late - time data for evidence of a supernova component . a type ib / c supernova like sn1998bw @xcite at a redshift of @xmath297 is expected to peak at approximately @xmath298 days after the burst . we find no evidence for a change in the luminosity of the host galaxy down to @xmath299 mag , which corresponds to a upper limit on the magnitude of the supernova of @xmath300 . at @xmath39 the central wavelength of the uvot @xmath163 band corresponds to @xmath301 . no observational data is available for the peak magnitude of sn1998bw at 2600 . swift_/uvot ultraviolet light curves are available for sn2007y @xcite , a sn ib / c with a peak absolute magnitude of uvw1 @xmath302 in the rest frame . this corresponds to a rest frame uvw1 of @xmath303 at the redshift of grb 090417b . this is similar to our limiting magnitude , so if we assume that sn2007y is typical of the sne ib / c associated with grbs we are not able to constrain the existence of a supernova associated with grb 090417b . all of the evidence presented in this paper points to grb 090417b being a dark burst because of localized extinction in the host galaxy . the host galaxy has a redshift of @xmath39 , and is clearly seen at optical and ultraviolet wavelengths . therefore lyman-@xmath11 absorption can not account for the missing optical afterglow . the @xmath1-ray light curve and spectrum during the first @xmath304 ks obey the conventional closure relationships and are consistent with synchrotron radiation from a relativistic fireball expanding into an external medium ( although the structure of that medium can not be determined with the available data ) . optical afterglows have been detected for many grbs with similar @xmath1-ray properties , so it unlikely that the lack of an optical detection is due to unusual physics during the afterglow phase of the burst . optical observations of grb 090417b started 378 s after the bat trigger , and follow - up observations were made from several observatories for several days after the burst . these observations went deep enough to have detected an optical afterglow if one had been present with a luminosity similar to that of other optical afterglows . therefore , the dark nature of grb 090417b can not be explained by a lack of observations . the dust hypothesis , however , can explain this dark burst . the @xmath1-ray spectrum is consistent with an column density of @xmath305 @xmath47 , which corresponds to an extinction of @xmath162 mag along the line of site in the host galaxy , assuming a milky way extinction law as explained in [ section : xrt_data ] . this value is in agreement with the minimum extinction , as explained in [ section : dark ] , that is needed to obscure the expected bright optical afterglow ( @xmath306 mag ) . an extremely high extinction ( @xmath30740 mag ) is also predicted by the dust scattering model , which is consistent with the @xmath1-ray data . such a large amount of dust along the line of sight to the grb naturally explains the lack of any detection at ultraviolet , optical , or near - infrared wavelengths while allowing a conventional @xmath1-ray afterglow to be observed . it is also consistent with the result of @xcite who found large mean extinctions for the putative host galaxies of several grbs . our result , that the extinction along the line of sight in the host grb 090417b is likely to be at least ten magnitudes in the uvot @xmath106 band , provides strong evidence that local dust concentrations in the host galaxy are responsible for at least some low - redshift dark bursts . unlike many grb host galaxies the dust in the host galaxy of grb 090417b appears to follow a milky way extinction law . in addition the host appears to be an @xmath227 galaxy with a star - formation rate that exceed @xmath308 @xmath309 yr@xmath30 . in other words , the host appears to have dust properties , and a luminosity that are similar to those of the milky way , and a star - formation rate that is consistent with what is seen in other grb host galaxies . grb 090417b is one of the few dark grbs where the association with a host galaxy is secure . in most cases associations between a grb and a galaxy have been made based on the probability of the nearest observed galaxy lying as close to the centre of the @xmath1-ray error circle as it does . there are a few exceptions , such as grb 000210 , grb 050713a and xrf 050416a . the host galaxy of grb 000210 was identified from the locations of its radio and @xmath1-ray afterglow . @xcite found a probability of a chance alignment of the galaxy and the burst of @xmath310 and conclude that if this is the host then the dark nature of grb 000210 is likely due to dust either at the location of the progenitor or along the line of sight . xrf 050416a occurred at a low redshift ( @xmath311 @xcite ) . although technically dark an optical afterglow was detected and was consistent with the standard fireball model ( e.g. , * ? ? ? * ; * ? ? ? the presence of an optical afterglow and the low line - of - sight extinction suggests that this burst should not be considered to be a dark grb . the line - of - sight extinction for xrf 050416a is @xmath23 mag @xcite , which is consistent with what is seen for most optically detected grb afterglows . @xcite note that grb 050713a also only barely qualifies as a dark burst under the @xcite definition . the internal line - of - sight extinction for this burst is unknown . however , an optical afterglow was seen @xcite , so it is unlikely that the extinction was large . the latter two of these bursts are technically dark based on the spectral slope between the @xmath1-ray and optical regimes ( @xmath15 ) criteria of @xcite . however , in these cases optical or infrared afterglows were detected , suggesting that these grbs were borderline cases of dark bursts . further evidence that dust is responsible for grb 090417b being dark comes from the late - time softening of the @xmath1-ray spectrum . this softening is consistent with what is predicted from the dust scattering model @xcite , and requires approximately 1540 mag of extinction at optical wavelengths . the effects of dust scattering do not become apparent in the @xmath1-ray spectrum until @xmath312 s after the burst . since @xmath1-ray spectra are usually extracted from early - time data ( when there are a large number of photons ) it is possible that the softening seen for grb 090417b has been present in the late - time @xmath1-ray spectra of other dark bursts , but has not been seen due to low count rates . for example , the standard catalogue of _ swift_/xrt spectral fits @xcite uses data taken within @xmath313 ks of the bat trigger , well before a softening due to dust scattering is expected to appear . @xcite point out that the host galaxies of dark grbs are not unusually reddened relative to the hosts of optically bright grbs . their analysis assumes that their statistical association of bursts with the nearest detected galaxy on the sky reveals the true hosts of these bursts . it is uncertain if this is valid for all of the grbs in their sample . however , it is likely that at least some ( if not all ) of these galaxies actually did host the associated grbs . although some of the hosts of dark grbs have been found to be unusually dusty compared to the hosts of optically - bright grbs ( e.g. , grb 030115 @xcite ) , there is no evidence that the hosts of dark grbs have systematically higher mean extinctions than the hosts of optically bright grbs . however , the mean extinction of the host is probably not the factor that determines if a particular burst is dark or not since any extinction due to material that is not along the line of sight to the burst will not affect our observations of the afterglow . what is important in making a burst optically dark or bright is the extinction along the line of sight . this extinction can be due to dust at the location of the grb , although it has been suggested that the ultraviolet and @xmath1-ray emission from grbs can destroy circumburst dust out to a few tens of pc from the progenitor @xcite . one way to get a dusty environment around a grb progenitor is the stellar wind from a wolf - rayet progenitor . collisions between wind - driven shells of material ejected during various stages of the wolf - rayet star s evolution can trigger the formation of large amounts of dust ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the radius of a wolf - rayet wind bubble is consistent with with the distance from the progenitor that we find for the dust along the line of sight to grb 090417b ( tens of pc ) . alternately , the extinction could be due to unrelated dust in the host that just happens to lie along the line of sight , such as a dusty star - forming region or a giant molecular cloud . giant molecular clouds have diameters of approximately 100 pc @xcite , but often contain dense cores where star formation occurs . a chance alignment of one of these cores along the line of sight to grb 090417b could account for the high extinction inferred from the x - ray data . for grb 090417b the late - time evolution of the @xmath1-ray spectrum can be explained if there is a sheet of dust at a distance of @xmath314 pc from the burst . this suggests that grb 090417b may have been dark because of either a dusty environment within 3080 pc of the progenitor ( e.g. * ? ? ? * ; * ? ? ? * ) , or a chance alignment of the progenitor with a region of high extinction in the host galaxy . the galactic extinction model of @xcite suggests that roughly 25% of the sight lines through the milky way have @xmath315 mag . the modelling of @xcite suggests that most sightline through a galaxy like the milky way , as seen from within the galaxy , have @xmath316 mag . however , their work does not include clumpiness in the dust distribution , which would lead to a larger fraction of sightlines having higher extinctions . this suggests that a non - negligible ( albeit uncertain ) fraction of grbs may be located along a sightline with more than approximately one magnitude of extinction in their hosts . when high - redshift dark bursts are accounted for the fraction of dark bursts is roughly comparable of the estimated fraction of highly extincted sightlines , which suggests that obscuration by dust along the line of sight may be responsible for some dark grbs . we have shown that the host of the dark grb 090417b is very likely to be sdss j135846.65 + 470104.5 , an @xmath227 galaxy at @xmath39 . this galaxy has an overall star - formation rate that is at least as great as that of the milky way , and an overall @xmath317-band extinction of @xmath318 mag , which is dustier than typical grb host galaxies , but not greatly different from the milky way . @xmath1-ray observations of grb 090417b show the normal temporal and spectral behaviour seen in the @xmath1-ray afterglows of many _ swift _ grbs . the data up to approximately 70 ks after the bat trigger are consistent with a relativistic fireball expanding into either a homogeneous ism or a wind - stratified external medium . the cooling frequency is below the @xmath1-ray band between 9.1 ks and 29.6 ks , and the @xmath294-corrected isotropic energy is @xmath319 erg . the physics of this burst do not appear to be unusual , and thus are unlikely to explain the dark nature of the burst . after approximately 70 ks the @xmath1-ray spectrum becomes significantly softer . we find that this can be explained using the dust scattering model of @xcite . the observed late - time spectral evolution can be produced by a sheet of dust approximately 3080 pc from the burst along the line of sight . the model predicts an extinction of @xmath30740 mag along the line of sight . this is consistent with the column density measured from the @xmath1-ray spectrum . therefore , we conclude that grb 090417b is probably dark because of a dense , localized layer of dust along the line of sight between us and the afterglow . we acknowledge the use of public data from the _ swift _ data archive . this paper is based in part on observations taken with the nordic optical telescope , operated on the island of santa miguel de la palma jointly by denmark , finland , iceland , norway , and sweden in the spanish observatorio del roque de los muchachos of the instituto de astrofsica de canarias . the authors would like to thank the anonymous referee for their comments , which improved this paper .
we use the observed anisotropic clustering of galaxies in the baryon oscillation spectroscopic survey ( boss ) data release 11 cmass sample to measure the linear growth rate of structure , the hubble expansion rate and the comoving distance scale . our sample covers 8498 and encloses an effective volume of 6 at an effective redshift of . we find , and when fitting the growth and expansion rate simultaneously . when we fix the background expansion to the one predicted by spatially - flatcdm model in agreement with recent planck results , we find ( 6 per cent accuracy ) . while our measurements are generally consistent with the predictions ofcdm and general relativity , they mildly favor models in which the strength of gravitational interactions is weaker than what is predicted by general relativity . combining our measurements with recent cosmic microwave background data results in tight constraints on basic cosmological parameters and deviations from the standard cosmological model . separately varying these parameters , we find ( 8 per cent accuracy ) and ( 16 per cent accuracy ) for the effective equation of state of dark energy and the growth rate index , respectively . both constraints are in good agreement with the standard model values of and . [ firstpage ] gravitation cosmological parameters dark energy dark matter distance scale large - scale structure of universe
1312.4899
galaxies map the distribution of the underlying dark matter field and provide invaluable information about both the nature of dark energy ( de ) and properties of gravity ( see e.g. * ? ? ? the shape of the two - point correlation function of the observed galaxy field , or of its fourier - transform the power spectrum , contains features such as baryon acoustic oscillations ( bao ) and the turn - over marking the transition between radiation dominated and matter dominated evolutionary phases @xcite . these features can be used to place tight constraints on relative abundances of different energy - density components of the universe ( radiation @xmath12 , dark matter @xmath13 , baryonic matter @xmath14 and de @xmath15 ) . presently , these ratios are measured to much higher accuracy in the cosmic microwave background ( cmb ; * ? ? ? therefore , for most cosmological models these features provide most information when used as a standard ruler . if the universe is statistically isotropic and homogeneous on large - scales , the correlation function and power spectrum should likewise be rotationally invariant . the observed two - point statistics instead exhibit a strong anisotropy with respect to the line - of sight ( los ) direction . two effects are responsible for this apparent anisotropy : the redshift - space distortions ( rsd ; * ? ? ? * ) and the alcock paczynski effect ( ap ; * ? ? ? the rsd arise in maps made from galaxies if distances are determined from measured redshifts assuming that they are only caused by the hubble flow . because of gravitational growth , the galaxies tend to infall towards high- density regions , and flow away from low - density regions , such that the clustering is enhanced in the los direction compared to the perpendicular direction . the observed redshifts thus have a component aligned with these flows . on large - scales where gravitational growth is linear , measuring the relative clustering in both los and transverse directions leads to a measurement of the logarithmic growth rate of structure @xmath16 where @xmath17 is a scale factor , @xmath18 is a measure of the amplitude of the matter power spectrum and @xmath19 is the linear growth function normalized such that @xmath20 ( see * ? ? ? * for a review of rsd ) . the magnitude of the large - scale velocity field traced by galaxies depends on the nature of gravitational interactions and measured values of @xmath21 can be used to constrain models of gravity ( see e.g. * ? ? ? galaxy clustering data measures the growth at low redshifts . combining this information with the accurate estimates of the amplitude of matter perturbations at @xmath22 provided by cmb allows for extremely strong constraints for deviations from the predictions of general relativity ( gr ) since even small changes in the growth of structure accumulate to a large offset over cosmic time ( for recent gr constraints see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? anisotropies are also observed due to the ap effect , which stems from the fact that we need to convert observed angular positions and redshifts of galaxies to physical coordinates in order to measure clustering statistics . if the fiducial cosmology used for this mapping is different from the true cosmology this will induce anisotropies in the measured clustering pattern even in absence of rsd . angular distortions are sensitive to the offset in the angular distance @xmath23 and distortions in the los direction depend on the offset in @xmath24 . measuring the ap effect provides accurate estimates of the angular distance and hubble parameter and can be used to constrain properties of de ( see e.g. * ? ? ? measurements of both angular and radial projected scales are usually reported in terms of the volume averaged distance @xmath25^{1/3},\ ] ] and the ap - parameter @xmath26 in the absence of rsd , the measured correlation function monopole would be sensitive mostly to isotropic scale dilation through @xmath27 and the quadrupole to anisotropic scale dilation through @xmath28 . most of the information on @xmath27 usually comes from the most pronounced feature in the correlation function the position of the bao peak in the monopole . it is therefore convenient to report results in terms of @xmath29 where @xmath30 is the sound horizon at the drag epoch which sets the bao scale ( for a review of bao and ap see e.g. * ? ? ? the rsd and ap are partially degenerate but have a different scale dependence which makes their simultaneous measurement possible @xcite . specifying cosmological models of background expansion or gravity helps to further break this degeneracy ( e.g. * ? ? ? * ) . measuring correlation function in different fiducial cosmological models and fitting the rsd signal in each can help to reduce the degeneracy as well @xcite . the rsd signal within the correlation function is difficult to model because of the significant contribution from nonlinear effects and higher order contributions from galaxy bias . a number of recent studies have shown that many current rsd models result in biased estimates of the growth rate ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in our work , we use the ` streaming model'-based approach developed in @xcite that has been demonstrated to fit the monopole and quadrupole of the galaxy correlation function with better than per cent level precision to scales above @xmath31 mpc , for galaxies with bias of @xmath32 . many distance - scale and rsd measurements have previously been made using spectroscopic survey data . recent highlights include the bao measurements from the 6df galaxy redshift survey ( 6dfgrs ; * ? ? ? * ) , sloan digital sky survey ii ( sdss - ii ; * ? ? ? * ) , sdss - iii baryon oscillation spectroscopic survey ( boss ) data release 9 sample ( dr9 ; * ? ? ? * ) , wigglez survey @xcite and sdss - iii boss dr10 and dr11 samples @xcite . the rsd signal has been measured in the 6dfgrs @xcite , the sdss - ii survey @xcite , the sdss - iii boss dr9 data @xcite and vimos public extragalactic redshift survey ( vipers ; * ? ? ? simultaneous fits to rsd and ap parameters have been performed for the wigglez survey @xcite , sdss - ii data @xcite and sdss - iii boss dr9 data @xcite . the analysis presented in this paper builds upon that of @xcite , who measured the rsd and ap simultaneously in the boss cmass dr9 sample , achieving a 15 per cent measurement of growth , 2.8 per cent measurement of angular diameter distance , and 4.6 per cent measurement of the expansion rate at @xmath33 . using these estimates @xcite derived strong constraints on modified theories of gravity ( mg ) and de model parameters . in this paper we perform a similar analysis on the cmass dr11 sample , which covers roughly three times the volume of dr9 . this paper is organized as follows . in section [ sec : data ] we describe the data used in the analysis . section [ sec : measurements ] explains how the two - dimensional correlation function is estimated from the data . section [ sec : covariances ] shows how we derive the estimates of the covariance matrix for our measurements . in section [ sec : model ] we describe the theoretical model used to fit the data . section [ sec : constraints ] presents and discusses our main results the estimates of growth rate , distance - redshift relationship and the expansion rate from the measurements . section [ sec : cosmoconstraints ] uses these estimates to constrain parameters in the @xmath6 cold dark matter ( @xmath6cdm ) model assuming gr ( @xmath6cdm - gr ) and possible deviations from this standard model . we conclude and discuss our results in section [ sec : conclusions ] . our measurements require the adoption of a cosmological model in order to convert angles and redshifts into comoving distances . as in @xcite we adopt a spatially - flat @xmath6cdm cosmology with @xmath34 and @xmath35 for this purpose . for ease of comparison across analyses , we follow @xcite and also report our distance constraints relative to a model with @xmath36 , @xmath35 , and @xmath37 , for which the bao scale @xmath38 mpc . the sdss - iii project @xcite uses a dedicated 2.5-m sloan telescope @xcite to perform spectroscopic follow - up of targets selected from images made using a now - retired drift - scanning mosaic ccd camera @xcite that imaged the sky in five photometric bands @xcite to a limiting magnitude of @xmath39 . the boss @xcite is the part of sdss - iii that will measure spectra for 1.5 million galaxies and 160.000 quasars over a quarter of the sky . we use the dr11 cmass sample of galaxies @xcite . this lies in the redshift range of @xmath40 and consists of 690826 galaxies covering 8498 square degrees ( effective volume of @xmath41 ) . most galaxies in the sample belong to the red sequence . about 25 per cent of them , however , would be classified as ` blue ' according to traditional sdss rest - frame colour cuts ( see e.g. * ? ? ? * ) . @xcite showed that there is no detectable colour dependence of distance scale and growth rate measurements in dr10 sample . [ fig : nz ] shows the redshift distribution of galaxies in our sample . the number density is of the order of @xmath42 peaking at @xmath43 . in northern and southern galactic hemispheres , computed assuming our fiducial cosmology . ] we measure the correlation function of galaxies in the cmass sample defined as the ensemble average of the product of over - densities in the galaxy field separated by a certain distance @xmath44 @xmath45 the overdensity as a function of @xmath44 is given by @xmath46 where @xmath47 is the expected average density of galaxies at a position @xmath48 and @xmath49 is an observed number density . we estimate the correlation function using the landy - szalay minimum - variance estimator @xcite @xmath50 where @xmath51 is the weighted number of galaxy pairs whose separation falls within the @xmath52 bin , @xmath53 is number of similar pairs in the random catalogue and @xmath54 is the number of cross - pairs between the galaxies and the objects in the random catalogue . [ fig : meas2d ] shows the two - dimensional correlation function of dr11 sample measured in bins of @xmath55 @xmath56 . both the ` bao ridge ' ( a ring of local maxima at approximately 100@xmath57 mpc ) and the rsd signal ( los ` squashing ' of the correlation function ) are detectable by eye . the random catalogue is constructed by populating the volume covered by galaxies with random points with zero correlation . we use a random catalogue that has 50 times the density of galaxies to eliminate extra uncertainty associated with the shot noise in the random catalogue . we weight each galaxy in the catalogue with three independent weights . first is the feldman kaiser peacock ( fkp ; * ? ? ? * ) weight @xmath58 $ ] . this approach downweights galaxies in high - density regions , achieving a balance between cosmic variance and shot - noise errors . the second weight @xmath59 accounts for the systematic effects associated with both the varying stellar density ( @xmath60 ; * ? ? ? * ) and seeing variations in the imaging catalogue used for targeting ( @xmath61 ; * ? ? ? the third weight corrects for the missed galaxies due to fibre collisions and redshift failures using the algorithm described in @xcite . the former is caused by the finite size of fibres that makes simultaneous measurement of spectra of two galaxies with small angular separation impossible . to correct for both of these effects , we upweight each galaxy by the number of its lost neighbours and the resulting weight is @xmath62 . since these effects are statistically independent , the total weight is a product of three @xmath63 . the weight of the pair is the product of individual weights for two galaxies . since the stellar and close - pair effects are absent in the random catalogue we apply only the fkp weight to them . the observed correlation function is a function of two variables : we use @xmath64 , the distance between galaxies , and @xmath65 , the cosine of the angle between their connecting vector and the los . the optimal choice of binning for the correlation function measurements depends on two competing effects . using small bin size retains more information , but since we estimate covariance matrices by computing a scatter of finite number of mock catalogues ( see section [ sec : covariances ] ) , using more bins deteriorates the precision at which the elements of the covariance matrices can be estimated . empirical tests performed on the mock catalogues suggest that the rsd signal is more or less insensitive to the binning choice , while the bao measurements are optimal at @xmath66 mpc ( for details see * ? ? ? we bin @xmath64 in 16 bins of @xmath67 mpc in size in the range of @xmath68 and @xmath65 in 200 bins in @xmath69 , and estimate the correlation function on this two - dimensional grid . the information in the correlation function below @xmath70 is strongly contaminated by non - linear effects , and the scales above @xmath71 have low signal - to - noise ratio and contribute little information . we compress the information in the two - dimensional correlation function by computing the legendre multipoles with respect to @xmath65 by approximating the integral with a discrete sum : @xmath72 where @xmath73 is the legendre polynomial of the order of @xmath74 . in the subsequent analysis we only use the monopole ( @xmath75 ) and the quadrupole ( @xmath76 ) moments . the higher order moments contain significantly less information and are more difficult to model . ( for the contribution of the higher order moments see e.g. * ? ? ? * ; * ? ? ? the rsd signal in the measured correlation function varies within the sample due to redshift evolution [ via the redshift dependence of @xmath77 and @xmath78 . if we keep track of the redshift of individual galaxy pairs in equation ( [ eq : ls ] ) , we effectively measure @xmath79 where summation is over individual galaxy pairs contributing to @xmath80 counts , @xmath81 is the correlation function at mean redshift of that galaxy pair and @xmath82 is the product of the weights of individual galaxies in the ith pair . thus our measurement is actually a weighted redshift - averaged correlation function . the evolving correlation function can be expanded into taylor series in redshift around some value of @xmath83 : @xmath84.\ ] ] keeping only the first - order term , we find @xmath85 and the second term disappears if we define @xmath86 if the derivatives of the correlation function of second order and higher are small , the redshift averaged correlation function is equal to the correlation function at an ` effective ' redshift given by equation ( [ eq : zbar ] ) . the ` effective ' redshift defined in equation ( [ eq : zbar ] ) is a function of scale . we adopt @xmath87 , a value which is close to the @xmath83 computed from the catalogue to better than 1 per cent precision for all scales in the range @xmath88 mpc . we checked that in our fiducial @xmath6cdm cosmology the contribution from the higher order terms in equation ( [ eq : xitaylor ] ) are indeed small for the expected theoretical variations in @xmath21 within the redshift range . we will therefore interpret our estimates as measurements of the correlation function at this effective redshift . [ fig : meas ] shows the measured monopole and quadrupole of the cmass galaxies along with @xmath89 errorbars ( see section [ sec : covariances ] for details of the error estimation ) . we will use these measurements in our analysis rather than the two - dimensional correlation function presented in fig . [ fig : meas2d ] . . the solid lines show predictions of our best - fitting model with @xmath90 , @xmath91 , @xmath92 , @xmath93 , @xmath94 , @xmath95 , @xmath96 and @xmath97 . ] to estimate the covariance matrix of our measurements we use a suite of 600 pthalo simulations . the simulations cover the same volume as the cmass sample and are designed to produce a similar bias ( for details of mock generation see * ? ? ? we compute the legendre multipoles from each individual mock catalogue and estimate the covariance matrix as @xmath98\left[\xi^k_{\ell ' } ( r_j)-\bar{\xi}_{\ell ' } ( r_j)\right],\ ] ] where the sum is over individual mocks and the average multipoles @xmath99 are also computed from the mocks . the unbiased estimator of the inverse covariance matrix is then given by @xmath100 where 32 is the number of bins used in the analysis ( for details see * ? ? ? [ fig : redcov ] shows the reduced covariance matrix ( diagonal elements normalized to one ) of our multipoles . as expected , the measured multipoles in the neighbouring @xmath64-bins are strongly correlated . the correlation between measured monopole and quadrupole at the same scale is up to 15 per cent on smaller scales . we will compute the likelihood of theoretical models as @xmath101 where @xmath102 @xmath103 are the set of parameters and @xmath104 are the theoretical predictions for the multipoles . in equation ( [ eq : chi2 ] ) we additionally rescale the inverse covariance matrix @xmath105 where @xmath106 is the length of vector @xmath103 . this accounts for the uncertainties in the determination of the inverse covariance matrix from the finite number of catalogues ( for details see * ? ? ? * ) . in our case , @xmath107 , @xmath108 and @xmath109 , which results in @xmath110 and @xmath111 . here even though the total number of fitted parameters is 8 , because the three ` shape ' parameters are constrained almost exclusively by the planck covariance matrix of equation ( [ eq : planckcovs ] ) which is not derived from our suite of 600 pthalo mocks and is assumed to be exact ] the multiplicative correction factor is then @xmath112 . in approximating the likelihood by equations ( [ eq : bosslik ] ) and ( [ eq : chi2 ] ) , we made two assumptions : that the errors on the monopole and quadrupole are drawn from a multivariate gaussian distribution ( equation [ eq : bosslik ] ) and that the dependence of inverse covariance matrix on free parameters is weaker than the dependence of the model [ @xmath113 in equation ( [ eq : chi2 ] ) ] . non - linear evolution will induce non - gaussianity . to check the validity of the first assumption , we estimate a skewness of @xmath114 in bins of @xmath64 from the 600 pthalo mocks using @xmath115^{3/2}}.\ ] ] fig . [ fig : skewness ] shows a histogram of the resulting distribution of sample skewness and the prediction made assuming that the distribution of @xmath116 is gaussian . the observed distribution is consistent with the assumption of gaussianity ; therefore , we will ignore the contribution of possible non - gaussian contributions to the likelihood of @xmath116 . the validity of our second assumption is helped by the fact that the signal - to - noise ratio is high and the mock catalogues were tuned to reproduce the observed clustering of cmass sample on average ( see * ? ? ? * for details ) . we use the ` streaming model ' to compute our theoretical template correlation function . within the streaming paradigm the correlation function in redshift space is derived by taking a real space , isotropic correlation function @xmath117 and convolving it along the los with a probability distribution function of the infall velocity @xmath118 of a galaxy pair at that separation . @xmath119 p(s_{|| } - r_{||})\mathrm{d}r_{||}\ ] ] where @xmath120 and @xmath121 are the components of a vector in the parallel and perpendicular to the los direction ; @xmath122 and @xmath123 are the analogous components in the real space . in the plane - parallel approximation we adopt here , @xmath124 . the function @xmath118 accounts for both quasilinear infall motions and the random small - scale velocities ( finger - of - god effect ; * ? ? ? following @xcite , we assume @xmath125 ^ 2/2(\sigma_{\rm in}^2(r,\mu)+\sigma^2_{\rm fog})\right)}{\sqrt{2\pi(\sigma_{\rm in}^2(r,\mu)+\sigma^2_{\rm fog})}}\ ] ] and compute the @xmath126 and @xmath127 values using the standard perturbation theory , while the correlation function in the configuration space @xmath128 is computed using lagrangian perturbation theory ( see * ? ? ? * for details ) . the parameter @xmath129 is an isotropic dispersion that accounts for motions of galaxies within their local environment that are approximately uncorrelated with the large - scale velocity field ; this parameter is varied within a broad prior consistent with the expected contribution from satellite galaxies ; see @xcite for further discussion . recently , @xcite extended the results of convolution lagrangian perturbation @xcite and combined them with the ` streaming model ' to obtain accurate predictions for the two - dimensional correlation function . this improved model is accurate for a wider range of biases than the original model . for the cmass sample , however , the original implementation of the model in @xcite remains accurate enough and is used in this analysis ( see also discussion following fig . [ fig : bfmocks ] ) . if the real geometry of the universe differs from the fiducial cosmology used to compute the correlation function , this will result in the additional distortions via the ap effect . to account for this , we rescale the redshift - space correlation function in equation ( [ eq : reidwhite ] ) as @xmath130 where @xmath131 and @xmath132 and @xmath133 are the hubble expansion rate and the angular distance in the fiducial cosmology . the model correlation function depends on the growth rate via @xmath134 and @xmath135 . the higher values of @xmath136 result in higher amplitude of both multipoles . the dependence on the hubble expansion rate and angular distance arise from the ap effect and are manifested as distortions of the multipole shapes . our model has been compared to n - body simulations and shown to fit the anisotropic clustering down to scales of @xmath137 mpc with per cent level precision @xcite . to check that the pthalo mocks adequately describe the rsd signature in the range of scales used in the analysis we fit our model to the mock measurements . for simplicity , we fix the shape of the linear power spectrum to the input value and use the input cosmology to compute radial and angular distances ( this is equivalent to fixing @xmath138 ) so that the only free parameters are @xmath136 , @xmath139 and @xmath140 . fig . [ fig : bfmocks ] shows the distribution of maximum - likelihood ( ml ) values of @xmath139 and @xmath136 recovered from the mock catalogues . the systematic offset between the mean of the ml values for @xmath136 and the input value of the mocks is of the order of 1 per cent , and the scatter in the ml values ( after appropriate rescaling as in * ? ? ? * ) is comparable to the errors produced in section [ sec : constraints ] . at least at the two - point level , this shows that the relevant systematic effects in the pthalos mocks are much less than our measurement precision and can be safely ignored . [ fig : kaisertest ] shows the mean values and @xmath89 errorbars recovered by fitting individual pthalo mocks with the non - linear ` streaming model ' as a function of minimal scale used in the analysis . the figure also shows the results if we use a linear theory model @xcite with no velocity dispersion nuisance parameter . we find that for the _ boss _ dr11 data , one would need to fit above @xmath141mpc to get unbiased estimates of the growth rate with the linear model . the non - linear model resulted in unbiased fits even when scales down to @xmath142mpc were used . interval of @xmath136 recovered by fitting the linear kaiser model ( blue hatched band ) and our ` streaming model ' ( red band ) to 600 pthalo mocks as a function of minimum scale used in the analysis . the black solid line denotes @xmath136 in the cosmology used to construct the mocks . ] the model predictions for @xmath143 depend on eight parameters . these are parameters determining the shape of the linear correlation function @xmath144 , the bias of galaxies @xmath139 , the linear growth rate @xmath136 , two ap parameters @xmath145 and @xmath146 , and fog velocity dispersion , @xmath147 . in linear theory , @xmath139 and @xmath136 are completely degenerate with @xmath148 , and observed clustering is only sensitive to their combination @xmath149 and @xmath21 . even though non - linear effects break this degeneracy , it is still present to high degree . to compute non - linear effects on the real - space correlation function as well as mean and variance of infall velocity , we need to specify a value of @xmath150 . in our analysis , we fix the value of @xmath151 , which is the best - fitting value to _ planck _ data within the @xmath6cdm - gr model . we checked that model predictions do not change significantly if we keep the values of @xmath149 and @xmath21 fixed and vary @xmath148 within @xmath152 of the _ planck _ constraints , and therefore , the recovered value of @xmath21 is not sensitive to the fiducial @xmath148 used to compute non - linear effects . when combining our measurements with _ data to constrain cosmological models , we do not fix a value of @xmath148 and compute it for each model accordingly ( see section [ sec : cosmoconstraints ] ) . the large - scale properties of the universe after inflation depend on several variables . first , we have the relative abundances of the main energy - density constituents radiation @xmath153 , baryons @xmath154 , dark matter @xmath155 and de @xmath156 . we must also specify the parameters describing initial conditions at the end of inflation the spectral index @xmath157 and the amplitude of curvature perturbations @xmath158 . finally , the behaviour of de can be fully described by its effective equation of state ( eos ) @xmath159 if the perturbations to de fluid are negligible . the energy density of radiation is determined with extremely high precision from the temperature of microwave background and is negligible at late times . in addition , the standard inflationary paradigm predicts the universe to be spatially flat to a high degree , which means that the de energy density can be expressed in terms of other components @xmath160 . the parameters @xmath161 , @xmath162 and @xmath157 are tightly constrained by cmb data in a way that is independent of late - time behaviour of de and gravity . the cmb also provides a measurement of a distance , to the last - scattering surface which depends on de , but since it utilizes only one integrated measurement of distance it results in highly degenerate constraints on @xmath159 and @xmath163 if used on its own . the low - redshift measurements from anisotropic galaxy clustering are strongly complementary to cmb information . the quantities @xmath164 and @xmath165 depend on @xmath166 and de properties , breaking degeneracies of cmb data . the @xmath136 in the @xmath21 measurement is sensitive to @xmath163 . relating @xmath148 to @xmath167 in a given model provides a strong additional test of both de and gravity . the fluctuations in the galaxy field ( @xmath148 measured by rsd ) are a result of initial fluctuations at recombination ( @xmath158 measured by cmb ) , and their relationship depends on the strength of gravity and expansion of the universe from @xmath168 to the redshift of the galaxy sample . even small offsets from gr and @xmath6 are amplified and result in large offsets at low redshifts . independent probes of distance and expansion rate , such as measurements of luminosity distance from supernovae type ia ( snia ) or direct measurements of @xmath165 , further enhance the cosmological constraints . the measured monopole and quadrupole ( see section [ sec : measurements ] ) along with the covariance matrix estimated from pthalo mocks ( see section [ sec : covariances ] ) are fitted with the predictions of the streaming model ( see section [ sec : model ] ) to derive constraints on the geometry and the growth rate at an effective redshift of @xmath169 . the theoretical template for the multipoles depends on the parameters @xmath170 $ ] . constraints on the shape of the correlation function ( @xmath171 ) from cmb data are significantly tighter than similar constraints obtainable from galaxy clustering only , and these constraints are largely independent of either the behaviour of de at low redshifts or the nature of gravity ( i.e. the values of @xmath172 , @xmath146 and @xmath21 which are of main interest here ) . to exploit this fact , we multiply the likelihood in equation ( [ eq : bosslik ] ) by a _ planck _ prior on this triplet @xmath173 with @xmath174 and the mean values of @xmath171 and the @xmath175 are given by _ temperature anisotropy data @xcite . we use the shape prior derived from the combination of _ planck _ temperature anisotropy data with the _ wmap _ low - multipole polarization likelihood which is @xcite @xmath176 @xmath177 to explore this parameter space we use the nested sampling method as implemented in the multinest software package @xcite . the free parameters of the model and their priors are listed in table [ tab : priors ] . .the priors on the model parameters . [ cols="^,^,^",options="header " , ] to derive constraints on de and mg parameters we will be using the marginalised likelihood of parameters @xmath29 , @xmath28 and @xmath21 , where @xmath178 is the sound horizon scale at the drag epoch . this marginalized likelihood can be approximated as a gaussian with mean @xmath179 and covariance matrix @xmath180 equations ( [ eq : mainmeans ] ) and ( [ eq : maincovs ] ) use values of @xmath181 derived by numerically integrating the recombination equations and integrating the sound speed up to the drag epoch . these values are related to the results derived from commonly used fitting formula of @xcite adjusted by a factor of @xmath182 . this ratio is independent of cosmology for a wide range of conventional cosmological models ( see e.g. * ? ? ? [ fig : dvffs8 ] shows the constraints on main cosmological parameters compared to the expectations from the _ planck _ data within standard @xmath6cdm - gr models along with dr9 results from @xcite . the dr11 results are in a good agreement with the _ planck _ predictions ; the @xmath183 difference between them is 1.6 for 3 degrees of freedom . equations ( [ eq : mainmeans ] ) and ( [ eq : maincovs ] ) represent the main results of our work and will be used later to constrain models of de and mg ( see section [ sec : cosmoconstraints ] ) . the companion papers , @xcite , @xcite , @xcite and @xcite use the same cmass dr11 data to constrain the distance redshift relation at @xmath33 . fig . [ fig : dvcompare ] shows our measurement of distance along with the result from bao only fits and previous similar measurements and _ planck _ predictions for spatially - flat @xmath6cdm model . from cmass dr9 and dr11 data sets . the blue band corresponds to 1@xmath184 uncertainty in _ planck _ prediction assuming @xmath6cdm . all measurements are mutually consistent . ] in fig . [ fig : dvcompare ] , the label 1d refers to the result derived by fitting the monopole of the correlation function only , while the label 2d refers to the result derived from the fit to the monopole and the quadrupole of the correlation function ( see * ? ? ? * for details ) . differ from our analysis in two important aspects . they apply ` reconstruction ' to the measured galaxy distribution to partially remove the nonlinear smearing of the bao feature , and marginalize over the broad - band shape of the correlation function , so that the estimate of the distance comes from the bao peak feature alone . @xcite and @xcite measured the distance redshift relationship using the legandre moments of power spectrum and correlation function , respectively . @xcite perform their analysis in fourier space . the @xcite analysis is in configuration space but uses a different range of scales and theoretical model than our work . despite differences in the applied methodology , the estimates are consistent within 1@xmath184 error bars . the growth rate , @xmath21 , has also been measured in the same redshift bin by ( * ? ? ? * dr11 ) , ( * ? ? ? * dr9 ) , ( * ? ? ? * dr11 ) and @xcite . the comparison of results is presented in fig . [ fig : comparefs8 ] . in the @xcite analysis , @xmath21 is a derived parameter computed by combining cmass data with _ planck _ assuming @xmath6cdm model ; their estimate is perfectly consistent with ours . the @xcite analysis is similar in the range of scales and theoretical modelling to the current paper , but performed on dr9 data set . all measurements are consistent with each other and are somewhat lower than the _ planck _ @xmath6cdm - gr expectations . from cmass dr9 and dr11 . the blue band corresponds to @xmath89 uncertainty in _ planck _ prediction assuming @xmath6cdm - gr . clustering measurements are mutually consistent and are lower than the cmb prediction . ] the fitting methodology adopted in this paper is identical to that used in our dr9 analysis @xcite , but some of the priors have been updated . we adopt a prior on the linear matter power spectrum shape from _ planck _ rather than _ wmap7 _ ; _ planck _ has substantially smaller errors , and so we expect the marginalization over the @xmath185 to contribute negligibly to our error budget in dr11 . we also adopted a slightly more conservative top - hat prior on @xmath186 , by increasing the allowed range from 0 40@xmath56 to 0 50@xmath56 , as the large - scale clustering data alone can - not well constrain this dispersion term ; we have checked that this change of prior range does not affect our best - fitting parameter values significantly . the effective area of dr11 is a factor of 2.5 larger than dr9 ; in the limit of negligible boundary effects , we would expect the covariance matrix on @xmath187 , and @xmath21 to be reduced by the same factor . a direct comparison indicates agreement at the @xmath188 level on the diagonals , with dr11 errors slightly larger than expected and with different off - diagonal structure . when projected on to @xmath21 ( at fixed @xmath189 and @xmath28 ) , which is the relevant case for the modified gravity constraints we present , our error in dr9 was 0.033 and is 0.028 in dr11 , while we would have expected 0.021 from the effective volumes this situation arises because , as we showed in table 2 of @xcite , the prior on @xmath186 reduces the uncertainty on @xmath21 in the fixed geometry case substantially . the statistical errors have shrunk significantly in dr11 , but we did not assume better prior knowledge on @xmath186 . fisher matrix analysis suggests that if @xmath186 parameter were perfectly known , the @xmath21 error would be reduced to 0.017 when the geometric and power spectrum parameters are held fixed . estimate and its effect on @xmath21 measurement see @xcite . ] in dr11 we obtain higher values for @xmath190 and @xmath21 , which brings us slightly closer to the values predicted by _ planck_. the @xmath183 offset between dr11 and dr9 results is just 0.3 per 3 degrees of freedom . to determine the separate contribution of monopole and quadrupole we perform the same fit to each individually . the monopole and quadrupole measurements on their own are unable to break the degeneracy between @xmath149 and @xmath21 and can only constrain combinations of the two . fig . [ fig : bf ] shows the constrains in @xmath149 @xmath21 derived from the two multipoles . the solid lines show the expected degeneracy directions based on the linear theory predictions . and @xmath21 from monopole and quadrupole separately . solid lines show expected directions of the principal components based on predictions of the linear theory . ] and @xmath146 from monopole and quadrupole separately . solid lines show expected directions of the principal components based on predictions of the linear theory . ] the quadrupole best constrains @xmath191 , as expected from the linear theory . the amplitude constraints from monopole are collinear to the combination @xmath192 , also as expected from linear theory . the ap parameters @xmath145 and @xmath146 show a qualitatively similar picture . individual multipoles can only constrain certain combinations of parameters . fig . [ fig : alphaepsilon ] presents constraints in the @xmath145 @xmath146 plane from the monopole and quadrupole separately . the solid lines show the expected degeneracy directions based on the linear theory predictions . the principal component of the monopole constraint coincides with @xmath193{\alpha_\bot^2\alpha_{||}}$ ] as expected from linear theory . the principal component of the quadrupole constraint is slightly tilted from the @xmath194 direction ; @xmath195 from the monopole and @xmath196 from the quadrupole . we next fit the monopole and quadrupole for the growth factor and ap parameters separately . first , we assume that the background expansion follows the predictions of spatially flat @xmath6cdm but allows the growth rate to be a free parameter . in this case , the parameters @xmath172 and @xmath146 can be computed from @xmath163 and @xmath197 . for this model , where the background expansion is assumed to be following the @xmath6cdm predictions , we find @xmath7 and @xmath198 . the constraint on growth improves to 6 per cent ( from 10 per cent ) and is perfectly consistent with the result of our more general fit . next , we assume that the growth rate follows the predictions of @xmath6cdm - gr , but let the expansion rate and the distance redshift relation vary . in this case @xmath21 is computed from @xmath163 but the @xmath172 and @xmath146 are free parameters . for this fit , we obtain @xmath199 and @xmath200 . constraints on @xmath172 move to a lower value and tighten to 2 per cent ( from 3 per cent ) , while constraints on @xmath146 move to a higher value and tighten to 1 per cent ( from 2 per cent ) . moments of the correlation function are measured with the best signal - to - noise ratio at small scales . the model that we use has been tested against numerical simulations with agreement at the per cent level down to @xmath201mpc . to determine explicitly the contribution of small scales on our fits , we redo the fit to the monopole and quadrupole keeping only scales above @xmath202mpc . the results of this fit and the comparison to the main results are shown in fig . [ fig : ls ] . , @xmath28 and @xmath29 from large - scales only ( green contours ) and from all scales ( red contours ) , along with predictions of @xmath6cdm - gr normalized by the _ planck _ all estimates are mutually consistent . ] the inclusion of scales between @xmath203mpc improves our constraints by approximately a factor of 2 . for the main parameters of interest we find @xmath204 , @xmath205 , @xmath206 , and @xmath207 . the biggest improvement is in the variables that are determined from the amplitudes of the multipoles such as @xmath149 , @xmath21 and @xmath28 . improvement in @xmath29 is more modest , because most of the information about this quantity is produced by the bao peak in the monopole at @xmath208mpc . small - scale clustering pushes @xmath21 and @xmath29 to higher values and @xmath28 to lower values . the two estimates , however , are highly consistent , the @xmath183 offset between the two being @xmath209 for 3 degrees of freedom . both sets of measurements are consistent with _ planck _ data . the @xmath183 difference between large - scale - only measurements and _ planck _ inferred values is 1.8 for 3 degrees of freedom , while the difference between large - scale - only measurements and the ones using all scales above 25@xmath57 mpc is 0.3 for 3 degrees of freedom . in following subsections , we constrain parameters of standard @xmath6cdm - gr model by combining our measurements with the cmb and previous , independent bao measurements . we also examine possible deviations from the standard model by considering phenomenological modifications to both @xmath6 and gr parts . as a cmb data set we adopt the recent measurements of cmb temperature fluctuations by the _ planck _ satellite @xcite supplemented by low-@xmath74 measurements of cmb polarization from the _ wmap _ misssion @xcite and the high-@xmath74 measurements from the atacama cosmology telescope ( * ? ? ? * act ) and the south pole telescope ( * ? ? ? * spt ) . for the rest of the paper , we will refer to this combination of cmb data as _ eplanck_. this affects the time of matter - radiation equality and angular - diameter distance to last scattering , as well as early integrated sachs wolfe effect and the lensing potential . ] . for our bao data compilation we use measurements from ( * ? ? ? * @xmath210 ) , ( * ? ? ? * @xmath211 ) . ] and ( * ? ? ? * @xmath212 ) . to sample cosmological parameter space , we use the monte carlo markov chain ( mcmc ) technique implemented by the cosmomc package @xcite . in a spatially flat @xmath6cdm - gr model , the expansion history of the universe and the growth of perturbations can be fully described by six parameters . we choose these to be @xmath213 $ ] . the mean values and 1@xmath184 confidence levels are listed in table [ tab : lcdm ] . parameter & _ eplanck _ & boss + _ eplanck _ & boss + _ eplanck _ + bao + @xmath214 & @xmath215 & @xmath216 & @xmath217 + @xmath218 & @xmath219 & @xmath220 & @xmath221 + @xmath157 & @xmath222 & @xmath223 & @xmath224 + @xmath225 & @xmath226 & @xmath227 & @xmath228 + @xmath229 & @xmath230 & @xmath231 & @xmath232 + @xmath197 & @xmath233 & @xmath234 & @xmath235 + by combining _ boss _ dr11 results with data , we are able to achieve a 1.6 per cent constraint on @xmath218 , a 1.3 per cent constraint on @xmath225 and a 1.2 per cent constraint on @xmath197 . after including bao data set , the constraints improve to 1.3 per cent on @xmath218 and a 1.0 per cent constraint on @xmath197 , while relative constraint on @xmath236 does not change . the constraints on @xmath237 and @xmath238 are dominated by the information from the _ eplanck _ data set . we now relax the assumption that the spatial curvature is zero and allow the @xmath239 parameter to vary along with @xmath240 . the posterior confidence regions on curvature and nonrelativistic matter density are shown in fig . [ fig : omokconstraints ] . and @xmath241 confidence levels in posterior likelihood . ] we find @xmath242 ( a 0.3 per cent constraint ) when combining _ boss _ dr11 with _ eplanck _ and @xmath243 ( a 0.3 per cent constraint ) when adding the bao compilation . in both cases the results are perfectly consistent with a spatially flat universe . alternative models of de predict a time - dependent eos @xmath159 . for a wide range of de models that do not exhibit sudden transitions or large amount of de at early times , for example models based on cosmological scalar fields , this time - dependence can be adequately parametrized by two parameters @xmath244 @xcite . ( for de models that do not belong to this family see e.g. * ? ? ? * ; * ? ? ? this reduces to our standard model for @xmath245 and @xmath246 . we first set @xmath247 to zero and check if there is an evidence for @xmath248 to differ from @xmath249 on average . the confidence level contours on @xmath248 and nonrelativistic matter density are shown in fig . [ fig : wconstraints ] . and nonrelativistic matter density from the combination of _ boss _ dr11 data with cmb and bao data sets . the contours correspond to 1@xmath184 and @xmath241 confidence levels in posterior . ] this analysis yields @xmath250 ( a 8 per cent constraint ) when _ boss _ dr11 is combined with _ eplanck _ data and @xmath251 ( a 6 per cent constraint ) when the bao compilation is added . in both cases , the results are perfectly consistent with a cosmological constant ( @xmath10 ) . our constraints on @xmath248 differ significantly from the dr9 results presented in @xcite , where we detected up to @xmath241 preference for @xmath252 . this change is mainly due to two differences . we now use _ eplanck _ as our cmb data set , which predicts a higher value for the non - relativistic matter density . also , our new measurements , although consistent with dr9 results , have moved in the direction that makes them more consistent with the cmb results ( see fig.[fig : dvffs8 ] ) . finally , we consider a model in which the spatial curvature is a free parameter and both @xmath253 and @xmath247 are allowed to vary . the constraints on this model are shown in fig . [ fig : fom ] . and @xmath247 from the combination of _ boss _ dr11 data with cmb and bao data sets . the contours correspond to 1@xmath184 and @xmath241 confidence levels in posterior likelihood . the @xmath6cdm prediction is consistent at the @xmath254 level . ] the dr11 data alone , even after combining with _ eplanck _ , is not able to break all the degeneracies of this large parameter space . when dr11 and _ eplanck _ are combined with the bao , we see a preference for larger values of @xmath253 and smaller values of @xmath247 . the @xmath6cdm value of @xmath245 and @xmath255 , however , is still within the @xmath241 confidence level . mg predict scale dependence of bias and growth rate even in the linear regime and the effect of small - scale screening mechanisms is difficult to model . this makes devising a completely self - consistent test of mg models a non - trivial task . a number of proposals for parametrizing families of mg models have been discussed recently ( see e.g. * ? ? ? * ; * ? ? ? these parametrizations , however , are difficult to correctly implement in practice for a few reasons . first , they rely on the linear theory and are not expected to work below scales of @xmath256mpc . secondly , they require a large number of free parameters and such a large parameter space can not be effectively constrained by current data . for these reasons , we follow the approach of @xcite and apply several few - parameter consistency tests to our measurements . we parametrize the growth rate as a function of @xmath163 using @xmath257^\gamma\ ] ] @xcite . this approach does not provide a fully self - consistent test of mg models , as mg models predict a more complex change in observables compared to gr . this parametrization is , however , easy to implement and provides a simple consistency test . in gr , we expect the @xmath258-index to be equal to 0.554 . measuring a significantly higher value would indicate a preference for a force weaker than gr gravity and vice versa . in our fits , we apply a hard prior of @xmath259 . when constraining deviations from gr , we fix de to be a cosmological constant . we also ignore the cmb power spectrum on large scales ( @xmath260 ) to ensure that cmb data are used only to constrain the background evolution . the parametrizations that we use are not physically motivated and are simply meant to describe effective gravity at low redshifts rather than provide a full model that works accurately at all redshifts up to last - scattering surface . index and nonrelativistic matter density from the combination of _ boss _ dr11 data with cmb , snia , bao and @xmath197 data sets . the contours correspond to 1@xmath184 and @xmath241 confidence levels in posterior likelihood . the best fit is consistent with gr at @xmath89 . ] constraints on @xmath258 and @xmath163 are shown in fig . [ fig : gammaconstraints ] . when combining _ boss _ dr11 with _ data , we recover @xmath261 ( a 16 per cent measurement ) . with the bao data set , we recover @xmath262 ( a 16 per cent measurement ) . the values are within @xmath263 confidence of gr values but favour a weaker gravity . next , we parametrize the linear equation of growth following the approach of @xcite as @xmath264 where @xmath265 is a matter overdensity , the overdot denotes a derivative with respect to @xmath266 , @xmath267 is the gravitational constant , and @xmath65 and @xmath268 are parameters describing deviations from gr . the gr limit is recovered when @xmath269 , where negative values of @xmath65 correspond to weaker than gr gravity and vice versa . the @xmath268 parameter dictates how rapidly the modifications are set larger values of @xmath268 corresponding to the modifications that appear at later times . since large values of @xmath268 correspond to models in which gravity is indistinguishable from gr until some low redshift when the modification suddenly becomes significant , they are basically unconstrained . we place a flat prior of @xmath270 to avoid this problem . the confidence level contours of @xmath65 and @xmath268 are shown in fig . [ fig : musconstraints ] . and @xmath268 from the combination of _ boss _ dr11 data with cmb and bao data sets . the contours correspond to 1@xmath184 and @xmath241 confidence levels in posterior likelihood . ] the gr predictions are within @xmath241 in posterior likelihood . similar to @xmath258-parametrization , the data again provide a mild preference for a weaker than gr gravity . this result is consistent with the dr9-based results reported in @xcite . we have used the anisotropic clustering of galaxies in the _ boss _ dr11 data set to simultaneously constrain the growth rate , the redshift distance relationship and the expansion rate at the redshift of @xmath33 . overall , our measurements are in good agreement with the results of the _ planck _ satellite propagated to low redshifts assuming @xmath6cdm - gr . by combining our measurements of @xmath136 , @xmath27 and @xmath28 with the cmb data we were able to derive tight constraints on basic cosmological parameters and parameters describing deviations from the @xmath6cdm - gr model . we were able to constrain the curvature of universe with 0.3 per cent precision , the de eos parameter @xmath248 with 8 per cent precision and the @xmath258-index for growth with 16 per cent precision . when we vary the background expansion within @xmath6cdm predictions of the _ planck _ data we measure the growth rate ( parametrized by @xmath258 ) to be weaker but consistent within @xmath263 of gr predictions . this preference for lower values of growth rate has also been observed in other similar low - redshift measurements ( see e.g. * ? ? ? * for discussion ) . our measurement of @xmath21 follows this trend but is closer to the gr predictions compared to the dr9 results of @xcite and the dr11 measurement of @xcite . similar measurements from a lower redshift ( lowz ) sample of _ boss _ galaxies will provide a complementary measurement of the growth rate in the de - dominated redshift range of @xmath271 , which will significantly strengthen the constraining power over possible gr modifications and can potentially increase the significance of the ` low growth rate ' signal . ls gratefully acknowledges support by the european research council . bar gratefully acknowledges support provided by nasa through hubble fellowship grant 51280 awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas 5 - 26555 . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national sci- ence foundation , and the u.s . department of energy office of sci- ence . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegiemellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , themichigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participationgroup , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . we acknowledge the use of mcmc data from planck legacy archive + ( http://www.sciops.esa.int/index.php?project=planck&page=planck_legacy_archive ) . numerical computations were done on the sciama high performance compute cluster which is supported by the icg , sepnet and the university of portsmouth . 999 albrecht a. et al . , 2009 , findings of the joint dark energy mission figure of merit science working group , arxiv:0901.0721 [ astro-ph.im ] alcock c. , paczynski b. , 1979 , nature , 281 , 358 anderson l. et al . , 2012 , mnras , 427 , 3435 anderson l. et al . , 2014 , mnras , 441 , 24 baker t. , ferreira p.g . , skordis c. , 2013 , prd , 87 , 024015 ballinger w.e . , peacock j.a . , heavens a.f . , 1996 , mnras , 282 , 877 bassett b.a . , hlozek r. , 2010 , in ruiz - lapuente p. , ed , dark energy : observational and theoretical approaches , cambridge university press , cambridge , p. 246 , arxiv:0910.5224 [ astro-ph.co ] battye r.a . , pearson j.a . , 2012 , jcap , 7 , 019 bennet c.l . et al . , 2013 , apjs , 208 , 20 beutler f. et al . , 2011 , mnras , 416 , 3017 beutler f. et al . , 2012 , mnras , 423 , 3430 beutler f. et al . , 2013 , mnras , 429 , 3604 bianchi d. et al . , 2012 , 427 , 2420 blake c. , kazin e.a . , beutler f. , davis t.m . , 2011 , mnras , 418 , 1707 blake c. , brough s. , colless m. , contreras c. , 2011 , mnras , 415 , 2876 blake c. , brough s. , colless m. , contreras c. , 2012 , 425 , 405 bloomfield j.k , flanagan e.e . , park m. , watson s. , 2012 , preprint , arxiv:1211.7054 [ astro-ph.co ] bolton a. et al . , 2012 , aj , 144 , 144 carlson j. , reid b.a . , white m. , 2013 , mnras , 429 , 1674 chuang c .- h . , wang y. , 2013 , mnras , 431 , 2634 chuang c .- h . et al . , 2012 , mnras , 2013 , 433 , 3559 chuang c .- h . et al . , 2013 , submitted to mnras , arxiv:1312.4889 [ astro-ph.co ] crocce m. , scoccimarro r. , bernardeau f. , 2012 , mnras , 427 , 2537 chevallier m. , polarski d. , 2001 , ijmpd , 10 , 213 das s. et al . , 2013 , preprint , arxiv:1301.1037 [ astro-ph.co ] dawson k. et al . , 2013 , aj , 145 , 10 doran m. , robbers g. , 2006 , jcap , 0606 , 026 eisenstein d.j . , hu w. , 1998 , apj , 496 , 605 eisenstein d.j . , j . , white m. , 2007 , apj , 664 , 660 eisenstein d.j . et al . , 2011 , aj , 142 , 72 elia a. , kulkarni s. , porciani c. , pietroni m. , matarrese s. , 2011 , mnras , 116 , 1703 feldman h.a . , kaiser n. , peacock j.a . , 1994 , apj , 426 , 23 feroz f. , hobson m.p . , bridges m. , 2009 , mnras , 398 , 1601 feroz m. , hobson m.p . , cameron e.,pettitt a.n . , 2013 , preprint , arxiv:1306.2144 [ astro-ph.co ] fukugita , m. et al . , 1996 , aj , 111 , 1748 gil - marin h. , wagner c. , verde l. , porcini c. , jimenez r. , jcap , 11 , 029 gunn j.e . et al . , 2006 , aj . , 116 , 3040 gunn j.e . et al . , 2013 , aj , 126 , 32 guzzo l. et al . , 2008 , nature , 451 , 541 hamilton a.j.s . , 1998 , preprint , arxiv : astro - ph/9708102 jackson j.c . , 1972 , mnras , 156 , 1 kazin e.a . , sanchez a.g . , blanton m.r . , 2012 , mnras , 419 , 3223 sanchez a. et al . , 2013 , mnras , 433 , 1202 sanchez a. et al . , 2014 , mnras , 440 , 2692 kaiser n. , 1987 , mnras , 227 , 1 landy s.d . , szalay a.s . , 1993 , apj , 412 , 64 lewis a. , bridle s. , 2001 , prd , 66 , 103511 linder e.v . , 2003 , prl , 90 , 091 linder e.v . , cahn r.n . , 2008 , app , 28 , 481 macaulay e. , wehus i.k . , eirksen h.k . , 2013 , prl , 111 , 161301 manera m. et al . , mnras 2013 , 428 , 1036 marulli f. , bianchi d. , branchini e. , guzzo l. , moscardini l. , angulo r.e . , mnras , 2012 , 426 , 2566 mehta k.t . , cuesta a.j . , xiaoying x. , eisenstein d.j . , padmanabhan n. , mnras , 2012 , 427 , 2168 meiksin a. , white m. , peacock j.a . , 1999 , mmras , 304 , 851 mueller e .- m . , bean r. , watson s. , 2013 , prd , 87 , 083504 okamura t. , taruya a. , matsubara t. , 2011 , jcap , 8 , 12 padmanabhan n. et al . , 2012 , mnras , 427 , 2132 percival w.j . et al . , 2014 , mnras , 439 , 2531 planck collaboration , 2013 , preprint , arxiv:1303.5076 [ astro-ph.co ] pogosian l. , silvestri a. , koyama k. , zhao g .- b . , 2010 , prd , 81 , 104023 rapetti d. et al . , 2013 , mnras , 432 , 973 reichart c.l . , shaw l. , zahn o. , 2012 , apj , 755 , 23 reid b.a . , white m. , 2011 , mnras , 417 , 1913 reid b.a . et al . , 2012 , mnras , 426 , 2719 reid b.a . et al . , 2014 , preprint , arxiv:1404.3742 [ astro-ph.co ] ross a.j . et al . , 2012 , mnras , 424 , 564 ross a.j . et al . , 2014 , 437 , 1109 samushia l. et al . , 2011 , mnras , 410 , 1993 samushia l. , percival w.j . , raccanelli a. , 2012 , mnras , 420 , 2102 samushia l. et al . , 2013 , mnras , 429 , 1514 simpson f. et al . , 2013 , mnras , 429 , 2249 smee s.a . et al . , 2013 , aj , 126 , 32 strateva i. et al . , 2001 , aj , 122 , 1861 taruya a. , nishimichi t. , saito s. , 2010 , prd , 82 , 063522 taruya a. , saito s. , nishimichi t. , 2011 , 83 , 103527 de la torre s. , guzzo g. , 2012 , mnras , 427 , 327 de la torre s. et al . , 2013 , a&a , 557 , 54 vlah z. , seljak u. , mcdonald p. , okamura t. , baldauf t. , 2012 , jcap , 11 , 009 wang l. , reid b.a . , white m. , 2013 , mnras , 437 , 588 weinberg d.h . , mortonson m.j . , eisenstein d.j . , hirata c. , reiss a.g . , eduardo r. , 2013 , phys . rep . , 530 , 87 wetterich c. , 2004 , plb , 594 , 17 zhao g .- b . , 2012 , prd , 85 , 123546 figure [ fig : gcshape ] displays the posterior likelihood of @xmath272 obtainable from _ dr11 _ data alone . the likelihood surface does not close even within @xmath273 of the cmb constraints . previous studies either fix the shape parameters to their cmb best - fit values ( e.g. * ? ? ? * ) , let @xmath274 vary and fix the rest ( e.g. * ? ? ? * ) , or marginalise over them by taking a prior centered around cmb best - fit values ( e.g. * ? ? ? * ) . we adopt a different approach and apply the cmb shape prior to our galaxy clustering likelihood . since later we will combine our results with planck data to obtain constraints on de and mg parameters one may be led to an erroneous impression that the cmb data is being double - counted . we demonstrate below that this is not the case . let @xmath275 be a cmb likelihood , where @xmath17 are shape parameters and @xmath139 are other parameters that may be related to de and gravity parameters of interest [ @xmath276 . let @xmath277 be galaxy likelihood , where @xmath278 are de and gravity dependent [ @xmath279 . for simplicity , assume @xmath17 , @xmath139 and @xmath278 to be scalars and that all likelihoods are multivariate gaussian . this demonstrates that direct combination of galaxy clustering and cmb data always produce stronger constraints on derived parameters and therefore the galaxy clustering measurements obtained by assuming a cmb prior on the shape can be combined with the cmb data without double counting the information .
the eigenvalue correlations of random matrices from the jacobi unitary ensemble have a known asymptotic behavior as their size tends to infinity . in the bulk of the spectrum the behavior is described in terms of the sine kernel , and at the edge in terms of the bessel kernel . we will prove that this behavior persists for the modified jacobi unitary ensemble . this generalization of the jacobi unitary ensemble is associated with the modified jacobi weight where the extra factor is assumed to be real analytic and strictly positive on $ ] . we use the connection with the orthogonal polynomials with respect to the modified jacobi weight , and recent results on strong asymptotics derived by k.t - r mclaughlin , w. van assche and the authors . * universality for eigenvalue correlations from the modified jacobi unitary ensemble * a.b.j . kuijlaars + _ department of mathematics , katholieke universiteit leuven , + celestijnenlaan 200 b , 3001 leuven , belgium + [email protected] + and + m. vanlessen + _ department of mathematics , katholieke universiteit leuven , + celestijnenlaan 200 b , 3001 leuven , belgium + [email protected]__ +
math-ph0204006
in the early sixties , dyson predicted that the local correlations between the eigenvalues of ensembles of random matrices , when their size tends to infinity , have universal behavior in the bulk of the spectrum . he expected that this universal behavior depends only on the type of the ensemble : orthogonal , unitary or symplectic . this constitutes the famous conjecture of universality in the theory of random matrices . for the classical ensembles ( hermite , laguerre and jacobi ) , this conjecture has been proven , see for example @xcite . for the unitary ensembles much more is known due to the connection with orthogonal polynomials , and the universality conjecture in the bulk of the spectrum is proved for a wide class of unitary ensembles , see @xcite . at the edge of the spectrum this universal behavior breaks down . for hermite ensembles , it is known that the local correlations ( at the soft edge ) can be expressed in terms of airy functions @xcite , and for jacobi and laguerre ensembles ( at the hard edge ) in terms of bessel functions @xcite . for example , for the jacobi unitary ensemble @xmath3 where @xmath4 is the jacobi weight , the eigenvalue correlations near @xmath5 are expressed in terms of the bessel kernel @xmath6 as @xmath7 . @xmath8 is the usual bessel function of the first kind and order @xmath9 . the order agrees with the exponent of @xmath10 in the jacobi weight . nagao and wadati ( * ? ? ? * @xmath11 ) expect that a universality result persists for more general jacobi - like ensembles , in the sense that the local form of the weight function near @xmath5 determines the eigenvalue correlation near @xmath5 . it is the aim of this paper to prove this universal behavior for a generalization of the jacobi unitary ensemble , which we call the modified jacobi unitary ensemble ( mjue ) . the mjue is given by ( [ jue ] ) with modified jacobi weight @xmath12 where @xmath13 and the extra factor @xmath1 is real analytic and strictly positive on @xmath2 $ ] . the modified jacobi ensemble is a probability measure on the space of @xmath14 hermitian matrices with all eigenvalues in @xmath15 . the mjue gives rise to a probability density function of the @xmath16 eigenvalues @xmath17 given by @xmath18 with @xmath19 and @xmath20 a normalizing constant . dyson @xcite showed , see also @xcite , that we can express the correlation functions @xmath21 @xmath22 in terms of orthogonal polynomials . denote the @xmath16th degree orthonormal polynomial with respect to the modified jacobi weight @xmath23 by @xmath24 , @xmath25 . then @xmath26 , where @xmath27 by the christoffel - darboux formula , we have @xmath28 which shows that asymptotic properties of @xmath29 are intimately related with asymptotics of the orthogonal polynomials @xmath30 as @xmath7 . in a previous paper with k.t - r mclaughlin and w. van assche @xcite , we studied the asymptotics of the polynomials that are orthogonal with respect to the modified jacobi weight . we used the riemann - hilbert formulation for orthogonal polynomials of fokas , its , and kitaev @xcite and the steepest descent method for riemann - hilbert problems of deift and zhou @xcite . in @xcite we concentrated on the asymptotics of the polynomials away from the interval @xmath2 $ ] , but the riemann - hilbert method gives uniform asymptotics in all regions in the complex plane . here we are interested in the behavior on @xmath2 $ ] , and in particular near the endpoints @xmath31 . the riemann - hilbert method was applied before to orthogonal polynomials by deift and co - authors @xcite . they studied orthogonal polynomials on the real line with varying weights , and used the asymptotics to prove the universality in the bulk of the spectrum for the associated unitary ensembles . we apply the same method to prove the universality at the edge of the spectrum for the mjue . our main result is the following . [ theoremuniversalityedge ] let @xmath23 be the modified jacobi weight ( [ definitiew ] ) and let @xmath29 be the kernel ( [ definitiekn ] ) associated with @xmath23 . then the following holds . 1 . for @xmath32 , we have as @xmath7 , @xmath33 the error term is uniform for @xmath34 in compact subsets of @xmath15 . 2 . let @xmath35 . then for @xmath32 and @xmath36 , we have as @xmath7 , @xmath37 the error term is uniform for @xmath34 in compact subsets of @xmath15 and for @xmath38 in compact subsets of @xmath39 . 3 . for @xmath40 , we have as @xmath7 , @xmath41 where @xmath42 is the bessel kernel given by ( [ besselkern ] ) . the error term is uniform for @xmath38 in bounded subsets of @xmath43 . note that the error term in ( [ universalnear1 ] ) holds uniformly for @xmath38 in _ bounded _ subsets of @xmath43 , not just in compact subsets . by symmetry , there is a corresponding universality result near @xmath44 . the eigenvalue density is the 1point correlation function @xmath45 , see for example @xcite . therefore , part ( a ) of the theorem yields the asymptotic eigenvalue density @xmath46 as @xmath47 . this result is in agreement with @xcite . the scaling in ( [ universalbulk ] ) has the effect that @xmath34 is the new origin and that the asymptotic eigenvalue density at @xmath34 is 1 . at the endpoints , ( [ densityofstates ] ) breaks down , and the eigenvalue density is @xmath48 as @xmath47 , near the endpoints , see for example @xcite . this explains the scaling in ( [ universalnear1 ] ) . part ( b ) of the theorem states the universality ( independent of the choice of @xmath49 and @xmath34 ) for @xmath29 in the bulk of the spectrum . it extends the result of nagao and wadati ( * ? ? ? * ( 4.19 ) ) for the case that @xmath50 . at the edge 1 of the spectrum we have a universality class for @xmath29 ( independent of the choice of @xmath51 and @xmath1 ) which is only affected by the local form of the modified jacobi weight near 1 , see part ( c ) . using theorem [ theoremuniversalityedge ] we can answer local statistical quantities concerning the eigenvalues . here we follow @xcite . the probability @xmath52 that there are no eigenvalues in the interval @xmath53 is given by @xmath54 where @xmath29 is the trace class operator with integral kernel @xmath55 acting on @xmath56 , and where @xmath57 is the fredholm determinant . for a fixed interval @xmath58 we have that @xmath59 , as @xmath47 . so , to understand the asymptotic behavior of @xmath60 at the edge of the spectrum , we will look at intervals near the edges which shrink with @xmath16 , and we are led to consider the asymptotic behavior of @xmath61 as @xmath47 , where @xmath62 . we have the following universality for @xmath60 at the edge 1 of the spectrum , depending on the parameter @xmath9 but independent of the choice of @xmath51 and @xmath1 . [ corollary2universalityedge ] for @xmath62 , we have @xmath63 where @xmath64 is the integral operator with kernel @xmath65 acting on @xmath66 , and @xmath67 is the fredholm determinant . as mentioned before , our main tool in proving theorem 1.1 is the asymptotic analysis of the riemann - hilbert problem for the orthogonal polynomials with respect to the modified jacobi weight , as developed in @xcite . we give an overview of this work in section 2 . this approach is able to give strong and uniform asymptotics for the orthogonal polynomials in every region in the complex plane , which we also review in section 2 . the proofs of theorem 1.1 and corollary 1.2 are given in section 3 . in this section we will recall the riemann hilbert problem ( rh problem ) from @xcite for the orthogonal polynomials for the modified jacobi weight , and the steps in the steepest descent method that are used to obtain the asymptotics on the interval . the analysis starts from a characterization of the monic orthogonal polynomials @xmath68 for the modified jacobi weight @xmath23 given by ( [ definitiew ] ) as a solution of a rh problem for a @xmath69 matrix valued function @xmath70 . this characterization of orthogonal polynomials is due to fokas , its , and kitaev @xcite . the conditions ( [ rhpyd ] ) and ( [ rhpye ] ) are needed to control the behavior near the endpoints , see @xcite for discussion . 1 . @xmath71 is analytic for @xmath72 $ ] . @xmath73 possesses continuous boundary values for @xmath32 denoted by @xmath74 and @xmath75 , where @xmath74 and @xmath75 denote the limiting values of @xmath71 as @xmath76 approaches @xmath34 from above and below , respectively , and @xmath77 3 . @xmath71 has the following asymptotic behavior at infinity : @xmath78 4 . @xmath71 has the following behavior near @xmath79 : @xmath80 o\begin{pmatrix } 1 & \log|z-1| \\ 1 & \log|z-1| \end{pmatrix } , & \mbox{if $ \alpha=0 $ , } \\[2ex ] o\begin{pmatrix } 1 & 1 \\ 1 & 1 \end{pmatrix } , & \mbox{if $ \alpha>0 $ , } \end{array}\right.\ ] ] as @xmath81 , @xmath82 $ ] . @xmath71 has the following behavior near @xmath83 : @xmath84 o\begin{pmatrix } 1 & \log|z+1| \\ 1 & \log|z+1| \end{pmatrix } , & \mbox{if $ \beta=0 $ , } \\[2ex ] o\begin{pmatrix } 1 & 1 \\ 1 & 1 \end{pmatrix } , & \mbox{if $ \beta>0 $ , } \end{array}\right.\ ] ] as @xmath85 , @xmath86 $ ] . the unique solution of this rh problem is given by @xmath87 -2\pi i \gamma_{n-1}^2 \pi_{n-1}(z ) & -\gamma_{n-1}^2 \int_{-1}^1 \frac{\pi_{n-1}(x)w(x)}{x - z } dx \end{pmatrix},\ ] ] where @xmath68 is the monic polynomial of degree @xmath16 orthogonal with respect to the weight @xmath23 and with @xmath88 the leading coefficient of the orthonormal polynomial @xmath30 . we apply a number of transformations @xmath89 to the original rh problem in order to arrive at a rh problem for @xmath90 , which is normalized at infinity , and whose jump matrices are uniformly close to the identity matrix . then , @xmath90 is uniformly close to the identity matrix , and by tracing back the steps we deduce the asymptotic behavior of @xmath73 . in the first transformation we turn the original rh problem into an equivalent rh problem for @xmath91 , which is normalized at infinity and with a jump matrix whose diagonal elements are oscillatory . let @xmath92 be the pauli matrix , and let @xmath91 be given by @xmath93 where @xmath94 for @xmath95 $ ] , so that @xmath96 is the conformal map from @xmath97 $ ] onto the exterior of the unit circle . then @xmath91 satisfies a rh problem which is normalized at infinity ( i.e. , @xmath98 as @xmath99 ) , and @xmath100 the jump matrix for @xmath91 factors into a product of three matrices . using this factorization , we transform the rh problem for @xmath91 into an equivalent rh problem for @xmath101 , with jumps on a lens shaped contour @xmath102 , as in figure [ figure1 ] . @xmath101 is defined in terms of @xmath91 by @xmath103 t(z ) \begin{pmatrix } 1 & 0 \\ -w(z)^{-1}\varphi(z)^{-2n } & 1 \end{pmatrix } , & \mbox{for $ z$ in the upper part of the lens , } \\[2ex ] t(z ) \begin{pmatrix } 1 & 0 \\ w(z)^{-1}\varphi(z)^{-2n } & 1 \end{pmatrix } , & \mbox{for $ z$ in the lower part of the lens . } \end{array}\right.\ ] ] for the third transformation , we need to construct parametrices in the outside region and near the endpoints @xmath31 . constructing the parametrix in the outside region , we need the szeg function @xmath104 associated with the weight @xmath23 given by @xmath105.\ ] ] the szeg function is a non - zero analytic function on @xmath106 $ ] such that @xmath107 for @xmath32 . the parametrix @xmath108 in the outside region is given by @xmath109 \frac{a(z ) - a(z)^{-1}}{-2i } & \frac{a(z ) + a(z)^{-1}}{2 } \end{pmatrix } d(z)^{-\sigma_3},\ ] ] where @xmath110 and @xmath111 . next , we define a parametrix @xmath112 in @xmath113 , which is the disk with radius @xmath114 and center @xmath5 , where @xmath115 is sufficiently small , by @xmath116 where @xmath117 and @xmath118 are scalar functions given by @xmath119 and @xmath120 in ( [ rhppsolution ] ) @xmath121 is a @xmath122 matrix valued function defined for @xmath123 , where @xmath124 is the contour shown in figure [ figure2 ] . for our purpose here , it suffices to know the expression of @xmath121 for @xmath125 . this is given in terms of the hankel functions @xmath126 and @xmath127 . for @xmath128 , we have @xmath129 \pi \zeta^{1/2 } \left(h_{\alpha}^{(1)}\right)'(2(-\zeta)^{1/2 } ) & \pi \zeta^{1/2 } \left(h_{\alpha}^{(2)}\right)'(2(-\zeta)^{1/2 } ) \end{pmatrix } e^{\frac{1}{2}\alpha\pi i \sigma_3 } .\ ] ] the factor @xmath130 in ( [ rhppsolution ] ) is given by @xmath131 @xmath130 is analytic in a full neighborhood of @xmath132 . there is a similar definition for the parametrix @xmath133 in a @xmath114 neighborhood @xmath134 of @xmath44 , see @xcite for details . we then have all the ingredients for the third transformation . we define @xmath135 \label{defru } r(z ) & = & s(z ) p^{-1}(z),\qquad \mbox{for } z \in u_{\delta } \setminus \sigma , \\[1ex ] \label{defrutilde } r(z ) & = & s(z ) \tilde{p}^{-1}(z),\qquad \mbox{for } z \in \tilde{u}_{\delta } \setminus\sigma . \end{aligned}\ ] ] then @xmath90 satisfies a rh problem with jumps on the system of contours @xmath136 shown in figure [ figure3 ] . note that @xmath90 depends on @xmath16 , but the contour @xmath136 does not depend on @xmath16 . the jump matrices for the rh problem for @xmath90 turn out to be uniformly close to the identity matrix with error term @xmath137 . then it follows that @xmath138 uniformly for @xmath139 , and also , see ( * ? ? ? * section 8.1 ) in particular formulas ( 8.19 ) and ( 8.20 ) , @xmath140 uniformly for @xmath141 . in the following we will also use that @xmath142 the asymptotic result ( [ asymptoticsr ] ) for @xmath90 may be refined and it is possible to obtain a full asymptotic expansion @xmath143 the matrix coefficients @xmath144 are explicitly computable in function of @xmath9 , @xmath51 , and the analytic factor @xmath1 in the modified jacobi weight . the first two coefficients are computed in @xcite . in @xcite we also derived asymptotic expansions for the polynomials @xmath30 and @xmath68 uniformly valid in compact subsets of @xmath145 $ ] , as well as asymptotic expansions for the leading coefficient @xmath88 and for the coefficients in the recurrence relation satisfied by the orthonormal polynomials . for example , we have by ( * ? ? ? * theorem 1.4 ) uniformly for @xmath82 $ ] , @xmath146 , \qquad \mbox{as } n \to \infty.\ ] ] the @xmath137 term can be developed into a complete asymptotic expansion in powers of @xmath147 . the riemann - hilbert method also leads to strong asymptotics on the interval @xmath15 and near the endpoints @xmath31 . while these results are closely related to the asymptotics of @xmath29 as given in theorem 1.1 , we do not actually rely on them in the proof of theorem 1.1 . therefore we state here the asymptotics of the orthogonal polynomials without proof . the proof is similar to the proof of theorem 1.1 , and in fact somewhat simpler . see also @xcite . as before , we let @xmath148 be the radius of the disks @xmath132 and @xmath134 . then we have for @xmath149 , @xmath150 , \label{asymptotiekpinbulk}\end{aligned}\ ] ] where @xmath151 and @xmath152 as @xmath7 , uniformly for @xmath153 . the @xmath137 terms have a complete asymptotic expansion , see @xcite . the function @xmath154 in ( [ asymptotiekpinbulk ] ) is given by @xmath155 where the integral is a cauchy principal value integral . we remark that , under various assumptions , asymptotic results on the interval of orthogonality have been established by many authors , see for example @xcite . for @xmath156 , the following is valid @xmath157 & & \nonumber \left[r_{11}(x ) \left(\cos \zeta_1(x ) j_{\alpha}(n \arccos x ) + \sin \zeta_1(x ) j_{\alpha}'(n \arccos x ) \right ) \right . \\ & & \qquad \left . - \frac{i}{d_{\infty}^2 } r_{12}(x ) \left(\cos \zeta_2(x ) j_{\alpha}(n \arccos x ) + \sin \zeta_2(x ) j_{\alpha}'(n \arccos x ) \right ) \right],\end{aligned}\ ] ] where @xmath151 and @xmath152 as @xmath158 , uniformly for @xmath159 , and where @xmath160 with @xmath161 given by ( [ functiepsi ] ) . there is an analogous expression for @xmath34 in the interval @xmath162 . in this section we prove theorem [ theoremuniversalityedge ] and corollary [ corollary2universalityedge ] . we follow the work of deift et al . @xcite . recall that @xmath163 where @xmath68 is the monic orthogonal polynomial of degree @xmath16 with respect to the modified jacobi weight @xmath23 . as in @xcite we replace the polynomials @xmath164 and @xmath68 by the appropriate entries of @xmath73 , see ( [ rhpysolution ] ) , to obtain @xmath165 thus , @xmath29 can be expressed in terms of the first column of @xmath73 . the asymptotic behavior of @xmath73 follows from the transformations @xmath166 described in section 2 , and the behavior ( [ asymptoticsr])([determinantr ] ) of @xmath90 . we will first express @xmath167 and @xmath168 in terms of @xmath90 . in the following , @xmath148 will be a small but fixed number . this number is the radius of the disks @xmath132 and @xmath134 used in the local rh analysis around @xmath31 . [ thasbulk ] we have for @xmath169 , @xmath170 with @xmath171 given by @xmath172 \frac{a(z ) - a(z)^{-1}}{-2i } & \frac{a(z ) + a(z)^{-1}}{2 } \end{pmatrix } e^{i\psi(z)\sigma_3},\ ] ] where @xmath173 the matrices @xmath174 and @xmath175 are uniformly bounded for @xmath149 as @xmath47 , and @xmath176 we use the series of transformations @xmath166 and we unfold them for @xmath76 in the upper part of the lens but outside the disks @xmath113 and @xmath179 , and then take the limit to the interval @xmath180 . thus , let @xmath76 be in the upper part of the lens but outside the disks @xmath113 and @xmath179 . we then have by ( [ yint ] ) , ( [ tins ] ) and ( [ defrbuiten ] ) @xmath181 inserting the expression ( [ rhpnsolution ] ) for @xmath108 into ( [ bulk1 ] ) , we obtain for the first column of @xmath73 @xmath182 \frac{a(z ) - a(z)^{-1}}{-2i } & \frac{a(z ) + a(z)^{-1}}{2 } \end{pmatrix } \begin{pmatrix } \frac{1}{d(z)}\ , \varphi(z)^{n } \\[2ex ] \frac{d(z)}{w(z)}\ , \varphi(z)^{-n } \end{pmatrix}.\ ] ] we now take the limit @xmath183 . since @xmath184 and @xmath185 , ( [ columnyinbulk ] ) now follows from ( [ prthasbulkeq ] ) . note that @xmath186 and @xmath187 are uniformly bounded for @xmath188 as @xmath47 . let @xmath189 be a neighborhood of @xmath2 $ ] such that @xmath190 is defined and analytic in @xmath189 , and let @xmath191 be a closed contour in @xmath192 $ ] encircling the interval @xmath2 $ ] once in the positive direction . via contour deformation , we may write @xmath161 in the form @xmath193 thus @xmath161 has an analytic extension to a neighborhood of @xmath15 . this implies that @xmath161 and its derivative are bounded on @xmath194 . from the explicit form ( [ definitiel ] ) of @xmath195 we then find that @xmath174 and @xmath175 are uniformly bounded for @xmath196 as @xmath47 . since @xmath197 it is easy to see from ( [ definitiel ] ) that @xmath198 . * of theorem 1.1 ( a ) * letting @xmath199 in ( [ kniny ] ) we get @xmath200 & = & \nonumber \frac{1}{2\pi i}\det \begin{pmatrix } \sqrt{w(x ) } y_{11}(x ) & \sqrt{w(x)}y'_{11}(x ) \\ \sqrt{w(x ) } y_{21}(x ) & \sqrt{w(x)}y'_{21}(x ) \end{pmatrix } \\[1ex ] & = & \label{prldeq1 } \frac{1}{2\pi i}\det \begin{pmatrix } 2^n \sqrt{w(x)}y_{11}(x ) & \frac{d}{dx}\left(2^n \sqrt{w(x)}y_{11}(x)\right ) \\ 2^{-n } \sqrt{w(x)}y_{21}(x ) & \frac{d}{dx}\left(2^{-n } \sqrt{w(x)}y_{21}(x)\right ) \end{pmatrix}. \end{aligned}\ ] ] by ( [ columnyinbulk ] ) the matrix in ( [ prldeq1 ] ) is equal to @xmath201 & & = l_+(x ) e^{in \arccos x \sigma_3 } \begin{pmatrix } 1 & -\frac{in}{\sqrt{1-x^2 } } \\ 1 & \frac{in}{\sqrt{1-x^2 } } \end{pmatrix } + \left(\frac{d}{dx } l_+(x)\right ) \begin{pmatrix } 0 & e^{in \arccos x } \\ 0 & e^{-in \arccos x } \end{pmatrix}. \end{aligned}\ ] ] since @xmath174 and @xmath202 have determinant one , it follows that @xmath203.\ ] ] the entries of @xmath174 and @xmath204 are uniformly bounded by lemma [ thasbulk ] . since @xmath205 , also the entries of @xmath206 are uniformly bounded . thus , we have uniformly for @xmath207 , @xmath208 \nonumber & = & \frac{n}{\pi \sqrt{1-x^2 } } + o(1 ) . \end{aligned}\ ] ] this proves part ( a ) of theorem 1.1 . * of theorem 1.1 ( b ) * let @xmath32 and @xmath36 . for the sake of brevity , we use @xmath209 and @xmath210 to denote @xmath211 and @xmath212 , respectively . we write @xmath213 from ( [ kniny ] ) we then have @xmath214 & = & \label{prthbulkeq1a } -\frac{1}{2\pi i(u - v ) } \det \begin{pmatrix } 2^n \sqrt{w(u_{x , n } ) } y_{11}(u_{x , n } ) & 2^n \sqrt{w(v_{x , n } ) } y_{11}(v_{x , n } ) \\ 2^{-n } \sqrt{w(u_{x , n } ) } y_{21}(u_{x , n } ) & 2^{-n } \sqrt{w(v_{x , n } ) } y_{21}(v_{x , n } ) \end{pmatrix}. \end{aligned}\ ] ] next , we use the expression ( [ columnyinbulk ] ) for @xmath167 and @xmath168 in ( [ prthbulkeq1a ] ) to obtain @xmath215 & & \nonumber + \ , \left . l_+(v_{x , n } ) \begin{pmatrix } 0 & e^{in\arccos v_{x , n } } \\ 0 & e^{-in\arccos v_{x , n } } \end{pmatrix } \right ] \\[2ex ] & = & \nonumber -\frac{1}{2\pi i(u - v ) } \det\left [ l_+(v_{x , n } ) \begin{pmatrix } e^{in \arccos u_{x , n } } & e^{in \arccos v_{x , n } } \\ e^{-in \arccos u_{x , n } } & e^{-in \arccos v_{x , n } } \end{pmatrix } \right . \\[1ex ] \label{prthbulkeq1 } & & \left . + \ , \left[l_+(u_{x , n})-l_+(v_{x , n})\right ] \begin{pmatrix } e^{i n \arccos u_{x , n } } & 0 \\ e^{-in \arccos u_{x , n } } & 0 \end{pmatrix } \right ] . \end{aligned}\ ] ] since @xmath216 , we then get @xmath217 \label{prthbulkeq1b } & & \left . + \ , l_+^{-1}(v_{x , n } ) \left[l_+(u_{x , n})-l_+(v_{x , n})\right ] \begin{pmatrix } e^{i n \arccos u_{x , n } } & 0 \\ e^{-in \arccos u_{x , n } } & 0 \end{pmatrix } \right ] . \end{aligned}\ ] ] since @xmath175 is uniformly bounded as @xmath47 , we have by the mean value theorem @xmath218 uniformly for @xmath149 and for @xmath38 in compact subsets of @xmath39 . since @xmath174 is uniformly bounded , and @xmath205 , we have that @xmath219 is uniformly bounded as well , so that @xmath220 \begin{pmatrix } e^{in\arccos u_{x , n } } & 0 \\ e^{-in\arccos u_{x , n } } & 0 \end{pmatrix } = \begin{pmatrix } o\left(\frac{u - v}{n}\right ) & 0 \\[1ex ] o\left(\frac{u - v}{n}\right ) & 0 \end{pmatrix}.\ ] ] thus by ( [ prthbulkeq1b ] ) and ( [ prthbulkeq2 ] ) @xmath221 e^{-in \arccos u_{x , n } } + o\left(\frac{u - v}{n}\right ) & e^{-in \arccos v_{x , n } } \end{pmatrix } \\[1ex ] \nonumber & = & -\frac{1}{2\pi i(u - v ) } \det \begin{pmatrix } e^{in \arccos u_{x , n } } & e^{in \arccos v_{x , n } } \\ e^{-in\arccos u_{x , n } } & e^{-in \arccos v_{x , n } } \end{pmatrix } + o\left(\frac{1}{n}\right ) \\[1ex ] \nonumber & = & - \frac{1}{2\pi i(u - v ) } \left ( e^{in ( \arccos u_{x , n}-\arccos v_{x , n } ) } - e^{-in ( \arccos u_{x , n}- \arccos v_{x , n } ) } \right ) + o\left(\frac{1}{n}\right ) \\[1ex ] & = & - \frac{\sin(n ( \arccos u_{x , n } - \arccos v_{x , n}))}{\pi(u - v ) } + o\left(\frac{1}{n } \right ) , \end{aligned}\ ] ] uniformly for @xmath149 and @xmath38 in compact subsets of @xmath222 . since @xmath223 also uniformly for @xmath149 and @xmath38 in compact subsets of @xmath39 , part ( b ) of theorem 1.1 follows . [ universalityedgetheorem1 ] for @xmath224 , we have @xmath225 with @xmath226 given by @xmath227 where @xmath90 is the result of the transformations @xmath166 of the rh problem , the matrix valued function @xmath108 is given by ( [ rhpnsolution ] ) , and the scalar functions @xmath228 and @xmath229 are given by ( [ deff(z ) ] ) and ( [ w^2(z ) ] ) , respectively . as in the proof of lemma [ thasbulk ] we unravel the transformations @xmath89 , but now for @xmath76 in the upper part of the lens and inside the disk @xmath113 . we then have by ( [ yint ] ) , ( [ tins ] ) , ( [ rhppsolution ] ) and ( [ defru ] ) @xmath234 since @xmath235 , we have by ( [ w^2(z ) ] ) that @xmath236 . inserting this into ( [ asymptotiekrond1eq1 ] ) we get for the first column of @xmath73 @xmath237 since @xmath76 is in the upper part of the lens and inside the disk @xmath113 , we have @xmath238 , see @xcite , and we thus use ( [ rhppsisolution2 ] ) to evaluate @xmath239 . from formulas 9.1.3 and 9.1.4 of @xcite we then have @xmath240 2\pi n f^{1/2}(z)j'_{\alpha}(2n(-f(z))^{1/2 } ) \end{pmatrix}.\ ] ] by ( [ eq536 ] ) and ( [ definitiem ] ) we have @xmath241 . inserting this and ( [ pr1 ] ) into ( [ asymptotiekrond1eq2 ] ) we get @xmath242 f^{1/2}(z)j'_{\alpha}(2n(-f(z))^{1/2 } ) \end{pmatrix}.\ ] ] we now take the limit @xmath243 . by ( [ deff(z ) ] ) , and since @xmath244 we have @xmath245 , so that @xmath246 . inserting this into ( [ asymptotiekrond1eq3 ] ) , we obtain ( [ universalityedgetheorem1eq ] ) . @xmath230 is analytic in @xmath113 since both @xmath247 , see ( * ? ? ? * proposition 6.5 ) , and @xmath90 are analytic in @xmath113 . so , we may write @xmath248 instead of @xmath249 in ( [ universalityedgetheorem1eq ] ) . by ( [ asymptoticsr ] ) and ( [ asymptoticsdr ] ) we have that @xmath250 and @xmath251 are uniformly bounded for @xmath232 as @xmath47 . since @xmath252 is analytic for @xmath232 and does not depend on @xmath16 , we have from ( [ definitiem ] ) that also @xmath226 and @xmath231 are uniformly bounded for @xmath232 as @xmath47 . we check that ( [ universalityedgetheorem1eq ] ) yields @xmath254 as @xmath255 , which is in agreement with the fact that @xmath167 and @xmath168 are polynomials , see ( [ rhpysolution ] ) . since @xmath256 as @xmath255 , we have by formula 9.1.10 of @xcite @xmath257 since @xmath230 is analytic near 1 , see lemma [ universalityedgetheorem1 ] , we have @xmath258 as @xmath255 . inserting ( [ newremarkedge ] ) into ( [ universalityedgetheorem1eq ] ) and noting that @xmath259 as @xmath260 , we then have indeed that @xmath261 and @xmath262 remain bounded as @xmath255 . [ universalityedgelemma1 ] let @xmath267 , @xmath268 , and @xmath266 . we then have as @xmath47 , @xmath269 \label{universalityedgelemma1eq2 } \lefteqn { j_\alpha(\tilde{u}_n ) = j_\alpha(\sqrt{u})+o\left(\frac{u^{\frac{\alpha}{2}+1}}{n^2}\right ) , } \\[1ex ] \label{universalityedgelemma1eq3 } \lefteqn { j'_\alpha(\tilde{u}_n ) = j'_\alpha(\sqrt{u})+o\left(\frac{u^{\frac{\alpha}{2}+\frac{1}{2}}}{n^2}\right ) . } \end{aligned}\ ] ] the error terms hold uniformly for @xmath270 in bounded subsets of @xmath43 . by formula 9.1.10 of @xcite we know that @xmath274 with @xmath275 an entire function . from ( [ universalityedgelemma1eq1 ] ) and taylor s formula we then get uniformly for @xmath270 in bounded subsets of @xmath43 , @xmath276 \nonumber & = & j_{\alpha}(\sqrt u)+o\left(\frac{u^{\frac{\alpha}{2}+1}}{n^2}\right ) , \end{aligned}\ ] ] which is ( [ universalityedgelemma1eq2 ] ) . the proof of ( [ universalityedgelemma1eq3 ] ) follows in a similar fashion , using @xmath277 and taylor s formula again . * of theorem 1.1 ( c ) * let @xmath278 and define @xmath279 we put @xmath280 from ( [ kniny ] ) we then have @xmath281 & = & \frac{1}{2\pi i ( u - v ) } \det \begin{pmatrix } 2^n \sqrt{w(u_n ) } y_{11}(u_n ) & 2^n \sqrt{w(v_n ) } y_{11}(v_n ) \\ 2^{-n } \sqrt{w(u_n ) } y_{21}(u_n ) & 2^{-n } \sqrt{w(v_n ) } y_{21}(v_n ) \end{pmatrix}. \label{prth1eq1a } \end{aligned}\ ] ] next , we replace the two columns in the determinant in ( [ prth1eq1a ] ) by the expression ( [ universalityedgetheorem1eq ] ) we found in lemma [ universalityedgetheorem1 ] . it follows that @xmath282 \frac{i}{2n } \tilde{u}_n j'_\alpha(\tilde{u}_n ) & 0 \end{pmatrix } + m(v_n ) \begin{pmatrix } 0 & j_\alpha(\tilde{v}_n ) \\[1ex ] 0 & \frac{i}{2n } \tilde{v}_n j'_\alpha(\tilde{v}_n ) \end{pmatrix } \right ] . \qquad\ ] ] we rewrite the matrix appearing in the determinant in ( [ prth1eq1b ] ) as @xmath283 \frac{i}{2n } \tilde{u}_n j'_\alpha(\tilde{u}_n ) & \frac{i}{2n } \tilde{v}_n j'_\alpha(\tilde{v}_n ) \end{pmatrix } + m(v_n)^{-1 } \bigl[m(u_n)-m(v_n ) \bigr ] \begin{pmatrix } j_{\alpha}(\tilde{u}_n ) & 0 \\[1ex ] \frac{i}{2n } \tilde{u}_n j'_\alpha(\tilde{u}_n ) & 0 \end{pmatrix } \right].\ ] ] now we use @xmath284 and the fact that @xmath226 is uniformly bounded for @xmath285 , see lemma [ universalityedgetheorem1 ] , to conclude that the entries of @xmath286 are uniformly bounded . by lemma [ universalityedgetheorem1 ] , we also have that @xmath287 is uniformly bounded so that @xmath288 . from lemma [ universalityedgelemma1 ] it follows that @xmath289 and @xmath290 uniformly for @xmath270 in bounded subsets of @xmath43 as @xmath47 . hence we have , uniformly for @xmath38 in bounded subsets of @xmath43 , @xmath291 \begin{pmatrix } j_\alpha(\tilde{u}_n ) & 0 \\[1ex ] \frac{i}{2n } \tilde{u}_n j'_\alpha(\tilde{u}_n ) & 0 \end{pmatrix } = \begin{pmatrix } o\left(\frac{u - v}{n^2}\ , u^{\frac{\alpha}{2}}\right ) & 0 \\[1ex ] o\left(\frac{u - v}{n^2}\ , u^{\frac{\alpha}{2}}\right ) & 0 \end{pmatrix}.\ ] ] it now follows that ( we use @xmath284 ) @xmath292 \tilde{u}_n j'_\alpha ( \tilde{u}_n)+o\left(\frac{u - v}{n}\ , u^{\frac{\alpha}{2}}\right ) & \tilde{v}_n j'_\alpha(\tilde{v}_n ) \end{pmatrix}.\ ] ] since @xmath293 and @xmath294 as @xmath47 , we then get uniformly for @xmath38 in bounded subsets of @xmath43 , @xmath295 \tilde{u}_n j'_\alpha(\tilde{u}_n ) & \tilde{v}_n j'_\alpha(\tilde{v}_n ) \end{pmatrix } + o\left(\frac{u^\frac{\alpha}{2}v^\frac{\alpha}{2}}{n}\right).\ ] ] in the determinant in ( [ kansintervalafschattingeq3 ] ) we can replace @xmath296 and @xmath297 by @xmath298 and @xmath299 respectively , and make an error which we could estimate using lemma [ universalityedgelemma1 ] . however , this estimate would not be uniform for @xmath300 close to zero . so we will be more careful . we bring in a factor @xmath301 into the first column of the determinant in ( [ kansintervalafschattingeq3 ] ) and a factor @xmath302 into the second . then we subtract the second column from the first to obtain @xmath303 u^{-\frac{\alpha}{2}}\tilde{u}_n j'_\alpha(\tilde{u}_n)- v^{-\frac{\alpha}{2}}\tilde{v}_n j'_\alpha(\tilde{v}_n ) & v^{-\frac{\alpha}{2}}\tilde{v}_n j'_\alpha(\tilde{v}_n ) \end{pmatrix } + o\left(\frac{u^\frac{\alpha}{2}v^\frac{\alpha}{2}}{n}\right).\ \ ] ] from lemma [ universalityedgelemma1 ] it follows that uniformly for @xmath34 in bounded subsets of @xmath43 , @xmath304 then it easily follows that the @xmath305-entry in the determinant in ( [ kansintervalafschattingeq4 ] ) is equal to @xmath306 similarly , if we use @xmath307 we find that the @xmath308-entry is @xmath309 from lemma [ universalityedgelemma1 ] it also follows that we may replace @xmath297 by @xmath299 in the second column at the expense of an error term @xmath310 . therefore , uniformly for @xmath38 in bounded subsets of @xmath43 , @xmath311 \nonumber & & \ , \times \ , \det \begin{pmatrix } u^{-\frac{\alpha}{2}}j_\alpha(\sqrt u)-v^{-\frac{\alpha}{2}}j_\alpha(\sqrt v ) + o\left(\frac{u - v}{n^2}\right ) & v^{-\frac{\alpha}{2}}j_\alpha(\sqrt v)+o\left(\frac{1}{n^2}\right ) \\[1ex ] u^{-\frac{\alpha}{2}}\sqrt u j'_\alpha(\sqrt u)- v^{-\frac{\alpha}{2}}\sqrt v j'_\alpha(\sqrt v ) + o\left(\frac{u - v}{n^2}\right ) & v^{-\frac{\alpha}{2}}\sqrt v j'_\alpha(\sqrt v)+o\left(\frac{1}{n^2}\right ) \end{pmatrix } \\[1ex ] & & \nonumber \qquad + \ , o\left(\frac{u^{\frac{\alpha}{2 } } v^{\frac{\alpha}{2}}}{n } \right ) \\[2ex ] & & \nonumber = \mathbb{j}_\alpha(u , v)+ \frac{u^{\frac{\alpha}{2}}v^{\frac{\alpha}{2}}}{2(u - v ) } \det \begin{pmatrix } u^{-\frac{\alpha}{2}}j_\alpha(\sqrt u)-v^{-\frac{\alpha}{2}}j_\alpha(\sqrt v ) & o\left(\frac{1}{n^2}\right ) \\[1ex ] u^{-\frac{\alpha}{2}}\sqrt u j'_\alpha(\sqrt u)- v^{-\frac{\alpha}{2}}\sqrt v j'_\alpha(\sqrt v ) & o\left(\frac{1}{n^2}\right ) \end{pmatrix } \\[1ex ] & & \label{kansintervalafschattingeq6 } \qquad + \ , o\left(\frac{u^\frac{\alpha}{2}v^\frac{\alpha}{2}}{n}\right ) . \end{aligned}\ ] ] since @xmath312 is an entire function we get by the mean value theorem that @xmath313 is bounded for @xmath38 in bounded subsets of @xmath43 , and similarly , that @xmath314 is bounded for @xmath38 in bounded subsets of @xmath43 . therefore , we have by ( [ kansintervalafschattingeq6 ] ) @xmath315 uniformly for @xmath38 in bounded subsets of @xmath43 , which completes the proof of part ( c ) of theorem 1.1 . since @xmath318 u^{-\frac{\alpha}{2}}\sqrt u j'_\alpha(\sqrt u)- v^{-\frac{\alpha}{2}}\sqrt v j'_\alpha(\sqrt v ) & v^{-\frac{\alpha}{2}}\sqrt v j'_\alpha(\sqrt v ) \end{pmatrix},\ ] ] and since @xmath319 and @xmath320 are entire functions of @xmath321 , it is easy to see , by the discussion following ( [ kansintervalafschattingeq6 ] ) , that for every @xmath322 , there exists a constant @xmath323 so that @xmath324 therefore , ( [ kansintervalgedrag ] ) is an immediate consequence of theorem [ theoremuniversalityedge ] . let @xmath328 and @xmath329 . since @xmath330 we have that @xmath325 is a finite rank operator and hence a trace class operator . now , for every @xmath331 we have by ( [ herhalingdefinitied_n ] ) @xmath332 so that @xmath325 is a positive operator . letting @xmath7 , we have that @xmath333 for every @xmath38 by theorem [ theoremuniversalityedge](c ) . by ( [ kansintervalgedrag ] ) , there is a constant @xmath323 independent of @xmath16 so that for @xmath334 $ ] , @xmath335 then by the dominated convergence theorem and ( [ d_npositief ] ) , @xmath336 so that @xmath42 is positive . since @xmath337 for @xmath334 $ ] we also have @xmath338 which implies that @xmath64 is a trace class operator . * of corollary [ corollary2universalityedge ] * it is well known that , see for example @xcite , @xmath339 for @xmath340 , we have that @xmath341 is continuous on @xmath342 \times [ 0,s]$ ] . then it follows as in @xcite that @xmath343 is equal to the fredholm determinant @xmath344 for @xmath345 , the integral kernel @xmath341 is not continuous , but satisfies an estimate @xmath346 for @xmath347 $ ] , for some constant @xmath323 . this is enough to establish ( [ pn = fredholmdeterminant ] ) also in this case . by theorem 3.4 of @xcite we have @xmath348 where @xmath349 is the trace norm in @xmath66 . so , in order to prove that @xmath350 converges to @xmath351 as @xmath47 , it is enough to show that @xmath325 tends to @xmath64 in trace norm . by theorem 2.20 of @xcite and the positivity of @xmath325 and @xmath64 , it then suffices to prove that @xmath352 weakly , and that @xmath353 as @xmath47 . so we have to prove that for @xmath354 , @xmath355 and that @xmath356 both ( [ weakconvergence ] ) and ( [ traceconvergence ] ) follow easily from the pointwise convergence @xmath357 , the uniform bound @xmath358 and the dominated convergence theorem . 99 m. abramowitz and i.a . stegun , handbook of mathematical functions , " dover publications , new york , 1968 . p. bleher and a. its , semiclassical asymptotics of orthogonal polynomials , riemann hilbert problem , and universality in the matrix model , _ ann . _ * 150 * ( 1999 ) , 185266 . p. deift , orthogonal polynomials and random matrices : a riemann hilbert approach , " courant lecture notes 3 , new york university , 1999 . p. deift , t. kriecherbauer , k. t - r mclaughlin , s. venakides , and x. zhou , uniform asymptotics for polynomials orthogonal with respect to varying exponential weights and applications to universality questions in random matrix theory . pure appl . math . _ * 52 * ( 1999 ) , 13351425 . p. deift , t. kriecherbauer , k. t - r mclaughlin , s. venakides , and x. zhou , strong asymptotics of orthogonal polynomials with respect to exponential weights , _ comm . pure appl . _ * 52 * ( 1999 ) , 14911552 . p. deift , t. kriecherbauer , k. t - r mclaughlin , s. venakides , and x. zhou , a riemann - hilbert approach to asymptotic questions for orthogonal polynomials , _ j. comput _ * 133 * ( 2001 ) , 4763 . p. deift and x. zhou , a steepest descent method for oscillatory riemann - hilbert problems . asymptotics for the mkdv equation , _ ann . _ * 137 * ( 1993 ) , 295368 . f. j. dyson , correlations between eigenvalues of a random matrix , _ comm . * 19 * ( 1970 ) , 235250 . fokas , a.r . its , and a.v . kitaev , the isomonodromy approach to matrix models in 2d quantum gravity , _ commun . * 147 * ( 1992 ) , 395430 . p. j. forrester , the spectrum edge of random matrix ensembles , _ nucl . b _ * 402 * ( 3 ) ( 1993 ) , 709728 . t. kriecherbauer and k.t - r mclaughlin , strong asymptotics of polynomials orthogonal with respect to freud weights , _ internat . . notices _ * 1999 * , no . 6 , ( 1999 ) , 299324 . kuijlaars , k.t - r mclaughlin , w. van assche , and m. vanlessen , the riemann hilbert approach to strong asymptotics for orthogonal polynomials , manuscript , 2001 . leff , class of ensembles in the statistical theory of energy level spectra , _ j. math . phys . _ * 5 * ( 1964 ) , 763768 . e. levin and d.s . lubinsky , orthogonal polynomials for exponential weights , " cms books in mathematics , vol . 4 , springer - verlag , new york , 2001 . d.s . lubinsky , a survey of general orthogonal polynomials for weights on finite and infinite intervals , _ acta appl . _ * 10 * ( 1987 ) , 237 - 296 . mehta , random matrices , " 2nd . ed . academic press , boston , 1991 . t. nagao and p.j . forrester , asymptotic correlations at the spectrum edge of random matrices , _ nucl . b _ * 435*(3 ) ( 1995 ) , 401420 . t. nagao and k. slevin , laguerre ensembles of random matrices : nonuniversal correlation functions , _ j. math * 34 * ( 1993 ) , 23172330 . t. nagao and m. wadati , correlation functions of random matrix ensembles related to classical orthogonal polynomials , _ j. phys . . japan _ * 60*(10 ) ( 1991 ) , 32983322 . t. nagao and m. wadati , eigenvalue distribution of random matrices at the spectrum edge _ j. phys . . japan _ * 62*(11 ) ( 1993 ) , 38453856 . p. nevai , `` orthogonal polynomials , '' memoirs amer . math . 18 , nr . 213 , amer . providence r.i . , 1979 . l. pastur and m. shcherbina , universality of the local eigenvalue statistics for a class of unitary invariant random matrix ensembles , _ j. stat . * 86 * ( 1997 ) , 109147 . m. shiroishi , t. nagao , and m. wadati , level spacing distributions of random matrix ensembles , _ j. phys . * 62*(7 ) ( 1993 ) , 22482259 . b. simon , `` trace ideals and their applications , '' cambridge university press , cambridge , 1979 . g. szeg , orthogonal polynomials , " fourth edition , colloquium publications , vol . 23 , amer . math providence r.i . , 1975 . v. totik , `` weighted approximation with varying weight , '' lect . notes math . 1569 , springer - verlag , berlin , 1994 . c. tracy and h. widom , level spacing distributions and the airy kernel , _ comm . * 159 * ( 1994 ) , 151174 . c. tracy and h. widom , level spacing distributions and the bessel kernel , _ comm . * 161 * ( 1994 ) , 289309 .
we discuss single inclusive hadron production from a high energy quark scattering off a strong target color field in the color glass condensate formalism . recent calculations of this process at the next - to - leading order accuracy have led to negative cross sections at large transverse momenta . we identify the origin of this problem as an oversubtraction of the rapidity divergence into the balitsky - kovchegov evolution equation for the target . we propose a new way to implement the kinematical restriction on the emitted gluons to overcome this difficulty .
1604.00225
hadronic reactions at modern collider energies reach a kinematical domain where gluon densities can be nonperturbatively large even at short distance scales where the qcd coupling is weak . a convenient effective theory of qcd in this regime is provided by the color glass condensate ( see e.g. @xcite ) , which describes the nonlinear small @xmath0 degrees of freedom in a hadron or nucleus as a classical color field . an ideal way to study these dense color fields is to probe them with simple dilute probes in a high energy collision , such as in deep inelastic scattering or forward particle production in proton - nucleus collisions . in the latter case the dilute probe is provided by the relatively well understood large @xmath0 partons of the probe proton which , at forward rapidity , scatter off the small @xmath0 color field of the target . several calculations @xcite of forward single inclusive particle production in this framework have provided a good description of available experimental data using the leading order expression for the cross section @xcite . as is often the case in qcd , the leading order calculations leave the overall normalization of the cross section quite uncertain . it would , therefore , be desirable to systematically go to higher orders in perturbation theory in these cross section calculations . similar developments towards higher order have recently taken place concerning the ( balitsky - kovchegov ( bk ) @xcite or jimwlk ) high energy evolution equations @xcite , and dis cross sections @xcite . an important advance in pushing the cgc framework to nlo accuracy was the calculation @xcite of nlo single inclusive particle production in forward proton - nucleus collisions ( see also the earlier works @xcite ) . we will here refer to the cross section formulae derived in refs . @xcite as the `` cxy '' result according to the authors . here it was shown that the divergences in the rapidity ( or longitudinal momentum ) and transverse momentum integrals appearing in the nlo calculation can be factorized into the bk and dglap evolution of the target and projectile , respectively . in a subsequent calculation @xcite it was shown that in the large transverse momentum limit the calculation reduces to the appropriate tree - level process in collinear factorization , although in this case without a factorization of the rapidity divergence . in the first numerical implementation @xcite of the factorization framework of @xcite the nlo corrections turned out to be large and negative at large transverse momenta of the produced particles , to the extent that the total cross section becomes negative . a solution to this problematic behavior has been suggested to lie in the detailed implementation of the factorization of the rapidity divergence @xcite or in a kinematical constraint that must be imposed on the phase space of emitted gluons @xcite . indeed a recent implementation @xcite of this phase space constraint has alleviated the problem , without however removing it completely . in this paper , we suggest an alternative way of implementing bk - factorization in the calculation of chirilli et al . @xcite that combines aspects of these previous works . our suggestion includes , as in @xcite , an explicit rapidity factorization scale in the `` hard functions '' of the nlo part of the cross section . since the dependence of the cross section on this scale cancels against the rapidity scale to which the target is evolved , the total cross section formulation is explicitly independent of this factorization scale . here we suggest implementing the `` kinematical constraint '' or , more precisely , ordering in light cone energy , by making this factorization scale dependent on the transverse momentum of the produced particle . we show that this allows for a renormalization prescription that makes the negative large momentum contribution to the cross section arbitrarily small . combined with a corresponding form of the bk equation this should in the future make it possible to resum the problematic contributions at large transverse momentum similarly as recently suggested for the nlo bk equation @xcite . for simplicity we will here only address the quark channel @xmath1 and perform numerical calculations only for the golec - biernat and wsthoff ( gbw ) @xcite parametrization of the dipole cross section . the gaussian @xmath2-spectrum at leading order in the gbw model has the advantage for the purpose of this paper of being very clearly distinct from the power - law behavior of the nlo contributions . this exacerbates the problem of negative cross sections and should therefore put any attempt to stabilize the perturbative expansion to a more stringent test . a fuller phenomenological analysis of single inclusive particle production would require implementing also the gluon initiated channel , and a more realistic bk - evolved dipole cross section . ultimately this should include a solution of the nlo bk equation @xcite and a simultaneous nlo fit to dis data . at present we leave these further steps for future work . this paper is structured as follows . we will first , in sec . [ sec : nlosinc ] briefly recall cxy results @xcite for the nlo corrections to the cross section in the quark - initiated channel and how the factorization into dglap and bk evolution is performed there . we will then discuss the introduction of an explicit rapidity factorization scale as advocated in @xcite and explicitly show the modification to the nlo spectrum resulting from a variation of this scale in sec . [ sec : rapscale ] . we then , in sec . [ sec : kc ] , discuss imposing the additional `` kinematical '' constraint of @xmath3-ordering ( see e.g. @xcite ) on the rapidity factorization , which we implement as a momentum dependence of the rapidity factorization scale . section [ sec : disc ] then concludes with a brief discussion of possible future steps . our starting point are the cxy formulae derived in ref . we will concentrate here on the quark channel , for which we write the cxy result as : @xmath4 where @xmath5 ^ 2 \\ { \mathcal{j}}({k_\perp},\xi ) & = & \int \frac{{\mathrm{d}}^2 { { \mathbf{q}}}}{(2\pi)^2 } \frac{2({{\mathbf{k}}}-\xi{{\mathbf{q}}})\cdot({{\mathbf{k}}}-{{\mathbf{q}}})}{({{\mathbf{k}}}-\xi{{\mathbf{q}}})^2({{\mathbf{k}}}-{{\mathbf{q}}})^2 } { \mathcal{s}}({q_\perp } ) -\int \frac{{\mathrm{d}}^2 { { \mathbf{q}}}}{(2\pi)^2 } \frac { { \mathrm{d}}^2{\mathbf{l}}}{(2\pi)^2 } \frac{2({{\mathbf{k}}}-\xi{{\mathbf{q}}})\cdot({{\mathbf{k}}}-{\mathbf{l}})}{({{\mathbf{k}}}-\xi{{\mathbf{q}}})^2({{\mathbf{k}}}-{\mathbf{l}})^2 } { \mathcal{s}}({q_\perp}){\mathcal{s}}({l_\perp } ) \\ { \mathcal{i}}_v({k_\perp},\xi ) & = & { \mathcal{s}}({k_\perp})\int \frac { { \mathrm{d}}^2 { { \mathbf{q}}}}{(2\pi)^2 } \left[\frac{{{\mathbf{k}}}-{{\mathbf{q}}}}{({{\mathbf{k}}}-{{\mathbf{q}}})^2 } - \frac{\xi{{\mathbf{k}}}-{{\mathbf{q}}}}{(\xi { { \mathbf{k}}}-{{\mathbf{q}}})^2 } \right]^2 \\ { \mathcal{j}}_v({k_\perp},\xi ) & = & { \mathcal{s}}({k_\perp } ) \left [ \int \frac{{\mathrm{d}}^2 { { \mathbf{q}}}}{(2\pi)^2 } \frac{2(\xi{{\mathbf{k}}}-{{\mathbf{q}}})\cdot({{\mathbf{k}}}-{{\mathbf{q}}})}{(\xi{{\mathbf{k}}}-{{\mathbf{q}}})^2({{\mathbf{k}}}-{{\mathbf{q}}})^2 } -\int \frac{{\mathrm{d}}^2 { { \mathbf{q}}}}{(2\pi)^2 } \frac { { \mathrm{d}}^2{\mathbf{l}}}{(2\pi)^2 } \frac{2(\xi{{\mathbf{k}}}-{{\mathbf{q}}})\cdot({\mathbf{l}}-{{\mathbf{q}}})}{(\xi{{\mathbf{k}}}-{{\mathbf{q}}})^2({\mathbf{l}}-{{\mathbf{q}}})^2 } { \mathcal{s}}({l_\perp } ) \right].\end{aligned}\ ] ] here we have slightly altered the notation of @xcite by including transverse momentum integrals in the functions @xmath6 and leaving out an overall integration over the impact parameter @xmath7 , thus our expression is for the multiplicity and not the cross section . the kinematical variables are defined as @xmath8 , @xmath9 , @xmath10 , @xmath11 , @xmath12 , @xmath13 , @xmath14 , and @xmath15 . most important for our discussion here is the momentum fraction @xmath16 : the fragmenting quark carries a fraction @xmath16 of the incoming quark longitudinal momentum . thus the incoming quark has a momentum fraction @xmath17 of the incoming proton , where @xmath18 is the probe momentum fraction in leading order kinematics . the radiated gluon in the nlo terms carries a longitudinal momentum fraction @xmath19 : i.e. the limit @xmath20 corresponds to the soft gluon emission that must be resummed into bk evolution of the target . these cross sections are all expressed in terms of the fourier - transform of the fundamental representation dipole operator @xmath21 the dipole is related to the notation of @xcite by an overall integration over the impact parameter @xmath22 in the following we leave out the explicit impact parameter dependence of @xmath23 from our notation . due to the unitarity of the wilson lines @xmath24 the dipole cross section satisfies the normalization condition @xmath25 the expressions here use the mean field approximation , replacing the expectation value of the product of two dipole operators by the product of two expectation values @xmath26 . the superscript @xmath27 refers to the fact that at this stage the dipole operator in the leading order part of ( [ eq : nlosigma ] ) is the unrenormalized `` bare '' dipole operator . as noted in @xcite , several important features are already visible in these expressions . first , the terms with an explicit coefficient @xmath28 ( i.e. @xmath29 and @xmath30 ) vanish in the limit @xmath31 but have collinear divergences in the transverse momentum integration . these are nicely treated in @xcite by dimensional regularization in the transverse momentum integrals and factorized into the dglap evolution of the quark distribution function @xmath32 and the fragmentation function @xmath33 . for these `` @xmath28-terms '' we will here follow the treatment of @xcite . the terms with a coefficient @xmath34 ( we will denote these as the `` @xmath35-terms '' in the following ) , i.e. @xmath36 and @xmath37 , have finite transverse integrals , one has to note that at large @xmath38 one can use the normalization ( [ eq : sumrule ] ) to perform the @xmath39 integral , leading to a cancellation between the uv - divergences in the two terms of @xmath37 . ] but are finite in the limit @xmath40 . thus they produce a rapidity divergence due to the explicit factor @xmath41 in the expression for the multiplicity . if one takes the large @xmath35-limit , there are additional cancellations between some @xmath28 and @xmath35-terms that are used in @xcite ; we will , however , not take this limit here . in the cxy calculation the rapidity divergence is subtracted by defining a renormalized dipole cross section as @xmath42 .\end{gathered}\ ] ] expressed in coordinate space this reduces to a more familiar looking form in terms of an integral form of the bk renormalization group equation @xcite @xmath43 .\end{gathered}\ ] ] as pointed out in @xcite , this subtraction really should include an explicit rapidity factorization scale . the remaining nlo cross section would then depend on this factorization scale , which cancels at this order in perturbation theory against the rapidity up to which the dipole cross section in the leading order cross section is evolved . this is completely analogous with the subtraction of the collinear divergences into dglap evolution in the @xmath28-terms , which leaves the nlo cross sections explicitly dependent on a transverse momentum factorization scale . let us now accordingly include the rapidity factorization scale and subtract the rapidity divergence as @xmath44 .\end{gathered}\ ] ] the original choice of @xcite simply corresponds to @xmath45 in our notation . with this subtraction , and the factorization of the transverse divergence into dglap evolution , the cross section now becomes @xmath46 \right\ } , \end{aligned}\ ] ] with @xmath47 \nonumber \\ & & + \frac{{\mathcal{s}}({k_\perp})}{2\pi } \left(\frac{(1+\xi^2)\ln(1-\xi)^2}{1-\xi}\right)_+ - 2\frac{1+\xi^2}{(1-\xi)_+ } i_{2 1}({k_\perp},\xi),\end{aligned}\ ] ] where @xmath48 is the quark - quark splitting function @xmath49 and the functions @xmath50 and @xmath51 are defined according to @xmath52.\end{aligned}\ ] ] in these expressions we used the plus prescription @xmath53 \nonumber \\ & & -g(1)\int^{x_0}_0 { \mathrm{d}}x f(x ) \ , , \end{aligned}\ ] ] where @xmath54 is singular at @xmath55 and @xmath56 is a regular function . now we can see that if @xmath57 is chosen to be very close to one ( i.e. if one makes sure to subtract only terms that have a very large energy logarithm ) , the contribution of the last @xmath35-term ( integrated from @xmath57 to 1 ) is negligible @xmath58 . the first two @xmath35-terms , on the other hand , yield a large logarithmic contribution @xmath59 from the upper limit of the integration . it is instructive to see how these expressions for the subtracted cross section behave in the opposite limits of small and large transverse momentum for the produced quark . at large transverse momentum the result can be easily obtained from @xcite , where the unsubtracted cross sections have been matched to collinear perturbation theory . in the large @xmath2 limit only the radiative corrections contribute and the leading behavior comes from @xmath60 where the integrated gluon distribution is @xmath61 with @xmath62 the transverse area of the target hadron . it is easy to see that if we replace the expression of @xmath29 in eq . ( [ eq : nlosigma ] ) by its large @xmath2 limit ( [ eq : largekt - cf ] ) , the @xmath28-term will yield a positive contribution . on the other hand , the large @xmath2 behavior of @xmath36 ( [ eq : largekt - nc ] ) means that the subtracted @xmath35-term in eq . ( [ eq : subtracted1 ] ) , @xmath63,\ ] ] will yield a contribution that behaves as a power in @xmath2 and is negative due to the growing @xmath16-dependence of @xmath64 in this limit . for values of @xmath57 close to 0 , the magnitude of this negative term is larger than the one of the ( positive ) @xmath28-term . therefore , if the leading order cross section falls rapidly at large @xmath2 ( as in the gbw parametrization ) , the leading large @xmath2 behavior comes from the nlo corrections and the cross section is negative . thus the negativity of the nlo cross section can be traced back to the fact that the unsubtracted cross section at large @xmath2 is proportional to @xmath16 . subtracting , as in cxy ( see ( [ eq : cxysub ] ) ) , the integral over the whole @xmath16-interval multiplied by the cross section at @xmath65 subtracts a very large finite contribution in addition to the divergent part . _ this oversubtraction of the rapidity divergence is what makes the cross section negative at large transverse momenta . _ this was the main point in @xcite , where the authors showed that reintroducing a part of this oversubtracted contribution again makes the cross section positive . when we increase @xmath57 towards 1 , the large negative contribution dies away . for practical purposes , this is not a formalism that we would wish to use for extremely small transverse momenta , since the independent vacuum fragmentation picture of hadron production is probably not a valid physical picture there . however , from a formal point of view it can be instructive to see what happens in the small @xmath2 limit . the fact that bk evolution preserves the dilute limit @xmath66 when @xmath67 , i.e. the sum rule ( [ eq : sumrule ] ) , tells us that the quantity @xmath68 , which is positive in the large @xmath2 limit , must be negative in some other regions of phase space . indeed , if we denote by @xmath69 the contribution that is added to @xmath70 in one rapidity step of bk evolution , the sum rule ensures that the @xmath71-integral of @xmath69 is zero . at large @xmath2 , @xmath68 is positive . therefore , if we increase @xmath57 we subtract a smaller but positive contribution and thus the cross section increases . on the other hand , at smaller @xmath2 we expect @xmath68 to be negative and the cross section to decrease with increasing @xmath57 . as both @xmath36 and @xmath37 are free of ir and uv divergences , we can estimate their leading order behavior by using @xmath72 with @xmath73 , which allows us to figure out the small @xmath2 limit of @xmath36 and @xmath37 as @xmath74 here @xmath75 and @xmath76 are some hard momentum scales which are much larger than @xmath2 and @xmath77 is chosen such that the integral of @xmath78 equals to @xmath79 $ ] with @xmath80 the integral of the function @xmath81 . from the above expressions , we see that both @xmath36 and @xmath37 have logarithmic divergences at @xmath82 and @xmath83 . it is known that @xmath84 should be finite at @xmath65 , and thus the @xmath85 terms in eqs . ( [ eq : jsmallk ] ) and ( [ eq : jvsmallk ] ) must have the same coefficient . therefore @xmath86 has to be @xmath87 at @xmath65 . this also tells us that , at @xmath65 , @xmath88 cancels out between the real and virtual terms . clearly , there is a @xmath88 behavior in @xmath84 when @xmath89 . in the gbw model it is possible to study analytically the behavior of the subtraction term . in this model the dipole cross section is given by @xmath90 which leads to @xmath91 for simplicity we will consider here that the saturation scale @xmath92 is a constant . in this model the real @xmath35 term reads @xmath93 , \ ] ] with @xmath94+\frac{e^{-{k_\perp}^2/{q_\mathrm{s}}^2}}{{q_\mathrm{s}}^2 } \ln(1-\xi)^2 , \ ] ] where @xmath95 is the exponential integral function , @xmath96 , and the virtual term reads @xmath97 .\ ] ] thus the subtraction term is @xmath98 .\ ] ] with this expression one can explicitly show that @xmath99=0 \ , , \ ] ] as required to satisfy the sum rule ( [ eq : sumrule ] ) . at small @xmath2 , @xmath36 and @xmath37 read @xmath100 and the subtraction term behaves like @xmath101 in fig . [ fig : subtraction ] , we show the behavior of @xmath102 as a function of @xmath2 for a fixed saturation scale @xmath103 gev . we observe that , as expected from the general limit ( [ eq : largekt - nc ] ) , the subtraction term is positive at large @xmath2 . the precise functional form @xmath104 of the small @xmath2 behavior , on the other hand , is specific to the gbw model . as a function of @xmath2 in the gbw model for @xmath103 gev . ] this confirms explicitly that when we increase @xmath57 , we are subtracting less of the positive contribution in the large @xmath2 region and less of the negative contribution at small @xmath2 . therefore we would expect that the subtracted multiplicity will increase at large @xmath2 and decrease at small @xmath2 in the gbw model with the increasing of @xmath57 . in this section we illustrate the features we have discussed so far using the gbw model to parametrize the dipole cross section @xmath105 . again we focus on the contribution of the quark channel @xmath1 . in this model , the saturation scale @xmath92 at a given @xmath0 is parametrized by @xmath106 where @xmath107 is the atomic number of the target and the values of the other parameters are @xmath108 , @xmath109 gev , @xmath110 , @xmath111 @xcite . in the following the @xmath0 value at which we evaluate @xmath92 is given by @xmath112 , where @xmath113 corresponds to the leading order kinematics : @xmath114 . since in the gbw model the dipole cross section has a simple gaussian form , it is possible to perform some integrals analytically . this avoids the need to deal with several oscillatory integrals in the numerical implementation which is thus simpler . it also enhances the difference in the behaviors of the leading and next - to - leading contributions at large @xmath115 . indeed , the leading order contribution is proportional to the fourier transform of @xmath105 , which is also gaussian , while at large @xmath115 the nlo corrections have a power law behavior , as can be seen from eqs . ( [ eq : largekt - cf ] ) and ( [ eq : largekt - nc ] ) . the expression for @xmath116 in the gbw model can be found in ref . @xcite in the large @xmath35 limit . here we need to keep @xmath35 finite because we want to have a clear separation between the @xmath28-terms , which are associated with the collinear divergence , and the @xmath35-terms which are associated with the rapidity divergence . in this case , the multiplicity reads @xmath117 + 2 { \mathcal{s}}({k_\perp } ) \left ( \frac{\left ( 1+\xi ^{2}\right ) \ln \left ( 1-\xi \right ) ^{2}}{1-\xi } \right)_{\!\ ! + } \nonumber \\ & \hspace{4.7 cm } + \ ! \frac{1}{\xi^2}\mathcal{p}_{qq}(\xi ) { \mathcal{s}}\left(\frac{{k_\perp}}{\xi}\right ) \ ! \ln \frac{{q_\mathrm{s}}^2}{\mu^2 e^{\gamma_e } } + \mathcal{l}\left(-\frac{{k_\perp}^2}{\xi^2 { q_\mathrm{s}}^2}\right)\right ] - 8\pi \frac{1+\xi^2}{(1-\xi)_+ } i_{2 1}({k_\perp},\xi ) \nonumber \\ & \hspace{4.7 cm } - 3 \delta(1-\xi ) { \mathcal{s}}({k_\perp } ) \left [ \ln \frac{{q_\mathrm{s}}^2}{{k_\perp}^2 e^{\gamma_e } } + \mathcal{l}\left(-\frac{{k_\perp}^2}{{q_\mathrm{s}}^2}\right)\right ] \bigg\ } \\ & \hspace{-2.1 cm } + \frac{{{n_\mathrm{c}}}}{2 } \frac{{\alpha_{\mathrm{s}}}}{2\pi^2 } \!\ ! \int \ ! \frac{\!{\mathrm{d}}z}{z^2}d_{h / q}(z ) \left\{\ ! \int_{\tau / z}^{{{\xi}_\text{f } } } \!\ ! { \mathrm{d}}\xi \frac{1+\xi^2}{1-\xi } \frac{x_p}{\xi } q\left(\!\frac{x_p}{\xi}\!\right ) \ ! { \mathcal{j}}({k_\perp},\xi ) \!-\ ! \int_{0}^{{{\xi}_\text{f } } } \!\ ! { \mathrm{d}}\xi \frac{1+\xi^2}{1-\xi } x_p q\left(x_p \right ) { \mathcal{j}}_v({k_\perp},\xi ) \!+\ ! \int_{{{\xi}_\text{f}}}^{1 } \!\ ! { \mathrm{d}}\xi \frac{1}{1-\xi } \left[{\mathcal{k}}(\xi)\!-\!{\mathcal{k}}(1)\right ] \right\ } \nonumber \end{aligned}\ ] ] where @xmath118 and the expressions for @xmath36 and @xmath37 in the gbw model can be read from eqs . ( [ eq : jcal_gbw ] ) and ( [ eq : jcalv_gbw ] ) , respectively . besides the absence of the impact parameter integration , there are two differences with the corresponding expressions given in ref . @xcite : first , we keep terms proportional to @xmath119 which vanish in the large @xmath35 limit taken in ref . @xcite . second , we modify the rapidity divergence subtraction by using the cutoff @xmath57 introduced previously . in fig . [ fig : dn_nocutoff ] we show the multiplicity @xmath120 as a function of @xmath121 in the gbw model for @xmath45 which corresponds to the choice made in ref . we take @xmath122 gev , @xmath123 , @xmath124 gev@xmath125 and @xmath126 . for the collinear pdfs @xmath32 and fragmentation functions @xmath33 we use the mstw 2008 @xcite and dss @xcite parametrizations , respectively , both at next - to - leading order . as observed in ref . @xcite , when @xmath45 the nlo multiplicity is negative when @xmath121 is larger than a certain value , of the order of the saturation scale . as discussed above , this negativity comes from the @xmath35-terms . this can be seen from the same figure where we also show the effect of including only the nlo corrections proportional to @xmath28 or @xmath35 . at large transverse momentum the @xmath28 corrections are positive while the @xmath35 corrections are negative and large enough to make the total nlo multiplicity negative . we observe that when including only the nlo contributions proportional to @xmath35 the multiplicity becomes negative very close to the point where @xmath127 . to see more clearly the behavior of the nlo corrections at small transverse momentum , we show in fig . [ fig : ratio_nocutoff ] the ratio of the nlo and lo multiplicity for @xmath128 gev when including only the @xmath28 or @xmath35 contributions or both . one can note that , already for values of @xmath121 of the order of 2 gev , both the @xmath28 and @xmath35 contributions are of the same order of magnitude as the lo term . in the gbw model at leading and next - to - leading order and when including only the @xmath28 or @xmath35 nlo corrections . the vertical dashed line corresponds to @xmath129 . ] , when including only the @xmath28 or @xmath35 nlo corrections or both . the vertical dashed line corresponds to @xmath129 . ] as discussed in sec . [ sec : rapscale ] , instead of subtracting the whole @xmath16 interval when subtracting the rapidity divergence ( [ eq : cxysub ] ) , one can introduce an explicit cutoff @xmath57 to determine which contributions are included in the renormalized dipole cross section ( [ eq : xifsub ] ) . we recall that , compared to the results shown above for @xmath45 , only the @xmath35-terms are affected by this procedure . in fig . [ fig : dn_cutoff ] , we show the multiplicity for several fixed values of @xmath57 as a function of @xmath121 . we observe that , for the reason exposed in sec . [ sec : analytical ] , values of @xmath57 close to 1 lead to a positive multiplicity up to larger values of @xmath121 . in particular , for @xmath130 , the multiplicity is positive up to @xmath131 gev . in fig . [ fig : ratio_cutoff ] , we show the ratio nlo / lo for several values of @xmath57 and @xmath128 gev . here we see that values of @xmath57 very close to 1 , corresponding to a positive multiplicity at large @xmath121 , lead to a nlo multiplicity smaller than at leading order at moderate @xmath121 . in the gbw model at nlo for different values of @xmath57 compared with the lo result . the vertical dashed line corresponds to @xmath129 . ] . the vertical dashed line corresponds to @xmath129 . ] in fig . [ fig : ratio_cutoff_qs_1 ] , we show the same ratio with a fixed value of the saturation scale , @xmath103 gev . here we see clearly that , as could be expected from the small @xmath2 behavior of the subtraction term ( [ eq : sub_smallkt ] ) , larger values of @xmath57 lead to smaller cross sections at small @xmath121 and larger cross sections at large @xmath121 . with @xmath103 gev . ] to conclude this section , it appears that the choice of the value of @xmath57 can have an important impact on the final results , both at small and large transverse momentum . in the following we propose a way to fix the value of this parameter based on physical considerations . it has recently been emphasized @xcite that , in order to have a stable bk evolution beyond leading order , the gluon cascade resummed by the evolution should be _ ordered in the light cone energy @xmath3 _ of the projectile . the requirement of ordering in @xmath3 can equivalently be thought of as an ordering in its conjugate variable @xmath132 , which is the light cone lifetime or `` ioffe time '' of fluctuations in the projectile @xcite . in many works , this feature is known as the `` kinematical constraint '' or as imposing `` exact kinematics '' @xcite , a terminology that we will comment on in sec.[sec : disc ] . imposing @xmath3-ordering has also been one ingredient in the program of `` small-@xmath0 resummation '' in the context of the linear bfkl evolution @xcite . since we are working in a frame where we consider the gluons as being emitted from the probe , the emitted gluon always has a smaller @xmath133 momentum than its parent quark ; the emissions are therefore naturally ordered in @xmath133 . the requirement of @xmath3 ordering , on the other hand , must be imposed separately . at leading order , the physical picture of the scattering is that of an incoming collinear quark ( i.e. with only a @xmath133 momentum component ) that acquires a transverse momentum and light cone energy ( @xmath71 and @xmath3 ) from the target . considering the production of a quark with a fixed @xmath71 , the light cone energy required is @xmath134 the fraction of the target momentum required to set the produced quark on shell is defined by the leading order kinematics as @xmath135 , where @xmath136 is the momentum of the target . thus we can write @xmath137 . we now want to implement @xmath3-ordering in the calculation of the single inclusive cross section . for this purpose let us consider the emission of a gluon from the incoming quark that is present in all of the contributing diagrams . labeling the transverse momentum of the incoming quark as @xmath38 and radiated gluon as @xmath39 ( cf . [ fig : q2qg ] ) , the light cone energy introduced from the gluon emission is @xmath138 \nonumber \\ & = \frac{x_g p^-}{{{\mathbf{k}}}^2 } \frac{({\mathbf{l}}-(1-\xi){{\mathbf{q}}})^2}{\xi(1-\xi)},\end{aligned}\ ] ] where @xmath19 is the momentum fraction of the emitted gluon . in some diagrams , the splitting happens before the interaction with the target and @xmath139 , in final state radiation diagrams @xmath140 . similarly , for some diagrams , but not all , the quark does not interact with the target after the emission and @xmath141 . in general , the momenta @xmath142 in eq . ( [ eq : kminusqg ] ) are integrated over in the diagram , whereas the transverse momentum @xmath71 is that of the produced quark and is kept fixed ( at a fixed fragmentation @xmath143 ) . it appears as an overall prefactor in ( [ eq : kminusqg ] ) when the incoming quark longitudinal momentum @xmath144 is expressed in terms of the target variables @xmath113 and @xmath136 and is therefore common to all diagrams contributing to the cross section . we now want to derive a renormalization group equation describing the target . this means that emissions in a certain kinematical regime have to be subtracted from the cross section and absorbed into a redefinition of the target , as was done in eq . ( [ eq : cxysub ] ) . in particular , this kinematical regime to be subtracted should include the limit @xmath145 . now the requirement of a @xmath3-ordering in the evolution means that the subtraction criterion should be a condition on @xmath146 . since @xmath147 for @xmath40 , this means that fluctuations around @xmath65 with @xmath146 larger than a certain factorization scale should be subtracted . thus , our renormalization scheme is defined by the condition that all the contributions with @xmath148 are to be subtracted from the cross section . most importantly , we want the subtracted region to include contributions with @xmath149 where @xmath150 is in principle an arbitrary factorization scale that will appear both in the leading order term as the rapidity up to which the target must be evolved , and in the hard factors of the nlo cross section . we emphasize that ( [ eq : kminusorder ] ) is not a ( momentum conservation ) kinematical constraint for the process , but a renormalization condition specifying which parts of phase space should be subtracted from the cross section . a natural choice that resums all the large energy logarithms into the evolution is to take @xmath151 . in the typical kinematical regime when all the transverse momenta are of the same order , @xmath152 for all @xmath16 , and we can safely subtract the whole @xmath16 interval as done in @xcite . however , when the produced quark momentum @xmath71 is much larger than the typical target scale @xmath92 , there are configurations where the @xmath3-ordering condition ( [ eq : kminusorder ] ) is not satisfied for all @xmath16 , because @xmath39 is integrated over in a range that includes typical target scales @xmath153 . thus we should , for large @xmath71 , only subtract values of @xmath16 close to 1 that satisfy @xmath154 or @xmath155 this line of argument leads us to our proposal to implement @xmath3-ordering or the `` kinematical constraint '' by subtracting a @xmath2-dependent fraction of the @xmath16 integral from the cross section as @xmath156,\end{aligned}\ ] ] with @xmath157 note that since there is a factorization scale @xmath158 in the limit , we have the freedom to change this scale with a factor of order 1 . this change should cancel against a corresponding change in the rapidity to which the dipole amplitude is evolved . an alternate , somewhat smoother form with the same parametric behavior would be @xmath159 where we can vary the factorization scale @xmath158 in an interval such as @xmath160 . in practice , the two choices ( [ eq : xifmin ] ) and ( [ eq : xifcontinuous ] ) lead to very similar results . therefore we will only show results obtained with the smoother choice ( [ eq : xifcontinuous ] ) . in figs . [ fig : dn_xifcontinuous ] and [ fig : ratio_xifcontinuous ] we show the multiplicity as a function of @xmath121 for three values of the ratio @xmath158 . in the three cases , the nlo multiplicity still becomes negative for some value of the transverse momentum . for @xmath161 or 2 this happens approximately at the same point as for @xmath45 , as can be seen by comparison with fig . [ fig : dn_nocutoff ] . on the other hand , for @xmath162 the multiplicity is positive up to a significantly larger value of @xmath121 of the order of 6 gev . for comparison we also show in fig . [ fig : dn_xifcontinuous ] the results obtained by following the approach of ref . @xcite , where the original cxy subtraction is used but a kinematical constraint is imposed which leads to an additional correction term @xmath163 . this additional term extends the @xmath121 range where the multiplicity is positive by about 0.5 gev , which is similar to what was obtained in ref . @xcite in the same kinematics . this is the main result of our paper : we have shown that it is possible to choose a value of @xmath158 which is still in its `` natural '' range and extends significantly the region where the nlo cross section has a reasonable physical behavior . in the case we have studied here the dependence on the exact choice for the renormalization scale is stronger than one would like , but we believe that this is due to two aspects of the calculation that can be improved in the future . first , we tried here to implement ordering in an effective way by relying only on external scales to fix @xmath57 . for more accurate results , one should instead impose the ordering ( [ eq : kminusorder ] ) inside the transverse momentum integrals in @xmath36 and @xmath37 . this could lead to sizeable differences with the results we have shown since , as can be seen from fig . [ fig : dn_xifcontinuous ] , our results are still very sensitive to the value of @xmath57 . second , we have used the gbw model to parametrize the dipole cross section . while this has practical advantages like enabling us to perform some integrals analytically , in this model the leading and next - to - leading order contributions have very different behaviors at large transverse momentum : the leading order term behaves like a gaussian , while the next - to - leading order corrections behave like a power law . therefore , at large transverse momentum , the behavior of the multiplicity is governed entirely by the nlo corrections , which is quite unnatural . on the other hand , using a more physical dipole cross section , such as one obtained by solving the balitsky - kovchegov @xcite equation , should make the leading order contribution behave more like a power law . this means that the end result would be less sensitive to the exact choice made for the parametrization of @xmath57 . taking a bk - evolved dipole cross section instead of the gbw one would also make the cross section formally independent of the factorization scale @xmath150 at this order in @xmath164 , which should significantly reduce the factorization scale dependence . ) of @xmath57 for different values of @xmath165 compared with the results at leading order and when using the correction term @xmath163 introduced in ref . @xcite . ] ) of @xmath57 for different values of @xmath165 . ] to summarize , we have in this paper proposed to modify the subtraction procedure of the rapidity divergence in the calculation of single inclusive hadron production in the hybrid formalism @xcite . it has been hoped that imposing ordering in the light cone energy of the probe could solve the issue of large negative nlo corrections to the cross section . we have here argued that the most natural way to implement this ordering is to impose it on the kinematics of the part that is subtracted from the cross section and absorbed into the target . to explicitly demonstrate the effect of our proposal , we have performed the calculations at finite @xmath35 for the gbw model . the part corresponding to the rapidity divergence is proportional to @xmath35 , whereas the parts associated with dglap evolution have a color factor @xmath28 . compared to the cxy formulation , our suggestion only modifies the @xmath35-terms and leaves the @xmath28-terms unchanged . indeed , we have shown that such a modification of the subtraction scale can lead to a more stable perturbative expansion for this cross section at high @xmath121 . by a suitable choice of the factorization scale @xmath150 , the stability of the perturbative expansion can be extended to arbitrarily high @xmath121 . this should be contrasted with the recent calculation of ref . @xcite , where a similar correction is obtained as an additional correction which in some cases still leaves the cross section negative at high enough @xmath121 . the main difference between our approach here and that of ref . @xcite is that there the kinematical modification is treated as a single correction term whereas here it is resummed to all orders by shifting the evolution variable in the bk equation . in some works ( e.g. @xcite ) , the kinematical constraint is explained as a requirement of keeping the momentum fraction in the target @xmath166 as less than @xmath167 . we would rather prefer to characterize this requirement by saying that the @xmath166 of the corrections that are resummed into the bk evolution should be larger than the @xmath113 determined by the leading order kinematics . the whole formalism here is based on the eikonal approximation , which is valid for only small enough @xmath166 . if the actual result for the cross section were to really depend on the dipole cross section at very large target @xmath166 , the whole formalism would be in grave trouble . fortunately this is not the case , since the contribution from large @xmath168 ( corresponding to @xmath169 ) is subtracted from the cross section and absorbed into the renormalization group evolution of the target : in fact it is only a tiny part of the subtraction . in spite of this different terminology , the actual equations in refs . @xcite lead to the same parametric dependence that we have here : the longitudinal factorization scale has to be modified by the presence of a large transverse momentum logarithm @xmath170 . in order to perform the whole calculation consistently at nlo accuracy , one needs to also solve the nlo version of the bk equation . to do this , the actual bk equation to be solved must be consistent with the subtraction procedure in the cross section . different subtraction schemes correspond to different versions of the bk equation . since the rapidity dependence of the dipole cross section is proportional to @xmath164 , a modification of the subtraction scale in eq . ( [ eq : dlzsub ] ) is a higher order @xmath171 effect , and the difference appears at the nlo level for the bk equation . in fact , it was recently shown in @xcite that a problematic double logarithmic correction to the nlo bk equation @xcite can indeed be resummed by redefining the evolution variable in a similar way as we have done in eq . ( [ eq : dlzsub ] ) . when written as a differential equation in @xmath172 , the modified bk equation is nonlocal in rapidity , which is inconvenient for a numerical solution . the authors of @xcite proposed a clever way of rewriting this in a local form , which has successfully been solved numerically @xcite . compared to the work in @xcite , our proposal here has been a very simplistic one . by explicitly taking @xmath92 as the scale of the logarithm , we have been able to write a subtraction procedure in such a way that we can reuse most of the cxy formulae with only slight modifications . to be consistent @xcite , the momentum scale in eq . ( [ eq : xifmin ] ) should actually be one that is integrated over in @xmath36 and @xmath37 . this would require a new computation of the regularized `` hard factors '' remaining after the subtraction . for a calculation that is consistent with recent nlo bk solutions it would be useful to carry out this subtraction for the resummation proposed in @xcite . this is , however , left for future work . we thank e. iancu , z. kang , b .- w xiao and d. zaslavsky for discussions . this work has been supported by the academy of finland , projects 267321 and 273464 . f. gelis , e. iancu , j. jalilian - marian and r. venugopalan , _ the color glass condensate _ , http://dx.doi.org/10.1146/annurev.nucl.010909.083629[_ann . nucl . part . * 60 * ( 2010 ) 463 ] [ http://arxiv.org/abs/1002.0333[arxiv:1002.0333 [ hep - ph ] ] ] . a. dumitru , a. hayashigaki and j. jalilian - marian , _ the color glass condensate and hadron production in the forward region _ , http://dx.doi.org/10.1016/j.nuclphysa.2005.11.014[_nucl . phys . _ * a765 * ( 2006 ) 464 ] [ http://arxiv.org/abs/hep-ph/0506308 [ arxiv : hep - ph/0506308 ] ] . j. l. albacete and c. marquet , _ single inclusive hadron production at rhic and the lhc from the color glass condensate _ , http://dx.doi.org/10.1016/j.physletb.2010.02.073[_phys . * b687 * ( 2010 ) 174 ] [ http://arxiv.org/abs/1001.1378 [ arxiv:1001.1378 [ hep - ph ] ] ] . tribedy and r. venugopalan , _ qcd saturation at the lhc : comparisons of models to p+p and a+a data and predictions for p+pb collisions _ , http://dx.doi.org/10.1016/j.physletb.2012.02.047[_phys . lett . _ * b710 * ( 2012 ) 125 ] [ http://arxiv.org/abs/1112.2445 [ arxiv:1112.2445 [ hep - ph ] ] ] . a. h. rezaeian , _ cgc predictions for p+a collisions at the lhc and signature of qcd saturation _ , http://dx.doi.org/10.1016/j.physletb.2012.11.066[_phys . lett . _ * b718 * ( 2013 ) 1058 ] [ http://arxiv.org/abs/1210.2385 [ arxiv:1210.2385 [ hep - ph ] ] ] . t. lappi and h. mntysaari , _ single inclusive particle production at high energy from hera data to proton - nucleus collisions _ , http://dx.doi.org/10.1103/physrevd.88.114020[_phys . rev . _ * d88 * ( 2013 ) 114020 ] [ http://arxiv.org/abs/1309.6963 [ arxiv:1309.6963 [ hep - ph ] ] ] . a. dumitru and j. jalilian - marian , _ forward quark jets from protons shattering the colored glass _ , http://dx.doi.org/10.1103/physrevlett.89.022301[_phys . * 89 * ( 2002 ) 022301 ] [ http://arxiv.org/abs/hep-ph/0204028 [ arxiv : hep - ph/0204028 ] ] . i. balitsky , _ operator expansion for high - energy scattering _ , http://dx.doi.org/10.1016/0550-3213(95)00638-9[_nucl . phys . _ * b463 * ( 1996 ) 99 ] [ http://arxiv.org/abs/hep-ph/9509348 [ arxiv : hep - ph/9509348 ] ] . y. v. kovchegov , _ small - x f2 structure function of a nucleus including multiple pomeron exchanges _ , http://dx.doi.org/10.1103/physrevd.60.034008[_phys . rev . _ * d60 * ( 1999 ) 034008 ] [ http://arxiv.org/abs/hep-ph/9901281 [ arxiv : hep - ph/9901281 ] ] . i. balitsky , _ quark contribution to the small-@xmath0 evolution of color dipole _ , http://dx.doi.org/10.1103/physrevd.75.014001[_phys . * d75 * ( 2007 ) 014001 ] [ http://arxiv.org/abs/hep-ph/0609105[arxiv:hep-ph/0609105 ] ] . i. balitsky and g. a. chirilli , _ next - to - leading order evolution of color dipoles _ , http://dx.doi.org/10.1103/physrevd.77.014019[_phys . rev . _ * d77 * ( 2008 ) 014019 ] [ http://arxiv.org/abs/0710.4330 [ arxiv:0710.4330 [ hep - ph ] ] ] . i. balitsky and g. a. chirilli , _ nlo evolution of color dipoles in n=4 sym _ , http://dx.doi.org/10.1016/j.nuclphysb.2009.07.003[_nucl . phys . _ * b822 * ( 2009 ) 45 ] [ http://arxiv.org/abs/0903.5326 [ arxiv:0903.5326 [ hep - ph ] ] ] . i. balitsky and g. a. chirilli , _ rapidity evolution of wilson lines at the next - to - leading order _ , http://dx.doi.org/10.1103/physrevd.88.111501[_phys . rev . _ * d88 * ( 2013 ) 111501 ] [ http://arxiv.org/abs/1309.7644 [ arxiv:1309.7644 [ hep - ph ] ] ] . a. kovner , m. lublinsky and y. mulian , _ jalilian - marian , iancu , mclerran , weigert , leonidov , kovner evolution at next to leading order _ , http://dx.doi.org/10.1103/physrevd.89.061704[_phys . rev . _ * d89 * ( 2014 ) 061704 ] [ http://arxiv.org/abs/1310.0378 [ arxiv:1310.0378 [ hep - ph ] ] ] . t. lappi and h. mntysaari , _ direct numerical solution of the coordinate space balitsky - kovchegov equation at next to leading order _ , http://dx.doi.org/10.1103/physrevd.91.074016[_phys . * d91 * ( 2015 ) 074016 ] [ http://arxiv.org/abs/1502.02400 [ arxiv:1502.02400 [ hep - ph ] ] ] . e. iancu , j. d. madrigal , a. h. mueller , g. soyez and d. n. triantafyllopoulos , _ resumming double logarithms in the qcd evolution of color dipoles _ , http://dx.doi.org/10.1016/j.physletb.2015.03.068[_phys . lett . _ * b744 * ( 2015 ) 293 ] [ http://arxiv.org/abs/1502.05642 [ arxiv:1502.05642 [ hep - ph ] ] ] . e. iancu , j. d. madrigal , a. h. mueller , g. soyez and d. n. triantafyllopoulos , _ collinearly - improved bk evolution meets the hera data _ , http://dx.doi.org/10.1016/j.physletb.2015.09.071[_phys . _ * b750 * ( 2015 ) 643 ] [ http://arxiv.org/abs/1507.03651 [ arxiv:1507.03651 [ hep - ph ] ] ] . t. lappi and h. mntysaari , _ next - to - leading order balitsky - kovchegov equation with resummation _ , http://dx.doi.org/10.1103/physrevd.93.094004[_phys . rev . _ * d93 * ( 2016 ) 094004 ] [ http://arxiv.org/abs/1601.06598 [ arxiv:1601.06598 [ hep - ph ] ] ] i. balitsky and g. a. chirilli , _ photon impact factor in the next - to - leading order _ , http://dx.doi.org/10.1103/physrevd.83.031502[_phys . _ * d83 * ( 2011 ) 031502 ] [ http://arxiv.org/abs/1009.4729 [ arxiv:1009.4729 [ hep - ph ] ] ] . g. beuf , _ nlo corrections for the dipole factorization of dis structure functions at low x _ , http://dx.doi.org/10.1103/physrevd.85.034039[_phys . rev . _ * d85 * ( 2012 ) 034039 ] [ http://arxiv.org/abs/1112.4501 [ arxiv:1112.4501 [ hep - ph ] ] ] . g. a. chirilli , b .- w . xiao and f. yuan , _ one - loop factorization for inclusive hadron production in pa collisions in the saturation formalism _ , http://dx.doi.org/10.1103/physrevlett.108.122301[_phys . * 108 * ( 2012 ) 122301 ] [ http://arxiv.org/abs/1112.1061 [ arxiv:1112.1061 [ hep - ph ] ] ] . g. a. chirilli , b .- w . xiao and f. yuan , _ inclusive hadron productions in pa collisions _ , http://dx.doi.org/10.1103/physrevd.86.054005 [ _ phys . * d86 * ( 2012 ) 054005 ] [ http://arxiv.org/abs/1203.6139[arxiv:1203.6139 [ hep - ph ] ] ] . t. altinoluk and a. kovner , _ particle production at high energy and large transverse momentum - the hybrid formalism revisited _ , http://dx.doi.org/10.1103/physrevd.83.105004[_phys . rev . _ * d83 * ( 2011 ) 105004 ] [ http://arxiv.org/abs/1102.5327 [ arxiv:1102.5327 [ hep - ph ] ] ] . j. jalilian - marian and a. h. rezaeian , _ hadron production in pa collisions at the lhc from the color glass condensate _ , http://dx.doi.org/10.1103/physrevd.85.014017[_phys . rev . _ * d85 * ( 2012 ) 014017 ] [ http://arxiv.org/abs/1110.2810 [ arxiv:1110.2810 [ hep - ph ] ] ] . a. m. stato , b .- w . xiao , f. yuan and d. zaslavsky , _ matching collinear and small @xmath0 factorization calculations for inclusive hadron production in @xmath173 collisions _ , http://dx.doi.org/10.1103/physrevd.90.014047 [ _ phys . * d90 * ( 2014 ) 014047 ] [ http://arxiv.org/abs/1405.6311[arxiv:1405.6311 [ hep - ph ] ] ] . a. m. stasto , b .- w . xiao and d. zaslavsky , _ towards the test of saturation physics beyond leading logarithm _ , http://dx.doi.org/10.1103/physrevlett.112.012302[_phys . * 112 * ( 2014 ) 012302 ] [ http://arxiv.org/abs/1307.4057 [ arxiv:1307.4057 [ hep - ph ] ] ] . z .- kang , i. vitev and h. xing , _ next - to - leading order forward hadron production in the small-@xmath0 regime : rapidity factorization _ , http://dx.doi.org/10.1103/physrevlett.113.062002[_phys . rev _ * 113 * ( 2014 ) 062002 ] [ http://arxiv.org/abs/1403.5221 [ arxiv:1403.5221 [ hep - ph ] ] ] . t. altinoluk , n. armesto , g. beuf , a. kovner and m. lublinsky , _ single - inclusive particle production in proton - nucleus collisions at next - to - leading order in the hybrid formalism _ , http://dx.doi.org/10.1103/physrevd.91.094016[_phys . rev . _ * d91 * ( 2015 ) 094016 ] [ http://arxiv.org/abs/1411.2869 [ arxiv:1411.2869 [ hep - ph ] ] ] . k. watanabe , b .- w . xiao , f. yuan and d. zaslavsky , _ implementing the exact kinematical constraint in the saturation formalism _ , http://dx.doi.org/10.1103/physrevd.92.034026[_phys . * d92 * ( 2015 ) 034026 ] [ http://arxiv.org/abs/1505.05183 [ arxiv:1505.05183 [ hep - ph ] ] ] . k. j. golec - biernat and m. wusthoff , _ saturation effects in deep inelastic scattering at low @xmath174 and its implications on diffraction _ , http://dx.doi.org/10.1103/physrevd.59.014017[_phys . rev . _ * d59 * ( 1998 ) 014017 ] [ http://arxiv.org/abs/hep-ph/9807513[arxiv:hep-ph/9807513 ] ] . g. beuf , _ improving the kinematics for low - x qcd evolution equations in coordinate space _ , http://dx.doi.org/10.1103/physrevd.89.074039 [ _ phys . _ * d89 * ( 2014 ) 074039 ] [ http://arxiv.org/abs/1401.0313[arxiv:1401.0313 [ hep - ph ] ] ] . a. martin , w. stirling , r. thorne and g. watt , _ parton distributions for the lhc _ , http://dx.doi.org/10.1140/epjc/s10052-009-1072-5 [ _ eur . phys . j. _ * c63 * ( 2009 ) 189 ] [ http://arxiv.org/abs/0901.0002[arxiv:0901.0002 [ hep - ph ] ] ] . d. de florian , r. sassot and m. stratmann , _ global analysis of fragmentation functions for pions and kaons and their uncertainties _ , http://dx.doi.org/10.1103/physrevd.75.114010[_phys . * d75 * ( 2007 ) 114010 ] [ http://arxiv.org/abs/hep-ph/0703242 [ arxiv : hep - ph/0703242 ] ] . l. motyka and a. m. stasto , _ exact kinematics in the small x evolution of the color dipole and gluon cascade _ , http://dx.doi.org/10.1103/physrevd.79.085016[_phys . rev . _ * d79 * ( 2009 ) 085016 ] [ http://arxiv.org/abs/0901.4949 [ arxiv:0901.4949 [ hep - ph ] ] ] . g. salam , _ a resummation of large subleading corrections at small x _ , http://dx.doi.org/10.1088/1126-6708/1998/07/019[_jhep_ * 9807 * ( 1998 ) 019 ] [ http://arxiv.org/abs/hep-ph/9806482 [ arxiv : hep - ph/9806482 [ hep - ph ] ] ] . m. ciafaloni , d. colferai and g. salam , _ renormalization group improved small x equation _ , http://dx.doi.org/10.1103/physrevd.60.114036 [ _ phys . rev . _ * d60 * ( 1999 ) 114036 ] [ http://arxiv.org/abs/hep-ph/9905566[arxiv:hep-ph/9905566 [ hep - ph ] ] ] . g. altarelli , r. d. ball and s. forte , _ resummation of singlet parton evolution at small x _ , http://dx.doi.org/10.1016/s0550-3213(00)00032-8[_nucl . * b575 * ( 2000 ) 313 ] [ http://arxiv.org/abs/hep-ph/9911273 [ arxiv : hep - ph/9911273 [ hep - ph ] ] ] . m. ciafaloni , d. colferai , g. salam and a. stasto , _ renormalization group improved small x green s function _ , http://dx.doi.org/10.1103/physrevd.68.114003[_phys . _ * d68 * ( 2003 ) 114003 ] [ http://arxiv.org/abs/hep-ph/0307188 [ arxiv : hep - ph/0307188 [ hep - ph ] ] ] .
this paper deals with the question of how to most effectively conduct experiments in partially observed markov decision processes so as to provide data that is most informative about a parameter of interest . methods from markov decision processes , especially dynamic programming , are introduced and then used in an algorithm to maximize a relevant fisher information . the algorithm is then applied to two pomdp examples . the methods developed can also be applied to stochastic dynamical systems , by suitable discretization , and we consequently show what control policies look like in the morris - lecar neuron model , and simulation results are presented . we discuss how parameter dependence within these methods can be dealt with by the use of priors , and develop tools to update control policies online . this is demonstrated in another stochastic dynamical system describing growth dynamics of dna template in a pcr model .
1209.4019
hidden markov models have proven their usefulness across a wide variety of applications . in many of these applications , the user or the experimenter will have some way of influencing the transitions of the underlying markov chain , as in markov decision processes , and such a process is called a partially observed markov decision process ( pomdp ) , see monahan @xcite . if we assume that the transition probability matrix is governed by some unknown parameters , an important problem is to understand how the process can be influenced to get data that is most informative about the parameters . we can think of this as experimental design for partially observed markov decision processes . we consider a pomdp @xmath0 . in this setting @xmath1 is an unobserved markov chain , where the transition probabilities depend in a parametric way on what control @xmath2 is chosen at time @xmath3 and an unknown parameter @xmath4 . the process @xmath5 is observed and depends on which state @xmath1 is in . our goal is to find ways to use the controls @xmath2 to improve parameter estimates of @xmath4 . since the maximum likelihood estimates for @xmath4 will be asymptotically efficient , our general strategy will be to use the controls to try to minimize the sample variance of the maximum likelihood estimates of @xmath6 . this will be achieved by maximizing a fisher information for @xmath4 . the controls are calculated using dynamic programming , a popular maximization algorithm from markov decision processes which outputs an adaptive control policy , i.e. the control chosen at time @xmath3 is based on observations up to time @xmath3 . the first attempt at using dynamic controls to maximize a fisher information was by lin et al . @xcite who proposed maximizing the fisher information that corresponds to direct observations of the underlying process @xmath1 , labeled the full information fisher information ( fofi ) , and using a filter to compute @xmath1 if it is not observed directly . this paper extends their work by making use of the pomdp structure and proposes maximizing a fisher information that is based on the observations @xmath5 , labeled the partial observation fisher information ( pofi ) . the control policies based on these two fisher informations , fofi and pofi , are compared , and we illustrate a setting in which the two control policies are quite different . the methods we use to calculate controls for maximizing fisher information will depend on the unknown parameter @xmath4 . we illustrate how this problem can partially be overcome by assuming a prior for @xmath4 to calculate a control policy before running the experiment . additionally we describe how , using data acquired as the experiment progresses , a posterior for @xmath4 can be used to calculate a more precise control policy . that is , parameter information from observations acquired at a time @xmath3 can be used to improve the policy used in what is left of the experiment . these methods will be based on the value iteration algorithm ( via ) , which is closely related to dynamic programming . the methods developed have application value beyond partially observed markov decision processes . lin et al . @xcite considered stochastic systems of the form @xmath7 where @xmath4 is the parameter of interest , to be estimated , @xmath8 is a control that can be chosen by the user , @xmath9 is the vector of state variables , @xmath10 is a vector valued function , @xmath11 a wiener process , and additionally @xmath12 is only observed partially or noisily . by discretizing time , state and observation spaces the process can be approximated by a pomdp , allowing us to use the methods developed in this paper to devise a control policy that maximizes information about the parameter @xmath4 . in order to illustrate our methods we present four examples , ordered by level of complexity , with the first two being pomdp s and the latter two continuous stochastic systems . we use the unknown @xmath4 to calculate controls in the first three examples to highlight the differences between pofi and fofi , but in the last example we examine means to deal with the dependence of the pofi and fofi policies on the parameter of interest . in the first example we hypothesize about the kind of systems in which we will observe large improvement in parameter estimation by using the pofi control policy over the fofi policy . following a discussion we construct a mock partially observed markov decision process , in which this improvement is shown using a simulation study . to illustrate the real - world applicability of design in discrete pomdp s we consider a realistic pomdp from experimental economics . the model will consist of a simple adversarial game similar to the rock - paper - scissor " game where one player tries to play in such a way that maximizes information about the other players strategy . the third example is a stochastic version of the morris - lecar neuron model , a dynamical system which models voltage in a single neural cell . this model is two dimensional , but only one dimension is observed . the model has multiple parameters and we investigate how the pofi and fofi control policies perform in estimating them . fourthly we consider an example from biology , a polymerase chain reaction ( pcr ) experiment where dna template is grown in liquid substrate . the population dynamics are modeled in a dynamical system with stochastic errors , and the aim is to estimate the half - saturation constant , a parameter which controls the saturation of the template . here we compare using a prior for @xmath4 and using via to calculate a control policy . in section @xmath13 we describe the pomdp setup , and then introduce dynamic programming , an algorithm from markov decision processes . this algorithm is then applied to evaluate control policies that maximize both the partial observation fisher information and the full observation fisher information . in section @xmath14 we show how the control policies can be applied to pomdp examples and then , in section @xmath15 , describe how stochastic dynamical systems can be transformed to pomdps by using discretization methods , and give an example of such a system . in section @xmath16 we discuss the problem of parameter dependence when running the dynamic programs , and how these issues can be dealt with . this is illustrated in the last example , and a discussion follows . we consider a markov decision process @xmath17 . in this setting @xmath18 is a markov chain , but the transition probabilities at time @xmath3 depend on a control @xmath2 chosen at that time . we assume a finite state space @xmath19 for the state process @xmath20 and that the controls available belong to some finite set @xmath21 . we let @xmath22 denote the size of @xmath19 and @xmath23 the size of @xmath21 . the transition probabilities are assumed to be parametric and we write @xmath24 short for @xmath25 where @xmath26 and @xmath27 . in addition to this we assume that the process @xmath18 is latent and we only observe the related observations @xmath28 whose relation to the @xmath18 can also depend on @xmath4 . we write @xmath29 short for @xmath30 , where @xmath31 and @xmath32 , and let @xmath33 denote the size of @xmath34 . this makes the system a partially observed markov decision process . it has a finite horizon @xmath35 in which we observe @xmath36 . we will use the short hand notation @xmath37 to denote @xmath38 , i.e. the observations between time @xmath39 and @xmath3 , and analogous notation for @xmath2 and @xmath1 . the objective is to use the controls to maximize the information we get about the parameter @xmath4 through the observed process @xmath40 . the parameter estimation is done using maximum likelihood and it is therefore natural to try to maximize the fisher information of our observed process which we can express as @xmath41\ ] ] which we label as the partial observation fisher information ( pofi ) , see appendix [ sec : express fi ] for details on its construction . when we consider continuous time dynamical systems the observation spaces will be continuous , but we will use this discretized fisher information as an approximation to the actual fisher information of the observations . in order to maximize the fisher information we employ the techniques of stochastic dynamic programming . assuming a reward function @xmath42 , our objective is to maximize the total expected reward @xmath43 $ ] by use of the controls . the essence of dynamic programming is that by starting at time @xmath44 and working backwards , we can compute an optimal policy that maps a state @xmath1 to a control @xmath2 that accounts for the choices of @xmath2 that we will make in the future . in a generic dynamic program we set @xmath45 and and then going backwards from @xmath46 solve @xmath47 \}\ ] ] where @xmath48 is called the value function , and we get the associated control @xmath49 \}\ ] ] for every state @xmath1 . this will give us a policy of what control to use at a certain state @xmath1 at a certain time @xmath3 . the use of these controls will maximize the total expected reward @xmath50 $ ] . we refer to puterman @xcite for a detailed description of dynamic programming . with this in mind , we set @xmath51 and we try to maximize the fisher information @xmath52 $ ] . note that in this instance the reward function depends on the entire history of observations and controls up to time @xmath3 . the value function in the corresponding dynamic program is @xmath53 \right \}\ ] ] and we denote it the fisher information to go . a problem here is that just in the first step of the dynamic program ( @xmath54 ) we would have to calculate the fisher information to go for @xmath55 many combinations of @xmath56 and @xmath57 . this is formidable for even modest dimensions . we therefore approximate the process by conditioning only on the last @xmath58 observations in the fisher information ; @xmath59 where @xmath60 is some prior that we assume for @xmath61 , although we generally suppress it in notation since we assume it is fixed . if @xmath62 we set @xmath63 to mean @xmath64 to ease notation . the reward becomes @xmath65 and @xmath66 \right \}$ ] the fisher information to go . the pseudocode for the corresponding dynamic program is given in appendix [ sec : dyn pofi ] . for this approximate dynamic program to be sensible we want the approximated fisher information to approach the true fisher information as @xmath39 increases . this holds given that the pomdp process satisfies certain technical mixing conditions , which we have stated in detail in assumption [ mixing ] in appendix [ sec : mixing ] . assume the conditions in assumption [ mixing ] hold . then , for @xmath67 and any control policy , we have @xmath68 [ fi approx ] where the constant @xmath69 and @xmath70 do not depend on @xmath39 or @xmath35 , see appendix [ sec : bounds ] . the proof requires extensions of work in cappe et al . @xcite and is given in appendix [ sec : bounds ] with further technical results reserved to appendix 2 . theorem [ fi approx ] states that @xmath71 approaches the true fisher information exponentially as @xmath39 increases , and is thus a viable approximation for the fisher information in the dynamic program . the runtime of the dynamic program however also grows exponentially in @xmath39 and we found that while setting @xmath72 , i.e. conditioning on one observation , gave poor results in some of our simulations , conditioning on two observations , i.e. @xmath73 , generally gave good results when compared to other control policies . setting @xmath74 increased runtime greatly and was in most of our applications infeasible without making more approximations to how the dynamic program is run . the exact effect of increasing @xmath39 is problem specific , and a thorough analysis of what @xmath39 to choose is beyond the scope of this paper . an alternative method to choose controls was proposed by lin et al . they considered constructing an optimal control policy for the fisher information that would apply if @xmath75 were observed directly ; @xmath76 this is labeled the full observation fisher information ( fofi ) . as noted before , when considering continuous time stochastic systems , the state space is continuous , but we use this fisher information as an approximation to the continuous state fisher information . an advantage of using fofi over pofi is that when running the dynamic program the markov property of the markov decision process @xmath77 allows us to only consider a maximization over the state space @xmath78 but not past values @xmath79 . the dynamic program for fofi is given in appendix [ sec : dyn fofi ] . however , maximizing fofi can lead to suboptimal controls since it is not the correct fisher information for the data . additionally , when the actual experiment is run we do not observe @xmath18 . instead we have to use the observed values to get a probability distribution ( a filter ) on the state @xmath1 , @xmath80 and use the control associated with the state that has the highest probability . the computation cost of running a dynamic program with fofi is @xmath81 , generally lower than that of pofi ; @xmath82 . the cost of estimating the state @xmath18 at runtime is @xmath83 at each time point @xmath3 . for details see b.1 in appendix 2 . after running an experiment , using one of the control policies , the parameter @xmath4 is estimated either via an em algorithm or by directly maximizing the loglikelihood . these two estimation methods had similar accuracy in our simulations , although the em algorithm was generally slower . convergence of the em algorithm is discussed in cappe et al . @xcite for hidden markov models and extends naturally to pomdp s . for the asymptotic properties of the mle we refer to cappe as well @xcite , where conditions for consistency and asymptotic normality in hidden markov models are given the central elements of their proof are the stationarity of the process @xmath84 along with certain forgetting properties of the filter , meaning that ignoring all but the past @xmath39 observations ( as we do ) yields an error in the filter distribution that decreases exponentially in @xmath39 . we note that if we employ a time - independent control policy ( as we do in section 5 ) , we obtain to a hidden markov model and can rely on @xcite if we assume stationarity . in appendix 2 we establish extensions of forgetting properties for pomdp models more generally , which points to a more general asymptotic theory for the mle in this case , but do not pursue this here . theorem [ fi approx ] shows that using @xmath71 is a good approximation to the partial observation fisher information for running a dynamic program . this provides a control policy that is an approximation to the optimal control policy . now consider using this approximate policy to run an experiment and then estimating @xmath4 by evaluating the mle . the asymptotic variance of this mle will be the the inverse of the partial observation fisher information , with controls from the approximate policy . it is therefore of interest to compare pofi with an optimal policy and pofi with the approximate policy . in appendix [ sec : best fi ] we show given that the mixing conditions in assumption [ mixing ] ( [ sec : mixing ] ) hold we have @xmath85 [ best fi ] where @xmath86 are the optimal controls , @xmath87 the approximate optimal controls and fi is the partial observation fisher information . the constant @xmath88 and @xmath70 do not depend on @xmath39 or @xmath35 , see appendix [ sec : best fi ] for details . that the asymptotic variance of the mle converges to the best possible asymptotic variance exponentially quickly in @xmath39 further supports our approximations . while the fofi strategy has been shown to be effective in lin et al . @xcite it is possible to define systems in which the strategy is not optimal and may in fact be worse than just using fixed or random controls . usually certain parts of state space will give more information about a parameter than others , given that the state space is perfectly observed . in these cases optimal controls would try to move the process to these states . however , if the state space is only partially observed , most information might be obtained in different parts of state space and the fofi controls become suboptimal . in cases like this pofi often does better than fofi . in this example , we demonstrate a system where fofi and pofi choose very different controls , and using a simulation study , we show that the controls chosen by pofi produce less variable parameter estimates . consider a discrete time markov chain @xmath1 with state space @xmath89 and a transition probability matrix @xmath90\ ] ] where the parameter of interest is @xmath91 $ ] and the control is @xmath92 . for @xmath93 or @xmath94 , choosing the control @xmath95 will increase the probability of the markov chain staying in its current state while choosing @xmath96 will increase the probability of it leaving its state . + now assume this process is nt observed directly but through a related process @xmath5 with state space @xmath97 whose transition probabilities depend on which state @xmath1 is in . we denote the transition probability matrices with @xmath98 given by @xmath99 , p_2= \left [ \begin{array}{c c } .5 & .5 \\ .5 & .5\\ \end{array } \right ] , p_3= \left [ \begin{array}{c c } 1-\frac{p}{2 } & \frac{p}{2 } \\ \frac{p}{2 } & 1-\frac{p}{2 } \\ \end{array } \right]\ ] ] if @xmath1 were observed we would get information about the parameter @xmath100 when @xmath1 leaves state @xmath101 and from @xmath5 when @xmath94 . the idea here is that since the fofi controls assume the whole state space is observed they might encourage @xmath1 to be in state @xmath101 , while the pofi controls that take into account what is actually observed might choose the controls more intelligently . indeed when calculating the controls according to fofi the long run control is to leave one s state " if @xmath94 and stay in one s state " if @xmath93 . it is harder to predict what kind of controls result from using pofi , but the control policy is given in table [ 6 state pofi ] . @xmath5 & 1 & 2 & 1 & 2 & 1 & 2 & 1 & 2 + @xmath102 & 1 & 1 & 2 & 2 & 1 & 1 & 2 & 2 + @xmath103 & 1 & 1 & 1 & 1 & -1 & -1 & -1 & -1 + to illustrate this difference , a simulation study was carried out to test which method performed better : the process @xmath1 was run for @xmath104 steps with @xmath105 , using controls chosen by pofi and again using those chosen by fofi . additionally we ran a simulation of the same length , but where the control was chosen randomly , with @xmath96 and @xmath95 having equal probability . then the parameter @xmath100 was estimated using an em algorithm . this was done 500 times to get an empirical distribution for the estimates of @xmath100 . the results are given in table [ 6 state table ] . estimates of @xmath100 using a pofi policy had the lowest mse and variability , but estimates based on the random policy had lower bias . estimates using a fofi policy were comparatively worse . fofi & .0109 & .1098 & .0121 + pofi & .0094 & .0867 & .0076 + random & .0059 & .0941 & .0089 + the following example describes an application of the above methods in the context of experimental economics . the problem is derived from sachat et . @xcite , in which we wish to model how humans change their game - playing strategies over time . .rewards in the gamble safe game . the first number is the reward for the row player and the second number the reward for the column player , given a certain outcome . [ cols="^,^,^",options="header " , ] in order to apply the methods described above to dynamical systems , we need to approximate them by a suitable partially observed markov decision process . we achieve this by discretizing time , state and observation spaces . in this paper , the continuous stochastic dynamical systems considered are of the form @xmath106 where @xmath4 is the parameter of interest , to be estimated , @xmath8 is a control that can be chosen by user , @xmath9 is the vector of state variables , @xmath10 is a vector valued function and @xmath11 a wiener process . the dynamical system is approximated on a fine grid of times @xmath107 and we obtain a discrete - time model @xmath108 where @xmath109 are independent normal random variables . we assume the underlying state variables @xmath1 are only observed partially or noisily . @xmath110 where @xmath111 . in order to approximate this as a markov chain , the state space is discretized in each dimension and the model is then thought of as moving between the different boxes . the probability of moving from box to box is approximated using the normal p.d.f . at the midpoints of the boxes . in the examples in this paper , only equidistant discretization is considered , but this restriction can be readily removed . if we label the two midpoints as @xmath112 and @xmath113 and the area of the second box as @xmath114 this probability is given as @xmath115 @xmath116 where @xmath117 is the dimension of @xmath9 . the probabilities are then normalized to make sure they sum to @xmath101 . if the controls @xmath2 can be chosen on a continuous scale then this scale has to be discretized as well . @xmath118 is then a markov decision process , and one can run the fofi dynamic program . for the pofi dynamic program the observation space needs to be discretized as well . the probability of what observation box is observed depends on in which box the underlying markov chain is in . if we label the midpoint of the underlying markov chain midpoint as @xmath119 and the midpoint of the observed process box midpoint as @xmath120 , and the area of the latter box as @xmath121 this probability is given as @xmath122 these probabilities are also normalized to sum to @xmath101 . the process @xmath123 is now a partially observed markov decision process and one can run the pofi dynamic program . the morris lecar model @xcite describes oscillatory electric behavior in a single neural cell , as regulated by flow of potassium and calcium ions across the cell membrane . these models are defined in terms of state variables @xmath124 and @xmath125 representing the voltage across the membrane and the flux of the potassium channel respectively . @xmath126 \dot{n}_t & = -\phi\cdot ( n_t - n_\infty(v_t ) ) / \tau_n(v_t ) \label{eq : ml - n}\end{aligned}\ ] ] where @xmath127 , @xmath128 and @xmath129 . we will write @xmath130 and @xmath131 as shortcuts equations ( [ eq : ml - v ] ) and ( [ eq : ml - n ] ) . the voltage between cells depends on potassium and calcium concentrations , and on the amount of leakage . the further these factors are away from their equilibriums @xmath132 the greater the rate of change in voltage . the multiplicative value @xmath125 changes the conductance of the potassium channel and is modeled through the second differential equation in which @xmath125 is driven towards a voltage - dependent equilibrium level defined by @xmath133 but converges to this at a much slower rate then the dynamics of @xmath124 . the neuron is stimulated by an external applied current , @xmath134 ( our control ) , and @xmath124 is measured . our goal is to maximize information about the parameters @xmath135 and @xmath136 , considered separately . we consider a stochastic version of this neural firing model , derived from @xcite , by adding @xmath137 and @xmath138 to equations ( [ eq : ml - v ] ) and ( [ eq : ml - n ] ) respectively , where @xmath139 and @xmath140 are independent wiener processes . stochastic models are important in this context in order to accommodate observable variation in the inter - spike interval where a deterministic model will require a fixed period ; see @xcite , for example . the first step is to discretize these equations with respect to time . we get that @xmath141 and @xmath142 where @xmath143 . we discretized @xmath124 onto the range @xmath144 $ ] and @xmath125 onto @xmath145 $ ] , after running a few trial versions of the model . both ranges where discretized into @xmath146 intervals . only @xmath124 is measured and it is measured noisily , @xmath147 where @xmath148 . the observation space was discretized to the same range as @xmath124 but into @xmath149 intervals . these approximations give rise to a partially observed markov decision process to which our methods can be applied . both fofi and pofi controls were calculated for this experiment , for the parameters @xmath150 , @xmath151 and @xmath136 , considered separately . the values for the parameters were set to be @xmath152 , @xmath153 , @xmath154 , @xmath155 , @xmath156 , @xmath157 , @xmath158 , @xmath159 , @xmath160 , @xmath161 , @xmath162 , @xmath163 and @xmath164 . the controls range was set to be @xmath165 $ ] and discretized to the set @xmath166 . an example of what the control policy looks like is given in figure [ fig : morris - lecar ] . a simulation study was run for each of the three parameters ; the system was simulated within the discretized markov chain framework with @xmath167 time steps and all schemes had 100 simulations . the parameter in question was estimated for each simulation using an em algorithm . as a baseline comparison we also ran a simulation study using a fixed control ( @xmath168 ) . the results are given in table [ morris - lecar ] . the difference between pofi and fofi turns out to be not very dramatic , likely due to the observations providing a great deal of information about the underlying state variables , which is when fofi performs well . & fofi & .4234 & 2.4722 & 6.2913 + & pofi & .4129 & 2.4068 & 5.9632 + & fixed & .9098 & 3.4240 & 12.551 + & fofi & .0613 & .3671 & .1385 + & pofi & .0158 & .3706 & .1376 + & fixed & .0249 & .6193 & .3841 + & fofi & .00485 & .01085 & .00014 + & pofi & .00257 & .01037 & .00011 + & fixed & .01357 & .02643 & .00088 + in the discussion above we calculated the dynamic program assuming knowledge of the parameter @xmath4 , the very thing we wish to estimate with maximal precision . since the dynamic programs we have considered are run before the experiment is started we generally wo nt have data to estimate @xmath4 . additionally , for the fofi simulations we have used @xmath4 directly to estimate @xmath1 within the filter to get the appropriate control , but this will not be possible in practice . there are a few ways of dealing with this . assuming some prior information one can use a prior for @xmath4 to run the dynamic program . to do this , we add one more expectation for @xmath4 at every time step @xmath3 , and then maximize the expected fisher information to get the best control . this strategy was employed in lin et al . @xcite . the rather obvious deficiency here , for both pofi and fofi , is that as the experiment runs , we get observations that can be used to improve our prior for @xmath4 , and could be used to get better controls , if we could brake the experiment and rerun the dynamic program . in some systems the time spent in each state is very short , too short to perform many calculations , making it valuable to have a look - up table " of controls . here the pofi controls have an advantage over the fofi controls , in the sense that they are of the look - up " kind , as fofi requires estimation of the underlying @xmath1 process , before the control can be looked up . in other systems , there is time to do some calculations between transitions . note , for example , that at time @xmath3 we have observed @xmath169 and this will allow us to calculate a posterior distribution @xmath170 for our parameter of interest . this posterior could then be used to run the dynamic program again , as described above , from time @xmath44 to time @xmath3 . this can be quite time consuming if done at each time step @xmath3 , so we propose a method that relies on the value iteration algorithm . a popular algorithm from the theory of markov decision processes is the value iteration algorithm ( via ) , see puterman @xcite . the theoretical motivation of via is similar to dynamic programming , but here the objective is to maximize an expected total reward @xmath139 that has a discounting factor @xmath171 , where @xmath172 , and the time horizon is assumed to be infinite ; @xmath173\ ] ] and @xmath139 is labeled as the expected total discounted reward . @xmath139 exists if @xmath174 is bounded , which is the case in the problems we consider . in puterman @xcite it is shown that an optimal control policy exists and it can be chosen to be time independent , i.e. to depend only on the state @xmath1 , and not the time @xmath3 . moreover this optimal control can be approximated using the value iteration algorithm described below . our experimental setting is neither discounted nor has it an infinite time horizon , but blackwell optimality guarantees that controls that maximize @xmath139 also maximize the expected average reward @xmath140 ( or its @xmath175 if the limit does nt exist ) ; @xmath176\ ] ] given that @xmath171 is chosen close enough to one . a blackwell optimal control policy exists if the state and action spaces are finite , which is the case in our setting . maximizing @xmath140 effectively amounts to maximizing the average input of each observation in our fisher information ; a reasonable strategy . how small @xmath177 needs to be is generally hard to determine , and choosing @xmath171 too high will cause the algorithm to converge slowly . see puterman @xcite chapter @xmath178 for more on blackwell optimality . in via we we calculate @xmath179 \right \}\ ] ] in a while - loop until @xmath180 converges to some fixed point , within some tolerance . convergence is guaranteed since each iteration of @xmath180 is a contraction mapping . our aim with via is to maximize what we in the pofi case label , the average partial observation fisher information @xmath181 ( or in the fofi case , the average full observation fisher information ) which is a reasonable quantity to maximize in order to obtain a time - invariant policy . we propose running via at every time step @xmath3 , but to use the posterior for @xmath4 , @xmath170 , which is conditioned on all the data observed so far , instead of using the prior for @xmath4 . this will give a control that maximizes the average fisher information , using all the parameter information that is available at time @xmath3 . instead of starting via at each time @xmath3 with @xmath182 , considerable time can be saved by using the last value vector @xmath180 from the previous run of via at time @xmath183 . this is because the posterior for @xmath4 often does nt change much between time steps , and the last @xmath180 from time @xmath183 thus being relatively close to the fixed point at time @xmath3 . let @xmath184 denote the value vector at time @xmath3 at the @xmath185th iteration of the @xmath3th via and let @xmath170 denote the posterior for @xmath4 given observations up till time @xmath3 . also , to ease notation , let @xmath186 . the pseudocode for this modified via using pofi is : set @xmath187 and @xmath188 @xmath189 @xmath190 and calculate and store @xmath191 @xmath192\ ] ] n = n+1 set @xmath193 now let @xmath194 @xmath195\ ] ] use control @xmath2 , and observe @xmath196 and then update the posterior for @xmath4 , @xmath197 updating fofi policies online using via can be done in a similar way . in the next example we compare using updated policies . polymerase chain reaction is a well established method to copy and multiply dna . we are interested in modeling the growth dynamics of dna template ( @xmath1 ) , for a fixed amount of substrate . the model we use is @xmath198 where @xmath199 . here @xmath1 is the amount of dna template , @xmath200 and @xmath201 the parameters of the model and @xmath2 the control , the percentage of template removed at each time point . we are interested in estimating the parameter @xmath201 , labeled the half - saturation constant . a good reference for pcr models is @xcite . we measure the amount of dna template at each time point , but with an error . our observations are @xmath202 and thus we have a dynamical system which when discretized becomes a partially observed markov decision process . the range for @xmath1 was set to be @xmath203 $ ] and then discretized into @xmath204 intervals , and @xmath5 was discretized to the same range , but only into @xmath205 intervals . the parameter values were set to be @xmath206 , @xmath207 , @xmath208 , @xmath164 and the possible values of the control @xmath209 . still with the objective of maximizing fisher information , we more realistically assumed priors for the parameters of the system , as discussed above . we conducted a simulation study using controls based on these priors for both pofi and fofi , and then compared their performance to controls that are updated online using via , also both for pofi and fofi . as a baseline comparison we also ran simulations using fixed controls and simulations where the true parameter is used ( unrealistically ) to calculate the control policy via dynamic programming as in the previous examples . for fixed controls we report the simulation with the lowest mse , which was when @xmath210 . the range for @xmath201 was set to be @xmath211 $ ] and then we discretized that interval into 10 points @xmath212 . we then considered a uniform prior on these points and a prior that is somewhat inaccurate , and puts the weight @xmath213 on the point @xmath214 and gives the others equal weight . the discounting factor for via was set to be @xmath215 . our simulation study had the time length @xmath216 and there were @xmath217 simulations for each case . the parameter @xmath201 was estimated using an em algorithm . the simulation results are given in table [ chemostat ] . c c & bias & st . dev . & mse + fofi & 0.1059 & 0.6598 & 0.4465 + pofi & 0.0053 & 0.6189 & 0.3831 + + & bias & st . dev . & mse + fofi & 0.0755 & 0.6374 & 0.4120 + pofi & 0.0516 & 0.7051 & 0.4998 + + fixed & .1264 & .7466 & 0.5734 + & & bias & st . dev . & mse + fofi & 0.0388 & 0.6180 & 0.3834 + pofi & 0.0766 & 0.5999 & 0.3658 + + & bias & st . dev . & mse + fofi & 0.0713 & 0.6787 & 0.4657 + pofi & 0.0954 & 0.6750 & 0.4648 + + & bias & st . dev . & mse + fofi & 0.0659 & 0.6235 & 0.3932 + pofi & 0.0323 & 0.6249 & 0.3916 + + we note that when we calculate the controls prior to the experiment ( no online updating ) , both the pofi and fofi controls are significantly better than using a fixed control , and pofi seems to do better than fofi when we use an uniform prior . interestingly in the fofi case , calculating the controls using the inaccurate prior does better then using the uniform prior , likely due to a reduction in prior variance , in spite of additional bias . accuracy increases in most cases when we allow for online updating using the via algorithm . starting the via with an uniform prior does better than starting with the inaccurate one , which is probably due to the via having to spend more time repairing " the prior . also , we note that via controls with uniform prior have a similar performance to a control policy using the true ( unknown ) parameter . additionally , in figure [ via_time ] , we see that using the previous final value vector as the starting value vector of via when going from time point @xmath3 to @xmath218 , does save considerable time , and more so as @xmath3 grows and the posterior for the parameter starts to change less . , for pofi using a uniform prior for the pcr model . in this paper we compared two possible ways to conduct experimental design in parametric pomdp s , based on using dynamic programming to maximize either the partial observation fisher information or the full observation fisher information . settings can arise where controls chosen by fofi are not optimal , due to focusing on the underlying process rather than the observed process , and in these cases controls chosen with pofi often perform better , as in the six state example and the adversarial game . in some of the examples analyzed they performed similarly . in recent years , there has been growing interest in statistical procedures within dynamical systems , such as parameter estimation and hypothesis testing , and many of these procedures could be performed more efficiently given good experimental design . in this paper we fully discretized the state and observational spaces to transform dynamical systems with stochastic errors into partially observed markov decision processes , allowing us to use the methods developed for pomdp s to our advantage . we also noted how the problem of parameter dependence can be overcome by averaging over a prior . additionally given that there is enough time between consecutive time steps , we showed how the controls can be efficiently updated online using observations gathered so far , by using a variant of the value iteration algorithm . this was demonstrated in the pcr example . finding controls that maximize information about parameters is a computationally challenging task . we have successfully demonstrated techniques for up to two dimensional systems , for a one dimensional parameter . adding dimensions in state , parameter or observation space quickly make the methods considered computationally intractable . considering a longer lag of past observations for pofi might also increase accuracy , but again at the cost of computation time . in order to extend these methods to higher dimensions one could possibly use techniques of approximate dynamic programming , see powell @xcite . we maximize @xmath59 with reward function @xmath65 and fisher information to go @xmath219 \right \}\ ] ] the pseudocode for this dynamic program is : + @xmath220 @xmath189 @xmath221 and calculate and store @xmath222 \right \}$ ] @xmath223 \right \}$ ] we maximize @xmath76 by setting the reward function as @xmath224 . the pseudocode for this dynamic program is : + @xmath225 @xmath189 @xmath1 and calculate and store @xmath226 \right \}$ ] @xmath227 \right \}$ ] in this section we find useful expressions for the fisher information of an experiment , that are needed to derive convergence arguments and set up dynamical programs . we use the short hand notation @xmath228 where the dependence on @xmath229 is suppressed . for data @xmath230 the fisher information for @xmath4 can be expressed in one or two derivatives @xmath231= e\left[\left(\sum_{t=0}^{t-1 } \dot{h}_{t+1}\right)^2 \right]\ ] ] and similarly the fisher information to go @xmath232 = e\left[\left(\sum_{t = k}^{t-1 } \dot{h}_{t+1}\right)^2 \middle|y_{0:k},u_{0:k-1}\right]\ ] ] where the equality is justified by both quantities being the fisher information for the same observations . we set @xmath233 . the fisher information to go can be calculated recursively ( in both one or two derivatives ) : @xmath234 = e\left[\left ( \dot{h}_{k+1}\right)^2+fi_{k+1}\middle|y_{0:k},u_{0:k-1}\right]\ ] ] [ fi recursive ] in the case of using two derivatives this follows from iterated expectation . in one derivative we have @xmath235\ ] ] @xmath236\ ] ] the cross term is @xmath237\ ] ] @xmath238\middle|y_{0:k},u_{0:k-1}\right]\ ] ] @xmath239\middle|y_{0:k},u_{0:k-1}\right]\ ] ] @xmath240 = 0\ ] ] thus @xmath241\ ] ] @xmath242 \middle|y_{0:k},u_{0:k-1}\right]\ ] ] @xmath243\ ] ] @xmath244\ ] ] and similarly @xmath245\ ] ] [ fi express ] this follows from using induction and lemma [ fi recursive ] . running an exact dynamic program , with @xmath246 as the value function , is not feasible because of the curse of dimensionality , leading us to certain approximations . we set @xmath247 where @xmath248 is the assumed distribution of @xmath249 and we will consider it to be fixed and known . allowing @xmath39 to be negative will ease notation when @xmath250 . we now set @xmath251\ ] ] @xmath252\ ] ] and denote it the approximated fisher information to go . that the formulation in one derivative is equal to the one in two derivatives follows from the individual parts of each sum having a fisher information interpretation . cappe et al . @xcite establish forgetting properties of the filter by assuming mixing conditions for hidden markov models . we use the same conditions , slightly modified to allow for controls , for pomdps ; modified strong mixing conditions . for each control @xmath253 there exist a transition kernel @xmath254 and measurable functions @xmath255 and @xmath256 from @xmath257 to @xmath258 such that for any @xmath259 , @xmath260 and @xmath261 , @xmath262 [ mixing ] cappe et al.s @xcite discussion on what models satisfy these conditions applies analogously to pomdp s . given these conditions we can prove the following , which is a modification of lemma @xmath263 in @xcite ; assume the strong mixing conditions in assumption [ mixing ] . then @xmath264 where @xmath265 and @xmath266 [ lemma 12.5.3 ] a proof is provided in appendix 2 . in this section we show that the approximated fisher information approaches the true fisher information exponentially as one conditions on more and more observations , while using the same controls . by corollary [ fi express ] the true fisher information is @xmath267 but since that is hard to optimize we consider @xmath268 see definitions for @xmath269 above . here we use fisher information in one derivative , but as noted above it is equivalent to using the formulation in two derivatives . also note that where @xmath270 we just set it to @xmath271 and use the initial distribution of @xmath272 . assume the mixing conditions in assumption [ mixing ] hold . then @xmath273 @xmath274 [ bound lemma ] set @xmath275 which sets an upper bound on the length of @xmath276 . note that @xmath277 also bounds @xmath278 since @xmath279 . now @xmath273 @xmath280 @xmath281 @xmath282 using theorem [ lemma 12.5.3 ] . setting @xmath283 gives the result , although that might not be the best bound . we now restate theorem [ fi approx ] from section @xmath284 . assume the conditions in assumption [ mixing ] hold . then , for @xmath67 and any control policy , we have @xmath285 where @xmath286 and @xmath287 is the bound from lemma [ bound lemma ] ; @xmath288 . [ fi approx append ] @xmath289 @xmath290 @xmath291 and by cauchy schwarz @xmath292 for the first parenthesis we use theorem [ lemma 12.5.3 ] to get @xmath293 and the second one is bounded by the lemma [ bound lemma ] . we get @xmath294 @xmath295 exactly the same arguments can be used to show that the approximated fisher information to go @xmath296 approaches the true fisher information to go as @xmath39 increases . a related problem we are interested in is how well controls that consider the last say @xmath39 observations do in comparison with controls that consider all past observations . that is , we want a bound on the best possible fisher information given controls that consider all past observations , compared with the best possible fisher information given controls that only consider the last @xmath39 observations . we first establish a baseline difference between two parts of the fisher information . assume the conditions in assumption [ mixing ] hold . then , for any control policy and any @xmath117 , we have @xmath297 where @xmath288 is the bound from lemma [ bound lemma ] . [ ind diff ] follows from lemma [ bound lemma ] and theorem [ lemma 12.5.3 ] as in the proof of theorem [ fi approx ] . as above , we assume that the fisher information to go ; @xmath298\ ] ] is approximated by @xmath299\ ] ] given that our controls are obtained by dynamic programming , we have that the optimal control at time @xmath3 is dependent on the optimal control obtained at time @xmath218 . let @xmath300 denote the set of optimal controls obtained in this manner , i.e. @xmath301 maximize @xmath246 . as argued these controls are hard to calculate and thus we resort to approximate optimal controls , here denoted @xmath302 , where @xmath303 maximize @xmath296 . we now restate theorem [ best fi ] from section @xmath304 on the loss of using approximate controls instead of exact ones in fisher information , in an experiment of length @xmath35 . given that the mixing conditions in assumption [ mixing ] hold we have @xmath305 where @xmath306 , and @xmath287 is the bound from lemma [ bound lemma ] . [ best fi append ] we analyze the difference by bounding errors in each step of the dynamic program inductively , starting at time @xmath54 and going backwards . set @xmath307 . we find that @xmath308 @xmath309 so far only using that @xmath310 maximizes @xmath311 and @xmath312 maximizes @xmath313 . @xmath314 @xmath315 by lemma [ ind diff ] . we now inductively assume @xmath316 where @xmath317 , and then get @xmath318 @xmath319 @xmath320 now moving from @xmath321 to @xmath322 we have @xmath323 since @xmath324 are the controls that maximize @xmath325 . by adding and subtracting the same quantity we get the following equivalent inequality @xmath326 @xmath327 @xmath328 line @xmath329 is bounded by @xmath330 by lemma [ ind diff ] and line @xmath331 by @xmath332 using @xmath333 and lemma [ ind diff ] . therefore @xmath334 @xmath335 and for the whole experiment we find @xmath336 we analyze the computational complexities of fofi and pofi . the computations required can be split into computations done prior to the experiment and computations that are required while running the experiment . a direct comparison is not completely fair since fofi requires computations at runtime while pofi does not as discussed below . we assume that the transition probability matrix @xmath338 is given . calculating @xmath339 is negligible compared to the calculations required for the dynamic program ; if we set @xmath340 then for a given time @xmath3 in the dynamic program we need to maximize @xmath341\ ] ] over @xmath342 for each @xmath78 , where @xmath343 is the value function from the previous step @xmath218 . this calculation requires adding @xmath344 and @xmath343 which are two @xmath345 tensors with cost @xmath346 . next we need a dot product between @xmath347 and @xmath348 over the @xmath349 dimension which has cost @xmath350 . finally maximizing over @xmath2 for each @xmath1 has cost @xmath351 . thus each step @xmath3 has cost @xmath352 and the dynamic program in total has cost @xmath81 in runtime a filter is required to estimate the state @xmath1 . the filter for time @xmath218 can be calculated via the following recursive formula @xmath353 and then normalizing . this requires @xmath354 dot products of vectors of length @xmath22 , with cost @xmath83 and the normalization has cost @xmath355 . thus we have @xmath83 computations at each time step @xmath3 during runtime . here the dynamic program maximizes the approximated partial observation fisher information , @xmath356\ ] ] first we note that @xmath357 is a @xmath358 tensor , and it can be calculated using bayes rule at the cost @xmath359 . calculating @xmath360 can also be done at the cost @xmath359 , but can also be effectively approximated using the finite difference approximation to the derivative . the cost analysis of the pofi dynamic program is just like the analysis of fofi . at a given time @xmath3 adding @xmath344 and @xmath343 has cost @xmath361 , the dot product between @xmath362 and @xmath363 has cost @xmath361 and the maximization has cost @xmath364 . this section is devoted to expanding hidden markov model theory to partially observed markov decision processes . we base it completely on cappe et al . @xcite and use their notation , only changing what is necessary . the purpose is to prove theorem 3 in appendix @xmath366 , which is a modified version of lemma @xmath263 in @xcite . in most cases the changes will amount to adding controls and seeing that the theory follows through , although the proof of theorem 3 has more substantial changes . let @xmath367 and @xmath368 be the state space and the observations space respectively . let @xmath369 be a transition kernel for our state space , where @xmath253 is a control , and @xmath21 is finite . also let @xmath370 be the transition kernel for moving from the state space to the observation space . we generally assume that the markov chain is initialized with distribution @xmath371 , and then runs for @xmath185 steps @xmath372 and that @xmath373 decisions are made on what controls @xmath253 to use . this results in @xmath185 observations @xmath374 and @xmath373 control @xmath375 . a standard result in hmm theory is that conditional on the observations @xmath385 the process @xmath386 still is a markov chain , although non - homogeneous , with a transition kernel called the forward smoothing kernel . we state the transition kernel here for our case , also conditional on the controls . the forward smoothing kernel allows us a convenient way of calculating the smoothing distributions . we first compute all the backward variables @xmath394 using the backward recursion given . we then note that @xmath395 can be calculated as @xmath396 and then we have the following recursion @xmath397 where @xmath398 are the forward kernels , and the last equation is a short hand way of writing the integral . to continue towards forgetting properties we need to introduce total variation ( see definition 4.3.1 in @xcite ) . let @xmath400 be a signed measure ( it can be negative ) and let @xmath401 where @xmath402 are ( positive ) measures . so if @xmath403 is the state space @xmath404 then we need to define the dobrushin coefficient ( see definition 4.3.7 in @xcite ) . let @xmath22 be a transition kernel from @xmath403 to @xmath257 . @xmath405 the dobrushin coefficient is sub - multiplicative ( see prop . 4.3.10 in @xcite ) . if @xmath406 are @xmath13 transition kernels we have @xmath407 when considering forgetting properties it seems logical to show that the filter @xmath414 depends less and less on the initial distribution of @xmath415 , as @xmath117 increases . specifically we have that when comparing initial distributions @xmath371 and @xmath416 : @xmath417 @xmath418 and since the dobrushin coefficient is sub - multiplicative @xmath422 and since the dobrushin coefficient @xmath423 satisfies @xmath424 we at least have that the difference between the @xmath13 filters is non - expanding . establishing forgetting properties thus amounts to showing @xmath425 for the forward smoothing kernels @xmath426 . note that so far no assumptions have been made on how quickly the hidden markov model mixes . those assumptions are made to get @xmath427 . in our case we have different transition kernels for each control . the weakest assumptions we can get away with is , if each transition kernel @xmath430 has a corresponding transition kernel @xmath431 and measurable functions @xmath432 and @xmath433 satisfying the strong mixing condition . by letting @xmath434 and @xmath435 we see that we can consider the same @xmath436 functions for each transition kernel @xmath430 . we restate the strong mixing conditions : modified strong mixing conditions . for each control @xmath253 there exist a transition kernel @xmath254 and measurable functions @xmath255 and @xmath256 from @xmath257 to @xmath258 such that for any @xmath259 and @xmath260 , @xmath437 lemma @xmath438 in cappe et al . @xcite uses the mixing conditions stated above to establish contracting bounds on the dobrushin coefficient . we restate the lemma for the pomdp case , where we also condition on the controls , and use the modified mixing conditions . a. for any non - negative integers @xmath117 and @xmath185 such that @xmath439 and @xmath261 , @xmath440 ( x ) \leq \prod_{j = k+1}^n \varsigma^+(y_j)\ ] ] b. for any non - negative integers @xmath117 and @xmath185 such that @xmath439 and any probability measures @xmath371 and @xmath416 on @xmath367 , @xmath441(x)}{\int \nu'(dx ) \beta_{k|n}[y_{k+1:n},u_{k : n}](x ) } \leq \frac{\varsigma^+(y_{k+1})}{\varsigma^-(y_{k+1})}\ ] ] c. for any non - negative integers @xmath117 and @xmath185 such that @xmath439 , there exists a transition kernel @xmath442 from @xmath443 to @xmath367 such that for any @xmath444 , @xmath445 , and @xmath446 , @xmath447(x , a)\ ] ] @xmath448 d. for any non - negative integers @xmath117 and @xmath185 , the dobrushin coefficient of the forward smoothing kernel @xmath449 $ ] satisfies @xmath450 ) \leq \rho_0(y_{k+1 } ) : = 1 -\frac{\varsigma^-(y_{k+1})}{\varsigma^+(y_{k+1})}\ ] ] if @xmath439 , and @xmath450 ) \leq 1 -\int \varsigma^-(y ) dy\ ] ] if @xmath451 . a. letting @xmath452 in the strong mixing conditions we find that for all @xmath253 @xmath453 we also have @xmath454 @xmath455 @xmath456 @xmath457 @xmath458 the other inequality is similar . b. using the recursion for the backward variables we find @xmath459 @xmath460 @xmath461\beta_{k+1|n}(y_{k+2:n},u_{k+1:n},dx_{k+1})\ ] ] @xmath462\beta_{k+1|n}(y_{k+2:n},u_{k+1:n},dx_{k+1})\ ] ] @xmath463 we get a similar inequality for @xmath255 . also note that the last integral does nt depend on @xmath371 , so it cancels when we take the ratio . the result follows . c. we have that @xmath464(x , a ) = \frac{\int_a q^{u_k}(x , dx_{k+1})g(x_{k+1},y_{k+1 } ) \beta_{k+1|n}(x_{k+1})}{\int q^{u_k}(x , dx_{k+1})g(x_{k+1},y_{k+1 } ) \beta_{k+1|n}(x_{k+1})}\ ] ] @xmath465 and we can set @xmath466 d. using @xmath467 we find that @xmath464(x , a ) \geq \frac{\varsigma^-(y_{k+1})}{\varsigma^+(y_{k+1 } ) } \lambda_{k|n}(y_{k+1:n},u_{k : n},a)\ ] ] and thus assumption @xmath468 holds and lemma @xmath469 gives @xmath470 a. we let @xmath371 and @xmath416 be two different initial distributions for @xmath471 . now for @xmath472 @xmath473 - \phi_{\nu',k|n}[y_{0:n},u_{0:n-1}]||_{tv}\ ] ] @xmath474||\phi_{\nu,0|n}[y_{0:n},u_{0:n-1 } ] - \phi_{\nu',0|n}[y_{0:n},u_{0:n-1}]||_{tv}\ ] ] @xmath475\ ] ] b. for any non - negative integers @xmath476 such that @xmath477 @xmath478 @xmath479 where @xmath371 is the initial distribution of @xmath471 . a. earlier we had @xmath480 and the first inequality now follows from the lemma @xmath438 part @xmath481 . the factor `` 2 '' follows from using the triangle inequality on the difference of two probability measures . b. this is just like part @xmath482 except we consider different initial distributions for @xmath483 . it is easily seen that @xmath493 is minimized as a function of @xmath4 at @xmath494 and thus @xmath495 assuming we can exchange derivatives with integration . the last equation is called fisher s identity . and @xmath499\ ] ] @xmath500 + e_{\theta'}[\log g(x_0,y_0;\theta)|y_{0:n},u_{0:n-1}]\ ] ] @xmath501\ ] ] we set @xmath502 and get @xmath503 @xmath504 + e_{\theta}[\frac{\partial}{\partial \theta}\log g(x_0,y_0;\theta)|y_{0:n},u_{0:n-1}]\ ] ] @xmath505\ ] ] we now wish to use the expression for @xmath508 derived in the last section . we have that @xmath509 but also @xmath510 this gives an alternative expression of @xmath511 . we get @xmath512 and for @xmath513 @xmath514 @xmath515\ ] ] @xmath516\ ] ] this expression can be generalized to starting the process at other values than zero ; @xmath517\ ] ] @xmath518\ ] ] @xmath519\ ] ] this is done in cappe et al . @xcite to extend the process to minus infinity ( @xmath520 ) . we do nt extend the process to infinity , but rather think of @xmath39 as indicating lack of information , that is assuming that the process starts at @xmath521 . assuming strong mixing conditions . then for @xmath513 cappe et al . @xcite prove the following inquality in the hmm case : @xmath522\right)^{1/2 } \frac{\rho^{(k+m)/2 - 1}}{1-\rho}\ ] ] we do nt extend the process to @xmath523 , but rather starting at @xmath471 and we prove the following inequality , also for @xmath513 @xmath524 where @xmath525 ( see theorem [ lemma 4.3.22 ] ) . [ lemma 12.5.3 ] just like in the proof of lemma @xmath263 in @xcite we match together different pairs of terms within the sums , depending on their index @xmath119 . more specifically for @xmath530 we match together the terms where @xmath530 in @xmath333 and @xmath331 . for @xmath531 we match the terms in @xmath333 with @xmath331 and the terms in @xmath329 with those in @xmath532 . for @xmath533 we match terms in @xmath333 with terms in @xmath329 and terms in @xmath331 with those in @xmath532 . that leaves @xmath534 in @xmath535 where we match @xmath333 and @xmath329 . if we look at the case where @xmath333 is matched with @xmath331 we have @xmath536 -e[\phi_{\theta}(x_{i-1},x_i , y_i , u_{i-1})|y_{1:k},u_{0:k-1}]||\ ] ] @xmath537 @xmath538|\ ] ] @xmath539 where @xmath540 $ ] is the forward smoothing kernel , and the inequality stems from proposition @xmath541 @xmath482 where the second line can be thought of as two different initial distributions for @xmath521 , and the kernel @xmath542 is bounded by @xmath101 . matching @xmath329 with @xmath532 is similar . for matching @xmath333 with @xmath329 and @xmath331 with @xmath532 we need a `` backwards bound '' ; @xmath543 @xmath544 that is established below , see theorem [ prop 12.5.4 ] . for matching @xmath331 with @xmath532 we get @xmath545\ ] ] @xmath546||\ ] ] @xmath547 @xmath548|\ ] ] @xmath549 where @xmath550 is the backwards smoothing kernel described below . matching @xmath333 with @xmath329 is a special case of the above . going back to our orginal objective , we have @xmath551 where @xmath552 is a sum over the pairs we considered above . now by minkowski s inequality we have @xmath553 now we have that @xmath554 where @xmath555 is the power of @xmath70 associated with @xmath556 . @xmath557 at this point cappe et al . @xcite argue that since in their case the process was started at inifinity and the process is homogeneous the expected value over @xmath558 is always the same by stationarity , and @xmath558 can be exchanged by @xmath559 . since argueing for stationarity is more of stretch for us , we also take the supremum over @xmath257 and remember that that set is also finite . @xmath560 @xmath561 we now deal with the sum of @xmath70 to different powers . from @xmath530 we have @xmath562 where we matched @xmath333 with @xmath331 . for @xmath563 we have @xmath564 where we matched @xmath333 with @xmath331 and @xmath329 with @xmath532 . for @xmath565 we have @xmath566 from matching @xmath333 with @xmath329 and @xmath331 with @xmath532 . finally for @xmath567 we have @xmath568 from matching @xmath333 with @xmath329 . this gives @xmath569 @xmath570 @xmath571 thus , finally we have @xmath572 the idea behind this proof is to replicate all the results derived so far for the backward smoothing kernel . that is , conditional on @xmath575 , @xmath576 and @xmath577 the time - reversed process @xmath403 is a non - homogeneous markov chain , where the conditional probability of moving from @xmath578 to @xmath483 given all the observations @xmath579 , controls @xmath580 and initial condition ends up only depending on @xmath581 , @xmath582 and the initial condition , and is governed by the backwards smoothing kernel given by @xmath583(x , f)\ ] ] @xmath584 just as we did in lemma @xmath438 we can show @xmath585\ ] ] @xmath586(x_j,\cdot \,)\ ] ] @xmath587\ ] ] where @xmath588(f)\ ] ] @xmath589 as we showed there this gives @xmath590 we now get that the 2 smoothers we are interested in can be thought of as smoothers of the reversed markov chain from @xmath591 to @xmath39 with 2 different initial distributions for @xmath592 , the starting position . we get @xmath593 @xmath594 @xmath595 @xmath596 ( where @xmath597 )
we consider the effects of impurities on topological insulators and superconductors . we start by identifying the general conditions under which the eigenenergies of an arbitrary hamiltonian belonging to one of the altland - zirnbauer symmetry classes undergo a robust zero energy crossing as a function of an external parameter which can be , for example , the impurity strength . we define a generalized root of , and use it to predict or rule out robust zero - energy crossings in all symmetry classes . we complement this result with an analysis based on almost degenerate perturbation theory , which allows a derivation of the asymptotic low - energy behavior of the ensemble averaged density of states for all symmetry classes , and makes it transparent that the exponent does not depend on the choice of the random matrix ensemble . finally , we show that a lattice of impurities can drive a topologically trivial system into a nontrivial phase , and in particular we demonstrate that impurity bands carrying extremely large chern numbers can appear in different symmetry classes of two - dimensional topological insulators and superconductors . we use the generalized root of to reveal a spiderweblike momentum space structure of the energy gap closings that separate the topologically distinct phases in superconductors in the presence of an impurity lattice .
1510.05909
ever since the concept of topological order in condensed matter systems picked up momentum during the last decade,@xcite there has been the question as to which kind of perturbations can cause topological phase transitions . in this paper , we investigate the possibility to utilize inhomogeneous perturbations , like those caused by impurity lattices , for this purpose . we point out that such perturbations are often naturally present in candidate materials for topological superconductors , since these are typically unconventional superconductors obtained by doping mott insulators.@xcite moreover , high quality topological insulators have recently been created by intentionally doping si into inas / gasb heterostructures.@xcite finally , a lattice of magnetic atoms placed on top of a superconductor may be a route towards the realization of a rich variety of topological phases.@xcite in principle , there is also a relation to the studies about so - called topological anderson insulators,@xcite disorder - driven topological superconductivity,@xcite and impurity bound states as a signature of a topologically nontrivial phase.@xcite to intuitively understand why impurity lattices may be of interest for the design of topologically nontrivial phases , a useful starting point is to realize that nontrivial topological invariants are associated with momentum space topological defects , @xcite which can be inserted into the system when the bulk energy gap closes . the impurities reduce the translational symmetry of the system and give rise to impurity bands , which can have complicated energy - momentum dispersion relations described by long - range hopping terms in the low - energy theory . this allows more freedom to deform the dispersion in such a way that many band inversions appear , giving rise to topological defects in momentum space . as a result these impurity bands can carry very large topological invariants . recently , rntynen and ojanen theoretically demonstrated the possibility of topological superconductivity with very large chern numbers in 2d ferromagnetic shiba lattices using this kind of approach . @xcite in the first part of this paper ( secs . [ sec : roots ] and [ sec : dos ] ) , we investigate whether the eigenstates of an arbitrary hamiltonian @xmath6 belonging to one of the ten altland - zirnbauer symmetry classes@xcite undergo robust zero - energy crossings as a function of @xmath7 . in this regard , there are two main results : ( i ) we define the generalized root @xmath8 of @xmath1 in such a way that @xmath9 , and @xmath8 is a polynomial function of the matrix elements of the hamiltonian . here , @xmath10 is the symmetry - induced degeneracy of eigenstates , when eigenstates at @xmath11 are considered degenerate . we show that @xmath8 is real in classes a , ai , aii , bdi , and d , it is complex in classes aiii , ci , and diii , and it can not be defined in classes c and cii . when @xmath12 is real , @xmath13 equals the parity of the topological invariant in @xmath14 dimensions.@xcite in the presence of an additional symmetry @xmath15=0 $ ] , @xmath16 , where @xmath17 is the chiral symmetry operator , @xmath12 becomes real also in classes aiii , ci , and diii.@xcite ( ii ) we use @xmath18 to predict [ @xmath18 is real ] or to rule out [ @xmath18 is complex ] robust zero - energy crossings in the respective symmetry classes . as a side product of this analysis we also derive the important result that if there are two parameters @xmath19 and @xmath20 , robust zero energy crossings can exist in the @xmath21-parameter plane also when @xmath22 is complex . we complement our exact algebraic formalism with almost degenerate perturbation theory@xcite applied to an eigenstate @xmath23 of @xmath24 and its symmetry partner eigenstates ( spes ) @xmath25 , @xmath26 , and @xmath27 , where @xmath28 , @xmath29 , and @xmath17 are the operators of time - reversal ( tr ) , particle - hole ( ph ) , and chiral symmetry , respectively . using this approach in combination with a scaling argument@xcite , we confirm the predicted universality of the exponent @xmath3@xcite characterizing the asymptotic behavior of the ensemble averaged density of states @xmath2 in the limit @xmath30 for each of the ten symmetry classes , respectively . this approach makes transparent why the exponent @xmath3 does not depend on the choice of the random matrix ensemble . in the second part of this paper ( sec . [ impurity - driven - trans ] ) , we study impurity - driven topological phase transitions . we discuss the relation between the zero - energy crossings in @xmath14 and topological phase transitions in @xmath31 for systems where the topological invariant @xmath32 is changed by band inversion at the high - symmetry points @xmath33 of the brillouin zone ( @xmath34 is a reciprocal lattice vector ) . our approach is to first consider a generic translationally invariant hamiltonian @xmath24 and a local perturbation @xmath35 causing a zero - energy crossing at @xmath36 . the robust zero - energy crossings are then utilized to provide an estimate for the impurity strengths necessary to drive topological phase transitions in the presence of an impurity lattice . for this purpose , we analyze @xmath37 , where @xmath38 and @xmath39 describes the impurity lattice formed by periodically continuing @xmath40 with a lattice constant @xmath41 . for each @xmath42 , a critical value @xmath43 exists , and they all coincide @xmath44 in the limit @xmath45 , where @xmath46 is the decay length of the zero - energy eigenstate wave function of @xmath47 . however , at finite @xmath41 , the @xmath43 being unequal give rise to a range of values @xmath7 where @xmath48 is nontrivial even though @xmath24 is trivial . we illustrate this general behavior analytically in the context of the 1d kitaev model@xcite ( equivalently the ssh model@xcite ) . the closing of the energy gap can also occur away from the high - symmetry points of the brillouin zone . in this case , one can consider @xmath49 as a parameter and use the zero dimensional @xmath50 to analyze the phase transitions . the main difference is that the zero - dimensional hamiltonian @xmath51 , where @xmath49 is considered to be a fixed parameter , usually belongs to a different symmetry class than the full higher dimensional hamiltonian containing all the momentum blocks . we have studied the impurity lattices in several two - dimensional models , which show this kind of gap closings away from the high - symmetry points . from these closings arises a rich variety of topological phases . in particular , we demonstrate that similarly as in the case of ferromagnetic shiba lattices,@xcite it is possible to obtain large chern numbers also in @xmath52 superconductors , but in this case one only needs nonmagnetic impurities instead of magnetic atoms . furthermore , we use @xmath50 to reveal a spiderweblike momentum space structure of the energy gap closings that separate the topologically distinct phases in @xmath5 superconductors in the presence of an impurity lattice . both the shiba lattice@xcite and the @xmath5 superconductor models belong to class d in the classification table for topological insulators and superconductors.@xcite however , the idea of using an impurity lattice to design topologically nontrivial phases with large chern numbers is more general . for this purpose we demonstrate that an impurity lattice can give rise to large chern numbers also in chern insulators belonging to class a in the classification table . [ cols="<,>,>,^,^,^,>,^,^,>,^,^",options="header " , ] there are three types of symmetries chiral , tr , and ph symmetry depending on the presence or absence of which a given hamiltonian matrix @xmath0 is said to belong to one of ten symmetry classes:@xcite [ eq : symmetries ] @xmath53 & = 0 , & t^2 & = \epsilon_t \label{eq : trs}\\ \mathrm{ph : } & & \{h , p\ } & = 0 , & p^2 & = \epsilon_p \label{eq : phs } \end{aligned}\ ] ] where @xmath54 , @xmath55 , @xmath56 is the operator of complex conjugation , and @xmath17 , @xmath57 , and @xmath58 are unitary matrices . tr and ph symmetry are characterized by the sign @xmath59 and @xmath60 , respectively . for a list of symmetry classes ( determined by the absence `` @xmath61 '' or the presence of various symmetries with @xmath59 , @xmath60 ) and an overview of results , see table [ tab : results ] . in the presence of a symmetry @xmath17 , @xmath28 , or @xmath29 , one finds for each eigenstate @xmath23 of @xmath0 so - called spes @xmath27 , @xmath25 , or @xmath26 , respectively . equations imply that @xmath25 ( @xmath27 and @xmath26 ) is an eigenstate of @xmath0 with the same ( with opposite ) energy as @xmath23 . however , tr symmetry with @xmath62 does not guarantee the existence of an orthogonal symmetry partner eigenstate ( spe ) , because the kramers theorem applies only to the case @xmath63 . therefore , in the present context such a symmetry should rather be understood as a constraint on @xmath0 . the number of spes @xmath64 for all ten symmetry classes are shown in table [ tab : results ] . the determinant of a matrix equals the product of its eigenvalues @xmath65 . hence , @xmath1 allows us to analyze zero - energy crossings and @xmath14 topological transitions . we show that in all symmetry classes , except for c and cii , one can explicitly define a function @xmath8 with two key properties : * @xmath8 is a ( complex valued ) polynomial expression in the matrix elements @xmath66 of the hamiltonian . * @xmath8 satisfies @xmath67 , where @xmath10 is the number of spes in the respective symmetry class . property ( ii ) defines in which sense it is justified to call @xmath8 the ( generalized ) @xmath10th root of @xmath1 . however , property ( i ) constitutes the important difference between @xmath8 and @xmath68{\det h}$ ] . most importantly , @xmath8 shares with @xmath1 the property that it vanishes if and only if the hamiltonian has a zero eigenvalue . before defining @xmath12 in sec . [ sec : definingq ] , we now explain its relation to the @xmath14 topological invariants . its relation to zero - energy crossings is detailed in sec . [ sec : qzeroenergycrossings ] . from @xmath67 and the possibility to calculate @xmath69 as the product of the hamiltonian s eigenvalues follows @xmath70 when @xmath12 is real . here , the prime indicates that only one member of the set of @xmath71 spes contributes a factor @xmath72 . in principle , @xmath12 is only defined up to a sign , but once the sign is chosen , it remains well defined when changing @xmath0 adiabatically . in consequence , we find that in those symmetry classes where the @xmath14 invariant @xmath32 exists , @xmath12 changes sign when @xmath32 changes by @xmath73 . on the other hand , in the presence of an additional symmetry @xmath74 in classes aiii , ci , and diii , @xmath0 is block diagonal with blocks @xmath75 ( @xmath76 ) , and if @xmath16 an invariant @xmath77 exists for each block ; cf . appendix [ sec : implications ] . then , @xmath12 changes sign when one of the @xmath78 changes by @xmath73 . formally , this corresponds to @xmath79 where x+@xmath74 means `` for hamiltonians from symmetry class x with the additional symmetry @xmath74 , @xmath16 '' . for examples of unitary symmetries in the context of topological order see refs . . [ [ classes - a - and - ai . ] ] classes a and ai . + + + + + + + + + + + + + + + + + since spes are absent , we can define @xmath80 . [ [ classes - bdi - and - d . ] ] classes bdi and d. + + + + + + + + + + + + + + + + + + hamiltonians in these symmetry classes obey ph symmetry with @xmath81 . in accordance with refs . , we define the real polynomial @xmath82 , where @xmath0 is a @xmath83 matrix . using @xmath84 and eq . it follows that @xmath85 and @xmath86 . thus @xmath87 is well defined . from the general property of the pfaffian @xmath88 , we then infer @xmath89 . [ [ class - aii . ] ] class aii . + + + + + + + + + + hamiltonians in aii obey tr symmetry with @xmath90 . we proceed analogously to the derivation of @xmath8 for classes bdi and d and thus define the real polynomial @xmath91 . [ [ classes - aiii - and - ci . ] ] classes aiii and ci . + + + + + + + + + + + + + + + + + + + + hamiltonians in these classes have a chiral symmetry . we focus on the case where @xmath92.@xcite this allows us to choose a basis such that @xmath93 since @xmath94 we can straightforwardly define the complex polynomial @xmath95 . chiral symmetry alone does not put any restriction on @xmath96 . albeit tr symmetry restricts @xmath96 in class ci , the phase of @xmath97 does depend on the parameters of the hamiltonian in both classes , ci and aiii . in the following , we show that the phase of @xmath97 becomes independent of the hamiltonian s parameters if there is a symmetry @xmath74 satisfying @xmath15 = \{c , u\ } = 0 $ ] . in the basis of eq . , @xmath98 implies @xmath99 with unitary matrices @xmath100 and @xmath101 . the symmetry @xmath15=0 $ ] implies @xmath102 . this can be used to show @xmath103 , which means that the complex phase of @xmath8 is independent of @xmath0 in the presence of the additional symmetry @xmath74 . hence , we define the real polynomial @xmath104 . [ [ class - diii . ] ] class diii . + + + + + + + + + + + a polynomial expression for the fourth root of @xmath1 is a central result of ref . . for completeness , we briefly summarize the derivation here . due to chiral symmetry , @xmath0 is of off - diagonal form , cf . tr symmetry with @xmath90 renders @xmath105 antisymmetric . this enables the definition @xmath106 . also in class diii , the complex phase of @xmath8 loses its dependence on the hamiltonian s parameters in the presence of a symmetry @xmath15 = \{c , u\ } = 0 $ ] . then , the real polynomial @xmath107 can be defined , where @xmath108 is a block of @xmath109 in the basis where @xmath17 is diagonal , cf . eq . . [ [ classes - c - and - cii . ] ] classes c and cii . + + + + + + + + + + + + + + + + + + in these classes , it is not possible to define @xmath8 as a polynomial in the matrix elements of @xmath0 . for a justification , see appendix [ sec : classescandcii ] . illustration of the real and imaginary parts of @xmath18 as a function of an external parameter @xmath7 . ( a ) in classes aiii , ci , and diii , @xmath18 takes complex values and generically does not become zero , independently of whether a parameter is considered on which @xmath12 depends linearly ( solid line ) or with some higher power ( dashed line ) . ( b ) in classes a , ai , aii , bdi , and d [ aiii , ci , and diii with additional symmetry @xmath74 which anticommutes with @xmath17 ] @xmath18 [ @xmath111 is always real . in these cases the change of sign of @xmath18 as a function of @xmath7 guarantees the existence of robust zero - energy crossing at some value @xmath112 . ] since @xmath9 , the zeros of @xmath8 coincide with the zeros of @xmath1 and the existence of zero - energy eigenstates of @xmath0 . for predicting zero - energy crossings , @xmath8 is more useful than @xmath1 because ( i ) it is a complex number in classes aiii , ci , and diii and thus has a phase degree of freedom and ( ii ) it accounts for symmetries and thus is a polynomial of lower degree than @xmath1 . in sec . [ sec : dos ] we supplement the following abstract discussion with additional more intuitive arguments . in ref . , the possibility to predict zero - energy crossings using @xmath8 was illustrated for a hamiltonian @xmath6 , where @xmath24 is from diii and @xmath35 is an on - site potential with impurity strength @xmath7 . while @xmath113 is a real fourth order polynomial in @xmath7 , @xmath18 is a complex first order polynomial . consequently , @xmath18 as a function of @xmath7 takes values on a straight line in the complex plane which generically does not contain the origin , so that zero - energy eigenstates will not occur for any @xmath7 ; cf . solid line in fig . [ fig : qh](a ) . more generally , @xmath8 describes a continuous curve in the complex plane under arbitrary continuous parameter changes of @xmath0 and will not go through the origin , so that it will not give rise to zero - energy states without fine tuning ; cf . dashed line in fig . [ fig : qh](a ) . as a result , in classes aiii , ci , and diii , where @xmath114 , changing a single parameter in @xmath0 will not induce robust zero - energy crossings . the critical impurity strength @xmath115 of an on - site potential impurity @xmath116 in a @xmath52 superconductor depends on the chemical potential @xmath117 . hence , zero - energy crossings appear not only by tuning @xmath7 , but also by tuning @xmath117 . ] illustration of the real and imaginary parts of @xmath22 as a function of an external parameter @xmath19 for two fixed values of another external parameter @xmath118 and @xmath119 in classes where @xmath8 takes complex values . if the two curves are on opposite sides of the origin as illustrated in the figure and assuming that for @xmath120 $ ] @xmath121 ( or @xmath122 ) is a continuously increasing / decreasing function of @xmath7 with @xmath123<0 $ ] and @xmath124>0 $ ] ( @xmath125<0 $ ] and @xmath126>0 $ ] ) , these curves can not be smoothly deformed into each other except if @xmath127 at a particular point @xmath128 and @xmath129 . therefore there exists a robust zero - energy crossing within the @xmath130-parameter plane . ] consider now the case where @xmath8 is real . focusing on @xmath6 , where @xmath40 is an arbitrary local perturbation , we find @xmath131 , where @xmath132 and @xmath133 depends on @xmath40 . clearly , when @xmath133 is odd , @xmath18 crosses zero at least once upon tuning @xmath7 from @xmath134 to @xmath135 [ fig . [ fig : qh](b ) ] . for even @xmath133 , an even number of zero - energy crossings will occur , including the possibility that @xmath12 does not cross zero . some remarks are in order . ( i ) although the zero - energy crossings are most easily analyzed as a function of @xmath7 , they appear also as a function of @xmath24 bulk parameters like the chemical potential . this is illustrated in fig . [ fig : ucinfinite ] , where we plot the critical impurity strength @xmath136 corresponding to a zero - energy crossing for an impurity @xmath116 , where @xmath137 annihilates a spinless electron on site @xmath138 , in an infinite @xmath52 superconductor . the unperturbed hamiltonian in momentum space reads @xmath139 with @xmath140 here @xmath141 and @xmath142 parametrize the nearest neighbor hopping and superconducting pairing , respectively , and @xmath117 is the chemical potential . @xmath115 was determined by performing a @xmath28-matrix calculation.@xcite ( ii ) in classes aiii , ci , and diii the presence of a symmetry @xmath74 is not a necessary but a sufficient condition to render the phase of @xmath8 independent of @xmath0.@xcite in fact , contrary to the set of equations @xmath15=0 $ ] , the necessary and sufficient condition @xmath143 is just a single equation . ( iii ) the argument for the existence of zero - energy crossings when @xmath12 is real can be generalized to interacting models where @xmath24 needs to be calculated self - consistently when the external parameter @xmath7 is varied . assuming that the coefficients @xmath144 become continuous functions of @xmath7 and there is no spontaneous symmetry breaking , the change of sign of @xmath18 still guarantees the existence of a robust zero - energy crossing . ( iv ) if @xmath145 it can still be used to predict robust zero - energy crossings . however , in this case one needs at least two external parameters @xmath130 , and the robust crossings can only be guaranteed to occur at particular critical points @xmath146 within a parameter plane . namely , consider the real and imaginary parts of @xmath22 as a function of an external parameter @xmath19 for two fixed values of another external parameter @xmath118 and @xmath119 . if the two curves are on opposite sides of the origin as illustrated in fig . [ fig : crossinginplane ] , under certain assumptions they can not be smoothly deformed into each other unless @xmath127 at a particular point @xmath128 and @xmath129 . therefore this kind of situation typically leads to the existence of a robust zero - energy crossing at a specific point within the @xmath130-parameter plane . there are relations between @xmath8 and the allowed couplings of spes as well as the density of states in random matrix ensembles . these relations are the subject of this section . we focus on the eigenstate @xmath23 of a hamiltonian @xmath24 which , together with its spes , is the one with energy closest to zero . subsequently , we study how the energy corresponding to this set of spes evolves as the strength of a perturbation @xmath40 from the same symmetry class as @xmath24 is tuned . therefore , we employ almost degenerate perturbation theory@xcite in the subspace spanned by these spes . moreover , we consider ensembles of random matrices from any of the ten symmetry classes . then , by adopting a scaling argument from ref . , we use our perturbative results to derive the universal asymptotic behavior of the respective ensemble averaged density of states in the zero - energy limit . as an immediate consequence of the defining symmetry properties eqs . , the matrix elements describing the coupling of spes are restricted as follows : [ eq : symmetryconstraints ] @xmath147 the possible couplings , restricted by these conditions , are depicted in fig . [ fig : spe ] for all ten symmetry classes . assuming that the energy @xmath148 of @xmath23 is sufficiently close to zero , the effect of a generic perturbation @xmath40 from the same symmetry class as @xmath24 on @xmath23 and its spes can be calculated via almost degenerate perturbation theory.@xcite the energies of the set of almost degenerate states in the presence of the perturbation is obtained from the eigenvalues of @xmath149 , where @xmath150 projects onto the subspace spanned by @xmath23 and its spes . as an example we consider symmetry class aiii , where the matrix elements for the projected perturbation read @xmath151 , and @xmath152 ; @xmath153 . hence , we find in the basis @xmath154 @xmath155 because @xmath23 couples to its chiral partner , one generically expects an avoided level crossing upon tuning the perturbation . exploiting the symmetry constraints [ eqs . ] , one can derive analogous results for all ten symmetry classes . the corresponding energies are always of the form @xmath156 with @xmath157 or @xmath158 depending on the symmetry class considered . the corresponding values of @xmath3 for all ten symmetry classes are shown in table [ tab : results ] , and as we show below @xmath3 is the universal exponent determining the low - energy asymptotic behavior of the ensemble averaged density of states . @xmath159 is to be understood as @xmath160 . @xmath161 is twice degenerate in classes aii , diii , and cii . in those classes with a ph symmetric spectrum one also gets @xmath162 . we now consider arbitrary random matrix ensembles from any of the ten symmetry classes . our perturbative results can then be used to calculate the universal asymptotic behavior of the respective ensemble averaged density of states @xmath163 in the limit @xmath164.@xcite to this end , we consider the sum @xmath165 to be a realization of a random matrix from the respective ensemble . the @xmath166 are then random variables with the corresponding probability distribution @xmath167 , @xmath168 , which we do not have to know in detail . @xmath148 is not independent of @xmath169 and thus has been absorbed in @xmath169 . the ensemble averaged density of states reads @xmath170 upon rescaling @xmath171 , we obtain @xmath172 . the symmetry constraints for each symmetry class were already taken into account in the hamiltonian @xmath173 and @xmath3 , respectively , so that generically @xmath174 will remain finite in the limit @xmath175 . hence , for energies less than the average level spacing @xmath176 , we obtain the asymptotic behavior @xmath177 the exponents @xmath3 that result from this argument are listed in table [ tab : results ] and agree with those stated in refs . . @xmath3 relates to @xmath8 in the following way : @xmath178 in classes c and cii where @xmath12 can not be defined . @xmath179 in classes aiii , ci , and diii where @xmath180 . @xmath159 when @xmath12 is real . finally , when a matrix ensemble with chiral symmetry is restricted to a subset of matrices @xmath0 satisfying @xmath15 = \{c , u\ } = 0 $ ] , the matrix element @xmath181 vanishes and @xmath3 is reduced to @xmath182 ; cf . table . [ tab : results ] . in this section we consider the topological order of a translationally invariant hamiltonian @xmath183 , where @xmath24 is gapped and @xmath39 describes a lattice of impurities with period @xmath41 determining the overall translational symmetry of the system . a general translationally invariant hamiltonian in dimension @xmath31 is block diagonal in momentum space , @xmath184.@xcite each hermitian block @xmath185 describes the physics of a @xmath14 system . in particular , there are @xmath186 high symmetry points @xmath42 in the brillouin zone , which satisfy @xmath187 with @xmath34 a reciprocal lattice vector , and each @xmath14 hamiltonian @xmath188 belongs to the same symmetry class as @xmath0 itself . hence , we can calculate @xmath189 [ using the definition of @xmath8 for the symmetry class of @xmath0 itself ] to determine zero - energy crossings of @xmath188 . in the case of the impurity lattice , the blocks in momentum space become @xmath190 . finding the zeros @xmath43 of @xmath37 yields information about band inversions at @xmath42 which often affect the topological invariants of the system . let us consider the case that @xmath47 has a zero - energy eigenstate , where @xmath40 is a single impurity of the set of impurities forming @xmath39 . its wave function decays on a length scale @xmath46 because @xmath35 is a local perturbation and @xmath24 is gapped . consider now @xmath191 in the limit @xmath192 . then , the boundaries of the unit cell have a vanishing effect such that @xmath193 independent of @xmath42 . however , at finite @xmath41 the @xmath43 being unequal give rise to a range of values @xmath7 where @xmath48 is nontrivial even though @xmath24 is trivial . in systems with sufficiently dilute impurity lattice ( compared to @xmath136 ) , impurity strengths of the order of @xmath115 are nevertheless needed to drive a topological phase transition . below , we employ the 1d kitaev model to analytically illustrate this general behavior . the closing of the energy gap can also occur away from the high - symmetry points of the brillouin zone . in general , this can happen at both @xmath49 points distinguished by unitary symmetries of the hamiltonian@xcite as well as arbitrary @xmath49 points , cf . sec . [ px - pysc ] . then , by considering @xmath49 as a parameter , one can still use the zero - dimensional @xmath50 to analyze the phase transitions . the main difference is that the zero - dimensional hamiltonian @xmath51 , where @xmath49 is considered as a fixed parameter , usually belongs to a different symmetry class than the full higher dimensional hamiltonian containing all the momentum blocks . we demonstrate that similar to the case of ferromagnetic shiba lattices,@xcite it is possible to obtain large chern numbers in @xmath52 superconductors , but in this case one only needs nonmagnetic impurities instead of the magnetic atoms . both of these models belong to symmetry class d. moreover , we demonstrate that an impurity lattice can give rise to large topological invariants in chern insulators belonging to symmetry class a. the general ideas presented so far can nicely be illustrated in the context of the kitaev model@xcite describing a superconductor of spinless fermions in @xmath194 @xmath195 to simplify the analysis we focus on @xmath196 . the hamiltonian @xmath24 belongs to class bdi with artificial tr symmetry @xmath197 in addition to the intrinsic ph symmetry of bogoliubov de gennes hamiltonians . for @xmath198 the system is in a trivial phase , and for @xmath199 the system supports majorana end modes . thus , there is a topological phase transition at @xmath200 . phase diagrams of the kitaev chain in the presence of impurity lattices as a function of impurity strength @xmath7 and @xmath201 for @xmath202 and different lattice constants @xmath41 . gray ( white ) areas denote topologically nontrivial ( trivial ) phases in case of impurity lattice constant @xmath203 . solid ( dashed ) lines are @xmath204 [ @xmath205 . solid red line is @xmath136 . ] we now consider the effect of an impurity lattice @xmath206 on the topology assuming that we start from inside the topologically trivial phase @xmath207 . here , @xmath41 is the lattice period . band inversions at the high symmetry @xmath208 points @xmath209 change the parity of the topological invariant.@xcite by determining the zeros of @xmath210 , we find that the critical impurity strengths are ( see appendix [ sec : kitaevchaincalc ] ) @xmath211 with a wave function decaying on the scale @xmath212 . when @xmath192 , we get @xmath213 in agreement with eq . . the phase diagrams for different values of @xmath41 are shown in fig . [ fig : phasediagramkitaev ] . for all values of @xmath41 , it is possible to turn the trivial phase to a topologically nontrivial phase with the help of impurity lattice . however for @xmath192 the trivial phase becomes nontrivial only in a very small region of parameters around @xmath214 . illustration of the dimensional crossover in the kitaev chain . a single impurity introduces a single subgap state ( first panel ) . upon increasing the density of impurities ( decreasing @xmath215 ) the wave functions for the impurity states start to overlap and one forms an impurity band the topology of this band changes and protected end states form , when the impurity concentration exceeds its critical value [ @xmath216 ; cf . @xmath217 is the wave function of the eigenstate with smallest positive energy @xmath218 . black ( red ) curves correspond to the particle ( hole ) component of @xmath217 . parameters : @xmath219 , @xmath220 ; cf . fig . [ fig : phasediagramkitaev ] . ] as illustrated in fig . [ fig : crossover ] , our results allow the following appealing interpretation : a single local impurity can cause a @xmath14 topological transition . taking several @xmath14 systems with an impurity of strength @xmath221 and coupling them together via hopping ( and pairing ) terms yields a system of dimensionality @xmath31 with good chances to be topologically different from the system one would obtain without impurities . in this section , we consider the two - dimensional @xmath52 superconductor described by the hamiltonian @xmath24 given by eq . . since @xmath24 belongs to class d , there exists a @xmath223-topological invariant , namely the sum @xmath17 of all chern numbers of the occupied bands . the topologically distinct phases of @xmath24 are @xmath224 for @xmath225 , @xmath226 for @xmath227 , and @xmath228 for @xmath229 . we will consider this type of superconductor in the presence of impurities described by the hamiltonian @xmath116 arranged on a square lattice with period @xmath41 to form a periodic perturbation @xmath230 . we demonstrate that the impurity lattice allows us to access extremely large chern numbers similarly as in the case of ferromagnetic shiba lattices.@xcite fig . [ fig : phasediagrampx ] shows the phase diagram of the hamiltonian @xmath231 as a function of @xmath7 and @xmath117 for @xmath232 and @xmath233.@xcite there exists a rich variety of phases , which can be accessed by tuning these parameters . the topological phase transitions separating these phases come in two variants . first , there are transitions accompanied by a gap closing at the high symmetry points where @xmath234 and @xmath235 . the corresponding phase transition lines @xmath43 were obtained from determining the zeros of @xmath37 . at these lines , the chern number changes by @xmath73 . notice that the gap closings at @xmath236 and @xmath237 coincide due to rotational symmetry . secondly , there are transitions where the gap closes at four momenta away from the high symmetry points . in these cases , the chern number changes by @xmath238 . moreover , the transition lines corresponding to changes of the chern number by @xmath239 always connect the thick lines corresponding to gap closings at high symmetry points ( see fig . [ fig : phasediagrampx ] ) . phase diagram of the @xmath52 superconductor in the presence of an impurity lattice as a function of impurity strength @xmath7 and chemical potential @xmath117 for @xmath240 and @xmath233 . thick lines are gap closings at the high symmetry points [ solid : @xmath241 , dashed : @xmath236 and @xmath237 , dash - dot : @xmath242 . at thin lines , the gap closes away from the high - symmetry points . numbers denote the sum of chern numbers of occupied bands in the respective phase . the thin red line is @xmath136 . the positions of the gap closings in the momentum space for the thin lines denoted as ( a)-(c ) and ( d)-(f ) are illustrated in fig . [ gapclosingsmomentumspace ] . ] chern curvature of the impurity band for different points in the phase diagram in fig . [ fig : phasediagrampx ] : ( a ) @xmath243 , @xmath244 , ( b ) @xmath245 , @xmath246 , ( c ) @xmath247 , @xmath248 , ( d ) @xmath249 , @xmath250 , ( e ) @xmath251 , @xmath246 and ( f ) @xmath245 , @xmath248 . the gap closings associated with topological phase transitions give rise to momentum space topological defects , which show up as an enhanced chern curvature in the vicinity of the gap closing point in the momentum space . ] as a function of @xmath7 for @xmath252 , @xmath240 , @xmath253 , and @xmath254 $ ] . the lines corresponding to @xmath245 and @xmath255 are on different sides of the origin , so that the hamiltonian realizes the situation illustrated in fig . [ fig : crossinginplane ] . namely , by tuning @xmath254 $ ] one smoothly deforms the curve so that there exists a particular point @xmath256 and @xmath112 where @xmath257 . this leads to a gap closing at @xmath258 and @xmath259 explaining one point in the topological phase - transition line denoted ( d)-(f ) in fig . [ fig : phasediagrampx ] . the whole phase - transition line is obtained by repeating this analysis for all @xmath260 , where @xmath261 $ ] . the unitary transformation @xmath262 block off - diagonalizing the hamiltonian is not unique . however , @xmath50 is unique up to an overall phase , and therefore the different choices of @xmath262 lead to the same figure up to the rotations of the @xmath121 and @xmath122 axis . ] the hamiltonian obeys reflection symmetries with respect to the @xmath263 axis , the @xmath264 axis , and the two diagonals between these axes . as a consequence , @xmath0 is also fourfold rotationally symmetric . therefore , the gap closings corresponding to changes of chern number by @xmath239 always occur either ( i ) along the vertical and horizontal lines connecting @xmath241 to @xmath265 and @xmath266 or ( ii ) along the diagonal lines connecting @xmath241 to @xmath267 points in the brillouin zone . the lines denoted ( a)-(c ) and ( d)-(f ) in fig . [ fig : phasediagrampx ] correspond to these two different situations . to illustrate that this is indeed the case , we notice that the gap closings associated with topological phase transitions give rise to momentum space topological defects , so that we can visualize these topological defects ( and gap closings ) by plotting the chern curvature of the impurity band in the vicinity of these phase transition lines as shown in fig . [ fig : curvatures ] . same as fig . [ fig : phasediagrampx ] but with parameters @xmath268 and @xmath269 . ] we can understand also these gap closings occurring away from the high - symmetry points with the help of the zero dimensional @xmath50 . for this purpose we first notice that along these special lines in the brillouin zone the zero - dimensional hamiltonian @xmath51 satisfies a chiral symmetry . the corresponding chiral symmetry operators can be written as @xmath270 for @xmath271 , @xmath272 for @xmath273 , @xmath274 for @xmath275 , and @xmath276 for @xmath277 . here @xmath278 , @xmath279 , @xmath280 , and @xmath281 are matrices within the unit cell , which mirror with respect to the @xmath263 axis , @xmath264 axis , @xmath282 line , and @xmath283 line , respectively . all these lines are defined in such a way that they run through the impurity site . therefore , as explained in sec . [ sec : definingq ] , we can use the chiral symmetry to write the hamiltonian in the block off - diagonal form @xmath284 and define a complex polynomial @xmath285 . it is easy to see that @xmath50 is always linear in @xmath7 but a nonlinear function of @xmath117 . therefore , there can exist some ranges of @xmath117 , where a situation depicted in fig . [ fig : crossinginplane ] is realized . in fig . [ gapclosingsdzero ] , we plot @xmath50 as a function of @xmath7 for @xmath252 and @xmath254 $ ] . the lines corresponding to @xmath245 and @xmath255 are on different sides of the origin , and by tuning @xmath254 $ ] one smoothly deforms the curve so that there exists a particular point @xmath258 and @xmath259 where @xmath257 . these values correspond to one point in the topological phase transition line denoted ( d)-(f ) in fig . [ fig : phasediagrampx ] . the whole phase transition line is obtained by repeating this analysis for all @xmath260 , where @xmath261 $ ] . because @xmath50 is a nonlinear function of @xmath117 there can exist several ranges of @xmath117 where similar situations are realized . however , all the phase transition lines where the chern number changes by @xmath239 ( connecting the thick lines corresponding to gap closings at high symmetry points in fig . [ fig : phasediagrampx ] ) can be explained with a similar analysis of the parameter values where @xmath257 . form a structure which resembles a spiderweb . the gap closings occurring along the lines @xmath271 , @xmath273 , @xmath275 , and @xmath277 form a support structure , and gap closing lines giving rise to changes of chern number by @xmath286 form curves which begin and end at these support lines . ] finally , we find that even larger chern numbers can be obtained by increasing the superconducting coherence length @xmath287 ( by decreasing @xmath142 ) and the lattice constant @xmath41 ; see fig . [ fig : phasediagramlargec ] . this is in agreement with ref . . there , it was argued that chern numbers on the order of @xmath288 can be obtained in impurity lattices where @xmath41 is much larger than the lattice constant . in fig . [ fig : phasediagramlargec ] , there are also phase transition lines where the chern number changes by @xmath286 . these phase transition lines always connect the phase transition lines corresponding to gap closings at the special lines @xmath271 , @xmath273 , @xmath275 , and @xmath277 . the positions of the corresponding eight gap closings in the @xmath49 space obey the rotational and reflection symmetries of the system . therefore , the positions of the energy gap closings in the momentum space corresponding to the topological phase transition lines form a structure , which resembles a spiderweb ( see fig . [ fig : kpositions ] ) . the gap closings occurring at high symmetry points and along the lines @xmath271 , @xmath273 , @xmath275 , and @xmath277 ( possessing chiral symmetries ) form the support structure of this spiderweb , and gap closing points giving rise to changes of chern number by @xmath286 form curves which begin and end at these support lines . the @xmath5 superconductor , similar to the ferromagnetic shiba lattice model considered in ref . , belongs to symmetry class d. however , it is clear that the particle - hole symmetry is not necessary for the appearance of large chern numbers since small perturbations breaking this symmetry can not affect the chern number due to the presence of the energy gap . in this section , we study a closely related model , generically not supporting particle - hole symmetry , which we call bilayer chern insulator . on one hand , this model allows us to establish how the breaking of the particle - hole symmetry affects the phase diagram . on the other hand , it allows us to generalize the interesting results concerning the impurity lattices also to topological insulators belonging to symmetry class a. phase diagram of the bilayer chern insulator in the presence of an impurity lattice as a function of @xmath7 and @xmath289 for @xmath290 , @xmath291 , @xmath292 , and @xmath293 . ( a ) for all values of @xmath289 and @xmath7 the chemical potential @xmath117 is tuned to be between the two impurity bands originating from the different layers . ( b ) the chemical potential is fixed to be @xmath294 . the gray areas of the parameter space are gapless . ] we assume that the hamiltonian for a bilayer chern insulator in the basis @xmath295 is @xmath296 here @xmath297 are the electron creation operators for the different layers , @xmath298 and @xmath299 are the dispersions for the energy bands in the different layers , @xmath300 are the hopping amplitudes , @xmath117 is the chemical potential , @xmath289 is the energy gap between bands ( @xmath301 means that the bands are inverted ) , and @xmath302 describes the interlayer tunneling amplitude . we assume that the energy band in layer 1 is electronlike with positive effective mass , whereas the energy band in layer 2 is holelike with negative effective mass . the interlayer tunneling is assumed to be odd in momentum , which is naturally the case if the electronlike band is made out of @xmath303 orbitals and the holelike band from @xmath304 orbitals . thus , our model describes a single spin block of the bhz model,@xcite which may in principle be realized by coupling a quantum spin hall insulator material to a ferromagnetic insulator to remove one of the spin blocks.@xcite furthermore , some of the quantum spin hall insulator materials , such as the inas / gasb quantum wells,@xcite naturally have the bilayer structure , which will be important in the following . namely , we will consider impurity lattices , where an impurity on lattice site @xmath305 acts asymmetrically on the two layers @xmath306 . here , @xmath307 describes the asymmetry of the impurity potential , and it can be controlled , for example , by placing different impurity atoms in the two layers at position @xmath305 . in the following we consider only the case @xmath308 . this gives rise to two impurity bands originating from the two different layers , and the band inversions occurring between these impurity bands can cause topological phase transitions . similarly as earlier we arrange the impurities on a square lattice with lattice period @xmath41 . for @xmath309 and ( a ) @xmath310 ( inside @xmath311 phase ) , ( b ) @xmath312 ( transition between @xmath311 phase and gapless phase ) , ( c ) @xmath313 ( inside gapless phase ) , ( d ) @xmath314 ( transition between gapless phase and @xmath315 phase ) . the other parameters are the same as in fig . [ fig : phasediagrambilayer](b ) . ] the phase diagram of the bilayer chern insulator in the presence of an impurity lattice as a function of @xmath7 and @xmath289 is shown in fig . [ fig : phasediagrambilayer ] for @xmath290 , @xmath291 , @xmath292 , and @xmath293 . as can be seen from fig . [ fig : phasediagrambilayer](a ) , despite the absence of particle - hole symmetry a rich variety of topologically distinct phases can be realized also in this model similarly as in the superconducting model analyzed in sec . [ px - pysc ] . in fig . [ fig : phasediagrambilayer](a ) , we have tuned the chemical potential to be between the impurity bands for all values of @xmath289 and @xmath7 in order to show the overall structure of the phase diagram for a large range of @xmath289 and @xmath7 values . however , there are also important differences to the @xmath5 superconductor which become apparent when the chemical potential remains fixed as in fig . [ fig : phasediagrambilayer](b ) . namely , in the absence of particle - hole symmetry , the topologically distinct phases generically are separated in parameter space by gapless phases . we can understand this generic feature with the help of @xmath50 : in symmetry class a , there are no symmetries present and therefore the zero dimensional invariant is @xmath316 . because @xmath316 is a second order polynomial in @xmath7 , we generically expect that the topological phase transitions occur as illustrated for the transition between the @xmath317 and @xmath311 phases in fig . [ gapclosingsdzerobilayer ] . inside the fully gapped topological phases , @xmath50 has the same sign everywhere in momentum space as illustrated for the @xmath317 phase in fig . [ gapclosingsdzerobilayer](a ) . by increasing @xmath7 one can reach the transition point between the @xmath317 phase and a gapless phase , and at this transition point @xmath318 at a specific point @xmath319 in the momentum space [ fig . [ gapclosingsdzerobilayer](b ) ] . however , by further increasing @xmath7 , @xmath50 generically changes sign in a region around @xmath319 [ fig . [ gapclosingsdzerobilayer](c ) ] indicating the appearance of zero - energy states ( fermi surface ) at the boundary of this region , i.e. , a gapless phase . when @xmath7 is further increased , so that one approaches the @xmath311 phase , the fermi surface shrinks again and finally one reaches a specific value of @xmath7 , where @xmath318 only at one point in the momentum space [ fig . [ gapclosingsdzerobilayer](d ) ] . this defines the transition point between the gapless phase and the @xmath311 phase . we thus find that generically the phase diagram for bilayer chern insulator contains both gapless phases and fully gapped phases with a rich variety of chern numbers . in addition to the gapless phases separating topologically distinct phases , there are also gapless phases describing semimetal phases with negative indirect gap . the relative amount of areas with gapless and fully gapped phases in the parameter space strongly depends on @xmath307 . in summary , we have identified the general conditions under which the eigenenergies of the hamiltonian undergo a robust zero - energy crossing as a function of external parameters in different altland - zirnbauer symmetry classes . we defined @xmath8 as a generalized root of @xmath1 and used it to predict or to rule out , respectively , robust zero - energy crossings in all symmetry classes . we complemented this result with a perturbation theory analysis , which allows for a derivation of the asymptotic low - energy behavior of the ensemble averaged density of states @xmath2 for all symmetry classes . finally , we have utilized @xmath320 to show that a lattice of impurities can drive a topologically trivial system into a nontrivial phase and revealed impurity bands carrying extremely large chern numbers in different symmetry classes of two - dimensional topological insulators and superconductors . this analysis makes it transparent that @xmath320 can be used as a powerful tool in the analysis of topological phase transitions in higher dimensional systems . the impurity bands carrying large chern number @xmath17 are interesting on their own . in the bilayer chern insulator , they will support a large number ( @xmath321 ) of topologically protected fermionic edge modes allowing dissipationless transport , and in topological superconductors there will be @xmath321 majorana edge modes . however , we also want to point out that the bandwidth of the impurity bands can be quite small , and it may be tuned by varying the distance between impurities . therefore , these impurity bands carrying large chern numbers can potentially also support interesting correlated phases in the presence of interactions . in particular , the interactions may lead to spontaneous symmetry breaking , such as superconductivity , ferromagnetism , or exciton condensation , and the physical properties , such as the charge of topological defects and superfluid density , are expected to depend on @xmath17.@xcite moreover , partially filled nearly flat bands with a large chern number in the presence of interactions may lead to exotic strongly correlated states of matter with interesting topological properties.@xcite l.k . acknowledges valuable discussions with h .- this work was supported by dfg grant ro 2247/8 - 1 , by the academy of finland through its center of excellence program , and by the european research council ( grant no . 240362-heattronics ) . in this section , we derive the block diagonal structure for @xmath0 satisfying a chiral symmetry in the presence of an additional symmetry @xmath15=\{c , u\}=0 $ ] , with @xmath322 . symmetries of that kind have been considered before for class diii in refs . . the complementary case where @xmath74 and @xmath17 commute was studied in ref . . the simultaneous eigenstates of @xmath0 and @xmath74 , @xmath323 and @xmath324 satisfy @xmath325 choosing the states @xmath326 as a basis of the hilbert space and sorting them appropriately , the operators @xmath74 , @xmath0 , and @xmath17 become block diagonal : @xmath327 here , @xmath328 denotes the degree of degeneracy of the eigenvalue @xmath329 . the specific form of @xmath17 follows from eq . and the property @xmath330 . each of the hamiltonian s blocks @xmath75 lacks chiral symmetry , although the full hamiltonian still has chiral symmetry which requires @xmath331 . therefore the number of spes per block is halved from the original value @xmath332 . consequently , the blocks @xmath75 belong to a different symmetry class than @xmath0 , namely to ( i ) a when @xmath0 belongs to aiii , ( ii ) a or ai when @xmath0 belongs to ci , ( iii ) a or ai when @xmath0 belongs to bdi , ( iv ) d or aii when @xmath0 belongs to diii , and ( v ) c or aii when @xmath0 belongs to cii . so in classes aiii , ci , and diii the symmetry @xmath74 prevents level repulsion and enables the definition of an invariant @xmath333 , which is the vector of invariants of independent blocks @xmath334 . this is not necessarily the case when @xmath0 is from cii , since in this case all blocks can belong to class c. in symmetry classes c and cii , it is not possible to calculate the generalized second and fourth root of @xmath1 , respectively . to show this , it suffices to demonstrate the impossibility of defining a polynomial expression @xmath8 in the matrix elements of a hamiltonian from class c , which would satisfy @xmath335 . from this , the impossibility of defining @xmath8 for arbitrary hamiltonians in class cii follows , because a simple hamiltonian in cii can , for example , be the direct sum @xmath336 of two hamiltonians from class c which are related to each other by tr , @xmath337 with @xmath63 . the simplest hamiltonian in class c reads @xmath338.@xcite here , @xmath339 and @xmath340 is the vector of pauli matrices . considering @xmath341 it becomes manifest that any candidate for @xmath8 in class c would involve the calculation of absolute values or square roots , like @xmath342 or @xmath343 , respectively . hence , there is no expression @xmath8 in class c that meets the criterion of being a polynomial expression in the hamiltonian s matrix elements . this section contains details about the derivation of the results for the kitaev chain with impurity lattice in sec . [ sec : kitaevchain ] . we consider the translationally invariant hamiltonian @xmath38 with @xmath24 given by eq . and @xmath206 . for the kitaev chain , bringing @xmath48 into the block off - diagonal form of eq . corresponds to transforming to the majorana basis @xmath344 where @xmath345 and @xmath346.@xcite in momentum space , the off - diagonal block @xmath347 at the high symmetry points @xmath209 reads @xmath348 with matrix dimension @xmath215 . it is then straightforward to determine the critical impurity strengths [ eq . ] from the condition @xmath349 . moreover , the corresponding normalized eigenvector @xmath350 satisfying @xmath351 is @xmath352 where @xmath353 . from these eigenvectors , one obtains the decay length @xmath212 . m. z. hasan and c. l. kane , http://dx.doi.org/10.1103/revmodphys.82.3045[rev . * 82 * , 3045 ( 2010 ) ] . qi and s .- c . zhang , http://dx.doi.org/10.1103/revmodphys.83.1057[rev . phys . * 83 * , 1057 ( 2011 ) ] . a. y. kitaev , http://dx.doi.org/10.1070/1063-7869/44/10s/s29[phys.-usp . * 44 * , 131 ( 2001 ) ] . kane and e.j . mele , http://dx.doi.org/10.1103/physrevlett.95.146802[phys . lett . * 95 * , 146802 ( 2005 ) ] . l. fu , c. l. kane , and e. j. mele , http://dx.doi.org/10.1103/physrevlett.98.106803[phys . . lett . * 98 * , 106803 ( 2007 ) ] . j. e. moore and l. balents , http://dx.doi.org/10.1103/physrevb.75.121306 [ phys . b * 75 * , 121306(r ) ( 2007 ) ] . r. roy , http://dx.doi.org/10.1103/physrevb.79.195322 [ phys . b * 79 * , 195322 ( 2009 ) ] . a. p. schnyder , s. ryu , a. furusaki , and a. w. w. ludwig , http://link.aps.org/doi/10.1103/physrevb.78.195125[phys . b * 78 * , 195125 ( 2008 ) ] . a. kitaev , http://dx.doi.org/10.1063/1.3149495[aip conf . proc . * 1134 * , 22 ( 2009 ) ] . t. hyart , a. r. wright , g. khaliullin , and b. rosenow , http://dx.doi.org/10.1103/physrevb.85.140510[phys . b * 85 * , 140510 ( 2012 ) ] . s. okamoto , http://dx.doi.org/10.1103/physrevb.87.064508[phys . b * 87 * , 064508 ( 2013 ) ] . d. d. scherer , m. m. scherer , g. khaliullin , c. honerkamp , and b. rosenow , http://dx.doi.org/10.1103/physrevb.90.045135[phys . b 90 , 045135 ( 2014 ) ] . you , i. kimchi , and a. vishwanath , http://dx.doi.org/10.1103/physrevb.86.085145[phys . b * 86 * , 085145 ( 2012 ) ] . a.m. black - schaffer , w. wu , and k. le hur , http://dx.doi.org/10.1103/physrevb.90.054521[phys . b * 90 * , 054521 ( 2014 ) ] . j. li , r .- chu , j. k. jain , and s .- q . shen , http://dx.doi.org/10.1103/physrevlett.102.136806[phys . lett . * 102 * , 136806 ( 2009 ) ] . c. w. groth , m. wimmer , a. r. akhmerov , j. tworzydo , and c. w. j. beenakker , http://dx.doi.org/10.1103/physrevlett.103.196805[phys . 103 , 196805 ( 2009 ) ] . i. adagideli , m. wimmer , and a. teker , http://dx.doi.org/10.1103/physrevb.89.144506[phys . b * 89 * , 144506 ( 2014 ) ] . w. qin , d. xiao , k. chang , s .- q . shen , and z. zhang , http://arxiv.org/abs/1509.01666[arxiv:1509.01666 ] . sau and e. demler , http://dx.doi.org/10.1103/physrevb.88.205402[phys . b * 88 * , 205402 ( 2013 ) ] . r .- j . slager , l. rademaker , j. zaanen , and l. balents , http://dx.doi.org/10.1103/physrevb.92.085126[phys . b * 92 * , 085126 ( 2015 ) ] . c. w. j. beenakker , j. m. edge , j. p. dahlhaus , d. i. pikulin , s. mi , and m. wimmer , http://dx.doi.org/10.1103/physrevlett.111.037001 [ phys . rev . * 111 * , 037001 ( 2013 ) ] . i. c. fulga , f. hassler , and a. r. akhmerov , http://dx.doi.org/10.1103/physrevb.85.165409[phys . b * 85 * , 165409 ( 2012 ) ] . f. haake , quantum signatures of chaos , 3rd ed . ( springer , berlin , heidelberg , 2010 ) , chap . 3 ; m.v . berry in les houches , session xxxvi , 1981 m chaotic behavior of deterministic systems ( north - holland , amsterdam 1983 ) . k. wlms , a. stern , and k. flensberg , http://dx.doi.org/10.1103/physrevlett.113.246401[phys . lett . * 113 * , 246401 ( 2014 ) ] . m. koshino , t. morimoto , and m. sato , http://dx.doi.org/10.1103/physrevb.90.115207 [ phys . b * 90 * , 115207 ( 2014 ) ] . except for the definitions of the hamiltonians @xmath24 in eqs . and , which do not contain an impurity lattice , all discussion in sec . [ impurity - driven - trans ] refers to the reduced brillouin zone in the presence of an impurity lattice . the chern numbers in figs . [ fig : phasediagrampx ] , [ fig : phasediagramlargec ] , and [ fig : phasediagrambilayer ] were obtained by connecting points in the gapped phases with @xmath357 to points at the lines @xmath358 and calculating the change @xmath359 of the impurity band s chern number at each gap closing line .
we study quantum degenerate ultra - cold fermi gases in the presence of artificial spin - orbit coupling and crossed zeeman fields . we emphasize the case where parity is violated in the excitation spectrum and compare it with the simpler situation where parity is preserved . we investigate in detail spectroscopic properties such as the excitation spectrum , the spectral function , momentum distribution and density of states for the cases where parity is preserved or violated . similarly , we show that thermodynamic properties such as pressure , chemical potential , entropy , specific heat , isothermal compressibility and induced spin polarization become anisotropic as a function of zeeman field components , when parity is violated . lastly , we discuss the effects of interactions and present results for the pairing temperature as the precursor for the transition to a superfluid state . in particular , we find that the pairing temperature is dramatically reduced in the weak interaction regime as parity violation gets stronger , and that the momentum dependence of the order parameter for superfluidity violates parity when crossed zeeman fields are present for finite spin - orbit coupling .
1211.5133
the study of quantum degenerate fermions has been in the forefront of research in ultra - cold atoms and molecules in recent years , where particular attention was paid to the so - called evolution from bcs to bec superfluidity . bringing ultra - cold fermions into quantum degeneracy and using feshbach resonances to tune interactions between colliding fermions opened the door for the exploration of their superfluid phases , and their thermodynamic and correlation properties . this ability to tune interactions and explore the limits of weak , strong and unitary interactions had impact not only in cold - atom physics , but also in condensed matter physics ( strongly correlated superconductors ) , nuclear physics ( superconductivity in quantum chromodynamics ) and astrophysics ( superfluidity in neutron stars ) . many advances in cold atoms followed after the development of new tools for their toolbox . for instance the experimental study of the evolution from bcs to bec superfluidity occurred after appropriate feshbach resonances for @xmath0li and @xmath1k were identified , and used to study the crossover problem for @xmath2-wave superfluids . in addition to @xmath2-wave feshbach resonances , both @xmath0li and @xmath1k also exhibit @xmath3-wave feshbach resonances , which could produce p - wave superfluids in particular in bec regime if @xmath3-wave feshbach molecules were stable . unfortunately , such @xmath3-wave molecules do not live sufficiently long for the creation of @xmath3-wave superfluids @xcite . in addition to the manipulation of interactions , it has been possible to extract experimentally detailed thermodynamic information of interacting ultra - cold fermions @xcite from local density images of trapped atoms with the help of the local density approximation @xcite . a further tool was developed recently through the production of tunable artificial spin - orbit fields @xcite that were created in @xmath4rb , a bosonic isotope of rubidium , by using a set of raman beams that allowed momentum transfer and mixing of two dressed spin states . in these experiments , the emergence of spin - orbit fields , controlled by the momentum transfer from the light fields to the atoms , was connected to the simultaneous existence of an artificial zeeman field controlled by the raman coupling . in such experiments , the artificial spin - orbit and zeeman fields were used to manipulate the effective interactions between bosons and thus produce new quantum phases of bosonic matter @xcite . in a subsequent experiment it was directly demonstrated that the effective interactions between bosons can acquire higher angular momentum components which are directly controlled by the artificial spin - orbit and zeeman fields . the effective single - atom hamiltonian created in these experiments is @xmath5 when expressed in the dressed state spin basis @xmath6 . here , @xmath7 represents the kinetic energy of an atom with momentum @xmath8 and spin state @xmath2 , @xmath9 represents the zeeman field along the quantization axis @xmath10 which is directly related to the raman coupling @xmath11 , @xmath12 and @xmath13 , where @xmath14 represents the detuning and @xmath15 represents a mixture of equal rashba @xcite and dresselhaus @xcite terms , which we label as erd spin - orbit coupling . the possibility of studying similar phenomena with ultra - cold fermions in the presence of spin - orbit coupling and zeeman fields @xcite lead to an explosion of theoretical research , which focused primarily on the case of rashba - only ( ro ) spin - orbit coupling @xcite which has been studied extensively in condensed matter physics in the context of non - centro - symmetric superconductors @xcite . some suggestions of possible realizations of artificial ro spin - orbit fields in ultra - cold fermions have appeared in the literature @xcite . however , in our group , we have focused mostly on the erd case @xcite , which is simpler to be realized experimentally as it was demonstrated for interacting bosons @xcite in the case of @xmath4rb , and more recently for non - interacting fermions such as @xmath1k in china @xcite and @xmath0li in the united states @xcite . in our recent work @xcite , we emphasized the importance of interactions in producing novel phases for spin - orbit coupled ultra - cold fermions in the presence of zeeman fields , and we analyzed in detail the erd case with a longitudinal zeeman field which is orthogonal to the spin - orbit field @xcite . in this paper , we generalize our previous analysis to include an additional zeeman field @xmath16 representing the detuning @xmath17 , which is perpendicular to longitudinal zeeman field @xmath18 representing the raman coupling @xmath11 . in this case , the simultaneous presence of @xmath16 and the erd spin - orbit field @xmath19 along the @xmath20 axis leads to the loss of parity for the eigenvalues of @xmath21 defined in eq . ( [ eqn : hamiltonian - single - atom ] ) above . this violation of parity in the energy spectrum leads to many detectable parity - violating properties in spectroscopic quantities such as the spectral function and momentum distribution , which are explicitly momentum - dependent , or in momentum - integrated quantities like the density of states when viewed as a function of zeeman fields @xmath16 and @xmath16 . furthermore , we demonstrate that this parity violation can also be seen and quantified in thermodynamic quantities such as the pressure , entropy , specific heat , chemical potential , compressibility and spin - polarization , which are discussed in detail for non - interacting ultra - cold fermions in the presence of erd spin - orbit and crossed zeeman fields @xmath16 and @xmath22 . furthermore , we also discuss the effects of parity violation when attractive @xmath2-wave interactions are present , and we pay particular attention to the pairing temperature and to the absence of parity in the order parameter tensor describing the superfluid order . the remainder of the paper is as follows . in section [ sec : magnetic - hamiltonian ] , we describe the magnetic hamiltonian in its general form , and particularize it to any linear combination of rashba and dresselhaus terms , while in section [ sec : hamiltonian ] we discuss the independent - atom ( non - interacting ) hamiltonian including both the kinetic energy and the magnetic parts . the eigenvalues and eigenvectors of the single - atom hamiltonian are discussed in section [ sec : excitation - spectrum ] , where a generalized helicity basis is introduced and fermi surfaces for various values of crossed zeeman fields are presented . particular attention is paid to parity violations for the eigenvalues and eigenvectors . in section [ sec : spectral - function ] , we show the momentum dependence for the spectral function at fixed frequency revealing the cases of weak and strong parity violations , while in section [ sec : momentum - distribution ] we describe the the spin - resolved momentum distributions which also reveal parity violations for various values of crossed zeeman fields . in section [ sec : density - of - states ] , we analyze the spin - dependent density of states for a few values of the crossed zeeman fields and fixed erd spin - orbit coupling , while in section [ sec : thermodynamic - properties ] , we define several thermodynamics properties which analyzed in the following sections , such as pressure and entropy in section [ sec : pressure - entropy ] , chemical potential in section [ sec : chemical - potential ] , isothermal compressibility in section [ sec : isothermal - compressibility ] and spin - polarization [ sec : induced - polarization ] . furthermore , we investigate the effects of parity violation on the effective interactions in section [ sec : effects - of - interactions ] and on the pairing temperature in section [ sec : pairing - temperature ] . lastly , we state our conclusions in section [ sec : conclusions ] emphasizing that several measurements can be made to detect parity violation in both non - interacting and interacting fermi gases in the simultaneous presence of an erd spin - orbit field , and crossed zeeman fields @xmath16 , @xmath22 , where @xmath16 is related to the detuning @xmath17 and @xmath22 to raman coupling @xmath11 . to describe quantum degenerate fermi systems in the presence of artificial spin - orbit and crossed artificial zeeman fields , we discuss first the type of magnetic hamiltonian to be used . generally speaking the coupling between magnetic fields and two spin - states is given by @xmath23 where @xmath24 is the effective magnetic field ( including zeeman and spin - orbit ) and @xmath25 , with @xmath26 being the pauli matrices . this leads to the magnetic hamiltonian @xmath27 although there are many possible forms of fictitious magnetic fields that can be created in the laboratory . we discuss here a few simpler magnetic field configurations . in principle , the external artificial zeeman field @xmath28 can have three components @xmath29 , in practice , we set @xmath30 , and @xmath31 because in current experimental setups @xmath14 , where @xmath17 is the laser detuning , and @xmath32 , where @xmath11 is the strength of raman coupling field . in addition , there are many possible types of spin - orbit contributions where @xmath33 can have three components @xmath34 . however , we consider particular forms of spin - orbit fields , which can be more easily realized in practice . the first type has the dresselhaus form @xmath35 where @xmath36 measures the strength of the dresselhaus field in units of velocity . the corresponding hamiltonian for such field is @xmath37 the second type has the rashba form @xmath38 where @xmath39 measures the strength of the rashba field in units of velocity . the corresponding rashba hamiltonian is @xmath40 either the dresselhaus or the rashba forms require spin - orbit fields along the @xmath41 and @xmath20 directions , which to be produced experimentally demand two orthogonal raman setups , such that momentum transfer occurs in two perpendicular directions . a linear combination of two forms leads to the field @xmath42 where the velocities @xmath43 the corresponding hamiltonian for such linear combination is @xmath44 the simplest type of spin - orbit field that can been created in the laboratory has the equal - rashba - dresselhaus ( erd ) form @xmath45 where @xmath46 , and @xmath47 , or equivalently @xmath48 . the corresponding hamiltonian for the erd spin - orbit field has the simple form @xmath49 taking into account an arbitrary superposition of rashba and dresselhaus spin - orbit coupling and a general uniform zeeman field @xmath50 , we can write the zeeman - spin - orbit hamiltonian as @xmath51 where the parallel component of the total field is @xmath52 and the transverse component is @xmath53,\ ] ] where @xmath54 and @xmath55 having presented the magnetic hamiltonian , we discuss next the hamiltonian including the kinetic energy of the atoms . the hamiltonian for non - interacting ultra - cold fermions with identical masses @xmath56 in the presence of spin - orbit and crossed zeeman fields can be written in second quantization as @xmath57 where the spinor @xmath58 describes the creation of fermion states with momentum @xmath8 and spin @xmath59 or @xmath60 . such hamiltonian describes two hyperfine states of fermi atoms such as @xmath61 or @xmath62 , and the corresponding hamiltonian matrix is @xmath63 where @xmath64 represents the kinetic energy of a fermion with mass @xmath56 , momentum @xmath8 and spin @xmath65 with respect to the chemical potential @xmath66 . we define the variables @xmath67 and the chemical potentials @xmath68 and notice that @xmath69 plays the role of the average kinetic energy , while @xmath70 plays the role of the parallel zeeman field @xmath71 including the external field @xmath22 and the internal field @xmath72 due to a possible initial population imbalance . in the limit that there is zero initial population imbalance , the chemical potentials can be set to @xmath73 and @xmath74 , while the kinetic energies reduce to @xmath75 and @xmath76 , where @xmath77 . the non - interacting hamiltonian has the simpler form @xmath78 which can be re - expressed as @xmath79 where @xmath80 in the remainder of the manuscript , we will discuss this explicit form of the hamiltonian and some interesting consequences . we define the total number of fermions as @xmath81 , and choose our energy , velocity and momentum scales through the fermi momentum @xmath82 defined from the total density of fermions @xmath83 where @xmath84 and @xmath85 with @xmath86 . this choice leads to the fermi energy @xmath87 and to the fermi velocity @xmath88 , which are the energy and velocity scales used throughout the manuscript . now , let us introduce the unitary matrix @xmath89 that diagonalizes the hamiltonian @xmath90 , such that @xmath91 is a diagonal matrix containing the eigenvalues of @xmath92 . the corresponding eigenvectors are the spinors @xmath93 where the unitary matrix @xmath89 has a momentum - dependent su(2 ) form and can be written as @xmath94 where the normalization condition @xmath95 is imposed to satisfy the unitarity condition @xmath96 leading to the following expressions @xmath97 where @xmath98 is taken to be real without loss of generality and @xmath99 is a complex function where the phase @xmath100 is defined via @xmath101 leading to @xmath102 $ ] . the eigenvalues of @xmath92 emerge as @xmath103 where @xmath104 is the eigenvalue where the momentum dependent effective field @xmath105 is aligned with the spin @xmath106 , and @xmath107 is the eigenvalue where the momentum dependent effective field @xmath105 is aligned with the spin @xmath108 . here , @xmath109 is the magnitude of the effective field . the respective eigenvectors are @xmath110 corresponding to the state @xmath111 , and @xmath112 corresponding to the state @xmath113 . [ htb ] for initially population balanced systems in the erd case , with the zeeman field having only @xmath16 and @xmath22 components , the magnitude of the effective magnetic field @xmath114 does not have well defined parity , which implies that the same is true for the eigenvalues @xmath115 . if either @xmath116 ( zero detuning ) or @xmath117 ( zero spin - orbit coupling ) , parity is restored for the eigenvalues . this can be seen in fig . [ fig : one ] , where the energy dispersions @xmath118 and @xmath119 are shown for several cases . in all panels of fig . [ fig : one ] the dashed lines represent the case where the erd spin orbit is finite @xmath120 , but the zeeman fields are zero @xmath121 . this case corresponds to shifted parabolic bands @xmath122 and @xmath123 where @xmath124 . the lower helicity band @xmath125 has two minima , which occur at @xmath126 and @xmath127 , respectively , while the upper helicity band @xmath125 has only one minimum occurring at @xmath128 . this can be seen by completing the squares and rewriting the dispersions of the helicity bands as @xmath129 and @xmath130 where @xmath131 is the characteristic kinetic energy associated with the spin - orbit coupling strength @xmath132 , which has units of velocity . in fig . [ fig : one]a , we show the case of @xmath133 , for @xmath116 and @xmath134 , and in fig . [ fig : one]b , we show the case of @xmath133 , for @xmath116 and @xmath135 . in both of these cases the energy dispersions are parity preserving , and the main fundamental difference between fig . [ fig : one]a and fig . [ fig : one]b is that the lower helicity band @xmath136 has two minima at finite @xmath137 as seen in fig . [ fig : one]a , but a single minimum at @xmath138 as seen in fig . [ fig : one]b . while the upper helicity band @xmath139 has only a single minimum at @xmath138 in both fig . [ fig : one]a and fig . [ fig : one]b . the two minima in the lower helicity band for fig . [ fig : one]a occur only for low zeeman fields @xmath140 . in figs . [ fig : one]c and [ fig : one]d , we show the helicity bands for @xmath141 , but @xmath142 and @xmath143 , respectively . in these cases , the energy dispersions are @xmath144 and @xmath145 and do not even parity as it is standard , since @xmath146 these dispersions are compared to the dispersions for the case of @xmath147 , but finite @xmath148 , which are shown as dashed black lines in fig . [ fig : one ] . the last examples described in fig . [ fig : one]e and fig . [ fig : one]f correspond to cases where both @xmath16 and @xmath22 are non - zero , having values @xmath149 for ( e ) and @xmath143 , @xmath135 for ( f ) . notice the presence of two minima for the lower helicity band in ( e ) , and the existence of only one minimum for the lower helicity band in ( f ) , but in both cases parity ( inversion symmetry ) is violated . [ htb ] in fig . [ fig : two ] , we show cross sections of the fermi surfaces ( fs ) in the @xmath150-@xmath151 plane at @xmath152 for erd spin - orbit coupling and a few values of crossed zeeman fields . the fermi surfaces have rotational symmetry about the @xmath150 axis , that is , in the @xmath151-@xmath153 plane , and their full three - dimensional structure can be visualized using this property . we show in figs . [ fig : two]a - c the case for zeeman fields @xmath154 and @xmath134 , where the fermi surfaces exhibit parity ( or inversion ) symmetry . these parameters correspond to the helicity bands shown in fig . [ fig : one]a , where the lower helicity band has two minima . while we show in figs . [ fig : two]d - f the case for zeeman fields @xmath149 , where the corresponding fermi surfaces do not have well defined parity or inversion symmetry . the specific values of the chemical potential @xmath155 and induced polarization @xmath156 are indicated in the captions . in fig . [ fig : two ] , notice also that as the chemical potential @xmath155 is changed below the bottom of the helicity band @xmath157 , the central pocket of the fermi surface disappears ( see magenta surface near zero momentum in fig . [ fig : two ] ) . further lowering of the chemical potential leads to the crossing of a local maximum of the helicity band @xmath125 , where the residual fermi surface break into two pockets ( see blue surfaces in fig . [ fig : two ] ) . this is reminiscent of the lifshitz transition @xcite in non - interacting metals , also called metal - to - metal or conductor - to - conductor transition , where under pressure or another external parameter the fermi surface of the system changes topology producing a major rearrangement of momentum states which lead to a drastic change in the density of states of the system . the thermodynamic potential @xmath11 in the vicinity of the usual lifshitz transition behaves as @xmath158 where @xmath159 is the regular ( analytic ) part , and @xmath160 is the prefactor of the non - analytic component . the isothermal compressibility is related to the second - derivative of the thermodynamic potential with respect to the chemical potential and behaves as @xmath161 where @xmath162 is the regular part and @xmath163 is the coefficient of the non - analytic component . according to ehrenfest s classification of phase transitions , the non - analyticity manifests itself only in the third derivative and is a third - order phase transition . however , the lifshitz transition is more commonly called the @xmath164-@xmath165 order transition in allusion to the specific @xmath166 power - law non - analyticity of @xmath11 in three dimensions . this topological transition is not characterized under landau s symmetry - based classification , since no symmetry is broken in the lifshitz case . this _ trivial _ lifshitz transition can be seen in fig . [ fig : two ] both for the parity - preserving and parity - violating examples . in fig . [ fig : three ] , we also show cross sections of the fermi surfaces ( fs ) in the @xmath150-@xmath151 plane at @xmath152 for erd spin - orbit coupling @xmath133 and a few values of crossed zeeman fields . the fermi surfaces have also rotational symmetry about the @xmath150 axis , that is , in the @xmath151-@xmath153 plane . we show in figs . [ fig : three]a - b the case for zeeman fields @xmath154 and @xmath135 , where the fermi surfaces exhibit parity ( or inversion ) symmetry . these parameters correspond to the helicity bands shown in fig . [ fig : one]b , where the lower helicity band has only one minimum . while we show in figs . [ fig : three]c - d the cases of the zeeman fields @xmath143 and @xmath135 where the corresponding fermi surfaces do not have well defined parity or inversion symmetry . the values of the chemical potential @xmath155 and induced polarization @xmath167 are @xmath168 , @xmath169 for fig . [ fig : three]a ; @xmath170 , @xmath171 for fig . [ fig : three]b ; @xmath168 , @xmath172 for fig . [ fig : three]c ; and @xmath170 , @xmath173 for fig . [ fig : three]d . notice that there is a fundamental difference between the fermi surfaces in figs . [ fig : two ] and [ fig : three ] in connection with their topology . the lower helicity band @xmath125 can have two simply connected fs for the parameters of fig . [ fig : two ] , but only one simply connected fs for the parameters of fig . [ fig : three ] . this means that there is only one _ trivial _ lifshitz transition in the case of fig . [ fig : three ] , while there are two _ trivial _ lifshitz transitions in the case of fig . [ fig : two ] . we call this transition for non - interacting systems _ trivial _ to contrast it with a more exotic , but related topological transition that can occur in @xmath3-wave @xcite or @xmath174-wave @xcite superfluids , where interactions play a fundamental role . a further characterization of the parity violation present in fermion systems with spin - orbit coupling and crossed zeeman fields can be made by analyzing additional spectroscopic quantities such as the spectral function to be discussed next . a very useful tool to probe momentum - resolved properties is the use of radio - frequency ( rf ) spectroscopy that can extract the spectral function @xcite yielding similar measurements to those encountered in photoemission spectroscopy of condensed matter physics . the resolvent ( or green ) operator matrix is defined as @xmath175 in momentum - frequency space . in the present case , the diagonal components of @xmath176 are @xmath177 for the spin @xmath59 component and @xmath178 for the spin @xmath60 component . the corresponding spectral function is @xmath179 , $ ] which in terms of the coherence factor @xmath98 and @xmath180 becomes @xmath181 component , and @xmath182 for the down - spin @xmath183 component . the spectral functions @xmath184 at frequency @xmath185 and temperature @xmath186 are shown in fig . [ fig : four ] for some values of the crossed zeeman fields and spin - orbit coupling . a small energy broadening @xmath187 is included and a logarithmic scale is used to help visualization . in the relevant panels of fig . [ fig : four ] , the blue dashed lines represent @xmath188 , and the red solid lines represent @xmath189 . additionally , the two left - most columns correspond to the @xmath190 component , and the two right - most columns describe the @xmath191 component . in figs . [ fig : four](a)-(d ) the zeeman fields are @xmath147 , the erd coupling is @xmath192 , the chemical potential is @xmath193 , and the induced polarization @xmath194 . in figs . [ fig : four](e)-(h ) the zeeman fields are @xmath154 , @xmath134 , the erd coupling is @xmath133 , and the chemical potential is @xmath195 , and the induced polarization is @xmath196 . in these two cases parity is preserved , since the eigenvalues @xmath197 and coherence factors @xmath98 and @xmath180 are invariant under momentum inversion for @xmath154 . in figs . [ fig : four](i)-(l ) the zeeman fields are @xmath149 , the erd coupling is @xmath133 , the chemical potential is @xmath198 , and the induced polarization is @xmath199 . parity is not preserved in the last case since @xmath200 ) . this is reflected in the absence of inversion symmetry , which is noticeable but weak . [ htb ] in fig . [ fig : five ] , we show the spectral functions @xmath184 at frequency @xmath185 and temperature @xmath186 for erd spin - orbit coupling @xmath133 . for varying chemical potentials , we choose the particular values of the crossed zeeman fields to be @xmath149 in order to emphasize the absence of inversion symmetry ( parity ) when @xmath201 . as before , a small energy broadening @xmath187 is included and a logarithmic scale is used to help visualization . the same color convention is used in the relevant panels of fig . [ fig : five ] , where the blue dashed lines represent @xmath188 , and the red solid lines represent @xmath189 . in addition , the two left - most columns describe the @xmath190 component , and the two right - most columns correspond to the @xmath191 component . in figs . [ fig : five](a)-(d ) the chemical potential is @xmath198 , the induced polarization is @xmath199 , and parity is weakly violated , but noticeable . in figs . [ fig : five](e)-(h ) the chemical potential is @xmath170 , the induced polarization is @xmath202 , and parity is more strongly violated . in figs . [ fig : five](i)-(l ) the chemical potential is @xmath203 , the induced polarization is @xmath204 , and parity is strongly violated . in the last two cases parity is violated more strongly , since the eigenvalues @xmath197 and coherence factors @xmath98 and @xmath180 are not invariant under momentum inversion for @xmath205 and are more sensitive to this violation for chemical potentials closer to the bottom of the helicity bands . to understand the momentum distribution for quantum degenerate fermi gases in the presence of spin - orbit and crossed zeeman fields , we look first at the expectation value @xmath206 of the number operators @xmath207 which describes the momentum distribution @xmath208 in the helicity basis . in this case , the momentum distribution is @xmath209 , where @xmath210 is the fermi function . the momentum distributions for the original spin states @xmath86 are defined as @xmath211 with the help of the unitary matrix @xmath212 defined in eq . ( [ unitary - matrix ] ) , which relates the creation and annihilation operators in the helicity and the standard spin basis the momentum distributions become @xmath213 component , and @xmath214 for the down - spin @xmath183 component . such expressions can be also obtained from the general relation @xmath215 between the momentum distribution and the spectral function for fermions . it is also convenient to obtain the momentum distribution sum @xmath216 and the momentum distribution difference @xmath217 the first distribution can be written in terms of the fermi functions only @xmath218 while the second distribution can be written in terms of the fermi functions and the components of the effective zeeman field @xmath219.\ ] ] in fig . [ fig : six ] , we show momentum distributions @xmath220 at @xmath186 and in the regime where the fermi system is largely degenerate , containing wide regions in momentum space where @xmath221 . the blue dashed lines represent cuts of @xmath220 along the @xmath222 direction , while the red solid lines represents cuts of @xmath220 along @xmath223 . the left - most columns correspond to spin @xmath59 , while the right - most columns represent spin @xmath60 . for reference , we show in figs . [ fig : six](a)-(d ) the case with zero spin - orbit coupling and without zeeman fields , corresponding to parameters @xmath192 , @xmath147 chemical potential @xmath224 . in this case , the momentum distributions for the two spin components are identical @xmath225 , meaning that the populations are balanced with @xmath226 , and induced polarization is @xmath194 . in figs . [ fig : six](e)-(h ) , we show momentum distributions for parameters @xmath133 , @xmath154 , @xmath134 and chemical potential @xmath227 . here , the momentum distributions acquire double _ plateaux _ structures along the direction @xmath223 due to the momentum shifts of the helicity bands as shown in fig . [ fig : one](a ) . additionally , the momentum distributions for different spin - components are no longer identical , such that @xmath228 , or @xmath229 , and the induced polarization is non - zero taking the value @xmath196 . in these two cases , parity is not violated and the momentum distributions are even functions of momentum under inversion symmetry . however , in figs . [ fig : six](i)-(l ) , we show momentum distributions for parameters @xmath133 , @xmath149 , and chemical potential @xmath230 , in which case the double _ plateaux _ structures are still preserved , population imbalance is present with @xmath229 and induced polarization @xmath199 . most importantly parity is weakly violated since @xmath231 . [ htb ] in fig . [ fig : seven ] , we show momentum distributions @xmath220 at @xmath186 for parameters @xmath133 , @xmath149 , and varying chemical potentials @xmath232 . we emphasize the regimes where parity is more strongly violated leading to momentum distributions without inversion symmetry : @xmath233 . the blue dashed lines represent cuts of @xmath220 along the @xmath222 direction , while the red solid lines represents cuts of @xmath220 along @xmath223 . for reference , we show in figs . [ fig : seven](a)-(d ) the case with @xmath198 and @xmath199 , where the fermi system is still largely degenerate , containing wide regions in momentum space with @xmath221 , and at the same time parity is violated only weakly . in figs . [ fig : seven](e)-(h ) , we show momentum distributions for @xmath170 and @xmath202 , while in figs . [ fig : seven](i)-(l ) , we show momentum distributions for @xmath203 and @xmath204 . in both cases , the fermi system is no longer degenerate , containing wide regions in momentum space where @xmath234 . in the last two cases the momentum distributions remain symmetric upon reflection along the @xmath151 or @xmath153 directions , but parity is more strongly violated leading to a highly asymmetric momentum distributions along the @xmath150 direction . [ htb ] having discussed the momentum distribution at low temperatures , we analyze next the density of states of quantum degenerate fermions in the presence of spin - orbit coupling and crossed zeeman fields . the density of states for spin @xmath2 can be written as @xmath235 in terms of the spectral function @xmath236 an analysis of the spin - dependent density of states is useful to provide the frequency ( energy ) dependence of the spin - polarization of the system . in fig . [ fig : eight ] , we show the density of states @xmath237 for various parameters with an energy broadening @xmath238 . we show specifically the case of zero spin - orbit coupling and zeeman fields in fig . [ fig : eight](a ) , which has the characteristic square - root frequency dependence for a three - dimensional system . in this case the system is not polarized and the tails below the bottom of the energy band are due to the finite energy broadening . we show two other situations for comparison corresponding to cases which are polarized with finite zeeman fields , as well as with a finite spin - orbit coupling . in fig . [ fig : eight](b)-(c ) , we show that the band edges shift to lower frequencies when the zeeman and spin - orbit fields are turned on . furthermore , in fig . [ fig : eight](b ) the spin - dependent density of states are shown for the case where parity is not violated , corresponding to @xmath239 , @xmath134 , and @xmath133 ; and in fig . [ fig : eight](c ) the spin - dependent density of states are shown for the case where parity is violated , corresponding to @xmath240 and @xmath133 . since momentum is integrated over , there is no clear signature of parity violation in @xmath237 as there is in momentum - resolved observables such as the spectral density @xmath241 , momentum distribution @xmath220 , or helicity dispersions @xmath136 and @xmath139 discussed earlier . however , we show the different spin - dependent density of states for comparison . the kinks present in these figures reflect the location of the maxima and minima of the helicity bands . having discussed the density of states , we present next an analysis of thermodynamic properties . [ htp ] from the energy spectrum of the hamiltonian @xmath242 defined in eq . ( [ eqn : hamiltonian - non - interacting ] ) , we can obtain the partition function as @xmath243,\ ] ] and the corresponding thermodynamic potential @xmath244 when no initial population imbalance is present , the total number of particles is fixed by @xmath245 which determines the chemical potential @xmath155 . in addition , because an external zeeman field is present , we can define the induced polarization @xmath246 where the number of particles in spin - state @xmath2 is @xmath247 next , we begin our analysis of thermodynamic variables that could be measured using the techniques already developed for ultra - cold fermions @xcite in the absence of spin - orbit coupling . in the discussion that follows , we will cover the pressure , the chemical potential , the isothermal compressibility and the induced magnetization ( spin - polarization ) . the pressure of ultra - cold fermions in the presence of spin - orbit coupling and crossed zeeman fields is @xmath248 where @xmath11 is thermodynamic potential , and @xmath249 is the helicity spin index . in fig . [ fig : nine ] , we show the scaled pressure @xmath250 as a function of the zeeman fields @xmath251 and @xmath252 for fixed spin - orbit coupling @xmath148 . notice that in the absence of spin - orbit and zeeman fields the pressure reduces to the standard results of an non - interacting fermi gas @xmath253 . the pressure is an even function of both @xmath16 and @xmath22 and is shown for a range of @xmath16 and @xmath22 varying from zero to @xmath254 . the spin - orbit coupling for figs . [ fig : nine](a)-(b ) is @xmath192 , and the pressure is completely isotropic in the @xmath16-@xmath22 plane , since the helicity bands @xmath255 are also isotropic in @xmath16-@xmath22 plane and parity is preserved . however , in fig . [ fig : nine](c)-(d ) , where @xmath133 , the pressure is anisotropic in the @xmath16-@xmath22 plane , since the helicity bands @xmath255 are now anisotropic in the @xmath16-@xmath22 plane and are not invariant under parity . [ htb ] we note in passing that the entropy @xmath256 of the system can be easily extracted , by rewriting the thermodynamic potential as @xmath257 , $ ] when expressed in terms of the fermi function @xmath258 . using the relation @xmath259 leads to the final result ( @xmath260 ) @xmath261 \right\},\ ] ] which is nothing but the entropy of a non - interacting fermi gas in the presence of spin - orbit coupling and crossed zeeman fields . we will not show plots of the entropy , or of the specific heat , which can also be easily obtained , but rather discuss next the chemical potential and its dependence on the crossed zeeman fields . the chemical potential @xmath155 in the grand - canonical ensemble is determined by fixing the average number of particles given in eq . ( [ eqn : number ] ) , which can be rewritten as @xmath262 where @xmath263 is the total momentum distribution defined in eq . ( [ eqn : momentum - distribution - sum ] ) . the behavior of @xmath155 as a function of the zeeman fields @xmath16 and @xmath22 is shown in fig . [ fig : ten ] , which uses the fact that @xmath155 is an even function of these variables . the range of the zeeman fields is also from @xmath264 to @xmath254 . the case where parity is preserved is shown if fig . [ fig : ten](a ) and ( b ) , corresponding to @xmath192 , such that the chemical potential is isotropic in the @xmath16-@xmath22 plane . similarly , in fig . [ fig : ten](c ) and ( d ) , we show the case corresponding to @xmath133 , where parity is violated for any finite zeeman component @xmath16 . as a result the chemical potential @xmath232 is anisotropic in @xmath16-@xmath22 plane , because the helicity bands are neither even nor odd in momentum space . this demonstrates that the anisotropy of @xmath155 is a measure of parity violation . another quantity that can be easily measured and that reveals a similar effect is the isothermal compressibility to be discussed next . the isothermal compressibility can be obtained from the knowledge of the pressure , as is defined as @xmath265 but in the grand - canonical ensemble , where we need to fix the average number of particles , the isothermal compressibility can be directly written as @xmath266 using the relation defined in eq . ( [ eqn : number - total ] ) and noticing that the partial derivative @xmath267 is directly related to the fermi function @xmath210 , it is possible to write the expression for the isothermal compressibility as @xmath268.\ ] ] the experimental extraction of the isothermal compressibility from measurements of density fluctuations was suggested theoretically several years ago both in harmonically confined systems @xcite and optical lattices @xcite , and early improvements in the detection schemes of density fluctuations @xcite became sufficiently sensitive to extract this information from experimental data . in a recent experiment @xcite using laser speckles , the isothermal compressibility and the spin susceptibility were measured as a function of interaction parameter via the fluctuation dissipation theorem throughout the evolution from bcs to bec superfluidity in balanced fermi systems . the atomic compressibility can be measured using the fluctuation - dissipation theorem relating the fluctuation in the density of particles @xmath84 , where @xmath269 is the average number of particles , and @xmath270 is particle - number operator . the relation between the isothermal compressibility @xmath271 and particle - number ( density ) fluctuations is given by the relation : @xmath272 we see no major technical impediment to use techniques that are sensitive to spin - dependent density fluctuations in population imbalanced fermi - fermi mixtures with equal masses @xcite , whether the imbalance is created initially via radio - frequency fields or via artificial spin - orbit and zeeman fields . such analysis was shown to be theoretically possible even for fermi mixtures with unequal masses @xcite , and preliminary experimental results for these systems @xcite seem to indicate that indeed the compressibility and spin susceptibility matrix elements can be directly extracted from the local density and density fluctuation profiles . thus , we show in fig . [ fig : eleven ] the isothermal compressibility @xmath271 as a function of both @xmath16 and @xmath22 for a range of @xmath16 and @xmath22 varying from zero to @xmath254 , at fixed temperature @xmath273 . the spin - orbit coupling for figs . [ fig : eleven](a)-(b ) is @xmath192 , and the compressibility is completely isotropic , as parity is preserved . however , in fig . [ fig : eleven](c)-(d ) , where @xmath133 , parity is broken for @xmath201 . this parity breaking is reflected in the helicity bands as discussed earlier , and manifests itself in the behavior of the compressibility versus @xmath274 via the anisotropy revealed in fig . [ fig : eleven](c)-(d ) . again such anisotropy is a reflection of the parity violation caused by the simultaneous presence of the erd spin - orbit field and @xmath16 . another important property that can be measured is the spin polarization as a function of the zeeman fields @xmath275 for fixed spin - orbit coupling , which is discussed next . we can also analyze the induced spin polarization in the presence of crossed zeeman fields and spin - orbit coupling . the spin polarization along the @xmath276 direction @xmath277 is given by the expectation value @xmath278 such general expression can be particularized for each component . for instance , the expectation value @xmath279 expressed in the original spin basis , can be rewritten in the helicity basis as @xmath280.\ ] ] analogously the expectation value of the spin - operator @xmath281 in the original spin basis is @xmath282\ ] ] in the original spin basis can be written as @xmath283\ ] ] in the helicity basis . lastly , the expectation value of @xmath284 in the original spin basis is @xmath285 which can be rewritten in the helicity basis as @xmath286.\ ] ] finally , the averages @xmath287 and @xmath288 can be expressed as real and imaginary parts of the transverse spin - polarization @xmath289 defined this way to be compatible with the definition of @xmath290 . the transverse spin - polarization takes the final form @xmath291.\ ] ] correspondingly the longitudinal spin polarization can be written as @xmath292,\ ] ] which is directly related to the induced population imbalance @xmath167 by the expression @xmath293 where @xmath294 is the total number of particles , as defined earlier . for erd spin - orbit coupling with field @xmath295 , the transverse field @xmath296 has only the y - component . this means that @xmath287 is identically zero for any value of @xmath16 and @xmath22 for any value of @xmath297 , given that @xmath12 . however , @xmath288 is not identically zero for the erd case above , unless parity is preserved in the helicity bands @xmath115 , which means @xmath116 . for any finite value of @xmath16 , @xmath288 is non - zero . this behavior is revealed in fig . [ fig : twelve ] , where the expectation values @xmath288 and @xmath298 of spin polarization are shown as a function of @xmath299 . in particular , we show in fig . [ fig : twelve ] the finite temperature ( @xmath186 ) induced spin polarizations per particle @xmath300 ( two left - most columns ) and @xmath301 ( two right - most columns ) for erd spin - orbit parameter @xmath192 from ( a ) through ( d ) , and for @xmath133 from ( e ) through ( h ) . in ( a ) and ( e ) , the red solid line represents @xmath300 as a function of @xmath251 at @xmath302 , and the blue dashed line represents @xmath300 as a function of @xmath303 at @xmath154 . in ( d ) and ( h ) , the red solid line represents @xmath301 as a function of @xmath251 at @xmath302 , and the blue dashed line represents @xmath301 as a function of @xmath252 at @xmath154 . having analyzed several thermodynamic properties for non - interacting quantum degenerate ultra - cold fermions , which already present some fundamental non - trivial properties such as the violation of parity , we discuss next the effects of interactions and how parity violation affects the pairing temperature of such fermions and the superfluid order parameter . in this section , we analyze briefly the effects of interactions in the presence of spin - orbit and crossed zeeman fields , with focus on the dependence of the pairing temperature with respect to the interaction parameter for given spin - orbit and zeeman fields . we also present a short discussion about the effects of parity violation on the superfluid order parameter . the interaction hamiltonian is @xmath304 where @xmath305 represents the strength of the contact interaction . only @xmath2-wave scattering is considered in regards to the original spin states @xmath59 and @xmath60 . converting the interaction term into momentum space leads to @xmath306 where the pair creation operator with center of mass momentum @xmath307 is @xmath308 and @xmath305 can be expressed in terms of the scattering length through the lippman - schwinger relation @xmath309 the interaction hamiltonian @xmath310 can be written in the helicity basis as @xmath311 where the indices @xmath312 cover @xmath106 and @xmath108 states . pairing is now described by the operator @xmath313 and its hermitian conjugate , with momentum indices @xmath314 and @xmath315 . the matrix @xmath316 is directly related to the matrix elements of the momentum dependent su(2 ) rotation into the helicity basis , and reveals that the center of mass momentum @xmath317 and the relative momentum @xmath318 are coupled and no longer independent . from eq . ( [ eqn : pairing - operator - helicity - basis ] ) it is clear that pairing between fermions of momenta @xmath319 and @xmath320 can occur within the same helicity band ( intra - helicity pairing ) or between two different helicity bands ( inter - helicity pairing ) . for pairing at @xmath321 , the order parameter for superfluidity is the tensor @xmath322 where @xmath323 leading to components : @xmath324 for total helicity projection @xmath325 ; @xmath326 for total helicity projection @xmath327 ; and @xmath328 for total helicity projection @xmath329 . it is very important to emphasize that for non - zero spin - orbit coupling and crossed zeeman fields @xmath16 and @xmath22 , the order parameter tensor @xmath330 does not have well defined parity . for instance , while @xmath331 and @xmath332 have odd parity , the matrix elements @xmath333 and @xmath334 do not have well defined parity . however , we may still define singlet and triplet sectors for the helicity basis , such that the singlet sector @xmath335/2 $ ] has even parity and the triplet sector defined by the components @xmath331 , @xmath336/2 $ ] and @xmath332 have odd parity for any value of the erd spin - orbit coupling @xmath148 and crossed zeeman fields @xmath251 and @xmath252 . the preservation of parity in the singlet and triplet sectors is also true for the rashba - only ( ro ) case , but the order parameter breaks time - reversal symmetry . within the mean field approximation , the hamiltonian matrix in the helicity basis is @xmath337 this hamiltonian matrix is traceless , therefore the sum of its eigenvalues is zero , however the eigenvalues of @xmath338 are not invariant under parity . by labeling the eigenvalues as @xmath339 , @xmath340 , @xmath341 and @xmath342 in decreasing order of energy , and using the tracelessness condition then the sum @xmath343 $ ] but each eigenvalue @xmath344 does not a well defined parity . typically these eigenvalues are even in momentum space , but not here because the parity violation induced by the simultaneous presence of the crossed zeeman fields and the spin - orbit coupling , thus , in the present case @xmath345 . however a generalized particle - hole symmetry applies leading to @xmath346 and @xmath347 the eigenvalues in this parity violating case can be obtained analytically for any mixture of rashba and dresselhaus terms from the determinant @xmath348 , $ ] which leads to the characteristic quartic equation @xmath349 for each momentum @xmath8 . here , the coefficient of the cubic term @xmath350 is the sum of the eigenvalues @xmath344 of the the hamiltonian matrix @xmath351 and therefore vanishes . the coefficient of the quadratic term is @xmath352 , while the coefficient of the linear term is @xmath353 the last coefficient is just the product of the four eigenvalues leading to @xmath354 . in the particular case of erd spin - orbit coupling with crossed zeeman fields , the coefficients become @xmath355 , and the coefficient of the quadratic term takes the form @xmath356 while the coefficient of the linear term is @xmath357 and lastly the coefficient of the zero - th order term is @xmath358 with @xmath359 here , @xmath360 has the same definition used in the paragraph that follows eq . ( [ eqn : hamiltonian - matrix - non - interacting ] ) , and is a measure of the kinetic energy with respect to the chemical potential . even in this simpler case of erd spin - orbit coupling , the precise analytical form of the eigenvalues in the presence of crossed zeeman fields is quite cumbersome , and we do not list them here explicitly . rather , we discuss next the consequences of parity violation on the pairing temperature of ultra - cold fermions in the presence of spin - orbit and crossed zeeman fields . from the excitation spectrum discussed above , we obtain the corresponding thermodynamic potential as @xmath361 from which the order parameter equation is determined via the minimization of @xmath362 with respect to @xmath363 , leading to @xmath364 where @xmath365 is the fermi function for energy @xmath366 . the contact interaction @xmath305 can be elliminated in favor of the scattering length @xmath367 via the lippman - schwinger relation defined in eq . ( [ eqn : lippman - schwinger - relation ] ) . the total number of particles @xmath81 is defined from the thermodynamic relation @xmath368 and leads to the corresponding number equation @xmath369,\ ] ] since the system is assumed to have no initial population imbalance . the self - consistent solutions of eq . ( [ eqn : order - parameter ] ) and ( [ eqn : number - saddle - point ] ) guarantee the existence of mean field solutions for the order parameter amplitude @xmath370 and the chemical potential @xmath155 as a function of the zeeman fields @xmath16 and @xmath22 , the spin - orbit coupling @xmath132 and scattering length @xmath367 . however , the thermodynamic stability of the solutions obtained has to be tested against the maximum entropy condition ( or minimum of the thermodynamic potential ) over the same parameter space spanned by the variables @xmath16 , @xmath22 , @xmath132 and @xmath367 , which determine the phase space of the present system . however , we discuss here only the effects of crossed zeeman fields @xmath16 , @xmath22 on the pairing temperature @xmath371 obtained by solving the mean - field self - consistent relations defined by eq . ( [ eqn : order - parameter ] ) and eq . ( [ eqn : number - saddle - point ] ) with the order parameter amplitude set to zero , i. e. , @xmath372 . in fig . [ fig : thirteen ] , we show the pairing temperature @xmath373 as a function of erd spin - orbit coupling @xmath374 for selected values of @xmath16 , @xmath22 and interaction parameter @xmath375 . the lines are guides to the eye given that the number of points does not form a dense set . in ( a ) @xmath376 and in ( b ) @xmath377 both at @xmath378 , showing the behavior of @xmath371 on the bcs side of unitarity . in ( c ) @xmath376 and in ( d ) @xmath379 both at @xmath380 , showing the behavior or @xmath371 on the bec side of unitarity . in ( a ) and ( b ) , the black - dotted line labels @xmath141 , the red - dashed line @xmath381 , the green - dash - dotted line @xmath134 , and the blue - solid line @xmath382 . however , in ( c ) and ( d ) , the black - dotted line labels @xmath383 , the red - dashed line @xmath384 , the green - dash - dotted line @xmath385 , and the blue - solid line @xmath386 . notice that the relative suppression of pairing with zero center of mass momentum in the bcs side @xmath387 is larger than in the bec side @xmath388 , since it relies strongly on pairing of states only close to the fermi wavevectors @xmath389 and @xmath390 . [ htb ] in fig . [ fig : fourteen ] , we show the pairing temperature @xmath391 versus scattering parameter @xmath375 for various values of the artificial zeeman field components @xmath16 and @xmath22 and particular values of the erd spin - orbit coupling @xmath132 . the black cross line corresponds to parameters @xmath192 , @xmath147 , where there are no erd spin - orbit coupling and no zeeman fields . the red solid line corresponds to parameters @xmath392 , @xmath147 . notice that these lines coincide , because the erd spin - orbit field can be gauged away producing exactly the same results for any value of @xmath148 so long as @xmath147 . the blue dashed - dotted line describes the case for parameters @xmath392 , @xmath154 , and @xmath385 , showing that the presence of zeeman field @xmath22 in the bcs regime produces an energy cost for pairing of fermions with opposite momenta and zero center - of - mass momentum , thus reducing the pairing temperature substantially . however , the purple dashed line describing the situation corresponding to @xmath392 , @xmath393 , and @xmath302 shows a much stronger suppression of the pairing temperature than for the case of @xmath392 , @xmath154 , and @xmath385 ( blue dot - dashed line ) , because it becomes much more difficult for fermions with opposite momenta to pair with zero center - mass - momentum due to the parity violation in the excitation spectrum of the fermions introduced by @xmath16 when @xmath394 . lastly , the green dashed line shows the case of @xmath392 , @xmath393 , and @xmath385 , where the combined effect of the zeeman energy cost and parity violation lead to a dramatic reduction of the pairing temperature in the bcs region and even near unitarity . another important point to emphasize in fig . [ fig : fourteen ] relates to the bending of @xmath371 in the two last cases corresponding to the dashed purple and dashed green lines . this bending indicate and instability of zero center - of - mass momentum pairing towards finite center - of - mass momentum pairing , and point of infinite slope in both curves indicates the separation of the two regimes . in the cases where the pairing temperature reflects the critical temperature of the system , the locations of infinite slope would correspond to a lifshitz point , and a small region to the left of the negative slope regime would correspond to a superfluid with finite center of mass momentum , which is favored due to parity violation in the helicity bands . for the parameters discussed the pairing temperature is not largely affected in the bec regime , since the binding of fermions is controlled by the emergence of two - body bound states with binding energy @xmath395 , and not by cooper pairing in the presence of a fermi sea . in the bec regime , an estimate of @xmath371 can be given by considering the chemical equilibrium condition @xmath396 , where @xmath397 is the chemical potential of the bosons formed by tightly bound fermions , and @xmath398 is the chemical potential of unbound fermions . since @xmath399 , both bosons and unbound fermions are highly non - degenerate , and behave like classical ideal gases , in which case the chemical potential can be directly calculated and used to obtain the relation @xmath400 to logarithmic accuracy . here , the binding energy @xmath401 is a function of the interaction parameter @xmath375 , the erd spin - orbit parameter @xmath148 and the zeeman fields @xmath251 and @xmath252 . the logarithmic term is an entropy correction that reduces the pairing temperature @xmath371 to a value much lower than the absolute value of the binding energy @xmath395 of two fermions . the pairing temperature @xmath391 is essentially the same for values of @xmath402 ( not shown in fig . [ fig : twelve ] ) , because the binding energy @xmath403 $ ] , in which case @xmath404 and the pairing temperature is @xmath405/ \ln [ 1/(k_f a_s)]^3 $ ] , which agrees with the numerical calculations in the regime of @xmath406 . we will not present here a discussion of the condensation or critical temperature @xmath407 as it requires a full calculation of fluctuations effects , which is now underway @xcite . however , in the extreme bec regime , where @xmath406 , the critical temperature @xmath407 can be obtained from the bose - einstein condensation temperature @xmath408^{2/3 } 2\pi / m_b$ ] , where the zeta function @xmath409 , the density of bosons @xmath410 is half the density @xmath411 of fermions , and @xmath412 is the mass of bosons which is a function of the mass @xmath56 of fermions , and the parameters @xmath132 , @xmath16 and @xmath22 . the ratio @xmath413 can be parameterized by the dimensionless ratios @xmath148 , @xmath251 and @xmath252 . in the limit of @xmath414 with finite @xmath148 , @xmath251 and @xmath252 , the boson mass @xmath415 , but corrections depending on the aforementioned ratios tend to make the mass heavier , and for finite but large @xmath375 this mass increase tends to reduce @xmath416 . however , the effect of interactions between the effective bosons is very subtle and the understanding of its dependence on the dimensionless ratios @xmath375 , @xmath148 , @xmath251 , and @xmath252 is being currently investigated @xcite . [ htb ] after a detailed discussion of the effects of parity violation on interactions , order parameter , and pairing temperature @xmath371 , which is is dramatically affected in the bcs regime , but largely unaffected in the bec regime for the parameter range investigated , we are ready to present our conclusions . we have analyzed the normal state of a degenerate fermi gas in the presence of artificial spin - orbit coupling and crossed zeeman fields . the specific form of the spin - orbit field chosen corresponds to a mixture of equal rashba and dresselhaus ( erd ) terms @xmath417 , which has been experimentally realized . the artificial zeeman field @xmath22 along a defined spin quantization @xmath10 corresponds to the raman intensity of the laser beams , and the crossed zeeman field @xmath16 pointing along the same direction the erd spin - orbit field corresponds to the frequency detuning from the atomic transition coupling two spin states . in such configuration , the eigenvalues of the non - interacting problem are obtained in a generalized helicity basis , where the helicity spin projection points either along or opposite to the quantization axis defined by the effective magnetic field @xmath418 in this case , we have shown that the presence of spin - orbit and crossed zeeman fields lead to parity violation in the excitation spectrum of the non - interacting fermi gas , which has immediate consequences for the fermi surface , spectral density and momentum distribution , which also do not have well defined parity . such parity violation emerges in momentum - resolved spectroscopic quantities because they are all functions of the magnitude @xmath419 of the effective field @xmath420 , which does not have well defined parity , since @xmath421 . in addition , we have shown that information on parity violation can be extracted from momentum - averaged thermodynamic properties such as pressure , entropy , chemical potential , compressibility , and spin - polarization as a function of crossed zeeman field components @xmath251 and @xmath252 for fixed spin - orbit coupling @xmath148 at fixed temperature @xmath422 . a signature of parity violation is the different behavior of any chosen thermodynamic quantity as a function of @xmath16 for @xmath141 and as a function of @xmath22 for @xmath116 . this anisotropy of thermodynamic properties in the @xmath423 plane for finite spin - orbit coupling @xmath148 is a direct reflection of the lack of parity ( inversion symmetry ) . lastly , we analyzed the effects of interactions for a degenerate fermi gas when parity is broken , and investigated how parity violation influences the order parameter of a uniform superfluid and the fermion pairing temperature . we noticed that the order parameter tensor for a uniform superfluid in the generalized helicity basis no longer possesses inversion symmetry , however singlet and triplet pairing in can still be defined this basis and preserves parity . furthermore , we found that the effects of parity violation are strong in the pairing temperature of fermions , because it becomes increasingly more difficult to pair states with zero center of mass momentum between helicity bands with progressively larger loss of inversion symmetry . j. p. gaebler , j. t. stewart , j. l. bohn , and d. s. jin phys . 98 * , 200403 ( 2007 ) . yasuhisa inada , munekazu horikoshi , shuta nakajima , makoto kuwat - gonokami , masahito ueda , and takashi mukaiyama , phys . * 101 * , 100401 ( 2008 ) . j. fuchs , c. ticknor , p. dyke , g. veeravalli , e. kuhnle , w. rowlands , p. hannaford , and c. j. vale , phys . a * 77 * , 053616 ( 2008 ) . r. a. w. maier , c. marzok , c. zimmermann , and ph . w. courteille , phys . a * 81 * , 064701 ( 2010 ) . n. navon , s. nascimbne , f. chevy , and c. salomon , science * 328 * , 729 , ( 2010 ) , mark j. h. ku , ariel t. sommer , lawrence w. cheuk , martin w. zwierlein science * 335 * , 563 ( 2012 ) . cheng and s .- k . yip , phys . rev . b * 75 * , 014526 ( 2007 ) . li han , c. a. r. s de melo , phys . a * 85 * , 011606(r ) ( 2012 ) . kangjun seo , li han , c. a. r. s de melo , phys . a * 85 * , 033601 ( 2012 ) . kangjun seo , li han and c. a. r. s de melo , arxiv:1110.6364v1 ( 2011 ) . kangjun seo , li han , c. a. r. s de melo , phys . lett . * 109 * , 105303 ( 2012 ) . pengjun wang , zeng - qiang yu , zhengkun fu , jiao miao , lianghui huang , shijie chai , hui zhai , and jing zhang , phys . 109 * , 095301 ( 2012 ) . lawrence w. cheuk , ariel t. sommer , zoran hadzibabic , tarik yefsah , waseem s. bakr , and martin w. zwierlein , phys . lett . * 109 * , 095302 ( 2012 ) . lifshitz , zh . . fiz . * 38 * , 1569 ( 1960 ) [ sov . jetp * 11 * , 1130 ( 1960 ) ] .
we present detailed three - dimensional modelling of the stellar luminosity distribution for the disks of 31 relatively nearby ( 110 mpc ) edge - on spiral galaxies . in contrast to most of the standard methods available in the literature we take into account the full three - dimensional information of the disk . we minimize the difference between the observed 2d - image and an image of our 3d - disk model integrated along the line of sight . thereby we specify the inclination , the fitting function for the z - distribution of the disk , and the best values for the structural parameters such as scalelength , scaleheight , central surface brightness , and a disk cut - off radius . from a comparison of two independently developed methods we conclude , that the discrepancies e.g. for the scaleheights and scalelengths are of the order of . these differences are not due to the individual method itself , but rather to the selected fitting region , which masks the bulge component , the dust lane , or present foreground stars . other serious limitations are small but appreciable intrinsic deviations of real disks compared to the simple input model . + in this paper we describe the methods and present contour plots as well as radial profiles for all galaxies without previously published surface photometry . resulting parameters are given for the complete sample .
astro-ph0004044
global parameters of galactic disks can be used to constrain the formation process as well as the evolution of disk galaxies . recently it has become possible to deduce parameters such as scalelength or central surface density from numerical or semi - analytical self - consistent galaxy formation models ( syer et al . @xcite , mo et al . @xcite , dalcanton et al . @xcite ) and compare the results with observed values . observationally the hubble deep field gives the opportunity to study morphological features and simple structural parameters for galaxies even at high redshifts ( takamiya @xcite , marleau & simard @xcite , fasano & filippi @xcite ) . + former statistical studies providing sets of homogeneously derived parameters for nearby galaxy samples are those of de jong ( @xcite ) using a two - dimensional and of courteau ( @xcite ) with a one - dimensional decomposition technique . + however , so far only a few statistical studies based on high quality ccd data ( de grijs @xcite , barteldrees & dettmar @xcite , hereafter paper i ) have addressed the actual three - dimensional structure of disk galaxies with regard to the stellar distribution perpendicular to the plane ( z - direction ) and taking into account an outer truncation _ ( cut - off radius ) _ for the disk , first introduced by van der kruit & searle ( @xcite ) . + while the stellar distribution perpendicular to the plane ( z - profile ) could result from various `` heating '' processes during the galactic evolution ( toth & ostriker @xcite , hernquist & mihos @xcite ) , the cut - off radius can be used to constrain either the angular momentum of a protocloud ( van der kruit @xcite ) or possible starformation thresholds in the gaseous disk ( kennicutt @xcite ) . + in the following , we have compiled parameters of galactic disks for a total of 31 edge - on galaxies in different optical filters of which 17 objects have been already discussed in paper i. one goal of our detailed comparison of several independent fitting procedures is to study the influence of the applied techniques on resulting disk parameters . it also provides the data base for statistical analysis addressing some of the above mentioned questions ( e.g. , pohlen et al . @xcite ) . for 14 objects surface photometric data are given for the first time in appendix a. the observations were carried out at the 42-inch ( 1.06 m ) telescope of the lowell observatory located on the anderson mesa dark side during several nights in december 1988 ( run identification : l1 ) and january 1989 ( l2 ) and at the 2.2 m telescope at eso / la silla during three runs in june 1985 ( e1 ) , march ( e2 ) and june ( e3 ) 1987 . at the 42-inch telescope we used a 2:1 focal reducer with the f/8 secondary , equipped with a ccd camera which is based on a thinned ti 800x800 wfpc 1 ccd with 15 @xmath2 m pixelsize resulting in a field of approximately 9@xmath3 with a scale of 0.7@xmath4pixel@xmath5 . images were taken with a standard johnson r filter . observations at the 2.2 m telescope were carried out with the eso ccd adapter using a 512x320 rca chip , giving an effective field size of @xmath6x@xmath7 and a scale of 0.36@xmath4pixel@xmath5 . for the eso observing runs we used the g , r , and i filters of the thuan and gunn ( @xcite ) system . exposures were mainly taken in the g or r band , and only seven galaxies were observed in all three filters . the northern sample observed at lowell was selected automatically in an electronic version of the ugc - catalog ( nilson @xcite ) searching for galaxies with an inclination class _ 7 _ matching the field size . after visual inspection to check the inclination and remove interacting and disturbed galaxies , the observed sample was chosen out of the remaining galaxies according to the allocated observing time . + for the southern sky there is no comparable catalog providing information of inclinations directly . using the axial ratios given e.g. in the eso - lauberts & valentijn catalog ( @xcite ) will introduce a selection bias preferring late type galaxies with lower b / d ratio ( guthrie @xcite , bottinelli et al . @xcite ) . one way to avoid this is extending the first selection to much lower axis ratios , comparable to @xmath8 , and then checking the inclination by eye . therefore we selected the galaxies according to the field size of about @xmath9 from a visual inspection of film copies of the southern sky survey ( see paper i ) . + table [ galaxies ] gives a list of the resulting sample used during our fitting process , with a serial number , the principal galaxy name , the used filter , the integration time in minutes ( _ 4 _ ) , and the run identification label , whereas the @xmath10 marks images already published in paper i. further parameters are taken from the rc3 catalogue ( de vaucouleurs et al . @xcite ) : the right ascension and declination , the rc3 coded hubble - type , the hubble parameter t , and the d@xmath11 diameter in arcminutes . in the case of eso 578 - 025 parameters are taken from the eso - uppsala catalogue ( lauberts @xcite ) . r l r c l c c l [email protected] [email protected] c [email protected] no.&name&filter&@xmath12 & & ra & dec&rc3 & & & @xmath13 & + & & & [ min ] & & & type & & & km s@xmath5 & + ' '' '' & & & & & & & & & & & + + 1 & eso 112 - 004 & r&40&e3@xmath10 & 002804.2&@xmath14580611 & .s.r6*.&5&6 & 1&32 & & + 2 & eso 150 - 014 & r&20&e3@xmath10 & 003637.9&@xmath14565424 & .l .. +*/&-0&7 & 1&91 & 8257&107&05 + 3 & ngc 585 & r&20&l1 & 013142.5&@xmath14005555 & .s .. 1*/&1&0 & 2&14 & 5430&72&15 + 4 & eso 244 - 048 & r&15&e3 & 013908.8&@xmath14470742 & .s .. 3./&3&0 & 1&38 & 6745 & 87&09 + 5 & ngc 973 & r&10&l1 & 023420.2&+323020 & .s .. 3 .. &3&0 & 3&72 & 4853 & 66&28 + 6 & ugc 3326 & r&30&l1 & 053936.0&+771800 & .s .. 6*.&6&0 & 3&55 & 4085&57&82 + 7 & ugc 3425 & r&30&l1 & 061442.0&+663400 & .s .. 3 .. &3&0 & 2&51 & 4057&57&04 + 8 & ngc 2424 & r&15&l1 & 074039.8&+391359 & .sbr3*/&3&0 & 3&80 & 3113&43&10 + 9 & ic 2207 & r&10&l2 & 074950.8&+335743 & .s .. 6*.&6&0 & 2&04 & 4793&65&10 + 10&eso 564 - 027 & r&30&e2 & 091154.4&@xmath14200703 & .s .. 6*/&6&3 & 4&07 & 2177&26&93 + 11&eso 436 - 034 & g&60&e3@xmath10 & 103244.2&@xmath14283646 & .s .. 3./&3&0 & 2&09 & 3624&46&02 + 12&eso 319 - 026 & g&30&e3@xmath10 & 113020.0&@xmath14410357 & .s .. 5./&5&3 & 1&48 & 3601&45&32 + 12&eso 319 - 026 & i&30&e3 + 12&eso 319 - 026 & r&30&e2 + 12&eso 319 - 026 & r&30&e3@xmath10 + 13&eso 321 - 010 & g&30&e3@xmath10 & 121142.2&@xmath14383253 & .s .. 0*/&0&0 & 1&86 & 3147&39&53 + 13&eso 321 - 010 & r&30&e3@xmath10 + 14&ngc 4835a & r&40&e3@xmath10 & 125713.6&@xmath14462243 & .s .. 6*/&6&0 & 2&57 & 3389&42&54 + 15&eso 575 - 059 & r&15&e2 & 130744.5&@xmath14192348 & .la.+?/&-0&8 & 1&86 & 4570&59&78 + 16&eso 578 - 025 & g&30&e2 & 140815.5&@xmath14200019 & & 1&0 & 1&60 & 6364 & 83&93 + 16&eso 578 - 025 & g&30&e3@xmath10 + 16&eso 578 - 025 & i&30&e2 + 16&eso 578 - 025 & r&30&e2 + 17&eso 446 - 018 & r&30&e2 & 140838.7&@xmath14293412 & .s .. 3./&3&0 & 2&34 & 4774 & 62&15 + 18&ic 4393 & r&30&e2 & 141749.5&@xmath14312056 & .s .. 6?/&6&0 & 2&40 & 2754 & 35&15 + 19&eso 581 - 006 & r&30&e3@xmath10 & 145803.1&@xmath14192329 & .sbs7p/&7&0 & 1&70 & 3119 & 40&91 + 20&eso 583 - 008 & r&30&e3 & 155750.5&@xmath14222947 & .s? .... &6&0 & 1&51 & 7399 & 97&98 + 21&ugc 10535 & r&25&e2 & 164600.0&+062800 & .s .. 2 .. &2&0 & 1&10 & 7586&102&48 + 22&ngc 6722 & r&10&e3@xmath10 & 190339.6&@xmath14645341 & .s .. 3./&3&0 & 2&88 & 5749 & 73&79 + 23&eso 461 - 006 & r&60&e3@xmath10 & 195155.9&@xmath14315852 & .s .. 5./&5&0 & 1&62 & 5949 & 78&26 + 24&ic 4937 & g&20&e1 & 200518.0&@xmath14561520 & .s .. 3./&3&0 & 1&95 & 2337 & 28&68 + 24&ic 4937 & i&20&e1 + 24&ic 4937 & r&30&e3@xmath10 + 25&eso 528 - 017 & g&30&e3@xmath10 & 203320.8&@xmath14270549 & .sb.6?/&5&7 & 1&59 & 6115 & 80&73 + 25&eso 528 - 017 & i&30&e3 + 25&eso 528 - 017 & r&60&e3@xmath10 + 26&eso 187 - 008 & r&30&e3@xmath10 & 204325.2&@xmath14561217 & .s .. 6./&6&0 & 1&51 & 4412 & 56&31 + 27&eso 466 - 001 & i&40&e3 & 214232.3&@xmath14292210 & .s .. 2./&2&0 & 1&38 & 7068 & 93&13 + 28&eso 189 - 012 & g&60&e3@xmath10 & 215538.7&@xmath14545233 & .sa.5*/&5&0 & 1&66 & 8398&109&04 + 28&eso 189 - 012 & i&20&e3 + 28&eso 189 - 012 & r&30&e3@xmath10 + 29&eso 533 - 004 & r&20&e1 & 221403.2&@xmath14265618 & .s .. 5*/&4&8 & 2&34 & 2594 & 33&54 + 30&ic 5199 & g&30&e3@xmath10 & 221933.0&@xmath14373201 & .sa.3*/&3&0 & 1&55 & 5061 & 65&78 + 30&ic 5199 & i&30&e3 + 31&eso 604 - 006 & r&30&e3@xmath10 & 231454.0&@xmath14205944 & .s .. 4./&4&0 & 1&86 & 7636&100&92 + + [ galaxies ] we applied standard reduction techniques for bias subtraction , bad pixel correction and flatfielding . following the procedure described in paper i the sky background was fitted for each image using the edges of the individual frames to reduce any large scale inhomogeneity in the field of view . for part of the data we tried to remove the foreground stars from the image , but even with sophisticated psf fitting using iraf - daophot routines we were not able to remove stars without any confusion . the remaining residuals were always of the order of the discussed signal . this technique could only be used to mask out the regions affected by stars . in order to increase the signal - to - noise ratio near the level of the sky background part of the data was filtered using a weighted smoothing algorithm ( see paper i ) . thereby the noise was reduced by about one magnitude measured with a three sigma deviation on the background , whereas the interpretation of the faint structure is hampered by this process . + we therefore conclude that the best way is to omit any additional image modifications , besides a rotation of the image to the major axis of the disk . most of the images were taken during non - photometric nights , therefore we tried a different way to perform photometric calibration . comparing simulated aperture measurements with published integrated aperture data led to the best possible homogeneous calibration of the whole sample . most of the southern galaxies were calibrated using the catalogue of lauberts & valentijn ( @xcite ) , whereas ned was used for all northern galaxies . we used equation ( 1 ) derived in paper i for the colour transformation of r and b literature values and the g measurements , and no correction between the r and r , and i and i band , respectively . due to the fact that the photometric errors , of the input catalogues from lauberts & valentijn ( @xcite ) as well as within the rc3 catalogue ( de vaucouleurs et al . @xcite ) are of the order of @xmath15 mag , we do not apply any further corrections . galaxies calibrated in this way are marked with _ l _ in column of table [ modelle ] , whereas for galaxies which did not have published values for their magnitude in the observed filter , we interpolated from calibrated images of the same night , by comparing the count rates for the sky value . these galaxies are marked with an _ i_. for a few nights no galaxy with published photometry was observed and in these cases we used interpolated night sky values from the same observing run . images calibrated in this way are marked with _ e _ in table [ modelle ] . the resulting zero points and central surface brightness values in these cases should be interpreted carefully , although the derived structural parameters , like scalelength and scaleheight , are not influenced by any uncertainty in the flux calibration . + appendix a shows the contour plots and selected radial profiles for the 14 objects ( 25 images ) not already published in paper i. in order to derive the intrinsic values of the scale parameters and to compare physical dimensions we tried to estimate distances for our galaxies . therefore we took published radial velocities corrected for the virgo centric infall from the leda database , and estimated the distance following the hubble relation with a hubble constant of @xmath16 km s@xmath5mpc@xmath5 . table [ galaxies ] gives the heliocentric radial velocities according to leda , and our estimated distances . our disk model is based upon the fundamental work of van der kruit and searle ( @xcite , @xcite , @xcite , @xcite ) . they tried to find a fitting function for the light distribution in disks of edge - on galaxies . these galaxies are , compared to the face - on view , preferred for studying galactic disks due to the fact , that in this geometry it is possible to disentangle the radial and vertical stellar distribution . their model include an exponential radial light distribution found for face - on galaxies ( de vaucouleurs , @xcite ; freeman , @xcite ) , a sech@xmath17 behaviour in @xmath18 , which is expected for an isothermal population in a plan - parallel system ( camm , @xcite ; spitzer , @xcite ) , and a sharp edge of the disk , first observed by van der kruit ( @xcite ) in radial profiles of edge - on galaxies . the resulting luminosity density distribution for this symmetric disk model is ( van der kruit & searle @xcite ) : @xmath19 @xmath20 being the luminosity density in units of [ @xmath21 pc@xmath22 , @xmath23 the central luminosity density , @xmath24 and @xmath18 are the radial resp . vertical axes in cylinder coordinates , @xmath25 is the radial scalelength and @xmath26 the scaleheight , and @xmath27 is the cut - off radius . + the empirically found exponential radial light distribution is now well accepted and it is proposed that viscous dissipation could be responsible ( firmani et al . @xcite , struck - marcel @xcite ; saio & yoshii @xcite , lin & pringle @xcite ) , although there is so far no unique explanation for the disk being exponential . an alternative description of the form @xmath28 proposed by seiden et al . ( @xcite ) did not get much attention , although it emphasizes the empirical nature of the exponential fitting function . + to avoid the strong dust lane and to follow the light distribution down to the region @xmath29 wainscoat ( @xcite ) and wainscoat et al . ( @xcite ) carried out nir observations using the much lower extinction in this wavelength regime compared to the optical . they found a clear excess over the isothermal distribution and proposed the z - distribution to be better fitted by an exponential function @xmath30 . according to van der kruit ( @xcite ) such a distribution would led to a sharp minimum of the velocity dispersion in the plane , which is not observed ( fuchs & wielen @xcite ) . therefore he proposed @xmath31 as an intermediate solution . de grijs ( @xcite ) extended this to a family of density laws @xmath32 following van der kruit ( @xcite ) , where the isothermal ( @xmath33 ) , and the exponential ( @xmath34 ) cases represent the two extremes . + therefore the luminosity density distribution can be written as : @xmath35 with h@xmath36 being the heaviside function . + in order to limit the choice of parameters we restrict our models to the three main density laws for the z - distribution ( exponential , sech , and sech@xmath17 ) . due to the choice of our normalised isothermal case @xmath26 is equal to @xmath37 , where @xmath38 is the usual exponential vertical scale height : @xmath39 in contrast to paper i and barteldrees & dettmar ( @xcite ) we define the cut - off radius at the position where the radial profiles become nearly vertical , corresponding to the mathematical description . they tried to avoid any confusion due to the lower signal - to - noise in the outer parts , by fixing the cut - off radius where the measured radial profile begin to deviate significantly from the pure exponential fit . the model of the two dimensional surface photometric intensity results from an integration along the line of sight of the three dimensional luminosity density distribution ( [ hatl ] ) with regard to the inclination @xmath40 of the galaxy . describing the luminosity density of the disk in a cartesian - coordinate grid k(@xmath41-@xmath42-@xmath18 ) with @xmath43 leads to the following coordinate transformation into the observed inclined system k@xmath44(@xmath45-@xmath46-@xmath47 ) with @xmath45 pointing towards the observer , whereas the rotation angle between the two systems corresponds to @xmath48 : @xmath49 taking into account this transformation we have to integrate equation ( [ hatl ] ) , obtaining an equation for the intensity of the model disk depending on the observed radial and vertical axes @xmath46 and @xmath47 on the ccd : @xmath50 together with equation ( [ hatl ] ) this gives : @xmath51 \nonumber \hat{l}_0 \int\limits_{-\infty}^{+\infty}\!\ ! \exp{\left(-\frac{\sqrt{(x'\ \sin i - z'\ \cos i)^2+y'^2}}{h}\right ) } f_n(z',z_0)\\ \ast\ , { \rm h}\left(r_{\rm co}-\sqrt{(x'\ \sin i - z'\ \cos i)^2+y'^2}\right ) \:dx ' \quad \label{numcalhatl}\end{aligned}\ ] ] therefore six free parameters fit the observed surface intensity on the chip ( @xmath46 , @xmath47 plane ) to the model : @xmath52 figure [ galacticmodels ] shows a sequence of computed models with an isothermal z - distribution ( @xmath53 ) for the exact edge - on case ( @xmath54 ) with characteristic values for the ratio @xmath55 : @xmath56 and for @xmath57 : @xmath58 keeping the cut - off radius @xmath27 and the total luminosity @xmath59 constant . the latter is causing a different central surface brightness @xmath60 for each model , ranging from 23.8 to 19.3 starting with @xmath61 mag@xmath62 for the reference model with @xmath63 and @xmath64 . all contour lines falling within the interval @xmath65 @xmath66 are plotted with a spacing of 0.5 . the first step is to divide the galaxy into its quadrants . the four images are then averaged according to their orientation , following van der kruit & searle ( @xcite ) and shaw ( @xcite ) . thereby larger foreground stars and asymmetrical perturbations in the intensity distribution are eliminated by omitting this region during averaging . smaller foreground stars are removed by median filtering . the average quadrant should result at least from three quadrants to get a representative image of the galaxy . this averaging additionally increases the signal - to - noise ratio . in order to determine the fitting area for modelling the disk component on the final quadrant , one has to avoid the disturbing influence of the bulge component and the dust lane . the region dominated by the bulge is fixed following wyse et al . ( @xcite ) defining the bulge component by `` light that is in excess of an inward extrapolation of a constant scale - length exponential disk '' . therefore the clear increase of the intensity towards the center which can be seen in radial cuts determines an inner fitting boundary @xmath67 . we tried to minimize the dust influence ( cf . section [ dust ] ) by placing a lower limit @xmath68 in the vertical direction by visual inspection . additionally we restricted the remaining image by a limiting contour line @xmath69 , where the intensity drops below a limit of 3@xmath70 on the background . + we are aware of the problem that these are rather rough definitions difficult to reproduce without quoting the exact values for @xmath71 . however , the final choice of the fitting area is a complex and subjective procedure depending on the intrinsic shape of each individual galaxy , the influence of their environment , and the quality of the image itself . therefore it is not possible to quote exact general selection criteria and to derive the structural parameters straight forward . one solution is to do it in a consistent way for a large sample , leaving the problem of comparing results from different methods ( cf . section [ comp ] ) . the numerical realisation of the fitting procedure minimizes the difference ( @xmath72 ) between the averaged quadrant and a modelled quadrant based on equation ( [ numcalhatl ] ) . @xmath73 @xmath74 is the intensity within the average quadrant ( observed intensity ) and @xmath75 of the modelled intensity . in contrast to shaw & gilmore ( @xcite , @xcite ) and shaw ( @xcite ) using a similar approach for their models we do not weight individual pixels . they weight the difference by the error in the surface brightness measure , which shaw ( @xcite ) derives from the averaging of the quadrants . however , this method implies an absolute symmetry for the disk in @xmath18 and @xmath42 , which is not the case for real galaxies . these kind of errors only reflect the asymmetry of the galaxy . using the observed errors in the surface brightness for weighting individual pixels does not result in a considerable advantage because they are nearly the same after smoothing . the minimal @xmath72 is found by varying five of the six free parameters of the model ( cf . . [ param ] ) , whereas the parameter @xmath27 is determinated by cuts parallel to the major axis . a significant decrease of the intensity extrapolated to @xmath76 gives the value of @xmath27 ( van der kruit & searle , @xcite ) , therefore it is important that the intensity at the cut - off radius is well above the noise limit . the other five parameters are determined by fixing the smallest @xmath72 . for the three different functions @xmath77 and every possible @xmath40 ( @xmath78 ) the remaining parameters ( @xmath23 , @xmath25 , and @xmath26 ) are varied with the `` downhill simplex - method '' ( nelder & mead , @xcite ; press et al . , @xcite ) until the global minimum of @xmath72 is found . @xmath75 is calculated by a numerical gaussian integration of equation ( [ numcalhatl ] ) . the possible inclination angles can be restricted from the dust lane of the galaxy ( paper i ) . tests of model disks with additional noise show that the `` downhill simplex - method '' found the input inclination @xmath40 , the used model @xmath77 and the other disk parameters within errors of @xmath79 , and @xmath80 1% . an estimation of the errors of the parameters for the best model disk can be made by inspection of the parameter space for @xmath72 around the smallest @xmath72 , with slightly different fitting areas , and different values of @xmath27 . @xmath77 is in almost all cases the same and the variation of @xmath40 is only small ( @xmath81 ) . the differences in @xmath25 and @xmath26 are about 15% ( in some cases up to 25% ) , and @xmath23 varies about a factor of 2 . within this method all disk parameters as well as the choice of the optimum function @xmath77 are estimated by a direct comparison of calculated and observed disk profiles by eye . + as a first step , the inclination @xmath40 is determined by using the axis ratio of the dust lane . depending on the shape of the dust lane it is possible to restrict the inclination to @xmath82 . the central luminosity density @xmath23 is calculated automatically for each new parameter set by using a number of preselected reference points along the disk . given a sufficient signal - to - noise ratio , the cut - off radius @xmath27 can be determined from the major axis profile . thus , the remaining fitting parameters are the disk scale length and height , @xmath25 and @xmath26 , as well as the set of 3 functions @xmath77 . the scale length @xmath25 is fitted to a number ( usually between 4 and 6 ) of major axes profiles ( left panel of fig . [ uwemod ] ) . the quality of the fit for the vertical profiles along the disk can be used as a cross - check for the estimated scalelengths ( right panel of fig . [ uwemod ] ) . the disk scaleheight @xmath26 is estimated by fitting the @xmath18-profiles , outside a possible bulge or bar contamination . the vertical disk profiles of most of the galaxies investigated enable a reliable choice of the quantitatively best fitting function @xmath77 . this is due to the fact that the deviations between different functions become visible at vertical distances larger than that of the most sharply - peaked dust regions . the first raw fitting steps are usually carried out by using a reduced number of both major and minor axis profiles simultaneously . afterwards , when a good first fitting quality is reached , a complete set of major and minor calculated and observed axes profiles are investigated in detail . at the end of this procedure a final , complete disk model is calculated using all previously estimated disk parameters . table [ modelle ] contains the best fit model for each image . together with the galaxy name , the filter , and the referring image , with integration time , and run i d , we list the inclination , the best fitting function for the z - distribution , the calibration index ( ref . section [ calib ] ) , and the central surface brightness of the model , without correcting for inclination . according to the distance tabulated in table [ galaxies ] , the cut - off radius @xmath27 is given in kpc and arcsec as well as the scalelength @xmath25 and the vertical scaleheight @xmath26 which is normalised to the isothermal case , being two times an exponential scaleheight @xmath38 . for the seven galaxies with available images in more than one filter , we do not see any correlation of fitted parameters with different wavelength , although we find the same inclination angle for the best fitted disk within the range of the errors . appendix a shows the best fitting model as an overlay to selected radial profiles for each image . the subsequent analysis of the distribution for the different parameters concerning the formation and evolution of galaxies will be given in forthcoming papers . l c c [email protected] lcc [email protected] [email protected] [email protected]@.l [email protected]@.l galaxy&filter&image&&@xmath77&cali.&@xmath83 & & & & & + & & & & & & mag@xmath62 & & & & & & & & + ' '' '' & & & & & & & & & & & + + eso 112 - 004 & r&40e3 & 87&5 & sech & l & 21.14 & 45&0 & & & 22&7 & & & 2&8 & + eso 150 - 014 & r&20e3 & 90&0 & sech & l & 22.00 & 64&1 & 33&27 & & 23&4 & 12&14 & & 5&5 & 2&85 + ngc 585 & r&20l1 & 88&0 & sech & e & 21.76 & 70&0 & 24&49 & & 32&7 & 11&44 & & 10&6 & 3&71 + eso 244 - 048 & r&15e3 & 87&0 & @xmath84 & l & 20.54 & 45&4 & 19&17 & & 13&7 & 5&78 & & 7&2 & 3&04 + ngc 973 & r&10l1 & 89&5 & @xmath84 & e & 20.83 & 105&0 & 33&74 & & 51&5 & 16&55 & & 12&4 & 3&99 + ugc 3326 & r&30l1 & 88&0 & sech@xmath17&e & 21.50 & 101&5 & 28&45 & & 70&4 & 19&74 & & 4&7 & 1&32 + ugc 3425 & r&30l1 & 87&0 & sech@xmath17&e & 21.01 & 80&5 & 22&26 & & 29&2 & 8&08 & & 7&4 & 2&05 + ngc 2424 & r&15l1 & 86&5 & @xmath84 & e & 20.52 & 112&0 & 23&40 & & 31&7 & 6&62 & & 11&9 & 2&49 + ic 2207 & r&10l2 & 86&5 & @xmath84 & e & 21.04 & 56&0 & 17&67 & & 38&2 & 12&06 & & 6&7 & 2&12 + eso 564 - 027 & r&30e2 & 88&0 & sech & l & 21.11 & 140&4 & 18&33 & & 50&9 & 6&65 & & 6&6 & 0&86 + eso 436 - 034 & g&60e3 & 88&0 & sech & l & 20.96 & 82&1 & 18&32 & & 22&6 & 5&04 & & 6&8 & 1&52 + eso 319 - 026 & g&30e3 & 86&5 & sech & e & 21.51 & 60&1 & 13&21 & & 14&2 & 3&12 & & 3&1 & 0&68 + eso 319 - 026 & r&30e2 & 86&0 & @xmath84 & i & 21.14 & 63&0 & 13&84 & & 14&7 & 3&23 & & 3&2 & 0&70 + eso 319 - 026 & r&30e3 & 88&0 & sech@xmath17&i & 21.94 & 61&9 & 13&60 & & 13&2 & 2&90 & & 2&8 & 0&62 + eso 319 - 026 & i&30e3 & 88&0 & sech & l & 21.01 & 64&8 & 14&24 & & 14&3 & 3&14 & & 3&4 & 0&75 + eso 321 - 010 & g&30e3 & 88&0 & sech & l & 20.11 & 64&4 & 12&34 & & 19&9 & 3&81 & & 4&7 & 0&90 + eso 321 - 010 & r&30e3 & 88&0 & sech & l & 19.54 & 64&8 & 12&42 & & 21&6 & 4&14 & & 5&0 & 0&96 + ngc 4835a & r&40e3 & 85&5 & @xmath84 & l & 20.92 & 90&0 & 18&56 & & 40&4 & 8&33 & & 7&9 & 1&63 + eso 575 - 059 & r&15e2 & 87&0 & sech@xmath17&l & 20.78 & 60&5 & 17&53 & & 22&6 & 6&55 & & 6&4 & 1&86 + eso 578 - 025 & g&30e2 & 86&5 & @xmath84 & l & 21.58 & 50&7 & 20&63 & & 17&0 & 6&92 & & 7&5 & 3&05 + eso 578 - 025 & g&30e3 & 86&5 & @xmath84 & l & 21.69 & 50&7 & 20&63 & & 17&0 & 6&92 & & 7&5 & 3&05 + eso 578 - 025 & r&30e2 & 86&0 & @xmath84 & l & 21.02 & 50&4 & 20&51 & & 14&9 & 6&06 & & 7&4 & 3&01 + eso 578 - 025 & i&30e2 & 86&0 & @xmath84 & l & 20.27 & 47&5 & 19&33 & & 20&3 & 8&26 & & 7&6 & 3&09 + eso 446 - 018 & r&30e2 & 86&5 & sech & l & 20.45 & 75&6 & 22&78 & & 23&7 & 7&14 & & 3&9 & 1&18 + ic 4393 & r&30e2 & 87&0 & @xmath84 & l & 20.30 & 75&6 & 12&88 & & 29&8 & 5&08 & & 5&8 & 0&99 + eso 581 - 006 & r&30e3 & 86&5 & @xmath84 & l & 21.33 & 55&8 & 11&07 & & 18&2 & 3&61 & & 5&6 & 1&11 + eso 583 - 008 & r&30e3 & 87&0 & @xmath84 & i & 20.84 & 56&2 & 26&70 & & 14&0 & 6&65 & & 3&8 & 1&81 + ugc 10535 & r&25e2 & 88&0 & sech@xmath17&i & 21.50 & 41&4 & 20&57 & & 11&4 & 5&66 & & 4&3 & 2&14 + ngc 6722 & r&10e3 & 86&5 & sech & l & 19.47 & 86&0 & 30&77 & & 21&6 & 7&73 & & 6&2 & 2&22 + eso 461 - 006 & r&60e3 & 87&5 & @xmath84 & l & 20.96 & 61&6 & 23&37 & & 21&1 & 8&01 & & 3&8 & 1&44 + ic 4937 & g&20e1 & 88&5 & sech@xmath17&l & 22.73 & 74&9 & 10&41 & & 93&8 & 13&04 & & 5&6 & 0&78 + ic 4937 & r&30e3 & 88&0 & sech@xmath17&l & 21.29 & 83&5 & 11&61 & & 32&1 & 4&46 & & 5&9 & 0&82 + ic 4937 & i&20e1 & 88&5 & sech & e & 19.92 & 78&1 & 10&86 & & 38&6 & 5&37 & & 7&2 & 1&00 + eso 528 - 017 & g&30e3 & 86&5 & @xmath84 & l & 21.98 & 57&6 & 22&54 & & 20&5 & 8&02 & & 3&5 & 1&37 + eso 528 - 017 & r&60e3 & 86&5 & @xmath84 & l & 21.28 & 55&1 & 21&56 & & 19&9 & 7&79 & & 2&7 & 1&06 + eso 528 - 017 & i&30e3 & 86&5 & sech & l & 20.89 & 50&0 & 19&57 & & 22&6 & 8&85 & & 3&2 & 1&25 + eso 187 - 008 & r&30e3 & 85&5 & @xmath84 & l & 20.77 & 50&0 & 13&65 & & 15&1 & 4&12 & & 4&4 & 1&20 + eso 466 - 001 & i&40e3 & 87&0 & @xmath84 & e & 19.52 & 52&6 & 23&76 & & 13&0 & 5&87 & & 8&2 & 3&70 + eso 189 - 012 & g&60e3 & 87&0 & @xmath84 & l & 21.50 & 56&2 & 29&81 & & 26&8 & 14&21 & & 3&8 & 2&02 + eso 189 - 012 & r&30e3 & 86&5 & @xmath84 & l & 20.71 & 56&5 & 29&97 & & 20&6 & 10&93 & & 3&4 & 1&80 + eso 189 - 012 & i&20e3 & 87&0 & sech & l & 20.44 & 54&7 & 29&01 & & 22&7 & 12&04 & & 3&2 & 1&70 + eso 533 - 004 & r&20e1 & 88&0 & @xmath84 & l & 20.32 & 68&4 & 11&12 & & 33&1 & 5&38 & & 7&7 & 1&25 + ic 5199 & g&30e3 & 86&5 & @xmath84 & l & 21.55 & 64&1 & 20&44 & & 19&9 & 6&35 & & 5&6 & 1&79 + ic 5199 & i&30e3 & 86&5 & @xmath84 & l & 19.97 & 57&6 & 18&37 & & 19&9 & 6&35 & & 4&9 & 1&56 + eso 604 - 006 & r&30e3 & 90&0 & sech & l & 21.22 & 70&6 & 34&54 & & 27&9 & 13&65 & & 3&8 & 1&86 + [ modelle ] the different fitting methods were independently developed within two diploma theses ( ltticke 1996 , schwarzkopf 1996 ) . the quality of the data basis for each project was the same . from the sample presented here there were five objects in common . these are used to compare the two methods and determine the quantitative difference of the derived parameters . r l c c c c cl galaxy&@xmath77 & @xmath40 & @xmath18 & @xmath25 & @xmath27 + & & [ 1.0ex][@xmath85 & [ 1.0ex ] [ @xmath86 & [ 1.0ex ] [ @xmath86 & [ 1.0ex ] [ @xmath86 + + & sech&88.0&5.0&21.6&64.8 + [ 1.1ex]eso 321 - 010 r \{&@xmath84&88.0&5.2&25.9&64.1 + & @xmath84&87.5&3.8&21.1&61.6 + [ 1.1ex]eso 461 - 006 r \{&@xmath84&87.5&3.6&16.2&61.2 + & @xmath84&86.5&6.1&26.8&83.5 + [ 1.1ex]ic 4937 r \{&sech&89.5&7.0&27.4&75.6 + & @xmath84&86.5&3.2&19.3&56.2 + [ 1.1ex]eso 189 - 012 r \{&@xmath84&88.0&3.6&21.3&56.9 + & sech & 88.5&4.1&39.8&67.7 + [ 1.1ex]eso 604 - 006 r \{&sech@xmath17&89.5&3.2&24.5&73.4 + table [ verglrlus ] shows the results for the five images . the mean deviation in the determined inclination is @xmath87 and 12.4% for the scaleheight ( ranging from 5.0%-26.6% ) whereas for three images different functions for the z distribution were used . the mean difference for the radial scalelength is 20.6% ( 2.1%-47.2% ) and 4.2% for the determination of the cut - off radius . + a subsequent analysis shows that it is not possible to ascribe the sometimes quite large discrepancies to the quality of the individual method . it turns out , that the main problem is the non - uniform determination of the fitting area . the intrinsic asymmetric variations of a real galactic disk compared to the model enforce a more subjective restriction of the galaxy image to the fitting region , whereby for example one has to exclude the bulge area and the dust lane . + this finding is in agreement with the study of knapen & van der kruit ( @xcite ) who compared published values of the scalelength and find an average value of 23% for the discrepancy between different sources . as already mentioned by schombert & bothun ( @xcite ) the limiting factor for accuracy of the decomposition is not the typical s / n from the ccd - telescope combination nor the errors in the determination of the sky background , but the deviation of real galaxies from the standard model . in our former study ( paper i ) with an earlier method to adapt equation ( [ numcalhatl ] ) , 20 of our 45 galaxy images have already been used . we decided to re - use them in this study to get models for as many galaxies as possible in a homogeneous way . additionally , paper i only presents the best fit values for the isothermal model , and uses a different definition of the cut - off radius . + only three galaxies are in common with the sample of de grijs ( @xcite ) : eso 564 - 027 , eso 321 - 010 , and eso 446 - 018 . the mean difference for the scalelength is 10.4 % ( ranging from 1.2%-20.3% ) and for the scale height ( normalised to the isothermal case ) 4.0% ( 0.0%-8.2% ) . for the remaining galaxies there are no models in the literature . our model only represents a rather simple axisymmetric three - dimensional model for a galactic disk , consisting of an one component radial exponential disk with three different laws for the density distribution in the z - direction and a sharp outer truncation . therefore it does not include additional components , such as bulges , bars , thick disks , or rings , and can not deal with any asymmetries . features like spiral structure or warps are not included , whereas reshetnikov & combes ( @xcite ) multiply their exponential disk by a spiral function introducing an expression to characterise an intrinsic warp depending on the position angle outside of a critical radius . + the choice of our fitting area tries to avoid the dust lane , possible only for almost edge - on galaxies , as a first step to account for the dust influence ( cf . section [ dust ] ) . examples of models including a radiative transfer with an extinction coefficient @xmath88 can be found in xilouris et al . ( @xcite ) . however , introducing more and more new components and features automatically increases the amount of free parameters . therefore we restricted our model to the described six parameters , to obtain statistically meaningful characteristics for galactic disks . + in the following we demonstrate that a simple disk model omitting the bulge component and the dust lane give indeed reasonable parameters . we have studied the influence of the bulge for some of our objects including the earliest type galaxy in our sample ( eso 575 - 059 ) presented here . we have subtracted our derived disk model from the galaxy and then tried to find the best representation for the remaining bulge by a de vaucouleurs @xmath89 or an exponential model . taking the slope of the vertical profile at @xmath90 and a fixed axis ratio , we have constructed the 2-dimensional model of the bulge . in agreement with andredakis et al . ( @xcite ) we find , that bulges of early type galaxies are better fitted by an exponential profile than by a @xmath89 . figure [ bulge ] shows the resulting vertical and radial cuts for eso 575 - 059 together with the models . despite the deviation between @xmath91 and @xmath92 which could be attributed to an additional component ( inner disk or bar ) , we do not find any evidence for changing our disk model due to the influence of the bulge . + therefore we conclude , that it is possible to nearly avoid any influence of the central component by fitting outside the clearly visible bulge region . dust disturbs the light profile by a combination of absorption and extinction and the net effect has to be calculated by radiation transfer models . therefore it is not obvious that outside the `` visible '' dust lane , which is excluded for the fitting area as a first step , the dust will not play a major role in shaping the light distribution . xilouris et al . ( @xcite ) , bianchi et al . ( @xcite ) , de jong ( @xcite ) , and byun et al . ( @xcite ) have recently addressed this problem in more detail . although they investigated the influence of the dust on the light distribution by quoting best fit structural parameters for the star - disk as well as the dust - disk , they did not quantify the influence on the star - disk parameters derived by standard fitting methods without dust . even kylafis & bahcall ( @xcite ) state within their fundamental paper on finding the dust distribution for ngc 891 that `` in order to avoid duplication of previous work , we will take ... ( the values for the star - distribution estimated by standard fitting methods ) '' . + we checked the influence of the dust on our determined parameter set , by studying simulated galaxy images with three different dust distributions . these _ dusty galaxies _ were kindly provided by simone bianchi who calculated images with known input parameters for the star and dust distribution with his monte carlo radiative transfer method ( bianchi et al . we have defined a _ worst _ , _ best _ , and _ transparent _ case , according to the dust distributions presented by xilouris et al . ( @xcite ) , of our mean stellar disk ( @xmath93 , @xmath94 , and @xmath95 ) . worst _ case is calculated with @xmath96 , @xmath97 , and @xmath98 , the _ best _ case with @xmath99 , @xmath100 , and @xmath101 , and a _ transparent _ case without dust . to be comparable we used the same method for selecting the fitting region by masking the `` visible '' dust lane and reserving a typical area for a possible bulge component using mean values for the _ transparent _ case . in contrast to our standard procedure we do not restrict the inclination range from the appearance of the dust lane , but specify the best @xmath72 model in the range @xmath102 in table [ dustvgl ] . for the models marked with a @xmath10 we pretend the correct input inclination . table [ dustvgl ] demonstrates , that even for the worst case we are able to reproduce the input parameters within the range of the typical 20% error discussed in section [ comp ] . it should be mentioned that for each case we overestimate the input scalelength and -height , whereas the determination of @xmath27 does not depend on the dust distribution . the implication on the distribution of the ratio @xmath55 will be discussed in a forthcoming paper . .comparison of the results for the stellar disk ( input : @xmath103 , @xmath104 , and @xmath105exp ) from our fitting procedure for the three different dust distributions . [ cols="<,^,^,^,^,^ , > , > " , ] trying to adapt a simple , perfect , and exact symmetric model to real galaxies always implies a compromise between the degree of any deviation and the final model ( section [ pec ] ) . the following list will provide some typical caveats found during the fit procedure which will characterise the quality of the specified model for individual galaxies . + * eso 112 - 004 * : warped , asymmetric , central part slightly tilted compared to disk , after fitting still remaining residuals + * eso 150 - 014 * : slightly warped , minor flatfield problems + * ngc 585 * : remaining residuals + * eso 244 - 048 * : possible two component system , slope of inner radial profile significantly higher than of an outer one , final model fits the inner parts + * ngc 973 * : one side disturbed by stray light of nearby star , seems to be radially asymmetric , remaining residuals + * ugc 3425 * : superimposed star on one edge + * ngc 2424 * : model does not fit very well without obvious reason + * eso 436 - 034 * : strong bulge component , possibly barred , hard to pinpoint final model , remaining residuals + * eso 319 - 026 * : outer parts show u - shaped behaviour , remaining residuals , therefore large ( @xmath106 ) difference in inclination angle + * eso 321 - 010 * : u - shaped , no clear major axis visible , therefore uncertain rotation angle , bar visible , bulge rotated against disk + * ngc 4835a * : strong residuals + * eso 446 - 018 * : the different sides of the disk are asymmetric visible in radial profiles and on the contour plot + * ic 4393 * : similar to ngc 4835a + * eso 581 - 006 * : galaxy shows typical late type profile , @xmath27 questionable , but nevertheless final model seems to fit well + * eso 583 - 008 * : disturbed by superimposed star , shows warp feature and a bar structure , @xmath27 questionable , remaining residuals + * ugc 10535 * : one side slightly extended + * ngc 6722 * : only one side observed , bulge rotated against disk , barred , strongly disturbed by dust absorption , radial extension visible , therefore @xmath27 should be treated with caution + * eso 461 - 006 * : minor flatfield problem seems to cause asymmetry , although model looks fine + * ic 4937 * : similar to ngc 6722 , dominating bulge , small disk , model significantly different compared to the i and r image , model possibly hampered by strong dust lane + * eso 578 - 025 * : bar visible + * eso 466 - 001 * : maybe two components , final model represents only inner part , outer part clearly different from normal disk component + * eso 189 - 012 * : slightly warped + * eso 533 - 004 * : similar to ngc 4835a , model fits the whole galaxy , leaving more or less no bulge component + * ic 5199 * : slightly radial asymmetric + * eso 604 - 006 * : only one side observed , bar structure visible the model limitations described above constrain the application of our fitting process . therefore we had to exclude about 20 galaxies from our original sample . they all show significant deviations from the simple geometry and an inclusion of their parameters obtained by forcing the model to fit the data will spoil the resulting parameter distribution . + one larger group classified mainly as s0 galaxies ( e.g. ngc 2549 , eso 376 - 009 , ngc 7332 , eso 383 - 085 , eso 506 - 033 ) shows a completely different behaviour of the luminosity distribution in the outer parts compared to the other galaxies . they all show an additional component , mainly characterised as an elliptical envelope . this is already visible in the contour plot , but becomes even more evident in a radial cut parallel to the major axis . in these cases the usual common curved decline of the profile ( e.g. eso 578 - 025 ) is missing , and is replaced by a more or less straight decline into the noise level , sometimes even by an upwards curved profile . fitting these luminosity distribution by our one component exponential disk with cut - off , will therefore naturally provide parameters qualitatively different compared to late type disks . this will be discussed in detail in a forthcoming paper . + another group consists of galaxies dominated mainly by their bulges , whereas the disk is only an underlying component , partly characterized as having thick boxy bulges ( dettmar & ltticke @xcite ) , e.g. ic 4745 , eso 383 - 005 , although there are also pure elliptical bulges ( e.g. eso 445 - 049 , ngc 6948 ) . + in the case of eso 383 - 048 and eso 510 - 074 the radial profiles clearly indicate that a more complex model will be needed to fit these kind of multicomponent galaxies . galaxies like ugc 7170 or eso 113 - 006 were excluded due to their strong warps , which made it impossible to fit the model in a consistent way . mainly late type galaxies such as eso 385 - 008 , ic4871 , ugc 1281 , or eso 376 - 023 show a too patchy and asymmetric light distribution , that any attempt to fit the profiles will give only very crude , low quality parameters . ugc 11859 and ugc 12423 were rejected due to their thin faint disks , which will maybe overcome by taking new images with longer integration times to get a higher signal - to - noise ratio , whereas ngc 5193a is completely embedded into the surface brightness distribution of its near companion . this work was supported by the _ deutsche forschungsgemeinschaft , dfg_. this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . we have made use of the leda database ( www-obs.univ-lyon1.fr ) . the authors wish to thank simone bianchi , who kindly provided us _ dusty - galaxy images _ produced with his radiative transfer code . andredakis y.c . , peletier r. , balcells m. , 1995 , mnras 275 , 874 barteldrees a. , dettmar r .- j . , 1989 , in : dynamics and interaction of galaxies , ed . wielen , springer - verlag , heidelberg , p.348 barteldrees a. , dettmar r .- j . , 1994 , a&as 103 , 475 ( paper i ) bianchi s. , ferrara a. , giovanardi c. , 1996 , apj 465 , 127 bottinelli l. , gouguenheim l. , paturel g. , de vaucouleurs g. , 1983 , a&a 118 , 4 byun y. i. , freeman k. c. , kylafis n. d. , 1994 , apj 432 , 114 camm g.l . , 1950 , mnras 110 , 305 courteau s. , 1996 , apjs 103 , 363 dalcanton j.j . , spergel d.n . , summers f.j . , 1997 , apj 482 , 659 de grijs r. , peletier r.f . , van der kruit p.c . , 1997 , a&a 327 , 966 de grijs r. , 1998 , mnras 299 , 595 de jong r.s . , 1996 , a&a 313 , 377 de jong r.s . , 1996 , a&as 118 , 557 de vaucouleurs g. , 1959 , in : handbuch der physik liii , ed . flgge s. , springer - verlag berlin , p. 275 de vaucouleurs g. , de vaucouleurs a. , corwin jr . buta r.j . , paturel g. , fouqu p. , 1991 , third reference catalogue of bright galaxies , springer - verlag new york dettmar r .- j . , ltticke r. , 1999 , in : asp conference series volume 165 , eds . gibson b.k . , axelrod t.s . , putman m.e . , p. 95 fasano g. , filippi m. , 1998 , a&as 129 , 583 firmani c. , hernandez x. , gallagher j. , 1996 , a&a 308 , 403 guthrie b.n.g . , 1992 , a&as 93 , 255 freeman k.c . , 1970 , apj 160 , 811 fuchs b. , wielen r. , 1987 , in : the galaxy , eds . gilmore g. , carswell , d. reidel publishing co. dordrecht , p.375 hernquist l. , mihos j.c . , 1995 apj 448 , 41 kennicutt r.c . , 1989 , apj 344 , 685 knapen j.h . , van der kruit p.c . , 1991 , a&a 248 , 57 kylafis n.d . , bahcall j.n . , 1987 , apj 317 , 637 lauberts a. , 1982 , the eso / uppsala survey of the eso(b ) atlas , eso lauberts a. , valentijn e.a . , 1989 , the surface photometry catalogue of the eso - uppsala galaxies , european southern observatory , garching lin d.n.c . , pringle j.e . , 1987 , apj 320 , l87 ltticke r. , 1996 , diploma thesis , ruhr - universitt bochum marleau f.r . , simard l. 1998 , apj 507 , 585 mo h.j . , mao s. , white s.d.m . , 1998 , mnras 295 , 319 nelder j.a . , mead r. , 1965 , computer journal , vol . 7 , p. 308 nilson p. , 1973 , uppsala general catalogue of galaxies , uppsala pohlen , m. , dettmar , r .- j . , ltticke r. , 2000 , a&a 357 , l1 press w.h . , fannery b.p . , teukolsky s.a . , vetterling w.t . , 1988 , in : numerical recipes in c , cambridge university press reshetnikov v. , combes f. , 1998 , a&a 337 , 9 saio h. , yoshii y. , 1990 , apj 363 , 40 schombert j.m . , bothun g. , 1987 , aj 92 , 60 schwarzkopf u. , 1996 , diploma thesis , ruhr - universitt bochum seiden p. e. , schulman l. s. , elmegreen b. g. , 1984 , apj 282 , 95 shaw m.a . , 1993 , mnras 261 , 718 shaw m.a . , gilmore g. , 1989 , mnras 237 , 903 shaw m.a . , gilmore g. , 1990 , mnras 242 , 59 spitzer l. , 1942 , apj 95 , 329 struck - marcell c. , 1991 , apj 368 , 348 syer d. , mao s. , mo h.j . , 1999 , mnras 305 , 357 takamiya m. , 1999 , apjs 122 , 109 thuan t.x . , gunn j.e . , 1976 , pasp 88 , 543 toth g. , ostriker j.p . , 1992 apj 389 , 5 van der kruit p.c . , 1979 , a&as 38 , 15 van der kruit p.c . , 1987 , a&a 173 , 59 van der kruit p.c . , 1988 , a&a 192 , 117 van der kruit p.c . , searle , l. , 1981a , a&a 95 , 105 van der kruit p.c . , searle , l. , 1981b , a&a 95 , 116 van der kruit p.c . , searle , l. , 1982a , a&a 110 , 61 van der kruit p.c . , searle , l. , 1982b , a&a 110 , 79 wainscoat r.j . , 1986 , in : ph.d . thesis , australian national university wainscoat r.j . , freeman k.c . , hyland a.r , 1989 , apj 337 , 163 wyse r.f.g . , gilmore g. , franx m. , 1997 , ara&a 35 , 637 xilouris e.m . , byun y.i . , kylafis n.d . , paleologou e.v . , papamastorakis p. , 1999 a&a 344 , 868 the following figures show in the top panel selected cuts parallel to the major axis ( full line ) as well as the best fit model ( dashed line ) according to table [ modelle ] . the bottom panel shows isophote maps of the surface brightness for the galaxies , rotated to the major axis . the faintest plotted contour is defined by a 3@xmath70 criterion of the background . table [ plots ] lists the name , image , and the vertical positions of the plotted profiles parallel to the major axes in arcsec . the contours are plotted equally spaced with 0.5 mag@xmath62 starting from the limiting surface brightness @xmath107 . & & & + + ngc 585 & r20l1 & 3,6,9,12 & 24.7 + eso 244 - 048 & r15e3 & @xmath143,@xmath149 & 24.8 + ngc 973 & r10l1 & @xmath142,@xmath144,@xmath146,@xmath148 & 23.8 + ugc 3326 & r30l1 & 3,6,9 & 24.8 + ugc 3425 & r30l1 & @xmath141,@xmath145,@xmath149 & 24.2 + ngc 2424 & r15l1 & 1,3,5,7 & 24.5 + ic 2207 & r10l2 & 2,6,10 & 24.1 + eso 564 - 027 & r30e2 & 2,4,6 & 25.2 + eso 319 - 026 & i30e3 & 2,4,6 & 24.4 + eso 319 - 026 & r30e2 & 2,4,6 & 24.7 + eso 575 - 059 & r15e2 & 3,6,9 & 24.4 + eso 578 - 025 & g30e2 & 2,4,6 & 25.2 + eso 578 - 025 & i30e2 & 2,4,6 & 23.3 + eso 578 - 025 & r30e2 & 2,4,6 & 24.5 + eso 446 - 018 & r30e2 & 2,4,6,8 & 25.1 + ic 4393 & r30e2 & @xmath143,@xmath1096,@xmath1099 & 25.1 + eso 583 - 008 & r30e3 & 2,4,6 & 25.0 + ugc 10535 & r25e2 & 2,4,6 & 25.5 + ic 4937 & g20e1 & @xmath142,@xmath144,@xmath146 & 25.1 + ic 4937 & i20e1 & @xmath142,@xmath144,@xmath146 & 23.3 + eso 528 - 017 & i30e3 & 2,4,6 & 24.2 + eso 466 - 001 & i40e3 & 1,3,5 & 23.8 + eso 189 - 012 & i20e3 & @xmath141,@xmath143,@xmath145 & 24.1 + eso 533 - 004 & r20e1 & @xmath143,@xmath146,@xmath149 & 24.6 + ic 5199 & i30e3 & @xmath1092,@xmath1094,@xmath1096 & 23.8 + +
accretion onto supermassive black holes produces both the dramatic phenomena associated with active galactic nuclei and the underwhelming displays seen in the galactic center and most other nearby galaxies . i review selected aspects of the current theoretical understanding of black hole accretion , emphasizing the role of magnetohydrodynamic turbulence and gravitational instabilities in driving the actual accretion and the importance of the efficacy of cooling in determining the structure and observational appearance of the accretion flow . ongoing investigations into the dynamics of the plunging region , the origin of variability in the accretion process , and the evolution of warped , twisted , or eccentric disks are summarized .
astro-ph0405170
most galactic nuclei are now believed to harbor supermassive black holes . these black holes are all accreting gas at a minimum from the interstellar medium proximate to the event horizon , and in some cases from dense disks of gas in orbit around the hole . the observational signatures of this accretion , however , differ dramatically from galaxy to galaxy . the black hole in the galactic center , for example , has an x - ray luminosity in quiescence of only @xmath0 , reaching @xmath1 during flaring states ( @xcite , 2003 ; @xcite ) . many nearby elliptical galaxies also show nuclear emission that is much weaker than might be expected on the basis of simple estimates of the black hole accretion rate ( @xcite ; @xcite ; @xcite ) . in stark contrast to these feeble displays , accretion onto black holes in quasars powers the most luminous steady sources in the universe . at a minimum , a theory of accretion needs to account for this dichotomy , and to explain at least some of the many secondary phenomena ( jets , outflows , variability , etc . ) associated with active galactic nuclei ( agn ) . more ambitiously , one might hope to understand the role of accretion in actually forming supermassive black holes . black hole formation is a difficult theoretical problem which , although currently untroubled by direct observations , is receiving increasing attention . in this chapter , i discuss selected aspects of the theory of accretion onto supermassive black holes , with an emphasis on the physical processes that drive accretion and determine the qualitative properties of the flow . in [ secpjatransport ] , i outline the mechanisms which can lead to angular momentum transport within an accretion flow , thereby allowing rotationally supported gas to flow inward and liberate gravitational potential energy . close to the black hole , turbulence driven by magnetohydrodynamic instabilities is probably the dominant mechanism for transport . further out , other processes , such as gravitational instabilities , are likely to assume that role . angular momentum transport , however , is only part of the story . although central to our understanding of accretion , knowledge of its origin does not suffice to explain why the galactic center looks nothing like a powerful agn . for that we need to consider the distinction between geometrically thin accretion disks , in which the gas can radiate efficiently and cool to sub - virial temperatures , and hot thick disks , which are radiatively inefficient and vulnerable to rapid mass loss . these questions are addressed in [ secpjaaccrate ] . subsequent sections examine the status of several open questions in the study of black hole accretion , including the dynamics of gas executing its final plunge into the hole , the origin of variability , and the evolution of disks that are warped or eccentric . in almost all circumstances , gas in the nuclear regions of galaxies has far too much angular momentum to be swallowed directly by the black hole . understanding the mechanisms that can lead to angular momentum transport in rotating flows is therefore the central problem in the theory of accretion onto supermassive black holes . at scales of @xmath2 pc or larger , this is a problem in galactic dynamics . following galactic mergers , for example , gravitational torques from bars and other transient structures can efficiently funnel gas into the nuclear regions ( shlosman , frank , & begelman 1989 ; @xcite ; @xcite ) . once within the black hole s sphere of influence , however , which for a galaxy of central velocity dispersion @xmath3 and black hole mass @xmath4 extends out to @xmath5 these galactic mechanisms become less efficient . within the sphere of influence , which reaches @xmath6 pc for a black hole of mass @xmath7 , the black hole s gravity overwhelms that of the host galaxy and dominates the potential , unless a comparably large mass of gas has managed to accumulate at such small radii . timescale arguments suggest that gas at @xmath8 probably _ can not _ trickle down through an accretion disk all the way to the vicinity of the black hole ( e.g. , @xcite ) . disks probably do form , however , at smaller radii of order 0.1 pc , where maser emission in the nucleus of ncg 4258 has an unmistakably disk - like geometry ( @xcite ) . at such radii , the specific angular momentum of a keplerian disk exceeds that needed for direct capture by the black hole by a factor of order @xmath9 . if that gas is to accrete , 99% of the angular momentum must either be redistributed within the disk angular momentum transport or lost entirely from the system . we will consider the known mechanisms which can accomplish this feat . in the inner disk , _ turbulence driven by magnetic instabilities _ is a potent source of angular momentum transport . although magnetohydrodynamic ( mhd ) disk turbulence has long been implicated in disk angular momentum transport and is mentioned prominently by shakura & sunyaev ( 1973 ) , current confidence in this conclusion rests on two more recent developments . first , balbus & hawley ( 1991 ) demonstrated that the introduction of a weak magnetic field renders accretion flows linearly unstable to a powerful local instability . subsequent numerical simulations ( @xcite ; @xcite ; @xcite ; @xcite ) showed that the instability rapidly develops into sustained turbulence , which transports angular momentum outward at a rate that is consistent with observational constraints derived from studies of accretion in mass transfer binaries ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ) . for a recent review of the role of mhd turbulence in accretion disks , see balbus ( 2003 ) . in the absence of magnetic fields , a differentially rotating disk with angular velocity @xmath10 is linearly stable to axisymmetric perturbations according to the rayleigh criterion if @xmath11 i.e. , if the specific angular momentum @xmath12 of the flow is an increasing function of radius . for a geometrically thin disk orbiting a point mass @xmath13 , the angular velocity is keplerian , @xmath14 the specific angular momentum @xmath15 , and the disk is hydrodynamically stable . numerical simulations support this conclusion ( @xcite ) . although real disks especially geometrically thick flows with significant pressure support can have angular velocity profiles that differ from the simple keplerian form , they too are invariably stable by the rayleigh criterion . matters are drastically different if the disk contains a magnetic field . analytic studies have shown that a weak magnetic field destabilizes astrophysical disks , provided that @xmath16 a condition which is almost always satisfied in real disks . this _ magnetorotational _ instability ( mri ) exists regardless of the initial magnetic field configuration ( @xcite ; @xcite ; balbus & hawley 1991 , 1992 ; @xcite ; @xcite ; @xcite ) , though the growth rates vary depending upon the magnetic field geometry , being fastest for vertical fields . proving the existence of the mri in the general case requires a moderately involved calculation , which can be found in the comprehensive review by balbus & hawley ( 1998 ) . analyses of much simpler systems , however , reveal most of the important physics . here i follow closely the treatment of balbus & hawley ( 2000 ) . consider a fluid element orbiting in a disk with a central gravitational potential @xmath17 . we will assume that pressure forces are negligible , as they will be , provided that perturbations to the disk velocity field are highly subsonic . in cylindrical polar coordinates @xmath18 , the equations of motion then read @xmath19 where the dots denote derivatives with respect to time , and @xmath20 and @xmath21 are forces that we will specify shortly . we now concentrate our attention on a small patch of the disk at radius @xmath22 that is corotating with the overall orbital motion at angular velocity @xmath23 . we define a local cartesian coordinate system @xmath24 via @xmath25 and substitute these expressions into equation ( [ pja : eq_motion ] ) above . discarding quadratic terms , the result is @xmath26 these equations describe the epicyclic motion of pressureless fluid perturbed from an initially circular orbit . if the disk contains a weak vertical magnetic field , perturbations to the fluid in the plane of the disk will be opposed by magnetic tension forces generated by the bending of the field lines . considering in - plane perturbations varying with height @xmath27 and time @xmath28 as @xmath29 , the magnetic tension force is @xmath30 , where @xmath31 is the displacement vector and @xmath32 is the alfvn speed . using this expression for @xmath33 and @xmath34 , and assuming the @xmath35 time dependence , equation ( [ pja : eq_hill ] ) becomes @xmath36 combining these equations yields a dispersion relation that is a quadratic in @xmath37 , @xmath38 \nonumber \\ + \left ( k v_a \right)^2 \left [ \left ( k v_a \right)^2 + { { { \rm d } \omega^2 } \over { { \rm d } \ln r } } \right ] = 0 \,.\end{aligned}\ ] ] the system is unstable if @xmath39 , which occurs when the third term in the dispersion relation is negative . instability therefore occurs if @xmath40 for a sufficiently weak field ( @xmath41 ) , or for long enough wavelength perturbations ( @xmath42 ) , we then obtain the aforementioned criterion for a disk to be unstable to the mri , namely @xmath43 in a real disk , the longest wavelength perturbation in the vertical direction will be of order the scale height @xmath44 of the disk . there will therefore be instability , provided that the magnetic field is weaker than some threshold value @xmath45 . the physical origin for this instability is fairly straightforward . adopting the same configuration that we have just analyzed , consider the effect of perturbing a weak vertical magnetic field threading an otherwise uniform disk . if the field remains frozen into the plasma , then field lines which connect adjacent annuli in the disk will be sheared by the differential rotation of the disk into a trailing spiral pattern . provided that the field is weak enough , magnetic tension is inadequate to snap the field lines back to the vertical . what tension there is , however , acts to reduce the angular momentum of the inner fluid element and boost that of the outer fluid element , providing angular momentum transport in the outward sense that is required to drive accretion . the pervasive nature of the mri , taken together with its rapid linear growth rate@xmath46 can be as large as @xmath47 , meaning that growth of magnetic field energy occurs on a dynamical timescale implies that well - ionized astrophysical disks will be unstable . analytic studies are unable , however , to investigate the most interesting consequence of the instabilities , namely , what happens when the mri reaches a saturated state and turbulence has developed throughout the flow ? quantities we would like to determine in this state include the @xmath48 component of the stress tensor , @xmath49 which includes both fluid ( reynolds ) and magnetic ( maxwell ) stresses . in this expression , @xmath50 is the fluctuation in the azimuthal velocity , @xmath51 and @xmath52 are alfvn velocities in the radial and azimuthal directions , respectively , and the angle brackets denote a density weighted average over height . if @xmath53 can be measured , in practice from a numerical mhd simulation , then it can be used to estimate the @xmath54 parameter introduced by shakura & sunyaev ( 1973 ) . they suggested the scaling @xmath55 where @xmath56 is the sound speed and @xmath54 is a dimensionless constant or parameter that measures the efficiency of angular momentum transport in disks . numerical simulations of the non - linear development of the mri have now been performed in both local and global geometries . local simulations ( @xcite , 1996 ; @xcite ; @xcite ; @xcite ; @xcite ) follow the evolution of the mri in a small patch of disk , with periodic boundary conditions in @xmath57 and ( in a modified form ) @xmath58 . global simulations ( @xcite ; @xcite ) have the advantage of being able to study large scale magnetic fields and are essential for the investigation of geometrically thick ( @xmath59 ) flows ( e.g. , @xcite ) . all such simulations have known limitations . the physical separation between the largest relevant scale ( that of the whole disk , or the wavelength of the most important mri modes ) and the dissipative scale ( either viscous or resistive ) is almost always too large to resolve numerically . moreover , many calculations rely on purely numerical effects at the grid scale to provide dissipation . discussions of the effect of these limitations can be found , for example , in balbus & hawley ( 1998 ) or schekochihin et al . nevertheless , it is encouraging that a variety of simulations , using different numerical techniques , agree in several critical aspects . the mri rapidly leads on a timescale of just a few orbital periods to a state of mhd disk turbulence in which angular momentum is transported outward . this is sustained , despite the presence of dissipation ( often at unphysically large scales ) within the numerical codes . the magnetic field energy remains smaller than the thermal energy in the disk . finally , the maxwell stress dominates by a large margin over the reynolds stress , though the latter also provides some outward transport . values of @xmath54 between @xmath60 and @xmath61 are representative of the bottom line of most simulations . the maintenance of a disk magnetic field , despite the presence of dissipation , implies that turbulence driven by the mri sustains a disk dynamo , just as fluid motions in stars and planets generate their own self - sustaining magnetic fields ( @xcite ; @xcite ; @xcite ) . the accretion disk case , however , is interestingly different from these better - observed systems . in a disk ( at least according to current belief ) there would be no turbulent motions in the absence of a magnetic field , which rather must generate the turbulent velocity field needed for its own amplification . this bootstrapping quality makes it harder to apply knowledge gleaned from the long history of work on solar and planetary dynamos to the disk case . how do the values of @xmath54 obtained from numerical simulations compare with those inferred from observations ? the best constraints come from modeling the episodic accretion events seen in the dwarf novae subclass of cataclysmic variables . in the widely accepted thermal disk instability model for dwarf nova outbursts ( @xcite ; @xcite ; @xcite ) , the eruptive behavior results from the inability of the disk to reach a steady - state at radii where hydrogen is partially ionized . under these conditions , there is no thermally stable vertical structure for a range of accretion rates , resulting in a limit cycle in which the disk alternates between a cool mass - accumulating phase and a hot draining phase . the timescales of the outbursts are directly related to the values of @xmath54 in the outburst and quiescent states . different authors ( e.g. , pringle et al . 1986 ; @xcite ; @xcite ; @xcite ) derive typical values of @xmath54 in the outburst state of @xmath62 , with quiescent values a factor of a few lower . this is in reasonably encouraging agreement with the values predicted from mhd simulations . moreover , as i will discuss later in the context of agn disks , the _ difference _ in @xmath54 between the outburst and quiescent states could be caused by a suppression of the mri when the disk is cool and the magnetic field is imperfectly coupled to the gas ( gammie & menou 1998 ) . of course , in many circumstances of interest , it is either inconvenient or unfeasible to determine the structure and evolution of disks via three - dimensional mhd simulations . in particular , the evolution of agn disks across the full range of radii spanned by real disks can not be simulated directly . this raises the following questions : under what circumstances are vertically - integrated one - dimensional disk models , or two - dimensional simulations using a parameterized viscosity , good approximations to the real disk physics ? for geometrically thin disks ( @xmath63 ) , the standard time - dependent one - dimensional treatment divides the problem into two parts ( e.g. , @xcite ) . first , one solves for the vertical structure of an isolated annulus of the disk using methods and equations analogous to those of stellar structure . the vertical structure calculation yields the functional dependence of the kinematic viscosity @xmath64 on the surface density @xmath65 , angular velocity @xmath23 , and @xmath54 . armed with this knowledge , we can then solve a one - dimensional diffusion equation for the time - dependence of the disk , @xmath66 \ , , \label{pja : eq_diffuse}\ ] ] subject to appropriate boundary conditions at the inner and outer edges of the disk . if thermal fronts are present ( or develop ) within the disk , this equation for @xmath65 needs to be supplemented with an explicit equation for the time - dependence of the central temperature @xmath67 ( e.g. , @xcite ) . provided that we restrict our attention to the axisymmetric evolution of geometrically thin disks , one - dimensional disk models constructed in this manner probably provide a reasonable approximation to the evolution of mri - active disks . at the most basic level , the mri is a local instability , and , as such , is in principle amenable to an @xmath54-type description ( @xcite ) . for the vertical structure part of the calculation , the main question is whether the distribution of the stress with height follows the simple scalings ( stress proportional to pressure ) normally employed as a generalization of the @xmath54 _ ansatz _ ( eq . [ pja : eq_ss ] ) . in fact , numerical simulations of the mri suggest that the stress may be roughly constant within the body of the disk ( up to @xmath68 ) , with a significant fraction ( perhaps a quarter ) of the magnetic energy escaping buoyantly from the disk to form a hot , strongly magnetized corona ( miller & stone 2000 ) . these simulation results differ from the simplest assumptions used in modeling vertical structure , but they do not invalidate the basic approach . indeed , it ought to be possible to compute vertical structure models and vital observational properties of the disk , such as the local spectrum ( e.g. , @xcite ; @xcite)using the computational results as a guide to where in the disk dissipation of energy occurs . this has not yet been done . at different times @xmath28 , where @xmath28 is measured in units of the orbital period at the initial center of the ring . the simulation is isothermal , with @xmath69 at @xmath70 , the initial center of the ring . the mri was triggered by threading an initially stable and axisymmetric ring of gas with a weak vertical magnetic field . a cylindrical disk approximation was used ( i.e. , no vertical stratification ) , and the evolution was computed using the zeus code ( stone & norman 1992 ) with @xmath71 grid points in @xmath58 , @xmath57 , and @xmath27 , respectively . as in the one - dimensional viscous disk solution to this problem , given by lynden - bell & pringle ( 1974 ) , the ring spreads out , with the mass flowing inward , while a fraction of the mass at large radii absorbs the angular momentum . in addition , however , there is obvious non - axisymmetric structure in the form of trailing spiral waves and relatively long - lived fluctuations in the magnitude of the derived stress . ] ( @xmath72 , 2 , 4 , and 6 orbits at the initial location of the ring ) . the simulated disk profile remains smooth.,title="fig : " ] -1.3truein the second fundamental assumption of one - dimensional models is the diffusive evolution of the surface density predicted by equation [ pja : eq_diffuse ] . this , too , is broadly compatible with what is known about the mri . figures [ pja : fig1 ] and [ pja : fig2 ] show the predicted evolution of a ring of gas in a keplerian potential , computed directly from a global three - dimensional mhd simulation using methods described by armitage ( 1998 ) . although large scale non - axisymmetric structure is evident in figure [ pja : fig1 ] , the diffusive evolution of the surface density that is predicted by the one - dimensional theory is nonetheless reproduced . the preceding discussion has emphasized that one - dimensional disk models based on the @xmath54 formalism can provide a reasonable description of the evolution of thin disks , especially if insights from numerical simulations of the mri are used to guide the detailed modeling . simulations , however , also reveal circumstances under which @xmath54 models fail to describe disks driven by magnetically induced turbulence . papaloizou & nelson ( 2003 ) and winters , balbus , & hawley ( 2003a ) have shown that averaging the magnetically induced stress over tens or even hundreds of orbital times is required to yield stable estimates of the average stress . in the context of a one - dimensional @xmath54 disk model , this implies that @xmath54 can not be regarded as a constant over shorter timescales . more generally , any attempt to extend @xmath54 models to treat two - dimensional flows ( for example , to study eccentric disks or disks strongly perturbed by binary companions ) is dangerous and not to be recommended . such generalizations frequently rest on the assumption that _ all _ the components of the stress tensor in a turbulent mhd disk flow behave in the same way as a navier - stokes viscosity . there is scant reason to believe that this is true , and indeed the limited numerical evidence collected to date suggests that the multidimensional evolution of magnetically driven disks differs significantly from that of their viscous counterparts ( @xcite ; winters , balbus , & hawley 2003b ) . so far , we have assumed that the disk orbits in the point mass potential of the black hole , ignoring the self - gravity of the disk itself . this approximation is generally valid in the inner regions of the disk and will obviously fail if the disk mass becomes comparable to the mass of the black hole ( perhaps during formation of the hole via accretion ) . even if the disk mass is relatively modest , however , self - gravity can still become important at large enough radii . to demonstrate this , we make use of the @xmath54-prescription and write the kinematic viscosity @xmath64 in the form @xmath73 for a steady disk , away from the boundaries ( e.g. , @xcite ) , @xmath74 which , together with equation [ pja : eq_alpha ] , specifies @xmath75 in terms of the accretion rate , sound speed , and @xmath54 parameter . the stability of a nearly keplerian gas disk against axisymmetric perturbations is measured by the toomre ( 1964 ) @xmath76 parameter , @xmath77 with instabilities developing as @xmath78 from above . making use of equations [ pja : eq_alpha ] and [ pja : eq_steady ] , we find that for a steady disk , @xmath79 since the temperature and sound speed in the disk are decreasing functions of radius ( at least if the heating is provided by accretion and/or irradiation from the central regions of the disk ) , the outer parts of a steady - state disk will inevitably reach the @xmath80 threshold for self - gravity to become important , provided that the disk is large enough . indeed , models of agn disks suggest that self - gravity typically sets in at @xmath81 pc , which is only @xmath82 schwarzschild radii for a @xmath83 black hole ( @xcite ; @xcite ; @xcite ) . interestingly , this radius is one to two orders of magnitude _ smaller _ than the radii from which maser emission has been observed in ngc4258 ( @xcite ) and ngc1068 ( greenhill & gwinn 1997 ) . this raises the immediate question : is the masing disk observed in these galaxies a roughly homogenous structure , or is it rather composed of discrete clumps ( shlosman et al . 1990 ; @xcite ; @xcite ) ? . under these conditions , a balance between heating and cooling is attainable , and the disk develops a stable spiral structure that acts to transport angular momentum outward . a shorter cooling time of @xmath84 _ ( right - hand panel ) _ leads to prompt fragmentation of the disk into multiple bound objects . note that the thickness of the simulated disk exceeded that appropriate for agn . figure adapted from rice et al . ( 2003 ; their figs . 2 and 4 ) . , title="fig : " ] -0.1truein once @xmath85 has been attained , self - gravity triggers the development of spiral structure in the disk . gravitational torques then act to transport angular momentum outward , in a manner superficially resembling the evolution of an ordinary viscous disk described by an @xmath54 model ( @xcite ) . such a correspondence rests on shaky theoretical foundations ( @xcite ) , because the non - local nature of gravitational angular momentum transport immediately violates the local property of @xmath54 models ( eq . [ pja : eq_ss ] ) . that said , examining the local properties of self - gravitating disks proves to be useful in delineating the circumstances under which a self - gravitating disk will fragment , rather than stably transport angular momentum ( @xcite ) . if , at a particular radius , the disk can cool via radiative losses on a timescale @xmath86 , then gammie ( 2001 ) showed that the boundary between stable transport and fragmentation lies at @xmath87 . for @xmath88 , a stable state can be reached in which @xmath89 and angular momentum is transported with an effective @xmath54 of @xmath90^{-1 } \,,\ ] ] where @xmath91 is a two - dimensional adiabatic index ( @xcite ) . as @xmath86 decreases , turbulence in the disk ( measured by the effective @xmath54 ) has to become increasingly violent in order to balance the radiative losses , and the overdensities of spiral structures increase . eventually , for @xmath92 , the disk fragments into bound objects . both local ( @xcite ; @xcite ) and global ( @xcite ; fig . [ pja : fig3 ] ) simulations broadly support these analytic results . application of these ideas to the outer regions of accretion disks in agn is complicated by the dominant role that irradiation plays in setting the thermal conditions in the outer disk . in particular , fluctuations in the disk temperature ( and , thus , @xmath76 ) will occur as the illumination from the central source varies , with the disk being most vulnerable to fragmentation when the source is in its lowest state . furthermore , there is no reason to expect that the disk at large radii is able to attain the steady - state described by equation [ pja : eq_steady ] . the most plausible scenario is that the inner non - self - gravitating disk is surrounded by a region in which self - gravity transports angular momentum in the stable , non - fragmenting regime . beyond that , at radii of order a pc or less , any disk formed from inflowing or outflowing gas is unstable to fragmentation ( @xcite ) . if correct , this suggests that agn need to be resupplied with gas of low specific angular momentum ( compared to that of a circular orbit at the sphere of influence ) in order to prevent fragmentation occurring in the outer disk . more speculatively , if fragmentation does occur , the edge of the accretion disk in agn could be a birth place for massive stars that would ultimately form stellar or intermediate mass black holes . if this process had occurred in the galactic center in the recent ( @xmath9310 myr ) past , it could leave a distinctive signature in the form of disk - like kinematics for some subset of massive stars ( @xcite ) . moreover , if some fraction of the stars ended up embedded in the inner disk , they would migrate inward on at most a viscous timescale . gravitational waves from the final inspiral of the stellar remnants with the supermassive black hole are potentially detectable with the _ laser interferometer space antenna _ ( _ lisa _ ; @xcite ; @xcite ) . in broad outline then , the mri is likely to provide the main mechanism of angular momentum transport in the inner disk , while self - gravity is potentially important at relatively large radii . a more subtle question is whether there is a region of the disk in which both processes are operative , or , conversely , a zone in which neither drives efficient angular momentum transport . this depends upon whether the mri is able to operate in the cool outer regions of the disk , where the gas has a low ionization fraction and the flux - freezing assumption of ideal mhd breaks down . for agn , a quantitative discussion of the effect of the low temperature on angular momentum transport processes has been given by menou & quataert ( 2001 ) . for a disk with sound speed @xmath56 and resistivity @xmath94 , the magnetic reynolds number is @xmath95 if @xmath96 is too low , then diffusion of magnetic fields due to the finite conductivity can suppress the mri ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ) . the critical value of @xmath96 ( which may also need to take account of the hall effect ) is probably of order unity or somewhat larger . identical reasoning suggests that ambipolar diffusion can also shut off the mri and prevent mhd disk turbulence from developing ( @xcite ) . applying these ideas to agn disks , menou & quataert ( 2001 ) found that the ionization state of the outer disk and hence the likelihood of the mri being operative was highly uncertain , depending upon ( among other things ) whether the outer disk was exposed to irradiation from the central source . if the mri does shut off at radii where @xmath97 , then an intermediate zone in the disk could be both magnetically and gravitationally inactive . such a `` dead zone '' would act to starve the inner disk of resupply from larger radii , until enough matter built up to initiate self - gravity or magnetorotational activity . complicated and poorly understood time - dependent accretion would probably result ( e.g. , @xcite ; @xcite ) . alternatively , if the regions of mhd turbulence and gravitational instability overlap , then there could be non - trivial interactions between the two angular momentum transport mechanisms . planet formation simulations ( winters et al . 2003b ) show that _ external _ gravitational torques can modify the properties of mhd disk turbulence , and the first simulations of self - gravitating magnetized disks ( @xcite ) likewise show that magnetic and gravitational torques are not simply additive . in the calculations of fromang et al . ( 2004 ) , the presence of mhd turbulence in a self - gravitating disk significantly reduced the rate of gravitational transport of angular momentum . finally , it remains possible that accretion is not always driven by angular momentum redistribution within the disk at all , but rather by angular momentum loss in an outflow . outflows and jets are certainly present in some agn , and if magnetically driven ( @xcite ) , they can potentially remove _ all _ of the angular momentum needed to allow accretion , while ejecting only a modest fraction of the mass . models in which outflows are entirely responsible for driving accretion are not popular , in part because of skepticism that the required magnetically launched outflows are present across a sufficiently broad range of disk radii . in particular , although there is near - universal agreement that magnetic fields are implicated in the formation of relativistic jets , the slower outflows that originate further away from the black hole may well be partially or entirely accelerated by radiation pressure acting on resonance lines ( @xcite ; @xcite ; @xcite ) . moreover , the stability of accretion arising from magnetic outflow torques is rather doubtful ( @xcite ; @xcite ; but see also @xcite and @xcite ) . nonetheless , the possibility that angular momentum loss via mhd outflows dominates over internal stresses in restricted regions of agn disks remains entirely viable . it constitutes a significant source of uncertainty in our understanding of the structure of the disk and when considering possible mechanisms for large - scale variability . the accretion rate onto a black hole is most usefully measured as a fraction of the accretion rate at which radiation pressure becomes important for the dynamics of the flow . for a spherically symmetric flow , radiation pressure acting on free electrons balances gravity at the eddington luminosity , @xmath98 where @xmath99 is the thomson scattering cross - section . if the luminosity is generated by accretion with radiative efficiency @xmath100 , then @xmath101 and the above expression defines an eddington accretion rate . taking @xmath102 , @xmath103 the accretion rate can then be expressed in dimensionless form as a fraction @xmath104 of the eddington rate . current theoretical understanding identifies @xmath105 as the most important parameter controlling the structure and observational appearance of black hole accretion flows . we will consider in turn the three cases : @xmath106 , @xmath107 , and @xmath108 . these cases , by definition , differ in the importance of radiation pressure for the dynamics of the flow . equally important , the ability of the flow to cool also varies systematically with @xmath105 . for @xmath106 , the critical question is whether the inflowing gas can radiate the gravitational potential energy liberated by accretion . most theoretical work suggests that the vast majority of the energy dissipated by mhd turbulence in a disk goes initially into heating the ions rather than the electrons ( @xcite ; @xcite ; @xcite ; @xcite ) , at least as long as the magnetic field remains sub - equipartition . sub - equipartition fields are the normal outcome of the non - linear phase of the mri in mhd simulations of accretion flows in the fluid limit ( for a discussion of how the mri itself is altered if the plasma is collisionless , see @xcite ) . radiative losses from bremsstrahlung radiation , synchrotron radiation , and comptonization of soft photons , on the other hand , primarily take energy from the electrons . in this situation , where the ions are heated simultaneously with the electrons being cooled , the extent of coupling between the two fluids obviously determines the state of the plasma . if the dominant coupling process is ordinary coulomb collisions , then the timescale for thermal equilibrium to be established between the ions and the electrons is ( @xcite ; @xcite ) @xmath109 here , @xmath110 is the electron temperature , @xmath111 is the boltzmann constant , @xmath112 is the proton ( ion ) mass , @xmath113 is the number density of ions , and @xmath114 is the electron mass . the coulomb logarithm has a value of @xmath115 . when the accretion rate is low , @xmath113 is small , and this , coupled with the high temperature near the black hole ( of the ions generally , and the electrons , as well , if we initially postulate a one - temperature plasma ) , implies a large value for @xmath116 . once @xmath116 becomes comparable or larger than the inflow time , electrons and ions can no longer maintain equilibrium via coulomb collisions . instead , a two - temperature plasma develops in which the ions stay close to the virial temperature while the electrons are substantially cooler . all current models for low @xmath105 accretion flows onto black holes are based on the two - temperature idea ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; and the numerous variants on these models ) , and would not be viable otherwise . in particular , any collective mechanism that coupled the ions to the electrons substantially more efficiently than coulomb collisions would undercut the foundations of two - temperature plasma accretion models . although no such mechanism is known , the well known complexities of laboratory plasmas suggest that the possible existence of more efficient coupling can not be entirely ruled out ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . although from a purely theoretical perspective the question of whether low @xmath105 accretion flows form a two - temperature plasma remains open , observations particularly those that indicate a remarkably low luminosity for the black hole at the galactic center ( @xcite)are hard to reconcile with almost any imaginable single temperature disk model ( see below ) . this has prompted most theorists to accept that accretion flows with @xmath117 develop a two - temperature structure , though the exact threshold below which a hot two - temperature flow supplants a cool thin disk is less certain . narayan & yi ( 1995b ) suggest that the critical accretion rate is around @xmath118 that is , one to two orders of magnitude below the eddington limit . once a two - temperature flow has developed , three model - independent conclusions follow . first , since the ion temperature @xmath119 remains comparable to the virial temperature , the flow can be _ geometrically thick _ , with @xmath120 . second , since most of the accretion energy remains locked up in the ions , the flow is _ underluminous _ , with an accretion efficiency much smaller than the @xmath121 values realized for thin disks . finally , since the gas is both hot and rapidly rotating , it is at most marginally bound to the black hole ( formally , the bernoulli constant , which is conserved for adiabatic flows , is positive ) . as a consequence , it is certainly plausible , and possibly inevitable , that a substantial fraction of the putative fuel for the black hole actually ends up outflowing from the galactic nucleus ( @xcite ; @xcite , 2004 ) . outflows further reduce the already low radiative efficiency ( now defined as @xmath122 , where the inflow rate is measured at large radius ) , because most of the gas then fails to sample the deep potential well close to the black hole . figure [ pja : fig4 ] shows semi - quantitatively ( not quantitatively , because the simulation depicted was two - dimensional and non - magnetic ) what the resulting flow looks like . -0.1truein apart from such generalities , no consensus has yet been reached as to the detailed structure of non - radiative accretion flows ( also described as advection dominated flows , since most of the liberated gravitational potential energy is advected with the fluid rather than being radiated ) . the problem lies in the fact that the structure of non - radiative accretion flows is expected to depend sensitively on the nature of energy and angular momentum transport . the numerical simulations needed to directly attack this issue are extraordinarily difficult . for an illustration of the difficulties , suppose that the black hole is accreting from the hot interstellar medium of an elliptical galaxy with @xmath123 and sound speed @xmath56 a few hundred kilometers per second . the ratio of the bondi radius , @xmath124 to the gravitational radius of the hole both of which must be resolved if we want to get the boundary conditions strictly correct is then @xmath125 , vastly exceeding the dynamical range accessible to any present three - dimensional simulation . two - dimensional simulations can do better but have their own limitations due to both the antidynamo theorem in axisymmetry ( @xcite ) and the fact that the mri has different properties in axisymmetry , as compared to three dimensions ( @xcite ; hawley et al . recent examples of simulations include work by stone & pringle ( 2001 ) , machida , matsumoto , & mineshige ( 2001 ) , hawley & balbus ( 2002 ) , proga & begelman ( 2003 ) , pen , matzner , & wong ( 2003 ) , and igumenshchev , narayan , & abramowicz ( 2003 ) . there is substantial agreement that outflows of various kinds occur generically as a side effect of non - radiative accretion , but opinions still differ on the steady - state radial structure of the flow , and on how to interpret the numerical results in terms of simpler analytic models . for prograde _ ( solid line ) _ and retrograde _ ( dashed line ) _ orbits . _ ( right panel ) _ efficiency of energy extraction if gas spirals slowly into @xmath126 before plunging silently into the black hole . plotted is @xmath127 , where @xmath128 is the binding energy of the marginally stable circular orbit for prograde and retrograde motion.,title="fig : " ] -2.5truein accretion rates above @xmath129 ( eq . [ pja : eq_adaf ] ) yield flows that are dense enough to provide good coupling between ions and electrons . provided that the accretion rate is not extremely high , i.e. , @xmath107 , the flow is then radiatively efficient and cools to form a geometrically thin ( @xmath63 ) disk . the disk extends from an outer radius , which is probably set by the considerations of gravitational stability described in the previous section , down to the marginally stable circular orbit at @xmath130 . interior to the marginally stable circular orbit , which lies at @xmath131 for a schwarzschild black hole , circular particle orbits are unstable any inward radial peturbation suffices to place the particle onto a plunging trajectory into the black hole . in the simple but possibly incorrect case where there is zero torque at @xmath126 ( non - zero torque is discussed in [ secpjadyn ] ) , the radiative efficiency is determined by the binding energy @xmath128 of gas at that radius . plots of these quantities are shown as a function of the dimensionless spin parameter ( @xmath132 for a hole with an angular momentum @xmath133 ) in figure [ pja : fig5 ] . the basic theory of thin disks is well known ( for reviews , see , e.g. , @xcite ; frank et al . 2002 ) . the main open issues ( at least for flat , circular disks ) concern the role of magnetic fields and radiation . the structure of magnetic fields near @xmath126 is of particular interest , since it determines not only the boundary condition at the disk s inner edge ( @xcite ) , but also the strength of magnetic fields threading the black hole itself . if the black hole is spinning , magnetic fields admit the possibility of tapping some of the rotational energy via the blandford - znajek ( 1977 ) process . spin energy extraction provides one way to power jets ( e.g. , @xcite ) , and is frequently suggested mostly on the basis of empirical arguments as the source of the radio - loud / radio - quiet dichotomy in agn ( e.g. , @xcite ) . as a counter to such claims , ghosh & abramowicz ( 1997 ) and livio , ogilvie , & pringle ( 1999 ) have argued that since dynamo generated disk fields are expected to be weak ( sub - equipartition ) , and dragging in ordered external fields is difficult ( lubow et al . 1994 ) , the efficiency of the blandford - znajek process is likely to be negligibly small for thin disks . these arguments are plausible , but i doubt that they represent the last word on the subject , not least because of observational indications for spin energy extraction in some agn ( @xcite ; @xcite , though note that even if spin energy _ is _ being extracted , this need not be the classical blandford - znajek process at work ) . as the accretion rate increases , thin disks are predicted to remain the preferred mode of accretion below some upper limit set by the effects of radiation pressure . for a spherical flow , that upper limit lies at @xmath134 . in disks , however , matters need not be so simple . once radiation pressure dominates gas pressure in the inner disk , the disk is prone to several instabilities , even in the absence of magnetic fields ( @xcite ; @xcite ) . when magnetic fields _ are _ included , it is found that the flow develops photon bubbles ( @xcite ; @xcite ; @xcite , 2003)low density radiation filled cavities that coexist with dense gas pressure dominated regions . in principle , radiation can escape from such a highly inhomogenous structure at rates above the eddington limiting luminosity , _ without _ being accompanied by catastrophic mass loss ( @xcite ; @xcite ; see also @xcite ) . this could permit the existence of stable , thin disks , with an emergent luminosity as large as @xmath135 ( @xcite ) . detailed calculations of the outcome of these instabilities in magnetized disks , and their implications for the structure of radiation pressure dominated flows even below the eddington limit , are now in progress ( @xcite ; @xcite ; @xcite ) . the final regime to consider is that of overfed accretion , with @xmath108 . by definition , radiation pressure is central to the dynamics of such flows , though it need not act to shut off accretion directly . at sufficiently high accretion rates , the opacity can be large enough that photons in the inner regions are advected inward faster than they can random walk away from the black hole . eventually , they are dragged across the event horizon along with the gas and are not radiated from the system . for a spherical accretion flow , it is straightforward to show that advection wins over diffusion inside a trapping radius given by @xmath136 where @xmath137 is the schwarzschild radius ( @xcite ) . the same behavior occurs for flows with non - zero angular momentum ( @xcite ) . the emergent luminosity from photon - trapped flows remains around @xmath138 ( e.g. , @xcite ; @xcite ) , so as @xmath105 increases , the radiative efficiency decreases inversely with the accretion rate . clearly , the similarities between the @xmath106 and the @xmath108 regimes are striking . in both cases , the radiative efficiency is low , for @xmath106 because of inefficient ion - electron coupling , and for @xmath108 because the high opacity prevents photons from escaping the fluid . in the simplest analysis , overfed accretion will then also result in a hot , geometrically thick disk , with the main difference from the underfed case being the lower value of @xmath91 appropriate for a radiation dominated gas . strong outflows could then be inevitable , and these might even reduce the accretion rate onto the hole back down to levels comparable to the eddington limit . once again , simulations of radiation dominated disks will be needed to test such speculations . in addition to providing an explanation for the dichotomy in the observed properties of supermassive black holes in agn and quiescent galactic nuclei , the existence of these different modes of accretion also has implications for the growth of black holes . accretion rates well below the eddington limiting value probably provide the black hole with a negligible amount of mass , despite the fact that this mode lasts much longer than more active phases ( @xcite ) . more surprisingly , attempts to flood the black hole with mass at rates vastly in excess of the eddington value may also fail if , as we have speculated , such high @xmath105 flows are as vulnerable to outflows as their low @xmath105 counterparts . the entire accretional growth of black holes may be restricted to occur via radiatively efficient accretion at rates that lie within an order of magnitude or so of the eddington limit ( probably extending _ above _ as well as below the nominal `` limit '' ) . for the most luminous agn , such a `` what you see is what you get '' model allows for a consistent match ( within substantial uncertainties ) between observations of the luminous output of agn and the local density of supermassive black holes ( @xcite ; @xcite ; @xcite ; following the basic approach of @xcite ) . it may also apply to the earlier growth phase of supermassive black holes . . @xmath139 peaks well outside the marginally stable circular orbit . the three upper curves show the effect of adding increasing amounts of torque at @xmath126 ( agol & krolik 2000 ) , with the labels showing the resulting increase @xmath140 in the radiative efficiency as a fraction of the original efficiency @xmath100 . no emission is assumed to originate from the plunging region.,title="fig : " ] -1.3truein the inner boundary of a radiatively efficient accretion disk around a black hole lies close to the marginally stable circular orbit . at @xmath141 , the force balance for a fluid element in a thin disk is predominantly between gravity and centrifugal force , with angular momentum transport driving a slow , subsonic inflow . interior to @xmath126 , circular orbits are unstable . in the standard model for black hole accretion disks ( @xcite ; @xcite ) , once gas reaches this plunging region , it flows into the black hole along geodesics , maintaining a constant energy and angular momentum . in this simple model , gas in the plunging region can have observational effects , mainly by reprocessing or reflecting radiation generated elsewhere in the system ( @xcite ; @xcite ) . the plunging region does not , however , alter the structure of the disk or the radiative efficiency of the flow . the inner boundary condition for the disk is simply that there should be zero torque at @xmath142 . early studies of black hole accretion assumed ( often explicitly ) that magnetic fields in the plunging region remained weak . provided that this is so , the inflowing gas rapidly attains supersonic ( and super - alfvnic ) radial velocities and becomes causally disconnected from the disk outside the marginally stable orbit . this justifies the zero - torque boundary condition at @xmath126 . if magnetic fields become strong , for example as a consequence of the shearing of frozen - in fields during the final inspiral , the argument fails . energy and angular momentum can then be exported from the plunging region to the disk , modifying its structure and emission ( @xcite ; @xcite ) . the same basic process can occur in either schwarzschild or kerr geometry . the consequences of a non - zero torque at the marginally stable orbit have been discussed in detail by agol & krolik ( 2000 ) . the most basic is that energy that would otherwise be lost down the hole can be transmitted from the plunging region to the disk , and subsequently radiated . as a result , the radiative efficiency of thin disk accretion could be larger than normally assumed ( fig . [ pja : fig5 ] ) . in particular , the standard value of @xmath143 for the yield of accretion onto schwarzschild holes may more usefully be regarded as a lower limit . this implies that although the qualitative trend for more rapidly spinning black holes to be more luminous at a given accretion rate than schwarzschild holes is very reasonable there are significant theoretical uncertainties that need to be borne in mind when attempting to determine the average spin of black holes from comparison of black hole masses and integrated agn output . for example , the relatively high inferred radiative efficiency ( @xmath144 ) derived by elvis , risaliti , & zamorani ( 2002 ) and yu & lu ( 2004 ) need not necessarily imply a rapid average spin for the supermassive black hole population . perhaps more interestingly , given that measuring @xmath100 to a factor of about two precision is obviously a challenging task , a non - zero torque at the disk inner edge makes a large change to the radial distribution of the disk emissivity . this is plotted in figure [ pja : fig6 ] , using expressions given in agol & krolik ( 2000 ) . unlike a `` standard '' disk , in which the local flux peaks well outside the last stable orbit , a torqued disk has a steeply rising emissivity profile right down to the marginally stable orbit ( at large radius , the contribution to the emissivity due to the torque at @xmath126 scales with radius as @xmath145 , which is steeper than the dissipation profile for an untorqued disk ) . this difference in the dissipation with radius is potentially observable . it is worth noting that similarly steep dissipation profiles are predicted in some models in which magnetic fields act to extract energy from a kerr black hole ( takahashi et al . 1990 ; @xcite ) . , as derived from numerical simulations with relatively high _ ( solid line ) _ and relatively low _ ( dashed line ) _ sound speed ( armitage & reynolds 2003 ) . _ ( right panel ) _ specific angular momentum @xmath12 at several different times for a simulated flow in which the disk magnetic field strength was relatively large ( reynolds & armitage 2001 ) . the inner disk is quite variable , but typically @xmath12 continues to decline interior to @xmath126.,title="fig : " ] -0.1truein numerical simulations have generally been supportive of the basic concept outlined by krolik ( 1999 ) . pseudo - newtonian calculations , which use a modified potential to mimic some of the most important relativistic effects ( @xcite ) , have confirmed that there can be significant non - zero torques at @xmath126 ( @xcite ; @xcite ; hawley & krolik 2001 , 2002 ) . as shown in figure [ pja : fig7 ] , this leads to a declining specific angular momentum profile that extends seamlessly through the last stable orbit to smaller radii . the same phenomenon has now been seen in general relativistic mhd simulations of black hole accretion ( @xcite ; @xcite ) . large amplitude ( factor @xmath146 ) fluctuations in the torque also occur and persist over a timescale of @xmath147 orbital periods . the importance of this torque for the dynamics of the disk may also vary with the temperature ( or , equivalently , thickness ) of the inner disk , with the zero - torque limit becoming a better approximation as the thickness decreases ( @xcite ; @xcite ; @xcite ; @xcite ) . the emission from agn is variable ( @xcite ; @xcite ) , as is the much weaker emission from the galactic center ( @xcite ; @xcite ; @xcite ; @xcite ) . although in some observationally important cases the _ correlations _ between variability in different wavebands can be understood through simple light travel time arguments ( @xcite ) , the basic origin of most of this variability is not well understood . here , i discuss what insights into the problem recent mhd calculations of disk accretion provide . the first question we might ask is , are there large amplitude changes in the luminosities of agn akin to those seen in most other disk accreting systems ? we have already discussed the outbursts of dwarf novae , which are attributed to a global thermal instability of the accretion disk . accretion onto some neutron stars and ( stellar mass ) black holes in low - mass x - ray binaries is similarly prone to thermal instability ( @xcite ) and can be modeled using a minimally modified version of the theory developed for dwarf novae ( @xcite ) . the same ideas have also been pressed into service to explain the outbursts of protostellar disks seen in fu orionis objects ( @xcite ; @xcite ) , though the identification of these events with thermal disk instabilities is less compelling . if thermal instabilities act on a global scale in agn , the timescales for the cycling between outburst and quiescence would be long probably @xmath148 or longer ( @xcite ) . transitions between `` on '' and `` off '' states might not occur on easily observable timescales , though they could leave traces in the form of intermittent activity from radio - loud agn ( @xcite ; @xcite ) . if such intermittent accretion is commonplace , there ought to exist a population of relatively quiescent galaxies in which the black hole is surrounded not by a radiatively inefficient flow , but rather by a temporarily inactive thin disk . detailed models of this type have been developed by siemiginowska , czerny , & kostyunin ( 1996 ) and by siemiginowska & elvis ( 1997 ) . disks in agn extend out to radii that encompass the zone in which hydrogen is partially ionized , so the basic ingredient necessary for thermal instabilities to occur is present ( @xcite ) . what is less clear is whether the efficiency of angular momentum transport ( i.e. , the value of @xmath54 ) varies between the ionized and neutral state in the agn environment . a substantial difference between @xmath149 and @xmath150 is needed if the whole disk is to flip between outburst and quiescent states . in dwarf novae , the required change probably occurs because the hot and cold states lie on opposite sides of the critical degree of ionization required for sustained mhd turbulence ( eq . [ pja : eq_rem ] ; @xcite ) . it has been argued ( @xcite ) that this is no longer the case in agn disks , leading to thermal instabilities that produce local flickering but not global outbursts . if so , any observational evidence for large scale outbursts in agn may require a different explanation . numerous possibilities for time dependent behavior suggest themselves in the outer disk , where there is an interplay of gravitational and mhd disk instabilities , though none have been worked out in any great detail . at frequencies between @xmath151 and @xmath152 ( i.e. , timescales of tens of minutes to around a year ) , x - ray observations provide good constraints on the properties of agn variability ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ; and references therein ) . the power spectral density @xmath153 derived from these observations is often consistent with a power law broken at a frequency @xmath154 , @xmath155 representative values for the two slopes are @xmath156 and @xmath157 ( vaughan & fabian 2003 ; @xcite ) . ( armitage & reynolds 2003 ) . the patchy emission is boosted on the approaching ( left hand ) side of the disk as a consequence of beaming . in constructing this image , the local disk emission has been assumed to scale with the vertically integrated magnetic stress . raytracing through the schwarzschild metric is then used to account for the relevant relativistic effects , including beaming , gravitational redshift , and light bending.,title="fig : " ] -0.0truein although a detailed comparison is beyond the capabilities of current numerical simulations , there appear to be good prospects for relating noise power spectra of this form to theoretical models of mhd disk turbulence . simulations of magnetically active disks show that the flow displays variability across a wide range of timescales , with physical quantities such as the magnetic stress and mass accretion rate showing temporal power spectra that are described by power laws ( @xcite ; @xcite ; @xcite ) . the patchy and rapidly fluctuating pattern of stress across the surface of the inner disk , which gives rise to this variability , is shown in figure [ pja : fig8 ] . if we assume that the local disk emission traces the magnetic stress , then the power spectra predicted from the simulations have the form shown in figure [ pja : fig9 ] . regardless of the inclination of the disk to the line of sight ( which affects the importance of relativistic effects such as beaming ) , the temporal power spectrum at frequencies comparable to those of the inner disk is around @xmath158 . individual annuli in the simulated disk , moreover , produce power spectra that break at around the orbital frequency to substantially steeper slopes of around @xmath159 ( @xcite ) . these results support the common assumption that the break frequency in observed systems scales with the orbital frequency at the marginally stable orbit , and hence with the mass of the black hole ( e.g. , @xcite ; @xcite ) . although there are significant uncertainties in many agn black hole masses , the observations of markowitz et al . ( 2003 ) suggest a correlation between black hole mass and break frequency . we note , however , that work to date has not been able to recover the observed values of @xmath160 and @xmath161the theoretical power spectra are steeper than observed . derived from a numerical simulation of accretion onto schwarzschild black holes ( armitage & reynolds 2003 ) . the frequency @xmath162 has been rescaled such that the orbital frequency at the radius of marginal stability corresponds to @xmath163 . shown here are the power spectra for @xmath164 , @xmath165 ( offset by @xmath9 ) , @xmath166 ( offset by @xmath167 ) , and @xmath168 ( offset by @xmath169 ) . mhd turbulence in the disk generates variability across a wide range of timescales , with quantities such as the predicted emission or mass accretion rate showing steep , approximately power law spectra.,title="fig : " ] -0.2truein flat disks , in which the gas follows close to circular orbits , are special cases of the more general situation in which the disk may be eccentric and/or warped out of a single plane . a common assumption is that an initially warped or eccentric disk will relax to a flat circular one `` on a viscous timescale '' . this is at best a gross simplification of the evolution of warped or eccentric disks , and may even be qualitatively wrong . as one might expect , the evolution of warped or eccentric disks is rather complex , and in this section i attempt only to provide a framework for understanding some of the key results . for more details , the interested reader is well advised to consult the recent and exhaustive studies by ogilvie ( 1999 , 2000 , 2001 ) , which include references to the earlier literature on the subject . observations provide clear motivation for considering the evolution of warped disks in agn . the masing disks in ngc4258 ( @xcite ) , ngc1068 ( @xcite ) , and the circinus galaxy ( greenhill et al . 2003 ) all appear to be warped , albeit to different degrees . this raises two questions . first , are the warps merely decaying features that reflect the complex angular momentum distribution of the infalling gas that formed the disk , or are they self - excited by some process intrinsic to the disk or agn itself ? second , if the warps are in fact decaying , how rapidly does that process occur compared to the timescale on which the gas would be accreted ? although easily formulated , both of these questions have proved to be difficult to answer . the basic physics revolves around the fact that in a warped disk , neighboring annuli are inclined relative to one another , and as a consequence , there is shear in the vertical direction as well as in the radial direction . dissipation of the energy associated with the out - of - plane motions is governed by the @xmath170 component of the stress , and this acts to flatten out the disk as , simultaneously , the @xmath171 stress drives accretion . the crucial point is that these stresses can be generated by entirely different physical processes ( for a concrete example , see @xcite ) and need not act on the same timescale , even roughly ( @xcite ) . the detailed physics of stresses within the disk will then determine whether the disk flattens before a significant fraction of the gas accretes , or whether instead the warp is advected inward along with the gas . to go further , we need to elaborate on the mechanisms that can allow annuli to exchange angular momentum . if the disk is thin enough , specifically , if @xmath172 then the evolution of the tilt as a function of radius , like the evolution of the surface density , is diffusive ( @xcite ) . if , conversely , @xmath173 then the warp is communicated through the disk by waves ( @xcite ) . assuming that @xmath174 in geometrically thin agn disks , then most parts of the disk ( except , perhaps , the innermost regions ) are likely to fall into the diffusive regime , which we will consider exclusively from now on . we will also ignore the complications of self - gravity ( papaloizou , terquem , & lin 1998 ) , even though , as already noted , the masing disks cited as motivation are probably self - gravitating . in the diffusive limit , the evolution of the disk can be described to a first approximation using the simple theory described by pringle ( 1992 ) . in this theory , the stresses within the disk are reduced to two viscosities ; @xmath175 , which describes the usual kinematic viscosity leading to inflow , and @xmath176 , which is related to the stresses which act to flatten out any warp . the timescales for accretion @xmath177 and for decay of the warp @xmath178 are related to these viscosities in the usual way : @xmath179 the relation between these two viscosities can be calculated either analytically ( @xcite ) or via numerical simulations ( @xcite ) . both approaches suggest that @xmath180 this is a striking result . it implies that warps are likely to decay rapidly and will require strong forcing if they are to persist for long periods . torques from the reprocessing of radiation from the central source by the disk provide one well studied forcing term , which could act to excite warps even in initially planar disks ( @xcite ; @xcite ) . the large ratio of @xmath181 implied by the above equation , however , severely limits the circumstances in which radiation driven warping can overcome the disk s intrinsic tendency to flatten out . torques from a disk wind are potentially more promising ( @xcite ; @xcite ) , though harder to evaluate in detail . further complications ensue if the black hole is rotating , even modestly . orbits whose angular momentum is inclined to the spin axis of a rotating hole will precess due to the lense - thirring effect ( `` frame dragging '' ) . this forced precession is only significant at small radii , where it will affect the inclination of the disk . again , the simplest case to analyze is the one in which the inner disk is in the diffusive regime of warp evolution . differential precession , acting on an initially flat but misaligned disk , will rapidly induce a sharp twist in the disk close to the marginally stable orbit . this will be flattened out by viscosity , with the consequence that the disk near a spinning hole will be driven into the equatorial plane of the hole , regardless of the inclination at large radii ( bardeen & petterson 1975 ) . over longer timescales , the torque between the disk and the hole will also act to realign the system until the angular momentum vector of the hole , and of the disk at _ all _ radii , are coincident . since normally @xmath182 , this balancing act will usually result in a large change in the spin axis of the black hole until it roughly matches the angular momentum vector of gas in the outer disk . how quickly this process occurs depends , again , largely on @xmath181 , with large values of this ratio leading to rapid alignment ( @xcite ) . in some circumstances , the inner accretion disk around a rotating black hole may be thick enough that warps display wave - like evolution . in this case , the inner disk may be able to remain misaligned with respect to the spin axis of the black hole , even in the presence of dissipation ( demianski & ivanov 1997 ; @xcite ; @xcite ) . this could have important consequences for the direction of jets launched from the inner regions of the disk . unlike in the case of warped disks , there is little observational evidence to suggest that agn disks or any disk that is not strongly forced , for example by a binary companion are significantly eccentric , though some sort of asymmetry is deduced in as many as 60% of agn with double - peaked balmer lines ( @xcite ) . theoretical studies , however , raise the interesting possibility that there could be circumstances in which isolated accretion disks spontaneously develop eccentricity . in particular , a two - dimensional disk model in which the viscosity follows a navier - stokes form , with shear viscosity @xmath183 and bulk viscosity @xmath184 , is unstable to the growth of eccentricity if @xmath185 that is , if the vertically integrated viscosity is increasing too rapidly with surface density ( @xcite ) . if the bulk viscosity is small , as is often assumed in accretion disk modeling , then this result implies that almost all disks ought to be unstable to the growth of eccentricity ( @xcite ; @xcite ) . earlier , of course , we specifically warned of the dangers of assuming that angular momentum transport in an mhd flow can be described via an effective navier - stokes viscosity . indeed , ogilvie ( 2001 ) shows that if angular momentum transport follows an alternative analytic prescription , which is motivated by the phenomenology of mhd turbulence , then the wholesale instability of disks to eccentricity can be avoided . the important point is that the behavior of eccentric disks which at a minimum will exist in galactic nuclei following the tidal disruption of stars ( @xcite ) , or when binary black holes are present is closely related to subtle aspects of turbulent transport within the disk . predictions as to their evolution can not yet be made with confidence . fromang , s. , terquem , c. , balbus , s. a. , & de villiers , j .- 2004 , in `` extrasolar planets : today and tomorrow '' , eds . j .- beaulieu , a. lecavelier des etangs & c. terquem . ( san francisco : asp conference series ) , in press ( astro - ph/0402373 )
we present the _ chandra _ study of the intermediately distant ( ) cluster of galaxies ms 0839.9 + 2938 . by performing both the projected and deprojected spectral analyses , we find that the gas temperature is approximately constant at about 4 kev in kpc . in the inner regions , the gas temperature descends towards the center , reaching kev in the central kpc . this infers that the lower and upper limits of the mass deposit rate are yr and yr , respectively within kpc where the gas is significantly colder . along with the temperature drop , we detect a significant inward iron abundance increase from about 0.4 solar in the outer regions to solar within the central kpc . thus ms 0839.9 + 2938 is the cluster showing the most significant central iron excess at . we argue that most of the excess iron should have been contributed by sne ia . by utilizing the observed sn ia rate and stellar mass loss rate , we estimate that the time needed to enrich the central region with excess iron is gyr , which is similar to those found for the nearby clusters . coinciding with the optical extension of the cd galaxy ( up to about kpc ) , the observed x - ray surface brightness profile exhibits an excess beyond the distribution expected by either the model or the nfw model , and can be well fitted with an empirical two- model that leads to a relatively flatter mass profile in the innermost region .
astro-ph0506260
the metallicity of the intra - cluster medium ( icm ) in clusters of galaxies is a sensitive tracer of both the history of the hierarchical clustering and the chemical evolutions of member galaxies and clusters themselves . since the icm has been enriched by the supernova explosions , the study of the metal abundance gradient in clusters , especially in those distant ones , may provide us with crucial constraints on the sn rate , stellar mass loss rate , and thus star formation rate . along with the metal enrichment , a tremendous amount of the supernova energy has been inputted into the icm , primarily in the form of gas dynamic energy that may be used to heat the icm . thus detailed measurements of the metal distribution in clusters as a function of redshift may also provide us with valuable hints to investigate how the icm has been heated , a puzzling question that has plagued astronomers for a long time since the absence of massive cooling gas has been confirmed in many clusters ( see , e.g. , makishima et al . 2001 for a review ) . in the past five years , the inward iron abundance increase , first discovered with _ ginga _ , _ rosat _ and _ asca _ in some nearby clusters that usually host a giant cd galaxy ( e.g. , koyama et al . 1991 ; fukazawa et al . 1994 ; ezawa et al . 1997 ; xu et al . 1997 ; ikebe et al . 1999 ; finoguenov et al . 2000 ) , has been revealed in more sources ( e.g. , a sample of 12 nearby rich clusters at @xmath13 that have cool cores ; de grandi et al . 2004 ) and in much more details by the high spatial resolution observations of _ chandra _ and _ xmm _ , which allows us to measure the abundances of iron and other metals with sufficient accuracy to deduce the sn ia and sn ii enrichments and their spatial variations . in clusters that exhibit a central iron excess , it has been found that the average iron abundance is typically @xmath14 solar within a radius of 3080 kpc , which descends to about @xmath15 solar in the outer regions . in several most neighboring sources , such as m87/the virgo cluster ( matsushita et al . 2003 ) , ngc 1399/the fornax cluster ( buote 2002 ) , ngc 5044/the ngc 5044 group ( buote et al . 2003 ) , ngc 1129/awm 7 ( furusho et al . 2003 ) , and ngc 4696/the centaurus cluster ( sanders & fabian 2002 ) , the average iron abundance is found to increase further to 0.71.5 solar within the central 1535 kpc , although in the innermost 2 kpc of m87 and ngc 4696 the iron abundance decreases down to @xmath16 solar to form a central dip . in m87 ( matsushita et al . 2003 ) , ngc 1399 ( buote 2002 ) , ngc 5044 ( buote et al . 2003 ) and a sample of 19 clusters ( tamura et al . 2004 ) , the spatial distributions of the silicon and sulphur abundances are likely to follow that of iron . the distribution of the oxygen abundance , on the other hand , is roughly uniform within the errors . these results strongly support the idea that there has been an excess sn ia enrichment in the cluster s inner region . for the time being , the spatial distributions of metal abundances in clusters at @xmath17 are less understood due to the instrumental limitations , although there is evidence that the same central iron abundance excess may also exist ( e.g. , allen et al . 2001 , 2002 ; iwasawa et al . 2001 ) . as a step to approach more distant clusters , in this paper we present the _ chandra _ study of the iron abundance distribution in ms 0839.9 + 2938 , an intermediately distant cluster located at @xmath0 ( stocke et al . the cluster was listed as an extended x - ray source and identified as a cluster in the _ einstein _ extended medium sensitivity survey ( gioia et al . it hosts a giant cd galaxy at its dynamic center , which is a weak , compact radio source with an intense , extended @xmath18 emission ( nesci et al . the advent of studying this source is that since its average gas temperature is low ( @xmath19 kev ; see 4.2 ) , it is easier to measure the abundance gradient accurately due to the large equivalent widths of fe-@xmath20 and -@xmath21 lines . we organize the paper as follows . in 2 , we describe the observation and data reduction . in 3 and 4 , we present the imaging and spectral analyses , respectively . in 5 , we investigate the spatial distributions of the gas and gravitating mass as well as gas cooling . finally , we discuss and summarize our results in 6 and 7 , respectively . throughout the paper , we adopt the cosmological parameters @xmath22 km s@xmath5 mpc@xmath5 , @xmath23 and @xmath24 , so that @xmath25 corresponds to @xmath26 kpc at the redshift of the cluster . in order to compare directly with previous results , we use the older solar abundances from anders & grevesse ( 1989 ) , where the iron abundance relative to hydrogen is @xmath27 in number , which is about 46% higher than the currently preferred meteoritic value ( mcwilliam 1997 ; grevesse & sauval 1999 ) . unless stated otherwise , the quoted errors are the 90% confidence limits . the _ chandra _ observation of ms 0839.9 + 2938 was carried out on january 29 , 2001 for a total exposure of 29.7 ks with the chips 2 , 3 , 6 , 7 , 8 , and 9 of the advanced ccd imaging spectrometer ( acis ) . the center of the cluster was positioned close to the nominal aim point on the s3 chip ( ccd 7 ) , so that most of the x - ray emission of the galaxy was covered by the s3 chip . the events were collected with frame times of 3.2 s and telemetered in the faint mode . the focal plane temperature was set to be -120 @xmath28c . in this work , we used the _ chandra _ data analysis package ciao software ( version 3.0 ) to process the data extracted from the s3 chip only . we kept events with _ grades 0 , 2 , 3 , 4 , and 6 , and removed all the bad pixels , bad columns , and columns adjacent to bad columns and node boundaries . in order to detect occasional background flares , whose effects are particularly significant on the backside - illuminated s1 and s3 chips , we examined the lightcurves of the background regions on the s3 chip both in 2.57 kev , where the background flares are expected most significant , and in 47 kev , where the source contamination is least . we found that there was almost no strong background flares that increased the count rate to over 20% more than its mean value . by excluding the high background intervals we obtained a net exposure of 27.1 ks for the analyses . we restrict our spectral study in 0.78.0 kev to avoid the calibration uncertainties at the low energy end and the instrumental background at the high energy end . in the spectral fittings , we have taken into account corrections for the acis quantum efficiency degradation and used the xspec v11.2.0 software . in figure 1 , we present both the smoothed 0.78.0 kev image of the central @xmath29 ( @xmath30 kpc ) of ms 0839.9 + 2938 in the logarithmic scale , and the corresponding dss optical image on which the x - ray contours are overlaid . the smoothed x - ray image was created by using the ciao tool csmooth with a minimum significance of 3 and a maximum significance of 5 , and has been background - subtracted and exposure - corrected . as can be seen clearly , the diffuse x - ray emission is strongly peaked at the cluster s center . within the innermost @xmath31 ( @xmath32 kpc ) , it appears to be elongated approximately in the northwest - southeast direction , where it follows the distribution of the optical lights . outside this region the diffuse emission is elongated in the north - south direction out to @xmath33 ( @xmath34 kpc ) . the x - ray emission of the cluster extends to at least @xmath35 ( @xmath36 kpc ) , where the background begins to dominate the flux . there is no apparent substructures in x - rays , which indicates that the cluster has not experienced any major merger events in the recent past , and is likely to be in a relaxed state . since the astrometric errors of the _ chandra _ observation are about @xmath25 , the position of the x - ray peak ( ra=08h42m56.0s , dec=+29d27m27.2s , j2000 ) is in excellent agreement with the optical center of the cd galaxy 2masx j08425596 + 2927272 ( becker et al . 1995 ) to within @xmath37 . based on the studies of brandt et al . ( 2000 ) and mushotzky et al . ( 2000 ) , we estimate that there may be 510 unrelated x - ray sources in the @xmath38 region of the cluster . by utilizing a wavedetect detection algorithm and a source detection threshold of 10@xmath39 , and then crosschecking the results both by using the celldetect detection algorithm and by eye , we detected 10 x - ray point sources in @xmath38 , among which less than 1 is expected to be a false detection . we find that two of them have optical and radio counterparts within @xmath25 . one is cgcg 150 - 019 , an sb galaxy located at @xmath40 ( wegner et al . 2001 ) , and the other is first j084251.9 + 292825 , a seyfert galaxy in ms 0839.9 + 2938 ( hutchings & edwards 2000 ) . in the analysis that follows , all the detected x - ray point sources were excluded . in figure 2 , we show the azimuthally - averaged surface brightness profile of the cluster in 0.78 kev that has been corrected with a weighted exposure map . all the detected point sources have been excluded , while the background has not been subtracted . we find that in @xmath41 kpc , the observed x - ray surface brightness profile can be well described ( @xmath42 ) by a model consisting of a standard @xmath12 component ( jones & forman 1984 ) and a spatially uniform background as @xmath43^{0.5 - 3\beta}+s_{\rm bkg},\\\ ] ] where @xmath44 is the core radius , @xmath12 is the slope , @xmath45 is the background , and @xmath46 is the normalization . the best - fitting parameters are @xmath47 @xmath48 kpc and @xmath49 . however , when we extrapolate the model inwards , the fit becomes worse ( @xmath50 @xmath48 kpc , @xmath51 , and @xmath52 ) , which significantly underestimates the data in @xmath53 ( @xmath54 kpc ) , leaving an obvious central excess emission above the model . such central excess emission has been seen in many clusters , e.g. , the centaurus cluster ( ikebe et al . 1999 ) , abell 1795 ( xu et al . 1998 ) , abell 1983 ( pratt & arnaud 2003 ) , and cl0024 + 17 ( ota et al . 2004 ) . it may be attributed to either a temperature drop , or a metal abundance increase , or an excess of the gas density , all within the central region . in 5.1 , we will present detailed model fittings to describe the observed x - ray surface brightness profile by taking into account both the temperature and abundance gradients obtained from the spectral analyses ( 4 ) . before the direct spectral fittings , we first study the two - dimensional hardness ratio distribution of the hot diffuse gas in the cluster . the hardness ratio hr is defined as the ratio of the background - subtracted and exposure - corrected counts in 1.48.0 kev to those in 0.71.4 kev . as can be seen in figure 3 , in @xmath55 kpc the hardness ratio is approximately constant at a value of @xmath56 , with typical errors of @xmath57 . assuming an average metallicity of 0.4 solar and the galactic absorption @xmath58 @xmath59 ( dickey & lockman 1990 ) , this is consistent with the hardness ratio of an isothermal plasma at @xmath60 kev . in @xmath61 kpc , there are two relatively colder ( 90% confidence level ) regions located at the southeast and northwest of the center , respectively . in the southeast colder region the gas temperature is estimated to be @xmath62 kev ( hr@xmath63 ) , which is close to that of the gas in the northwest colder region ( @xmath64 kev , hr@xmath65 ) . in @xmath66 kpc , where the hardness ratio is @xmath67 with typical errors of @xmath68 , the gas appears to be significantly colder than in the outer regions , indicating the existence of a plasma at @xmath69 kev . these results agree very well with those obtained in the detailed spectral fittings ( 4.3 ) . we extracted the total spectrum of the cluster using a circular region filter with a radius of @xmath70 ( @xmath71 kpc ) , which is centered at the x - ray peak , excluding all the detected point sources . the background spectrum was extracted from a separate region on the s3 chip that is far away enough from the cluster . we also have used the @xmath72 blank fields for background , and obtained essentially the same results as those shown below . at the boundary of the so defined cluster region , the count flux is about 3 times more than that of the background . the resulting background - subtracted spectrum contains a total of about 14,100 counts in 0.78.0 kev , showing a strong line feature at @xmath73 kev that corresponds to a rest - frame energy of about 6.7 kev where the k@xmath74 lines of he - like iron should reside . the appearance of this fe k@xmath74 feature indicates that the average gas temperature of the cluster should be within 2.75.2 kev . we have fit the spectrum with a model consisting of a single absorbed isothermal component ( table 1 ) . first we choose the apec model ( smith et al . 2001 ) as the isothermal component , and fix the redshift and absorption to @xmath0 and the galactic value , respectively . the fit is marginally acceptable ( @xmath75 ) , which gives an average gas temperature of @xmath76 kev , and an average metal abundance of @xmath77 solar . in order to improve the fit we have attempted to apply the vapec model instead of the apec model . for ms 0839.9 + 2938 , the redshift is large enough to move the ly@xmath74 lines of silicon to lower energies to avoid the calibration uncertainties caused by the mirror ir edge . thus in the fitting , we leave both the fe and si abundances free , and tie the abundance of ni to that of fe . although we have applied the contamination file from the latest caldb to correct for the acis deficiency in low energies , the degradation of energy resolution still does not recover . for an approximate estimation , we tie the abundances of other @xmath78-elements ( o , ne , mg , s , al , ar and ca ) together . the model yields a gas temperature of @xmath79 kev , and abundances of @xmath80 solar , @xmath81 solar , and @xmath82 solar that is mostly dominated by the s abundance . although the vapec fit ( @xmath83 ) is better than the apec fit , there are still residuals remaining at about 0.9 kev , where the model underestimates the data . these residuals may mostly caused by the excess fe-@xmath20 emissions arising from the colder gas in the central region , whose existence has been inferred in the study of the hardness ratio distribution ( 4.1 ) . in either the apec or vapec fitting , even if the absorption is left free , the obtained column density is still consistent with the galactic value , and the fit is not improved . we also have examined if the fit to the total spectrum can be further improved by the use of a model consisting of two isothermal spectral components , both subjected to a common absorption ( table 1 ) . for this purpose we employ two vapec components , whose absorption is again fixed to the galactic value because allowing it to vary does not improve the fit . the metal abundances of each vapec component are grouped in the same way as in the single vapec fitting , and the abundances of the same element of the two vapec components are tied together . we perform the f - test and find that at the 95% confidence level the two - temperature fit ( @xmath84 ) is better than the one - temperature apec or vapec fit . the obtained gas temperatures for the two components are @xmath85 kev and @xmath86 kev , and the metal abundances are @xmath87 solar , @xmath88 solar , and @xmath89 solar . with these best - fit parameters we calculate that the unabsorbed 0.510 kev flux of the cluster is @xmath90 erg @xmath59 s@xmath5 , corresponding to a luminosity of @xmath91 erg s@xmath5 . in order to investigate the temperature and abundance gradients , we divided the cluster into five annular regions , all centered at the x - ray peak . as the first step , we study the projected spectra extracted in these annuli . the background spectrum was again extracted in a region on the s3 chip far away enough from the cluster ; the use of the _ chandra _ blank fields yields essentially the same results . assuming that the gas in each annulus is in the thermal equilibrium , we fit the spectra using a single absorbed apec model with the absorption fixed to the galactic value ; allowing the absorption to vary does not improve the fits . as is shown in figure 4 and table 2 , the fit is acceptable for each region . the gas temperature is found to be approximately constant at @xmath92 kev outside about @xmath93 kpc , while it descends to @xmath94 kev in the central @xmath3 kpc , as has been expected in the study of the hardness ratio distribution ( 4.1 ) . on the other hand , the average metal abundance , which is mainly determined by iron , increases towards the center from a roughly constant value of 0.4 solar in the outer regions to about 0.8 solar within the central @xmath3 kpc . instead of the apec model , we have also attempted to apply the vapec model ( fig . 4 and table 2 ) . for better statistics , all the abundances of the @xmath78-elements are tied together , and the absorption is again fixed to the galactic value . we find that the fits are acceptable too , and the obtained gas temperatures agree nicely with those obtained with the apec model . in the regions outside @xmath95 kpc , the iron abundance varies slowly in a range of 0.250.42 solar , while in the central region , it increases drastically to about 0.8 solar . the spatial distribution of the average abundance of the @xmath78-elements , however , is not well determined due to large errors . next , we perform the deprojected analysis of the spectra extracted in the inner four regions by using the standard `` onion peeling '' method ( fig . 4 and table 2 ) . assuming that the gas temperature and abundance are uniform in each spherical shell , and that the gas density follows the best - fit profile derived in 5.1 , we subtract emissions from the outer shells projected on the inner shells . we fit the deprojected spectra with either the apec or vapec model whose absorption is fixed to the galactic value . we find that the deprojected spectra can be well fitted with either of the models . the gas temperatures obtained with the apec model agree very well with those obtained with the vapec model , and both the average metal abundance of the apec model and the iron abundance of the vapec model show significant central increase in the innermost region as has been found in the projected analysis . for each model the best - fit gas temperatures and metal abundances obtained in the projected and deprojected analyses are consistent with each other . in the innermost region , the deprojected temperatures are only slightly lower than , but still consistent with the projected values . as we have shown above , both the projected and deprojected spectral analyses indicate that in ms 0839.9 + 2938 there is a central temperature drop within a radius of about @xmath96 kpc . according to the deprojected apec fit , the gas temperature descends from @xmath97 kev in @xmath98 kpc to @xmath99 kev in @xmath100 kpc , which can be approximated by a power - law profile @xmath101 , where @xmath102 kev , @xmath103 and @xmath104 kpc ( fig . 4a ) . in the same region where the cold gas is present , the iron abundance shows a clear tendency to increase . in the projected analysis , both the iron - dominated metal abundance of the apec model and the iron abundance of the vapec model in the central @xmath3 kpc region are significantly higher than their counterparts in @xmath105 kpc at the 90% confidence level . they are also apparently higher than the abundances in @xmath106 kpc and @xmath107 kpc , respectively , with the 90% confidence ranges overlap with each other only slightly . in the deprojected analysis , although the obtained errors are much larger , the average metal abundance and iron abundance still show a tendency to increase inward . at the 90% confidence level , their values inferred for the @xmath100 kpc region are significantly higher than those for the @xmath105 kpc region . in order to crosscheck these results with better statistics , we divide the cluster into the inner ( @xmath108 kpc ) and outer ( @xmath109 kpc ) regions , and fit their deprojected spectra with the an absorbed vapec model ( table 3 ) . in the fitting the absorption is fixed to the galactic value , and the metal abundances are grouped in the same way as in the fittings of the total spectrum . we find that the fits are acceptable for both regions . at the 90% confidence level , the central temperature decrease and abundance increase can be confirmed . the silicon abundance and the average abundance of other @xmath78-elements , however , show no tendency to vary between the two regions . since the iron abundance is determined as much by the fe - l emission lines kev as by the fe - k lines , the obtained inward iron abundance increase might be an artifact , which is actually caused by the enhancement of the fe - l blend due to an inward temperature drop that is even steeper than has been modeled . we have examined this possibility . first we exclude the fe-@xmath20 lines completely from the spectra and use the fe-@xmath21 lines only to measure the iron abundance . the obtained iron abundance does show a tendency to increase inwards , although it is less significant due to poorer statistics . second , we fit the deprojected spectrum of the innermost region by using a single absorbed vapec model with the absorption fixed to the galactic value and the iron abundance fixed to 0.47 solar ( the upper limit of the iron abundance in the third annular region ; table 2 ) . we do obtain a lower temperature of @xmath110 kev as is expected . however , the fit appears to be worse and should be ruled out at the 90% confidence level in terms of f - test , showing obvious spectral residuals in @xmath111 kev as well as at @xmath112 kev , where the fits to the fe-@xmath21 and fe-@xmath20 lines are worse . this indicates that a central abundance increase is much more preferred than an even steeper temperature gradient . third , we study the two - dimensional fit - statistic contours of temperature and iron abundance at the 68% , 90% and 99% confidence levels for the innermost region and the third region , respectively , all obtained in the deprojected vapec fitting ( fig . 5 ) . as can be clearly seen the central iron abundance is higher than that of the third region at a significance of 99% . consequently , we conclude that the possibility of the detected central iron abundance increase being an artifact should be excluded . note that the coexistence of the temperature drop and abundance increase may indicate that they are physically correlated , since the higher the metallicity is , the faster the gas cools down . assuming that 1 ) the gas is ideal and in the thermal equilibrium state , and 2 ) the spatial distribution of the gas is spherically asymmetric , we deduce the three - dimensional gas density distribution @xmath113 by fitting the observed radial x - ray surface brightness profile @xmath114 , which was extracted in 0.78.0 kev , azimuthally - averaged , and exposure - corrected . the calculations was performed in a projected way by using @xmath115 where @xmath116 is the cooling function calculated by taking into account both the temperature and abundance gradients as is modeled by the best - fit deprojected apec parameters , and @xmath45 is the background . first we assume that the gas density follows the @xmath12 distribution @xmath117 ^{-3\beta/2},\ ] ] where @xmath118 is the core radius and @xmath12 is the slope . we find @xmath119 kpc and @xmath120 . the fit , however , is unacceptable ( @xmath121 ) , showing strong residuals in @xmath122 kpc where the model significantly underestimates the data ( fig . 2 ) . in fact , the gas density derived in such a way is lower than that obtained in the deprojected spectral analysis ( 4.3 ) by about 45% . this confirms the result in 3.2 , although the central brightness excess above the model is less prominent since the effects of the central temperature drop and abundance increase are both counted in , which enhance the model emission in the central region . next we examine if @xmath114 can be reproduced by adopting the universal gravitating mass distribution as has been found in the cosmological n - body simulations ( navarro et al . ( 1997 ) @xmath123 where @xmath124 is the gravitating mass density , @xmath125 is the scale radius , @xmath126 is the critical density of the universe at the observed redshift , and @xmath127 is the characteristic density contract that can be expressed in term of the equivalent concentration parameter _ c _ as @xmath128}.\ ] ] as is shown in makino et al . ( 1998 ) and wu et al . ( 2000 ) , under the assumptions of 1 ) spherical symmetry , 2 ) hydrostatic equilibrium , and 3 ) ideal and isothermal gas , the gas distribution can be given in an analytic form @xmath129 where @xmath130 is the dimensionless radius , @xmath131 is the index , and @xmath132 is the average molecular weight for a fully ionized gas . note that in eq.(6 ) a background density at infinity has been subtracted to avoid the divergence in integrating the x - ray emission along the line of sight . as are shown in figure 2 and table 4 . we find that , comparing with the @xmath12 model , the fit is improved in the central region , giving @xmath133 kpc , @xmath134 , and @xmath135 . however , it still systematically underestimates the data in @xmath136 kpc . in order to improve the fit in the central region , we have attempted to add an additional @xmath12 component to the @xmath12 model to account for the central excess emission , for which the gas density distribution is @xmath137 ^{-3\beta_{1 } } + n_{g2,0}^{2 } \left [ 1 + ( r / r_{c2})^{2 } \right ] ^{-3\beta_{2 } } \right\ } ^{1/2}.\ ] ] this empirical two - component model has been successfully applied to many clusters of galaxies as well as some groups of galaxies ( see , e.g. , makishima et al . 2001 and references therein ) . we find that the model can best fit the observed surface brightness profile throughout the spatial range spanned by the cluster and yield an acceptable fit . the best - fit parameters are @xmath138 kpc and @xmath139 for one @xmath12 component , which mostly describes the surface brightness profile in the outer region , and @xmath140@xmath48 kpc and @xmath141 for another , which describes the central excess brightness . the distribution of the diffuse x - ray emission in different energy bands is roughly symmetric outside to about @xmath35 ( @xmath36 kpc ) in ms 0839.9 + 2938 , indicating that the cluster is nearly relaxed . in this case the x - ray imaging spectroscopy can be considered as a reliable tool to deduce the mass distributions in the cluster . assuming that the gas is ideal and the cluster is in the hydrostatic equilibrium state , and utilizing the best - fit temperature gradient obtained in the deprojected analyses and the gas density profile deduced with the two-@xmath12 model , we calculate the three - dimensional gravitating mass distribution by using @xmath142 where @xmath143 is the total mass within the radius @xmath144 , g is the gravitational constant , and @xmath145 is the proton mass . here the effects of non - thermal turbulent , magnetic pressure , and cosmic ray pressure are not taken into account . the resulting distribution of the total gravitating mass is shown in figure 6 , along with the 90% errors determined by performing the monte - carlo simulations that account for the range of temperature and gas profiles allowed by the data . the total mass distribution shows an excess beyond that predicted by the @xmath12 model in @xmath146 kpc , which is roughly the optical boundary of the cd galaxy ( nesci et al . 1989 ) and the region where the excess x - ray emission , temperature drop , and iron abundance increase are observed . the excess mass ( @xmath147 ) is comparable to the mass of the cd galaxy as is inferred from its optical luminosity ( @xmath148 within @xmath11 kpc ; nesci et al . 1989 ) . at the virial radius @xmath149 ( @xmath150 kpc ) , within which the average mass density is 200 times the current critical density of the universe , the extrapolated gravitating mass is @xmath151 . in figure 6 , we also show the calculated gas mass fraction along with the 90% errors . it increases from about 0.05 at the center to about 0.1 at @xmath152 kpc , and then keeps as a constant in the outer regions . for @xmath153 ( spergel et al . 2003 ) , we obtain @xmath154 , which is self - consistent with the value adopted in this work . with the gas density @xmath113 deduced with the best - fit two-@xmath12 model ( 5.1 ) , we estimate the gas cooling time as @xmath155 ( cf , sarazin 1986 ) , where both continuum and line emissions are included ( fig . we find that @xmath156 yr at the center , which is much shorter than the age of the universe at the cluster s redshift . the cooling radius @xmath157 , defined as the radius where the cooling time is equal to the age of the universe , is @xmath158 kpc . this is roughly the radius within which the gas temperature begins to descend inwards . in order to determine the mass deposit rate of the cooling processes , we fit the spectrum extracted in the two innermost annuli ( @xmath159 kpc ) where the gas temperature is significantly lower than that of the outer regions . we use the cooling flow model mkcflow , which is absorbed by a column density fixed to the galactic value ( fig . we find that when the low temperature @xmath160 of mkcflow is fixed to 0.08 kev , the model yields a rather poor fit to the data ( @xmath161 ) , giving rise to too much cool gas and thus overestimating the data in @xmath162 kev . when @xmath160 is left free , the fit becomes acceptable ( @xmath163 ) , but it requires @xmath164 kev , which may suggest that the spectrum is dominated by a hot component . we then add an additional ambient isothermal component ( mekal ) into the model , tying its temperature to @xmath165 of mkcflow . when @xmath160 is fixed to 0.08 kev , we obtain @xmath166 kev and @xmath167 yr@xmath5 ; when @xmath160 is free , we obtain @xmath168 kev , @xmath169 kev , and @xmath170 yr@xmath5 . whether or not @xmath160 is fixed , the mkcflow+mekal fit is acceptable ( @xmath171 and @xmath172 , respectively ) , but @xmath160 and @xmath173 are far from being well constrained . on the other hand , the two - temperature apec model also gives a good fit ( @xmath171 ) to the data . the derived best - fit low and high temperatures are @xmath174 kev and @xmath175 kev . thus , given the current data quality and instrumental capabilities , the two - phase isothermal model can not be distinguished from the single mkcflow model with a free @xmath160 , or the mekal+mkcflow model . although not well constrained , the cooling flow model can not be ruled out for this cluster . the inward iron abundance increase found in many nearby clusters and groups around / in the giant , dominating elliptical galaxy at the center seems to follow a somewhat universal profile . actually , when we compare the iron abundances in the inner ( @xmath176 ) and outer regions ( @xmath177 and @xmath178 , respectively ) of the clusters in which an inner abundance enhancement is observed ( buote et al . 2003 ; nevalainen et al . 2001 ; pratt @xmath179 arnaud , 2003 ; takahashi @xmath179 yamashita 2003 ; finoguenov et al . 2004 ; tamura et al . 2004 ) , we find that within the errors the ratio @xmath180 is roughly a constant as a function of either the icm temperature or the cluster s b - band luminosity ( fig all these results strongly imply that the observed central iron excess has probably been caused by the same enrichment process associated with the central dominating galaxy , which is a cd galaxy in most cases . meanwhile , the high spatial resolution observations of the nearby giant ellipticals m87 ( matsushita et al . 2003 ) and ngc 5044 ( buote et al . 2003 ) revealed that the spatial distribution of oxygen abundance is approximately uniform out to about 4070 kpc , inferring that the metals synthesized in sne ii are evenly distributed in the cluster / group . thus , the central iron abundance increase should be ascribed mainly to sne ia . by performing both the projected and deprojected spectral analyses , we find that in ms 0839.9 + 2938 the measured average iron abundance is about 0.4 solar in the outer regions , while it increases to @xmath181 solar within the central @xmath182 kpc at the 90% confidence level . reports about detections of such a significant central iron excess in clusters at @xmath9 are rare in literatures . in 4c+55.16 ( @xmath183 ; iwasawa et al . 2001 ) the metal abundance increases from @xmath16 solar in outer regions to @xmath184 solar within the central @xmath185 kpc . in abell 2390 ( @xmath186 ; allen et al . 2001 ) , it increases from @xmath187 solar in @xmath188 kpc to @xmath189 solar within @xmath190 kpc . however , note that the results of the latter two cases were acquired by performing projected analysis at the 68% confidence level , and the abundance is an average value for both iron and @xmath78-elements , although it is iron - dominated . in ms 0839.9 + 2938 , the excess iron mass within @xmath191 kpc is estimated to be @xmath192 , which is similar to that of abell 2390 ( @xmath193 within @xmath194 kpc , based on the data of allen et al . the derived @xmath195 obviously exceeds the iron enrichment caused only by the stellar winds and ram pressure stripping of metal - enriched gas ( e.g. , trager et al . 2000 ; ciotti et al . therefore , we speculate that a significant part of the excess iron may have been blown into the icm directly by the sn ia explosions . by reproducing the observed iron - to - silicon abundance ratio in the central @xmath196 kpc of ms 0839.9 + 2938 ( fe / si=@xmath197 solar at the 68% confidence level ) , where a clear iron abundance increase is seen , we have attempted to estimate the sn ia contribution to the iron enrichment . the current supernova models give very different yields to the metals ( gibson et al . when we adopt the theoretical sn ia yields obtained with the w7 model and the weighted sn ii yields in nomoto et al . ( 1997 ) , where an imf slope of @xmath198 ( salpeter 1955 ) and a stellar mass range of @xmath199 were assumed for the sn ii progenitors , we find that the sn ia contribution to the total iron mass is @xmath200 , corresponding to an iron abundance of @xmath201 solar . this fraction is consistent with those found in ngc 5044 ( @xmath202% ; buote et al . 2003 ) , m87 ( @xmath203% ; matsushita et al . 2003 ) , ngc 1399 ( @xmath204 ; buote 2002 ) , and the rgh 80 group ( @xmath204 ; xue et al . we are not able to derive reasonable results by using the sn ia yields obtained with either the wdd1 or wdd2 model in nomoto et al . ( 1997 ) , because both models give rise to too much silicon . using the same method as above we estimate that for the @xmath205 kpc region , where the observed fe / si ratio is @xmath206 solar ( 68% confidence ) , the sn ia contributions is @xmath207 , or @xmath208 ( w7 model ) . by comparing these results with those for the inner region , we find that the spatial distribution of the sn ii contribution to iron is roughly constant at about @xmath209 solar , which is close to that in finoguenov et al . ( 2002 ; @xmath210 solar ) and matsushita et al . ( 2003 ; @xmath211 solar ) . it is crucial to examine if there has been enough time for sne ia to enrich the cluster s central region with the observed amount of excess iron . since the sn ia enrichment of iron in the icm is determined both by iron blown out directly into the icm during the sn ia explosions and by iron lost in the stellar winds , we calculate the enrichment time @xmath212 ( b@xmath213hringer et al . 2004 ) using @xmath214,\ ] ] where @xmath215 is the iron mass contributed by sne ia , @xmath216 ( @xmath217 ; nesci et al . 1989 ) is cluster s b - band luminosity , @xmath218 is the direct sn ia iron - enriching rate in units of @xmath219 yr@xmath5 @xmath220 , and @xmath221 is the iron - enriching rate due to the stellar mass loss in units of @xmath219 yr@xmath5 @xmath220 . as is shown above , we assume that sne ii evenly enriched the gas throughout cluster , yielding an iron abundance of 0.08 solar , so by subtracting the sn ii contribution , we obtain @xmath222 . in sn ia explosions the rate of iron mass that has been directly transferred into the icm is @xmath223 where @xmath224 is the iron yield per sn ia event , and @xmath225 is the sn ia rate in units of snu ( 1 snu = 1 supernova @xmath226 century@xmath5 ) . in calculations , we choose @xmath227 and 0.7 @xmath219 , respectively , and adopt either a temporally constant @xmath225 of @xmath228 snu ( cappellaro et al . 1999 ) , or an observation - constrained time - dependent @xmath225 that increases with the redshift until @xmath229 and then drops ( dahlen et al . 2004 ) . in the stellar winds , the iron loss rate is @xmath230 where @xmath231 is the iron mass fraction in stellar winds , and @xmath232 is the age in units of 15 gyr ( ciotti et al . we list the calculated enrichment times and redshifts at which the sn ia enrichment is expected to start in table 5 . we find that when the supernova rate was a constant in the past , the derived enrichment time is too close to or even longer than the hubble time at the cluster s redshift ( @xmath233 yr ) . when the observed time - dependent supernova rate provided in dahlen et al . ( 2004 ) is adopted , the derived enrichment times are 7.9 gyr and 6.4 gyr for @xmath234 and @xmath235 @xmath219 , respectively . for comparison , we also perform similar calculations for the inner @xmath236 kpc region of abell 2390 and find that the resulting enrichment time is 6.610.7 gyr . these results agree with those deduced for the virgo cluster , the centaurus cluster , the perseus cluster and abell 1795 ( @xmath237 gyr ; b@xmath213hringer et al . 2004 ) , indicating that the excess iron can be entirely produced by the central brightest galaxy ( de grandi & molendi 2001 ; de grandi et al . since only few sne ia are observed at @xmath238 , we predict that the most distant clusters showing a significant central iron excess should be detected at no farther than @xmath239 . the derived x - ray temperature , x - ray luminosity and virial mass at @xmath149 of ms 0839.9 + 2938 are @xmath240 kev , @xmath241 @xmath242 erg s@xmath5 , and @xmath243 , respectively . within the errors , these values are consistent with the observed luminosity - mass relation ( e.g. , markevitch et al . 1998 ) and mass - temperature relation ( e.g. , sanderson et al . 2003 ) after the effect of gas cooling is compensated . using the gas density obtained with the best - fit two-@xmath12 model ( 5.1 ) and the analytical power - law temperature profile ( 4.4 ) , we calculate the azimuthally averaged gas entropy @xmath244 in the cluster and plot its spatial distribution in figure 9 . in @xmath245 ( @xmath246 kpc ) , the profile can be approximated by a power - law form with an index of 1.1 , as has been expected from shock heating that occurred during the spherical collapse ( tozzi & norman 2001 ) . at @xmath247 , the derived gas entropy is @xmath248 kev @xmath249 , which is consistent with the entropy - temperature relation for a sample of 66 relaxed clusters presented in ponman et al . we present the _ chandra _ acis study of the intermediately distant ( @xmath0 ) cluster of galaxies ms 0839.9 + 2938 . by performing both the projected and deprojected spectral analyses , we find that the emission - measure weighted gas temperature is approximately constant at about 4 kev in @xmath250 kpc . in the inner regions , the gas temperature descends towards the center , reaching @xmath2 kev within the innermost @xmath3 kpc . along with the temperature drop , we detect a significant inward iron abundance increase from about 0.4 solar in the outer regions to @xmath8 solar within the central @xmath3 kpc . besides ms 0839.9 + 2938 , another cluster at @xmath251 that has been found to show a similar , strong central iron excess is abell 2390 ( @xmath186 ; allen et al . 2001 ) . we argue that most of the excess iron is contributed by sne ia . by using the observed sn ia rate and stellar mass loss rate , we estimate that in ms 0839.9 + 2938 the time needed to enrich the central region with excess iron is @xmath10 gyr , which is similar to those found for the nearby clusters . in almost the same region where the temperature drop and abundance increase are seen , the observed x - ray surface brightness profile shows an excess beyond the distribution expected by either the @xmath12 model or the nfw model , and can be well fitted with an empirical two-@xmath12 model . the origins of all these phenomena can be correlated with each other ( cf , makishima et al . 2001 ) . the excess amount of gas , a part of which comes from the supernova explosions with heavy metals , is trapped around the central dominating galaxy to form the observed surface brightness excess . the relatively high gas density and high metallicity both speed up the gas cooling process , causing the observed inward gas temperature drop . this work was supported by the national science foundation of china ( grant no . 10273009 & 10233040 ) , the ministry of science and technology of china ( grant no . nkbrsf g19990754 ) , and shanghai key projects in basic research ( no . 04jc14079 ) . allen , s. w. , ettori , s. , fabian , a. c. 2001 , mnras , 324 , 877 allen , s. w. , schmidt , r. w. , fabian , a. c. 2002 , mnras , 335 , 256 anders , e. , grevesse , n. 1989 , gecoa , 53 , 197 becker , r. h. , white , r. l. , helfand , d. j. 1995 , apj , 450 , 559 b@xmath213hringer , h. , matsushita , k. , churazov , e. , finoguenov , a. , ikebe , y. 2004 , a@xmath179a , 416 , 21 brandt , w. n. , hornschemeier , a. e. , schneider , d. p. , garmire , g. p. , chartas , g. , hill , g. j. , macqueen , p. j. , townsley , l. k. , burrows , d. n. , koch , t. s. , and 2 coauthors 2000 , aj , 119 , 2349 buote , d. a. 2002 , apj , 574 , l135 buote , d. a. , lewis , a. d. , brighenti , f. , mathews , w. g. 2003 , apj , 595 , 151 cappellaro , e. , evans , r. , turatto , m. 1999 , a@xmath179a , 351 , 459 ciotti , l. , pellegrini , s. , renzini , a. , dercole , a. 1991 , apj , 376 , 380 dahlen , t. , strolger , l .- g . , riess , a. g. , mobasher , b. , chary , r .- r . , conselice , c. j. , ferguson , h. c. , fruchter , a. s. , giavalisco , m. , livio , m. , and 3 coauthors 2004 , apj , 613 , 189 de grandi , s. , molendi , s. 2001 apj , 551 , 153 de grandi , s. , ettori , s. , longhetti , m. , molendi , s. 2004 , a@xmath179a , 419 , 7 dickey , j. m. , lockman , f. j. 1990 , ara@xmath179a , 28 , 215 ezawa , h. , fukazawa , y. , makishima , k. , ohashi , t. , takahara , f. , xu , h. , yamasaki , n. y. 1997 , apj , 490 , l33 finoguenov , a. , david , l. p. , ponman , t. j. 2000 , apj , 544 , 188 finoguenov , a. , matsushita , k. , bhringer , h. , ikebe , y. , arnaud , m. 2002 , a@xmath179a , 381 , 21 finoguenov , a. , henriksen , m. j. , briel , u. g. , de plaa , j. , kaastra , j. s. 2004 , apj , 611 , 811 fukazawa , y. , ohashi , t. , fabian , a. c. , canizares , c. r. , ikebe , y. , makishima , k. , mushotzky , r. f. , yamashita , k. 1994 , pasj , 46 , l55 furusho , t. , yamasaki , n. y. , ohashi , t. 2003 , apj , 596 , 181 gibson , b. k. , loewenstein , m. , mushotzky , r. f 1997 , mnras , 290 , 623 gioia , i. m. , maccacaro , t. , schild , r. e. , wolter , a. , stocke , j. t. , morris , s. l. , henry , j. p. 1990 , apjs , 72 , 567 grevesse , n. , sauval , a. j. 1999 , a@xmath179a , 347 , 348 hutchings , j. b. , edwards , l. 2000 , aj , 119 , 1100 ikebe , y. , makishima , k. , fukazawa , y. , tamura , t. , xu , h. , ohashi , t. , matsushita , k. 1999 , apj , 525 , 58 iwasawa , k. , fabian , a. c. , allen , s. w. , ettori , s. 2001 , mnras , 328 , l5 jones , c. , forman , w. 1984 , apj , 276 , 38 koyama , k. , takano , s. , tawara , y. 1991 , nature , 350 , 135 makino , n. , sasaki , s. , suto , y. 1998 , apj , 497 , 555 makishima , k. , ezawa , h. , fukuzawa , y. , honda , h. , ikebe , y. , kamae , t. , kikuchi , k. , matsushita , k. , nakazawa , k. , ohashi , t. , and 3 coauthors 2001 , pasj , 53 , 401 markevitch , m. 1998 , apj , 504 , 27 matsushita , k. , finoguenov , a. , bhringer , h. 2003 , a@xmath179a , 401 , 443 mcwilliam , a. 1997 , ara@xmath179a , 35 , 503 mushotzky , r. f. , cowie , l. l. , barger , a. j. , arnaud , k. a. 2000 , nature , 404 , 459 navarro , j. f. , frenk , c. s. , white , s. d. m. 1997 , apj , 490 , 493 nesci , r. , gioia , i. m. , maccacaro , t. , morris , s. l. , perola , g. , c. , schild , r. , e. , wolter , a. 1989 , apj , 344 , 104 nesci , r. , perola , g. c. , wolter , a. 1995 , a@xmath179a , 299 , 34 nevalainen , j. , kaastra , j. , parmar , a. n. , markevitch , m. , oosterbroek , t. , colafrancesco , s. , mazzotta , p. 2001 , a@xmath179a , 369 , 459 nomoto , k. , iwamoto , k. , nakasato , n. , thielemann , f .- k . , brachwitz , f. , tsujimoto , t. , kubo , y. , kishimoto , n. 1997 , nupha , 621 , 467 ota , n. , pointecouteau , e. , hattori , m. , mitsuda , k. 2004 , apj , 601 , 120 ponman , t. j. , sanderson , a. j. r. , finoguenov , a. 2003 , mnras , 343 , 331 pratt , g. w. , arnaud , m. 2003,a@xmath179a , 408 , 1 salpeter , e. e. 1955 , apj , 121 , 161 sanders , j. s. , fabian , a. c. 2002 , mnras , 331 , 273 sanderson , a. j. r. , ponman , t. j. , finoguenov , a. , lloyd - davies , e. j. , markevitch , m. 2003 , mnras , 340 , 989 sarazin , c. l. 1986 , reviews of modern physics , 58 , 1 smith , r. k. , brickhouse , n. s. , liedahl , d. a. , raymond , j. c. 2001 , apj , 556 , 91 spergel , d. n. , verde , l. , peiris , h. v. , komatsu , e. , nolta , m. r. , bennett , c. l. , halpern , m. , hinshaw , g. , jarosik , n. , kogut , a. , and 7 coauthors 2003 , apjs , 148 , 175 stocke , j. t. , morris , s. l. , gioia , i. m. , maccacaro , t. , schild , r. , wolter , a. , fleming , t. a. , henry , j. p. 1991 , apjs , 76 , 813 takahashi , s. @xmath179 yamashita , k. 2003 , pasj , 55 , 1105 tamura , t. , kaastra , j. s. , den herder , j. w. a. , bleeker , j. a. m. , peterson , j. r. 2004 , a@xmath179a , 420 , 135 tozzi , p. , norman , c. 2001 , apj , 546 , 63 trager , s. c. , faber , s. m. , worthey , g. , gonz@xmath252lez , j. j. 2000 , aj , 119 , 1645 wegner , g. , thorstensen , j. r. , kurtz , m. j. , brown , w. r. , fabricant , d. g. , geller , m. j. , huchra , j. p. , marzke , r. o. , sakai , s. 2001 , aj , 122 , 2893 wu , x .- 2000 , mnras , 316 , 299 xu , h. , ezawa , h. , fukazawa , y. , kikuchi , k. , makishima , k. , ohashi , t. , tamura , t. 1997 , pasj , 49 , 9 xu , h. , makishima , k. , fukazawa , y. , ikebe , y. , kikuchi , k. , ohashi , t. , tamura , t. 1998 , apj , 500 , 738 xue , y .- j . , b@xmath213hringer , h. , matsushita , k. 2004 , a@xmath179a , 420 , 833 lccccccr [ tbl : totalspectrum ] apec & 4.04 ( fixed ) & @xmath253&@xmath254&@xmath255&@xmath255&@xmath255&107.6/91 + & @xmath256&@xmath257&@xmath258&@xmath255&@xmath255&@xmath255&102.8/90 + vapec&4.04 ( fixed ) & @xmath259&@xmath255 & @xmath260&@xmath261&@xmath258&96.4/89 + & @xmath262&@xmath263&@xmath255 & @xmath264&@xmath265&@xmath266&90.5/88 + 2@xmath255vapec&4.04 ( fixed)&&@xmath267&@xmath268&@xmath269&87.7/87 + & & & & & & + llccccccccc 1&0.0 - 37.3 & @xmath270&&44.0/43&&@xmath271&&34.9/43 + 2&37.3 - 74.1 & @xmath272&&40.8/44&&@xmath273&&34.2/44 + 3&74.1 - 130.1 & @xmath274&&36.2/43&&@xmath275&&32.3/43 + 4&130.1 - 230.0&@xmath276&&41.0/38&&@xmath277&&45.7/43 + 5&230.0 - 444.1&@xmath278&&35.9/42&&@xmath255 & & @xmath255 + + & & & & + & & @xmath279 & @xmath280&@xmath281&@xmath282 & & @xmath279 & @xmath280&@xmath281 & @xmath282 + & & ( kev)&(solar ) & ( solar ) & & & ( kev)&(solar ) & ( solar ) & + 1&0.0 - 37.3 & @xmath283&@xmath284&@xmath285&43.9/42&&@xmath286&@xmath287&@xmath288&34.8/42 + 2&37.3 - 74.1 & @xmath289&@xmath290&@xmath291&40.6/43&&@xmath292&@xmath293&@xmath294&34.1/43 + 3&74.1 - 130.1 & @xmath295&@xmath296&@xmath297&35.5/42&&@xmath298&@xmath299&@xmath300&30.8/42 + 4&130.1 - 230.0&@xmath301&@xmath302 & @xmath303&37.2/37&&@xmath304&@xmath305 & @xmath306&41.3/42 + 5&230.0 - 444.1&@xmath307&@xmath308&@xmath309&30.1/41&&@xmath255 & @xmath255 & @xmath255 & @xmath255 + llcccccr inner&0.0 - 74.1 & 4.04 ( fixed ) & @xmath310&@xmath311&@xmath312&@xmath313&101.2/96 + & & @xmath314&@xmath315&@xmath316&@xmath317&@xmath318&100.2/95 + outer&74.1 - 444.1 & 4.04 ( fixed ) & @xmath319&@xmath320&@xmath321&@xmath322&79.1/98 + & & @xmath323&@xmath324&@xmath325&@xmath326&@xmath327&79.1/97 + cccccccr + @xmath328 ( 10@xmath329 cm@xmath329)&@xmath331 ( @xmath48 kpc)&@xmath332&@xmath333 ( 10@xmath329 cm@xmath329)&&@xmath334 ( @xmath48 kpc)&@xmath335&@xmath282 + 21.7@xmath336 & 65.9@xmath337 & 0.660@xmath338&40.4@xmath339&&15.5@xmath340 & 0.586@xmath341 & 46.1/59 + fig.1 ( a ) : central @xmath29 ( @xmath30 kpc ) of ms 0839.9 + 2938 in 0.78 kev , which is plotted in the logarithmic scale . the image has been background - subtracted , exposure - corrected , and smoothed with a minimum significance of 3 and a maximum significance of 5 . x - ray intensity contours are spaced from @xmath348 to @xmath349 photons @xmath59 s@xmath5 pixel@xmath350 . the inner @xmath351 region , which is marked by a box , is magnified in ( b ) , where the contour levels are from @xmath352 to @xmath353 photons @xmath59 s@xmath5 pixel@xmath350 . ( c ) : corresponding dss optical image for the same sky field as in ( a ) on which the x - ray contours are overlaid for comparison . fig.2 azimuthally - averaged radial surface brightness profile of ms 0839.9 + 2938 in 0.78 kev . the background is assumed to be spatially uniform and is represented by the dotted line . in @xmath41 kpc , the surface brightness profile can be well described by the @xmath12 model ( dashed line ) , showing an obvious central emission excess . the fits of the two-@xmath12 , @xmath12 and nfw models to the whole cluster region , and the fit of the @xmath12 model to the outer region only are shown with the solid , dashed , dot - dashed and dot - dot - dot - dashed lines , respectively . fig.3 two - dimensional distribution of the hardness - ratio of the diffuse emission in ms 0839.9 + 2938 , which is defined as the ratio of the background - subtracted and exposure - corrected counts in 1.48.0 kev to those in 0.71.4 kev . the cross represents the peak of x - ray emission , and the circle ( @xmath354 kpc ) marks the region in which significant temperature drop and abundance increase are detected . the image is plotted in linear scale , and has been smoothed in the same way as in figure 1 . fig.4 ( a ) : projected ( solid ) and deprojected ( dashed ) radial distributions of the gas temperature , which are obtained with the absorbed apec model . the best - fit temperature profile for the deprojected distribution is shown as a dotted line . ( b ) : the same as ( a ) but for the absorbed vapec model . ( c ) : projected ( solid ) and deprojected ( dashed ) radial distributions of the average metal abundance , which are obtained with the absorbed apec model . ( d ) : the same as ( c ) but for the iron abundance obtained with the absorbed vapec model . fig.5 confidence contours at the 68.3% , 90% and 99% confidence levels for the gas temperature and iron abundance derived with the deprojected vapec model for the innermost ( @xmath355 kpc ) and the third ( @xmath356 kpc ) annular regions . fig.6 upper and middle panels : calculated radial distributions of the total gravitating mass , gas mass and gas mass fraction , along with the 90% error limits ( shown as dotted lines ) . the total mass was calculated by using both @xmath12 and the best - fit two-@xmath12 model . the gas mass and gas fraction were calculated by using the best - fit two-@xmath12 model . lower panel : the calculated cooling time as a function of radius . the horizontal , dashed line represents the age of the universe at the redshift of cluster . radii within which the gas temperature drops and iron abundance increases significantly ( @xmath357 ) , and at which the gas temperature begins to decrease inwards ( @xmath358 ) , are indicated by vertical lines . fig.7 upper and middle panels : spectrum extracted in @xmath159 kpc , where the gas temperature is significantly lower than that of the outer regions , along with the best - fit two - temperature apec model and residuals . lower panel : model ratios of the single mkcflow model and mekal+mkcflow model to the two - temperature apec model . @xmath160 of the mkcflow component is fixed to 0.08 kev . fig.8 ratios of the iron abundance in cluster s inner region to that in the outer region z@xmath359/z@xmath360 as a function of the cluster s ( a ) average gas temperature and ( b ) optical luminosity for nearby clusters ( solid ; tamura et al 2004 ; girardi et al . 2000 ; david et al . 1995 ; pratt & arnaud 2003 ) and two intermediately distant clusters , i.e. , ms 0839.9 + 2938 ( dashed ) and abell 2390 ( dot - dashed . this is the average abundance ratio rather than iron abundance ratio ; allen et al . the ratio for the nearby clusters is shown as a horizontal dotted line .
the parallel computational complexity of the evolution model is studied . it is shown that histories can be generated by a massively parallel computer in a time that is polylogarithmic in the length of the history . in this parallel dynamics , histories are built up via a nested hierarchy of avalanches . stated in another way , the main result is that the _ logical depth _ of producing a history is exponentially less than the length of the history . this finding is surprising because the self - organized critical state of the model has long range correlations in time and space that appear to imply that the dynamics is sequential and history dependent . the parallel dynamics for generating histories is contrasted to standard dynamics . standard dynamics and an alternate method for generating histories , conditional dynamics , are both shown to be related to p - complete natural decision problems implying that they can not be efficiently implemented in parallel .
cond-mat0105296
the model @xcite is among the simplest systems to display self - organized criticality and avalanche behavior . starting from arbitrary initial conditions , the model spontaneously reaches a state with long range correlations in space and time that are set up by a series of avalanche events . the approach to criticality , the critical state of the model , and the properties of avalanches in the model have been extensively studied during the past decade . there are now precise values of critical exponents and well - tested scaling laws . on the other hand , there is not yet a theory of the critical state except for the case of mean field theory and near zero dimension . in this paper we investigate the complexity of the histories generated by the model . these histories manifest long time correlations with early random events influencing distant , later events . however , the past may determine the future in ways which are more or less complex . it is the purpose of this paper to give a precise meaning to the notion of a complex history and then to show that histories produced by the model are not complex in this sense . the framework in which we will define the complexity of a history is computational complexity theory . computational complexity theory measures the computational resources needed to solve problems . among those resources are _ parallel time _ or equivalently _ depth _ , which measures the number of steps needed by a massively parallel computer to solve a problem . if the solution to a problem can be re - organized so that many computations are carried out simultaneously and independently then a great deal of computational work can be accomplished in few parallel steps . on the other hand , it may be that the sequential nature of the problem precludes reorganizing the work into a small number of parallel steps . the theory of parallel computational complexity provides algorithms for efficiently solving some problems in parallel and provides evidence supporting the proposition that solutions to other tasks can not be substantially accelerated by parallelization . a powerful feature of this theory is that it yields essentially the same results independent of the choice of the model of computation . thus , measures of parallel computational complexity reveal the logical structure of the problem at hand rather than the limitations of the particular model of parallel computation . we will define the complexity of histories generated by a stochastic model such as the model in terms of the parallel time required to generate a sample history . if , through parallelization , the dynamics of the model can be reorganized so that many processors running for a few steps can generate a statistically valid history then we will say the histories generated by the model are not complex . if on the other hand , there is little benefit from parallelization so that a history must be generated in a step by step fashion , then the model generates complex histories . the model illustrates the point that scale invariant spatial and temporal correlations are not sufficient to guarantee complexity in the sense defined here . indeed , it is the strictly hierarchical nature of the avalanches in the model that permits us to generate them in a small number of highly parallel steps . the present work fits into a series of studies of the computational complexity of models in statistical physics . although previous studies focused mainly on the computational complexity of generating spatial patterns , in many cases the analysis can also be applied to reveal the complexity of the history of the growth of the pattern . for example , two dynamical processes for diffusion limited aggregation ( dla ) have been shown to be p - complete @xcite . these results suggest that these processes for generating dla clusters can not be re - organized to greatly reduce the parallel time needed for the generation of an aggregate ( however , see [ ] for a parallel algorithm that achieves a modest speed - up for dla ) . the growth of an aggregate can be viewed as a history and the same arguments can be employed to give evidence that such histories can not be significantly parallelized . similarly , predicting the behavior of abelian sandpiles for @xmath0 has been shown to be p - complete @xcite . by contrast , the patterns generated by a number of growth models such as the eden model , invasion percolation and the restricted solid - on - solid model were shown to require little parallel time @xcite and these results carry over immediately to fast parallel algorithms for generating histories for these models . in sec . [ sec : comp ] we give an overview of selected aspects of computational complexity theory relevant to the present work and in sec . [ sec : bsmodel ] we review the model . in sec . [ sec : parbs ] we present parallel dynamics for the model and in sec . [ sec : simpar ] we present simulation results showing that histories can be efficiently simulated using parallel dynamics . in sec.[sec : bspc ] we show that the standard dynamics for the model and a new dynamics based on conditional probabilities both yield p - complete problems . the paper concludes in sec . [ sec : conc ] . computational complexity theory determines the scaling of computational resources needed to solve problems as a function of the size of the problem . introductions to the field can be found in refs . . computational complexity theory may be set up in several nearly equivalent ways depending on the choice of model of computation . here we focus on parallel computation and choose the standard _ parallel random access machine _ ( pram ) as the model of computation . the main resources of interest are _ parallel time _ or _ depth _ and number of processors . a pram consists of a number of simple processors ( random access machines or rams ) all connected to a global memory . although a ram is typically defined with much less computational power than a real microprocessor such as pentium , it would not change the scaling found here to think of a pram as being composed of many microprocessors all connected to the same random access memory . the processors run synchronously and each processor runs the same program . processors have an integer label so that different processors follow different computational paths . since all processor access the same memory , provision must be made for how to handle conflicts . here we choose the _ concurrent read , concurrent write _ ( crcw ) pram model in which many processors may attempt to write to or read from the same memory cell at the same time . if there are conflicts between what is to be written , the processor with the smallest label succeeds in writing . the pram is an idealized and fully scalable model of computation . the number of processors and memory is allowed to increase _ polynomially _ ( i.e. as an arbitrary power ) in the size of the problem to be solved . communication is idealized in that it is assumed that any processor can communicate with any memory cell in a single time step . obviously , this assumption runs up against speed of light or hardware density limitations . nonetheless , parallel time on a pram quantifies a fundamental aspect of computation . during each parallel step , processor must carry out independent actions . communication between processor can only occur from one step to the next via reading and writing to memory . any problem that can be solved by a pram with @xmath1 processors in parallel time @xmath2 could also be solved by a single processor machine in a time that is no greater than @xmath3 since the single processor could sequentially run through the tasks that were originally assigned to the @xmath1 processors . on the other hand , it is not obvious whether the work of a single processor can be re - organized so that it can be accomplished in a substantially smaller number of steps by many processor working independently during each step . two examples will help illustrate this point . first , suppose the task is to add @xmath4 finite - precision numbers . this can be done by a single processor in a time that scales linearly in @xmath4 . on a pram with @xmath5 processor this can be done in @xmath6 parallel time using a binary tree ( for simplicity , suppose @xmath4 is a power of @xmath7 ) . in the first step , processor one adds the first and second numbers and puts the result in memory , processor two adds the third and fourth numbers and puts the result in memory and so on . after the first step is concluded there are @xmath5 numbers to add and these are again summed in a pairwise fashion by @xmath8 processors . the summation is completed after @xmath9 steps . this method is rather general and applies to any associative binary operation . problems of this type have efficient parallel algorithms and can be solved in time that is a power of the logarithm of the problem size , here @xmath4 , that is , _ polylog _ time . a second example is the problem of evaluating the output of a boolean circuit with definite inputs . a boolean circuit is composed of and , or and not gates connected by wires . suppose the gates are arranged in levels so that gates in one level take their inputs only from gates of the previous level and give their outputs only to gates of the next level so there is no feedback in the circuit . the number of levels in the circuit is referred to as the _ depth _ of the circuit . at the top level of the circuit are the true or false inputs . at the bottom level of the circuit are one or more outputs . given some concisely encoded description of the circuit and its inputs , the problem of obtaining the outputs is known as the _ circuit value problem _ ( cvp ) . clearly , we can solve cvp on a pram in parallel time that is proportional to the depth of the circuit since we can evaluate a single level of the circuit in a single parallel time step . on the other , there is no known general way of speeding up the evaluation of a boolean circuit to a time that is , say polylog in the depth of the circuit . for the case of adding @xmath4 numbers , the logical structure of the problem is sufficiently simple that we can substitute hardware for time to obtain the answer in polylog parallel steps . for cvp , hardware can not be used to substantially decrease the depth of the problem . at the present time , the statement that cvp can not be solved in polylog parallel time is not mathematically proved . instead , one can show that cvp is _ p - complete_. to understand the meaning of p - completeness we must first introduce the complexity classes * p * and * nc * and the notion of _ reduction_. * p * consists of the class of problems that can be solved by a pram in polynomial time with polynomially many processors . * nc * consists of the class of problems that can be solved in polylog time on a pram with polynomially many processors . a problem @xmath10 is reduced to a problem @xmath11 if a pram with an oracle for @xmath11 can be used to solve @xmath12 in polylog time with polynomially many processors . an oracle for @xmath11 is able to supply the pram a solution to an instance of @xmath11 in a single time step . intuitively , if @xmath10 can be reduced to @xmath11 then @xmath10 is no harder to solve in parallel than @xmath11 . a problem in * p * is p - complete if all other problems in * p * can be reduced to it . it can be proved that cvp is p - complete . furthermore , if it could be shown that there is any problem in * p * that is not in * nc * then it would follow that cvp is not in * nc * . however , the conjecture that @xmath13 , though widely believed , is not yet proved . since reductions are transitive , showing that another problem @xmath10 is p - complete can be proved by reducing cvp to @xmath10 . later in the paper we will take this approach to prove that two problems representing different dynamics are p - complete . in addition to supplying a p - complete problem , boolean circuits serve as an alternate model of computation . for a given computational problem , and for a given problem , there is a boolean circuit that solves the problem . in order to solve problems of any size , a family of boolean circuits is needed , one for each problem size . furthermore , to be comparable to a pram running a single program , the family of circuits should satisfy a _ uniformity _ condition that bounds the complexity of the computation needed to specify each circuit in the family . the resources of circuit depth ( number of levels in the circuit ) and circuit width ( maximum number of gates in a level ) are roughly comparable to the pram resources of parallel time and number of processor , respectively . indeed the class * nc * can be defined as the set of problems that can be solved by uniform circuit families whose depth scales polylogarithmically in the problem size and whose width scales polynomially in the problem size . * p * is the class of problems solvable by circuits of polynomial width and depth . in this section we review the model and its behavior . a detailed discussion can be found in ref . . the model defines histories of configurations on a @xmath14-dimensional lattice . at each lattice site @xmath15 and discrete time @xmath16 , there is a number @xmath17 $ ] . the _ standard _ dynamics for the model creates a history sequentially with one new time slice determined in each step of the dynamics . a single time step in standard dynamics for the model consists of locating the _ extremal _ site @xmath18 , which has the minimum number . the extremal site and its @xmath19 nearest neighbors are given new @xmath20 s chosen randomly from the uniform distribution on @xmath21 $ ] . the @xmath20 s on the remaining sites are unchanged . let @xmath22 be the extremal value at the extremal site @xmath23 at time @xmath16 . histories can be organized into _ avalanches _ or causally connected regions of activity in the @xmath24-dimensional space - time . avalanches can be classified by bounding the extremal value during the avalanche . the collection of renewed sites from @xmath16 to @xmath25 is an @xmath26-avalanche if @xmath27 and @xmath28 but for all @xmath29 with @xmath30 , @xmath31 . note that it is also possible to have an @xmath26-avalanche whose duration is a single time step ( i. e. @xmath32 and no times @xmath29 ) if @xmath27 and @xmath33 . an @xmath26-avalanche can be partitioned into a sequence of one or more @xmath34-avalanches for @xmath35 . repeated use of this idea allows us to partition avalanches into a nested sequence of subavalanches . the partitioning of avalanches into subavalanches is one of the key ideas in the parallel algorithm described below . the properties of avalanches are entirely independent of the environment in which they exist so long as the environment is large enough to contain the avalanche . indeed we can grow the _ backbone _ of an avalanche without regard to the environment . the backbone of an @xmath26-avalanche is the set of sites @xmath15 and the corresponding @xmath36 for which @xmath37 . ( note that the last time step in an avalanche is not part of the backbone ) . the backbone of an @xmath26-avalanche for @xmath38 is shown in fig . [ fig : f13 ] . the only information we need about the environment is that all environmental sites have random numbers uniformly distributed in the interval @xmath39 $ ] . the values of these numbers is unimportant and the full lattice of random numbers does not have to be realized . during the construction of an @xmath26-avalanche , sites must be renewed . instead of assigning random numbers in the interval @xmath21 $ ] as is done in standard dynamics , one can declare a renewed site to be part of the environment with probability @xmath40 or , with probability @xmath26 , declare it part of the backbone . if the site is part of the backbone it is then given an explicit random number in the interval @xmath41 . the construction of an @xmath26-avalanche backbone begins with the renewal of the origin and its neighbors and then continues until there are no remaining sites with number less than @xmath26 . this approach for making an avalanche backbone is called the branching process in [ ] and is an efficient method for carrying out large - scale simulations of the model . -avalanche for @xmath38 in the one - dimensional model . black dots represent positions where the random numbers are less than @xmath26 . immediately before and at the last time of the avalanche all random numbers are greater than @xmath26 . ] the model manifests self - organized criticality : independent of the initial state , after a long time the system approaches a stationary state where the average avalanche size diverges and the distribution of avalanche sizes follows a power law . in the critical state almost all @xmath20 s exceed a critical value , @xmath42 ( for the @xmath43 , @xmath44 ) . if @xmath45 is the duration of an avalanche then , in the critical state , the probability of an avalanche of duration @xmath45 decays as @xmath46 ( @xmath47 for @xmath43 ) . the average duration @xmath48 of @xmath26-avalanches diverges as @xmath49 with @xmath50 for @xmath43 . the @xmath26-avalanche size distribution , @xmath51 has the scaling form @xmath52 with @xmath53 , @xmath54 and @xmath55 related by the usual exponent relation @xmath56 . the scaling function @xmath57 goes to a constant as @xmath58 and decays rapidly to zero for @xmath59 . thus the upper cut - off @xmath60 in the @xmath26-avalanche size distribution obeys the scaling law @xmath61 ( @xmath62 for @xmath43 @xcite ) . the spatial extent of @xmath26-avalanches also diverges in the critical state . let @xmath63 be the number of sites covered by an @xmath26-avalanche , then @xmath64 . the approach to the critical state is characterized by a _ @xmath65 , which is a stepwise increasing function of time and is the maximum value of @xmath66 for @xmath67 . times when the gap increases mark the end of an avalanche . if the gap first has the value @xmath68 at time @xmath16 then all @xmath20 s in the system are greater than or equal to @xmath68 at time @xmath16 and a @xmath68-avalanche begins at time @xmath69 . on average , the gap approaches the critical value as a power law , @xmath70 where @xmath71 is the number of sites in the system . in the parallel dynamics for the model we will first identify the successive values of the gap and then fill in the intervals between them with the appropriate avalanches . in this section we present a parallel approach for constructing histories . at the core of this approach is a procedure for hierarchically building avalanches in which small avalanches are pasted together to yield larger avalanches . the large scale structure of the history is controlled by the sequence of gap values . each value of the gap determines an avalanche that needs to be constructed . the algorithm proceeds in three stages . in the first stage , described in sec . [ sec : gap ] , the gap sequence is obtained and in the second stage , described in sec . [ sec : parf ] , the required avalanches are constructed . finally , in the third stage , all space - time points that are not yet determined are explicitly given numbers . although we have not run the parallel algorithm on a parallel computer , we have measured essential aspects of its behavior using a conventional simulation and the results are reported in sec . [ sec : simpar ] . the key finding is that we can produce histories in a parallel time that is polylogarithmic in the length of the history . in this subsection we show how to construct a sequence of gap values , @xmath72 , @xmath73 for a system with @xmath71 lattice sites ( independent of dimension ) . suppose that at some time the gap increases from @xmath72 to @xmath74 . this event occurs at the last step of a @xmath72-avalanche so , before the selection of the extremal value that increases the gap , the distribution on every site of the lattice is uniform on the interval @xmath75 $ ] . thus @xmath74 is a random variable given by @xmath76 where @xmath77 is a random variable that is the minimum over @xmath71 random numbers uniformly distributed on @xmath21 $ ] and @xmath78 is the xor operation for real numbers . since @xmath78 is associative we can unroll eq.([eq : gap ] ) to obtain , @xmath79 where @xmath80 is the initial value of the gap and the @xmath81 are identically distributed independent random variables chosen from the same distribution as @xmath77 described above . using a standard parallel minimum program , each @xmath81 can be generated on a probabilistic pram in @xmath82 parallel time . a standard parallel prefix algorithm @xcite can then be used to evaluate all @xmath83 values of the gap in @xmath84 parallel time . in this section we show how to generate an @xmath26-avalanche in parallel time that is polylog in @xmath85 . the idea of this algorithm is to generate in parallel a large number of independent small avalanches and then to paste groups of these together to create a smaller number of larger avalanches . these are in turn pasted together to create yet larger avalanches and so on in a hierarchical fashion . the algorithm is controlled by a pre - determined list of @xmath86 values , @xmath87 . these values specify the types of avalanches to be manufactured . the sequence of @xmath86 converges to @xmath42 as @xmath88 with @xmath89 a constant between @xmath90 to @xmath91 . this choice for @xmath92 implies that for large @xmath4 , @xmath93 so that the average number of @xmath94-avalanches needed to construct an @xmath95-avalanche is asymptotically constant ( and equal to @xmath96 ) . the length of the sequence and its penultimate element @xmath97 is chosen so that @xmath98 . since @xmath99 , @xmath100-avalanches always have @xmath101 and consist of the origin and its @xmath19 nearest neighbors together with their associated @xmath20 values . in parallel we can make @xmath102 , @xmath100-avalanches in constant time using @xmath102 processors . we now construct @xmath103-avalanches by pasting together one or more @xmath100-avalanches . the origin of the first @xmath100-avalanche is also the origin of the @xmath103-avalanche . if the minimum of the @xmath104 @xmath20 s of the @xmath100-avalanche is greater than @xmath103 then we are done and the @xmath100-avalanche is also an @xmath103-avalanche . if not , a second @xmath100-avalanche is attached to the first @xmath100-avalanche with its origin lined up with the extremal site of the first avalanche . again , we check whether the minimum @xmath20 among all the sites covered by the pair of @xmath100-avalanches is greater than @xmath103 and , if so , we are done . the pattern for constructing @xmath105-avalanches from an @xmath94-avalanche should now be clear : @xmath94-avalanches are pasted together until all the covered sites have @xmath106 . the process of pasting together @xmath94-avalanches is a generalized version of the standard process . indeed , pasting together @xmath100-avalanches is exactly the standard process . all that we need to know about each @xmath94-avalanche is its origin and the final values of @xmath20 on the sites the avalanche covers . during the construction of an @xmath105-avalanche , when a new @xmath94-avalanche is attached , its origin is lined up with the extremal site among all of the sites currently covered by the growing @xmath105-avalanche . the @xmath20 values on the covered sites of the newly added @xmath94-avalanche renew the sites covered by this avalanche and , if some of these sites were previously part of the environment , the total number of covered sites of the growing @xmath95-avalanche is increased . if the covered sites all have @xmath107 , the @xmath95-avalanche is finished . the pasting together of @xmath94-avalanches must be carried out sequentially so that if the number of @xmath94-avalanches needed to construct a single @xmath105-avalanches is @xmath108 then it will take @xmath108 parallel steps to grow the avalanche . finally , once an @xmath26-avalanche is constructed , it is straightforward to find its backbone by eliminating those sites for which @xmath109 . on average @xmath96 @xmath94-avalanches must be pasted together to form one @xmath105-avalanche , that is @xmath110 . however , the @xmath111 step in the parallel construction of an @xmath26-avalanche is not complete until _ all _ the required @xmath105-avalanches have been made . thus , the time required for this step will be set by the maximum value of @xmath108 . as @xmath26 approaches @xmath42 , and for fixed @xmath4 we need to make increasingly many @xmath105-avalanches so that the tail of the @xmath108 distribution is explored . as we shall see in the next section , the expected value of the maximum of @xmath108 increases logarithmically in the number of @xmath112-avalanches that are constructed . as a result , the average parallel time required to generate an @xmath26-avalanche is predicted to be proportional to @xmath113 . the discussion thus far has posited an effectively infinite environment for the avalanche to grow in , however this is not required . the same hierarchical construction can just as well be used to make avalanches in periodic boundary conditions in a system of size @xmath114 . additional complications arise for other boundary conditions due to the breaking of translational symmetry and this situation will not be discussed here . finally , we are ready to describe how to construct a full history in a system with @xmath71 sites and periodic boundary conditions starting from the initial condition , @xmath115 . first , following the procedure of sec . [ sec : gap ] we construct a sequence of @xmath83 gap values , @xmath116 . then , using the parallel algorithm described above , for each gap value , @xmath72 , we construct the backbone of a @xmath72-avalanche for all @xmath117 . the site @xmath15 where @xmath118 is chosen randomly from among all the sites of the lattice and this site serves as the spatial origin for the @xmath72-avalanche . the starting time @xmath119 of the @xmath72-avalanche is obtained by summing over the durations of the preceding avalanches , @xmath120 where @xmath121 is the duration of the @xmath122-avalanche . successive avalanche backbones are concatenated to form the backbone of the history which includes all the avalanche backbones and the intervening extremal sites that have the gap values . thus far only the backbone of the history is specified while sites in the environment are conditioned by the extremal values but not fully specified . a definite , statistically valid , history is constructed from the backbone by fixing all space - time points @xmath123 not in the backbone . at the last time , @xmath124 , the gap has the values @xmath125 . the randomly chosen extremal site , @xmath126 is part of the backbone and takes the value @xmath125 . all other sites at time @xmath124 are chosen from the uniform distribution on @xmath127 $ ] . from these values we now work backwards in time independently for each site . consider site @xmath15 , we have that @xmath128 until the most recent time that @xmath15 was part of the backbone . during the time intervals that site @xmath15 is part of the backbone , its value is determined . for each connected time interval , @xmath129 $ ] that site @xmath15 is part of the environment , @xmath36 takes a single value . there are two possibilities . the interval may come to end because @xmath15 is an extremal site where the gap increases from @xmath130 to @xmath72 that is , @xmath131 and @xmath132 . in this case , @xmath118 throughout the interval . the second possibility is that site @xmath15 is renewed at time @xmath133 and becomes part of the @xmath72-avalanche backbone at that time . in this case , @xmath134 is chosen uniformly from the distribution @xmath75 $ ] and has this value during the interval from @xmath135 to @xmath136 . working backwards in this way a definite history is re - constructed from the backbone . it should be noted that the construction of a definite history from the backbone can be carried out in parallel in polylog time since each site can be treated independently and since intervals during which a site is not in the backbone can be identified using a parallel graph connectivity algorithm . we have studied the performance of the parallel algorithm for constructing avalanches in the one - dimensional model . we collected statistics on @xmath108 , the number of @xmath94-avalanches needed in the construction of @xmath105-avalanches during the manufacture of a single @xmath97-avalanche . let @xmath137 be the maximum value of @xmath108 required during the construction of the @xmath97-avalanche . the parallel running time @xmath138 , up to a constant of proportionality , for constructing an @xmath97-avalanche is obtained by summing @xmath139 , @xmath140 the simulation works by running standard dynamics and , on the fly , keeping track of @xmath108 and @xmath141 . the following pseudo - code sketches a single step of an algorithm that collects statistics on the distribution of @xmath108 and computes the running time for constructing @xmath97-avalanches . it is based on the observation that an @xmath95-avalanche is concluded when the extremal value @xmath142 . arrays @xmath108 and @xmath141 are initialized to zero . 1 . execute one time step of standard dynamics and find the extremal value , @xmath143 . 2 . * for * @xmath144 to @xmath145 * do * + * if * @xmath146 * then * * @xmath147 * * if * @xmath148 * then * @xmath149 * collect statistics for @xmath108 . * @xmath150 + * end do * 3 . * if * @xmath151 * then * the @xmath97-avalanche ends . calculate the parallel time , @xmath2 by summing @xmath141 . + * else * go to step 1 . the re - organization of the sequential history of an avalanche into a hierarchy of subavalanches is illustrated in figs . [ fig : min ] and [ fig : fseven ] for the case of an @xmath152-avalanche . in fig.[fig : min ] the sequence of extremal values is shown together with horizontal lines indicating bounds @xmath153 . figure [ fig : fseven ] shows a tree representing the hierarchy of subavalanches making up the @xmath152 avalanche . the @xmath95 level of the tree represents the @xmath95-avalanches needed in the construction of the final @xmath152-avalanche . the number next to each node represents the ending time of a subavalanche . for example , the @xmath154 level of the tree shows that the @xmath152-avalanche is composed of three @xmath154-avalanches ending at times 3 , 8 and 9 . the total parallel time is the sum of the maximum degree of the tree at each level and is 13 in this case . vs. time @xmath16 during an @xmath152-avalanche . horizontal lines are the values @xmath153 . ] -avalanche of fig.[fig : min ] into subavalanches . the @xmath95-level of the tree represents partitioning the @xmath152-avalanche into @xmath95-avalanches and the numbers on the nodes represent the ending times of the @xmath95-avalanches . roughly speaking , sequential time runs to the right and parallel time runs upward . ] in our simulations , for each @xmath155 up to 40 we constructed @xmath156 @xmath97-avalanches . the sequence of avalanche sizes is given by eq . [ eq : fn ] with @xmath157 and @xmath158 . figure [ fig : time ] shows the @xmath159 vs. @xmath160 . a fit of the data to the functional form @xmath161 over the range @xmath162 yields @xmath163 . the fit is reasonable with @xmath164 . we also collected data for @xmath165 and @xmath166 over a smaller range of @xmath155 . for @xmath165 we found that @xmath167 ( @xmath168 ) for @xmath169 while for @xmath166 we found that @xmath170 ( @xmath168 ) for @xmath171 . since @xmath172 , a conservative conclusion from the data is that the average parallel time @xmath173 for constructing an @xmath26-avalanche behaves as @xmath174 or , in terms of the duration of the avalanche@xmath175 . for constructing an @xmath97-avalanche vs. @xmath176 . the straight line is a linear fit to the data for @xmath177 . ] the time requirements of the parallel algorithm can be understood by considering the probability density , @xmath178 of @xmath108 . recall that @xmath108 is the number of @xmath94-avalanches needed in the construction of an @xmath95-avalanche and that its average is @xmath179 . suppose that the tail of @xmath178 behaves as @xmath180 . given these assumptions we can estimate the maximum value , @xmath181 , of @xmath108 in the construction of an @xmath97-avalanche . the estimate is obtained by requiring that the expected number of @xmath95-avalanches manufactured times @xmath182 is order unity . the expected number of @xmath95-avalanches made during the construction of an @xmath97-avalanche is @xmath183 so we have @xmath184 summing over @xmath4 yields @xmath185 in qualitative agreement with the numerical results . we measured @xmath178 during our simulations . the logarithm of the histogram of @xmath108 for several values of @xmath4 is shown in fig.[fig : distr ] . it is clear that the behavior of @xmath178 is more complicated than simple exponential decay . there is a short time fast decay and , for larger @xmath4 , a long time slow decay . it is possible that the asymptotic behavior of @xmath178 for the two largest values of @xmath4 in fig.[fig : distr ] both display exponential decay with the same value of @xmath186 . it is also possible that @xmath186 is a slowly decreasing function of @xmath4 or that the asymptotic decay is slower than exponential . if @xmath186 decreases as a power of @xmath4 or if the tail of @xmath187 is a stretched exponential it will still be the case that @xmath159 is polynomial in @xmath155 and thus polylogarithmic in @xmath85 . on the other hand , if the tail of @xmath187 is a power law then @xmath159 would be exponential in @xmath155 . the data @xmath187 does not suggest a power law tail and the data for @xmath159 appears to rule out exponential growth however more numerical work is needed to fully understand the asymptotic behavior of these quantities . for several @xmath95-avalanches . ] we have seen in the previous section that the construction of histories can be arranged to be carried out in parallel in a time that is polylog in the length of the history . there are however alternative dynamics for producing histories with the correct statistical properties . in this section we show that two of these dynamics have the property of p - completenesss . the property of p - completeness implies that it is not possible to run these dynamics in parallel in polylog time . the two dynamics considered here are the standard dynamics by which the model is usually defined and a conditional dynamics where explicit values are not assigned to sites until it is necessary . thus , if we knew only about standard or conditional dynamics we would conclude that the model generates complex histories . parallel dynamics shares features of both standard dynamics and conditional dynamics . although standard dynamics can not be implemented in parallel in polylog time , in sec . [ sec : parstan ] we show how to use the parallel construction of avalanches to achieve power law speed - up . standard dynamics formalizes the usual approach to generating histories . initially , the lattice is described by values @xmath188 . a list of numbers @xmath189 , n=1,2 , \ldots , ( 2d+1 ) t_{\rm max}\}$ ] controls the dynamics . at each time step , the extremal site is located . the extremal site and its @xmath19 neighbors are renewed using the next group of @xmath104 @xmath190 values . periodic boundary conditions are assumed . figure [ fig : standyn ] shows an example of @xmath43 standard dynamics . the numbers along the top row of the figure are @xmath188 and the numbers in the three columns on the right are the groups of @xmath190 s . successive states of the lattice are given in successive rows in the lower left section of the figure . each row is obtained by aligning the three @xmath190 values of the row under the minimum @xmath191 value of the previous row . are to the right of the vertical line . during each time step , the extremal site and its two neighbors are renewed by the triple of numbers to the right of the vertical line . ] p - completeness for a natural decision problem associated with standard dynamics is proved by a reduction from the circuit value problem ( cvp ) . the inputs of the standard bak sneppen decision problem are the initial site values , @xmath188 , the renewal numbers , @xmath192 , a specified site @xmath193 , time @xmath124 and bound @xmath194 . the problem is to determine whether @xmath195 . in order to effect the reduction from cvp we need to show how to build logical gates and wires that carry truth values . we will show how to build and gates , or gates and wires in the one - dimensional model . this set of gates allows us to build monotone circuits . since monotone cvp is p - complete only for non - planar circuits , our construction constitutes a p - completeness proof only after it is extended to @xmath196 models . after we present the basic construction for @xmath43 we will discuss how to extend the construction to @xmath196 and the prospects for proving p - completeness for @xmath43 . in order to embed a circuit in standard dynamics , most of the sites of the lattice will have the value 1 . a truth value is represented by a pair of neighboring sites @xmath15 and @xmath197 such that within the pair , one site has the value one and the other has a value less than one . if the left member of the pair is a one the pair represents true , otherwise it represents false . a truth value is activated when its smaller member is the extremal site of the lattice . once the truth value is activated , it can be moved by appropriate choice of the triple of @xmath190 s . the triple @xmath198 where @xmath199 is a suitable chosen number , @xmath200 will move a truth value to the right while the triple @xmath201 will move the truth value to the left . and gates are realized by transporting two truth values next to one another . suppose that initially , two truth values @xmath202 and @xmath203 are next to each other as shown in the fig . [ fig : stanand ] . the numbers @xmath199 , @xmath186 , @xmath204 , @xmath205 and @xmath206 are ordered @xmath207 and , at the time the gate is activated , @xmath206 is less than all the other numbers in the lattice . the top panel in fig . [ fig : stanand ] shows the situation where @xmath208 true and @xmath209 true . three time steps later , @xmath210 represented by the middle two sites of the lattice is true . the top panel in fig . [ fig : stanand ] shows the situation where @xmath208 true and @xmath209 false in which case @xmath211 false as desired . the case @xmath208 false and @xmath209 false is the same as @xmath212 true except the states in the part of the lattice representing the gate are shifted one lattice constant to the left so that @xmath211 false as desired . the final case @xmath208 false and @xmath209 true is left as an exercise for the reader . and @xmath203 are represented by the two pairs of sites above the horizontal line . the output @xmath210 is represented by the middle two sites along the bottom row to the left of the vertical line . the various numbers are ordered @xmath213 . the top pictures shows the situation with both inputs are true yielding an output of true . the bottom picture shows the situation where the left input is true and the right input is false yielding false . ] or gates are slightly more complicated than and gates . initially , the truth values are brought together until they are separated by one site as shown in fig . [ fig : stanor ] the numbers are ordered @xmath214 and , at the time the gate is activated , @xmath215 is less than all the other numbers in the lattice . the top panel in fig . [ fig : stanor ] shows the situation where @xmath208 true and @xmath209 true . six time steps later , @xmath210 , represented by the second and third columns to the left of the vertical line , is true . the situation @xmath208false and @xmath209false is seen to be correct by translational invariance . the case where @xmath208 false and @xmath209 true is shown in the bottom panel of fig . [ fig : stanor ] . verifying the case @xmath208 true and @xmath209 false is left as an exercise for the reader . and @xmath203 are represented by the two outer pairs of sites above the horizontal line . the output @xmath210 is represented by the second and third sites to the left of the vertical line . the various numbers are ordered @xmath214 . the top pictures shows the situation with both inputs true . the bottom picture shows the situation where the left input is false and the right input is true , also yielding true . ] we also need a gadget for fan - out . figure [ fig : stanfan ] shows how two copies , @xmath210 and @xmath216 can be obtained from @xmath202 . the figure shows the case where @xmath208 true . the case where @xmath208 false is seen to be correct by translational invariance . is represented by the middle pair of sites above the horizontal line . the outputs @xmath210 and @xmath217 are represented by the two pairs of sites to left of the vertical line . the numbers are ordered @xmath218 . ] overall timing of the circuits is controlled by the numbers representing truth values and used in gates . since , at each stage , the smallest number on the lattice is selected , truth values represented by small numbers are activated first . a truth value @xmath202 that is active , meaning it is represented by a number @xmath199 that is the minimum number on the lattice , can be put into storage for a pre - determined period by choosing @xmath190 values @xmath219 with @xmath186 larger than all the values associated with logic variables that are active until @xmath202 is active again . in this way all the required @xmath191 and @xmath190 values can be calculated with a fast parallel computation , in advance knowing only the activation time of each logic gate . and and or gates , fan - out , timing and wires ( transporting truth values ) are sufficient for constructing arbitrary monotone boolean circuits . since planar monotone circuits are not p - complete , the preceding construction does not show that the @xmath43 standard bak sneppen problem is p - complete . it is straightforward to construct all the gadgets and arrange timings within @xmath196 standard dynamics by simply padding the set of @xmath190 values with 1 s in the directions transverse to the planes in which the action of the gadgets takes place . since truth values can be transported in any direction , non - planar monotone circuits can be embedded in @xmath196 systems . the conclusion is that the @xmath196 standard bak sneppen problem is p - complete . thus there can be no exponential speed - up of standard dynamics through parallelism ( unless it should turn out that @xmath220 , a highly unlikely prospect ) . although the current proof does not yield a p - completeness result for the planar standard bak sneppen problem , we do not see a general reason preventing the construction of non - monotone circuits and we can not rule out the possibility that the planar problem is also p - complete . at least for @xmath196 , the p - completeness result precludes the possibility of solving the standard bak sneppen problem in polylog time . nonetheless we can achieve a sublinear parallel time solution to the problem by invoking the hierarchical construction of avalanches from parallel dynamics . the history produced by standard dynamics can be viewed as a sequence of avalanches and these avalanches can be efficiently constructed from the list of @xmath190 values using the parallel methods of sec . [ sec : parf ] . the road block in achieving exponential speed - up for standard dynamics does not lie in the construction of avalanches , rather it is due to the presence of explicit initial values instead of initial probabilities defined by @xmath221 and made explicit only at the end of construction . the presence of explicit initial values means that the environment of each avalanche is specified and the sequence of gap values can not be determined in advance of making the avalanches . the accelerated algorithm for the standard bak sneppen problem proceeds as follows . the first gap value , @xmath222 is just the minimum among the initial values . the algorithm then calls a parallel avalanche subroutine to produce a @xmath221-avalanche from the initial sequence of @xmath190 values . this avalanche ( the full avalanche , not just the backbone ) renews the sites that it covers while the rest of the environment is unaffected . at the ending time , @xmath223 of the @xmath221-avalanche , the extremal site is determined and its values determines the next gap value , @xmath224 . generally , suppose that the gap increases at times @xmath225 , then @xmath226 and the parallel avalanche subroutine is called to produce a @xmath227-avalanche using the remaining list of @xmath190 values . the origin of this avalanche is the extremal site @xmath228 and the avalanche renews all the sites it covers . as the gap approaches the critical value , @xmath42 , the algorithm becomes increasingly efficient since larger and larger avalanches may be made in parallel . the parallel avalanche subroutine takes as inputs a gap value @xmath72 and the remaining @xmath190 values , @xmath229 values . its output is the @xmath72-avalanche that would have resulted if standard dynamics were used with the same sequence of @xmath190 values . the procedure is similar to that described in sec.[sec : parf ] and uses the same hierarchy of @xmath95-avalanches but care must be taken that the @xmath190 values are used in the correct order . each block of @xmath104 @xmath190 values is an @xmath100-avalanche . in parallel we must now group this sequence of @xmath100-avalanches into @xmath103-avalanches . the procedure for grouping avalanches is illustrated using fig.[fig : fseven ] . each node on the bottom level of the tree represents blocks of @xmath104 @xmath190 values . in parallel we now independently start from each of these nodes and carry out standard dynamics until an @xmath103-avalanche is complete . for example , starting from bottom level node 2 it takes two time steps to build an @xmath103-avalanche denoted by the 3 on the @xmath103 level of the tree but it only takes one time step starting from bottom level node 3 since this @xmath100-avalanche is also an @xmath103-avalanche . the time that standard dynamics must be run to obtain each of the @xmath103-avalanches defining the nodes on the @xmath103-level is just the maximum in - degree of this level . thus far we have made @xmath103-avalanches starting from each bottom level node , now we must determine which of these avalanches to keep to reproduce the result of standard dynamics acting on the @xmath190 sequence . the correct @xmath103-avalanche to associate with each node in the @xmath103-level is the one that includes all of the children of the node . this choice insures that all @xmath190 values are used . for example , the @xmath103-avalanche labeled 3 includes steps 2 and 3 rather than 3 alone . having assembled the correct set of @xmath103-avalanches we can now group these into @xmath230-avalanches and so on . the new ingredient in going from the @xmath94 to @xmath95-avalanches when @xmath231 is that the generalized version of the standard dynamics is used as discussed in sec . [ sec : parf ] . as before , @xmath72-avalanches can be constructed in a time that is polylog in the @xmath45 , the duration of the avalanche . the speed - up obtained due to the parallel avalanche subroutine can be estimated by the scaling laws for the @xmath232 and @xmath233 . up to logarithmic factors , it takes constant parallel time to produce a @xmath68-avalanche . combining eqs.[eq : gamma ] and [ eq : gapt ] we have @xmath234 where @xmath2 is parallel time and @xmath16 is sequential time . integrating the differential equation yields , @xmath235 so that power law speed - up is obtained . the important difference between parallel and standard dynamics is that in parallel dynamics , sites that are not part of avalanches are specified probabilistically until the last stage of the computation while , for standard dynamics , all sites are explicitly determined on each time step . this suggests the idea that an efficiently parallelizeable dynamics might result from minimizing the number of explicitly determined values until the last stage of the computation . this idea lead to the conditional dynamics described and analyzed in the next two sections . in standard dynamics with the usual choice of initial conditions , lattice sites are assigned random numbers uniformly distributed on @xmath21 $ ] and the lattice site with the smallest number is selected for replacement . in parallel dynamics only the backbone is fully specified until the end of the computation . in _ conditional dynamics _ this idea is carried as far as possible . extremal space - time points are fully specified and conditions are placed on all other space - time points until the end of the calculation . conditional dynamics is similar to the run time statistics approach in ref .. initially , each site is equally likely to be extremal . a site @xmath236 is selected at random from among all the sites of the lattice . suppose there are @xmath237 sites on the lattice . the first extremal number , @xmath238 is chosen from the distribution of the minimum of @xmath71 numbers distributed uniformly on @xmath21 $ ] . although we have not yet assigned numbers to any other sites we now know that these sites are conditioned to be greater than @xmath238 . the next step in the standard process is to renew @xmath236 and its neighbors with numbers chosen on the interval @xmath21 $ ] . again , at this stage , we do not need to explicitly know any of the numbers on the lattice except to say that all the numbers on the lattice except @xmath236 and its neighbors are randomly chosen in the interval @xmath239 $ ] while @xmath236 and its neighbors are randomly chosen on the interval @xmath21 $ ] . we can characterize the condition on site @xmath15 at time @xmath16 by the number @xmath240 which indicates that the random number at site @xmath15 is chosen from the uniform distribution on the interval @xmath241 $ ] . we have that @xmath242 for @xmath236 and all its neighbors while @xmath243 for all other sites . the extremal site @xmath244 at @xmath245 is a weighted random choice among the @xmath71 sites with site @xmath15 weighted by @xmath246 . the extremal value @xmath247 is chosen from the distribution of the minimum of @xmath71 numbers , where the @xmath248 number is chosen from the uniform distributions on @xmath249 $ ] . the minimum numbers @xmath238 and @xmath247 now determine the conditions @xmath250 and so on . specifically , @xmath23 and its neighbors are renewed at time @xmath69 so that @xmath251 for @xmath23 and its neighbors . for all other sites , @xmath252 . one way to implement the selection of the extremal site at time @xmath16 given the conditions @xmath240 is to assign random numbers @xmath253 on the interval @xmath21 $ ] to each space - time point . from these numbers , compute @xmath254 . note that @xmath255 is uniformly distributed on @xmath241 $ ] . the extremal site @xmath23 is the site with the minimum among the @xmath256 s and @xmath257 . note that all the random numbers @xmath253 can be assigned in advance . it is this realization of conditional dynamics that is shown to be p - complete in sec.[sec : pc ] . after the conditional dynamics has run for a time @xmath258 we are left with the conditions @xmath259 together with a list of extremal sites @xmath260 and extremal values @xmath261 . from these data we can reconstruct a history by working backwards . the approach is very similar to the construction of a full history from the backbone in the parallel dynamics . the final conditions , @xmath262 can be used to obtain final values : @xmath263 are randomly chosen from the uniform distributions on @xmath264 $ ] . for each @xmath15 , these final values are then the correct values backwards in time through the latest time when site @xmath15 is renewed . suppose site @xmath15 is last renewed at time @xmath265 that is @xmath266 or @xmath15 is a neighbor of @xmath267 . if @xmath266 then @xmath268 and this value holds until the next earliest time that @xmath15 is renewed . if site @xmath15 is renewed at time @xmath16 because it is the neighbor of the extremal site @xmath267 then @xmath269 is chosen from the uniform distribution on @xmath270 $ ] and this value holds until the next earliest time that @xmath15 was renewed . working backwards in this way a definite history is constructed from the set of extremal values , extremal sites and conditions . it should be noted that the reconstruction of a definite history from the data generated by the conditional dynamics can be carried out in parallel in polylog time . during conditional dynamics , the values of the @xmath206 s become spatially non - uniform however at times when the gap increases , almost all the @xmath206 s are equalized . if @xmath68 increases at time @xmath271 then for all @xmath15 , @xmath272 except , of course , at the extremal site @xmath267 and its neighbors where the @xmath206 s are reset to zero . p - completeness for conditional dynamics with @xmath273 is proved by a reduction from the monotone circuit value problem . to carry out the reduction we need a way to implement and and or gates and non - local fan - out . we will explain how the reduction works for the one - dimensional model . since non - local fan - out can be implemented , the p - completeness proof holds even for the one - dimensional problem . the inputs of the conditional bak sneppen decision problem are the initial conditions , @xmath274 , the random numbers , @xmath275 , a specified site @xmath193 , time @xmath124 and bound @xmath194 . the problem is to determine whether @xmath276 . in the reduction from cvp , truth values reside at specified sites and are represented by the values of @xmath240 . if @xmath277 , the site has the value true at time @xmath16 and if @xmath278 the site has the value false . the number @xmath279 is chosen to be small enough that the total time duration to evaluate the circuit , @xmath124 still yields small values , thus we set @xmath280 . sites with truth values are separated by background sites where @xmath281 . if site @xmath15 has an initial truth value this is represented by an initial value , @xmath282 or 1 . sites that represent outputs of gates or fan - outs have @xmath283 . in addition to sites representing truth values , there is a single site called @xmath284 with values @xmath285 and @xmath286 . a site @xmath15 carrying a truth value can be _ read _ at time @xmath16 by setting @xmath287 . if site @xmath15 is true at time @xmath16 it is selected since @xmath288 whereas , if @xmath15 is false at time @xmath16 then @xmath289 and @xmath284 is selected at time @xmath16 . finally , for each and gate and each fan - out , we need two neighboring auxiliary sites @xmath290 and @xmath291 with @xmath292 and @xmath293 . first , we show how to fan - out a truth value at site a at time @xmath16 to a new site , a@xmath294 at time @xmath295 using the gadget shown in fig.[fig : fan ] . in fig.[fig : fan ] , white squares have @xmath296 , black squares have @xmath297 and gray squares have small but non - zero @xmath298 values . the gray squares in column @xmath284 have @xmath299 and the gray square in column @xmath291 at time @xmath300 has @xmath301 . the value of @xmath302 depends on the truth value of a. if a is true then @xmath303 but if a is false then @xmath304 . at time @xmath16 , site @xmath290 is selected . first suppose a is true . then , since @xmath305 and @xmath306 we have that site a is selected at time @xmath69 . the selection of site a means that @xmath307 and therefore , @xmath308 so that site @xmath291 is selected at time @xmath300 . the selection of @xmath291 at time @xmath300 insures that a@xmath294 is set to true for any time later than @xmath300 since a@xmath294 is refreshed at time @xmath295 and @xmath309 . in fig . [ fig : fan ] we show a@xmath294 read at time @xmath295 but this is not necessary , it can also read at any time after @xmath295 . now suppose a is false . then site @xmath284 is selected at time @xmath69 and , consequently , @xmath310 . thus site @xmath284 is selected at time @xmath300 and site a@xmath294 fails to be refreshed so that @xmath311 and a@xmath294 is false . . ] the non - local fan - out just described can be easily extended to produce a non - local and gate as shown in fig . [ fig : and ] . as before , @xmath312 . the inputs a and b may be anywhere on the lattice , and the output a@xmath313b appears immediately to the right of the auxiliary sites @xmath290 and @xmath291 . suppose first that both a and b are true , then @xmath314 and @xmath315 so that @xmath291 is selected at time @xmath295 and a@xmath313 b is refreshed at time @xmath316 . thus the site a@xmath313b is correctly set to true at time @xmath316 . on the other hand , suppose that a is false and b is true . then @xmath317 . thus site a@xmath313b fails to be refreshed so that a@xmath313b is correctly set to false . the other two possibilities for a and b are easily seen to work in the same way . , site a@xmath313b is set to a and b. ] the or gate is local in space but non - local in time and is shown in fig.[fig : or ] . initially site a@xmath318b is set to false however , at time @xmath300 , a@xmath318b is correctly set to a or b since if either a or b are true , then a@xmath318b is refreshed and can be selected at any later time to be the input to a fan - out or and gate . figure [ fig : or ] shows a and b read at time @xmath16 and @xmath300 however any time separation between the reading a and b is permitted , the only constraint on the or gate is that both a and b must be read before a@xmath318b is read . , site a@xmath318b is set to a or b. ] these gadgets are sufficient to reduce monotone cvp to the one - dimensional conditional bak sneppen problem and show that the latter is p - complete . note that the non - local character of the gadgets allows non - planar monotone cvp to be reduced to the one - dimensional problem , in contrast to the situation for standard dynamics where the p - completeness holds only for @xmath196 . our main result is that histories can be efficiently generated in parallel . specifically , simulations and analytic arguments suggest that a history of length @xmath124 can be simulated on a pram with polynomially many processors in average parallel time @xmath319 with the actual asymptotic behavior close to @xmath320 . the exponential speed - up achieved by parallelization is the result of re - organizing the history into a nested hierarchy of independent avalanches . the construction of a single @xmath26-avalanche can be carried out in average parallel time @xmath321 which is exponentially less than the expected duration of the avalanche @xmath322 . how is it possible to create long range correlations in space and time very quickly in parallel ? first of all , it must be emphasized that the ground rules for parallel computation with a pram allow for non - local transmission of information in a single time step . to see how space - time correlations can be set up in a parallel time that is polylog in the correlation length or time , consider one step in the hierarchical construction of avalanches . for example , suppose two @xmath94-avalanches are concatenated to yield a single @xmath95-avalanche . the correlation length and time are thereby increased by constant factors in a fixed number of parallel steps . non - local transmission of information is needed to align the origin of the second @xmath94-avalanche with the final extremal site of the first @xmath94-avalanche and it is this alignment of the two @xmath94-avalanches that increases the correlation length and time . the result is that spatio - temporal correlations grow exponentially in the number of parallel steps . in the parallel construction , independent _ collection of @xmath94-avalanches are concatenated to form an @xmath95-avalanche . nonetheless , avalanches have a temporal structure and exhibit aging @xcite . aging is consistent with the independence of subavalanches for the following reason . the last extremal value in the construction of an @xmath95-avalanche must exceed @xmath95 and this extremal value is more likely to come from the last @xmath94-avalanche used in the construction . thus , extremal values and other properties of avalanches display aging . another example where critical correlations are set up much faster in parallel than might be expected are cluster monte carlo algorithms @xcite for critical spin systems . each cluster sweep can be accomplished in polylog parallel time on a pram with polynomially many processors using a parallel graph connectivity algorithm . the number of sweeps required to reach equilibrium scales as a small power ( typically much less than unity ) of the system size . thus long range critical correlations are set up in a parallel time that is much less than the correlation length . this is quite different than physically realistic local dynamics where critical correlations require a time that is at least quadratic in the correlation length . the intuition that histories are generated by an inherently sequentially process is not entirely wrong . in fact , standard dynamics was shown here to be associated with a p - complete decision problem for @xmath196 . standard dynamics takes as its inputs the initial values on the lattice and a sequence of @xmath323-tuples of numbers that are used to renew the extremal site and its neighbors at each time step . the p - completeness result implies that it is almost certainly not possible to parallelize standard dynamics so that the resulting history is generated in polylog time . the key problem that prevents full parallelization of standard dynamics is that all site values are explicitly defined at every step in the construction . in contrast , parallel dynamics does not assign explicit values except in the avalanche backbones until the end of the calculation . for the model , predicting the future starting from explicit initial conditions is harder than sampling a typical history . conditional dynamics is a third method for generating histories that has the smallest set of space - time sites explicitly defined until the end of the calculation . nonetheless , conditional dynamics is also associated with a p - complete decision problem and so can not be efficiently parallelized . parallel dynamics shares features in common with both conditional and standard dynamics and is intermediate in the degree to which it avoids explicit specification of site values until the end of the computation . the results presented here reflect characteristics of the model that are independent of the pram model of computation in which they were presented . the primary result of the paper can equivalently be stated in terms of families of boolean circuits . specifically , we can envision a hard - wired device composed of logical gates and random bit generators . the gates are wired in a feedforward direction so that , in generating a history , each gate is used only once . when the circuit is activated it produces a statistically correct history . for each system size @xmath114 and time @xmath124 we need a different circuit . because of the equivalence of prams and circuit families , our result can be stated in terms of the logical depth of the circuit . logical depth is the longest sequence of gates between any of the random bit generators and the output representing the history . the existence of an efficient parallel algorithm implies that there is a uniform family of the circuits for generating histories whose depth scales polylogarithmically in @xmath124 , independent of @xmath114 . the number of gates in the circuit is bounded by a power of @xmath324 . the actual running time for any real circuit generating histories would , of course , be polynomial and not polylogarithmic in @xmath124 because of the need for wires connecting distant gates required for establishing the long range spatio - temporal correlations . a variety of non - equilibrium models in statistical physics have been employed to study the spontaneous emergence of complexity . besides the model , other examples include diffusion limited aggregation @xcite ( dla ) , invasion percolation and sandpiles . besides their application to specific physical phenomena , these models have a broad appeal because they are governed by simple microscopic rules and yet they display self - organized criticality . it has been argued that they perhaps shed light on far more complex phenomena found , for example , in biology or economics . surely , however , biological and economic systems generate histories that have polynomial rather than polylogarithmic logical depth . perhaps one criterion for a model of the spontaneous emergence of complexity is that it should generate histories that require more than polylogarithmic depth to simulate . neither the model nor invasion percolation @xcite satisfy this requirement . on the other hand , both diffusion limited aggregation and the bak , tang and wiesenfeld ( btw ) sandpile model @xcite are related to p - complete problems @xcite which strongly suggests that the clusters or avalanches associated with these models can not be simulated in polylog depth . however , neither model generates histories much deeper than a power of the system size . in the case of dla , the cluster size can not exceed the system size so that the length of the history is bounded by @xmath325 where @xmath326 is the fractal dimension of the cluster . sandpile models are well - defined for arbitrarily long times . however , the btw model has an abelian property , which allows avalanches to be combined in any order . thus , it is possible to rearrange long histories so they are generated in polylog parallel time . from a given initial condition , it requires @xmath327 steps to relax the system @xcite . a sequence of avalanches due to randomly dropping sand on single sites can be organized into a binary tree using the same idea as adding numbers in parallel . the result is that a history of length @xmath328 can be simulated in parallel in time @xmath329 . by contrast , we find that histories can be simulated in parallel time @xmath319 independent of system size . this work was supported by nsf grants dmr-9978233 . we thank ray greenlaw , cris moore and stefan boettcher for useful discussions .
the physical processes involved in diffusion of galactic cosmic rays in the interstellar medium are addressed . we study the possibility that the nonlinear mhd cascade sets the power - law spectrum of turbulence which scatters charged energetic particles . we find that the dissipation of waves due to the resonant interaction with cosmic ray particles may terminate the kraichnan - type cascade below wavelengths cm . the effect of this wave dissipation has been incorporated in the galprop numerical propagation code in order to asses the impact on measurable astrophysical data . the energy - dependence of the cosmic - ray diffusion coefficient found in the resulting self - consistent model may explain the peaks in the secondary to primary nuclei ratios observed at about 1 gev / nucleon .
astro-ph0510335
the galactic cosmic rays the gas of relativistic charged particles with high energy density can not always be treated as test particles moving in given magnetic fields . in particular , the stochastic acceleration of cosmic rays by mhd waves is accompanied by the damping of the waves , since the wave energy is dissipated . the rate of wave damping on cosmic rays through the cyclotron resonance interaction was first estimated by @xcite . if we exclude cold hi regions ( where the waves are damped by collisions of ions with neutral atoms ) and also perhaps regions of the interstellar gas with very high temperature @xmath1 k and weak magnetic field ( where landau damping on thermal particles is high ) , then one finds that this mechanism of dissipation could dominate over other known mechanisms in the interval of wavelengths @xmath2 cm to @xmath3 cm . the non - resonant interaction of diffusing cosmic rays with magnetosonic waves @xcite is important only at large wavelengths and is not important for the present investigation . we also do not consider the large body of instabilities in cosmic rays that may arise because of the non - equilibrium distribution of charged energetic particles caused by possible strong anisotropy or large gradients and which may amplify waves in the background plasma see e.g. @xcite and @xcite for a review of such processes . cyclotron wave damping on cosmic rays changes the wave spectrum in the interstellar medium , which in turn affects the particle transport since the cosmic ray diffusion coefficient is determined by the level of turbulence which scatters charged particles . thus in principle the study of cosmic ray diffusion requires a self - consistent approach . we shall see below that the effect of cosmic rays on interstellar turbulence should be taken into account at energies below a few gev / nucleon , and this may result in a considerable increase in the particle diffusion coefficient . this picture will be also verified by its consistency with observations of interstellar turbulence . consistency of this picture with observations of interstellar turbulence is also discussed it is remarkable that the interpretation of cosmic ray data on secondary nuclei may require this effect . the secondary nuclei are produced in cosmic rays in the course of diffusion and nuclear interactions of primary nuclei with interstellar gas . the @xmath4h , @xmath5he , li , be , b and many other isotopes and elements are almost pure secondaries . their abundance in cosmic ray sources is negligible and they result from the fragmentation of heavier nuclei . cosmic ray antiprotons and the major fraction of positrons are among the secondary species as well . the ratio of secondary to primary nuclei such as has a peak at about @xmath6 gev / n and decreases both with increasing and decreasing energy , e.g. @xcite . the high energy behavior is naturally explained by particle scattering on interstellar mhd turbulence with a power - law spectrum in wavelength , but the required sharp increase of the diffusion coefficient at small energies has usually been considered as improbable since it needs a drastic `` physically unjustified '' bending down of the wave spectrum at wave numbers @xmath7 @xmath8 , see e.g. @xcite , @xcite . alternative explanations of the peaks in secondary to primary ratios without invoking peculiarities in the cosmic ray diffusion coefficient have been suggested . the two most popular of these involve diffusive - convective particle transport in the hypothetical galactic wind @xcite and the stochastic reacceleration of cosmic rays by interstellar turbulence @xcite . these processes are included in the numerical computations of cosmic ray transport in the galaxy in the frameworks of the galprop code @xcite . the situation with the interpretation of the energy dependence of secondaries in cosmic rays is still uncertain . the peaks in the secondary to primary nuclei ratios calculated in diffusion - convection models seem to be too wide . models with reacceleration reproduce the shape of the peaks but the absolute flux of antiprotons turns out to be too low compared to observations @xcite . this is why the effect considered in the present work is so important . we investigate a self - consistent model of cosmic - ray diffusion in interstellar turbulence where the wave damping on energetic particles is taken into account , and implement this effect in the galprop code . to make the first calculations reasonably tractable we keep only the most essential features of cosmic - ray diffusion in random magnetic fields and use the simplest description of the nonlinear wave cascade in the interstellar turbulence . the steady state transport equation that describes diffusion and convective transport of cosmic ray protons and nuclei in the interstellar medium is of the form ( see e.g. , * ? ? ? * for discussion ) : @xmath9-\frac{\partial}{\partial p}\left ( p\frac{\nabla\mathbf{u}}{3}\psi\right ) + \frac{\partial}{\partial p}\left ( \dot{p}_{\mathrm{loss}}\psi\right ) + \frac{\psi}{\tau}=q . \label{diffeq}\ ] ] here @xmath10 is the particle distribution function in momentum @xmath11 normalized on total cosmic ray number density as @xmath12 , @xmath13 is the spatial diffusion coefficient , @xmath14 is the diffusion coefficient in momentum , @xmath15 is the velocity of large - scale motions of the interstellar medium ( e.g. the velocity of a galactic wind),@xmath16 is the momentum loss rate for the energetic charged particles moving through the interstellar medium , @xmath17 is the time scale for nuclear fragmentation . if needed , the supplementary term which describes radioactive decay can be added to eq . ( [ diffeq ] ) . the spatial boundary condition for @xmath18 is @xmath19 , corresponding to the free exit of cosmic rays from the galaxy to intergalactic space where their density is negligible . the region of cosmic ray diffusion is a cylinder of radius @xmath20 kpc and total width @xmath21 ( @xmath22 kpc ) . the source term @xmath23 includes both the direct production of primary energetic particles accelerated from the thermal background in galactic sources ( e.g. supernova remnants ) and the contribution to the nuclei considered via the processes of nuclear fragmentation and radioactive decay of heavier nuclei . the typical value of the diffusion coefficient found from the fit to cosmic ray and radioastronomical data is @xmath24 cm@xmath4/s at energy @xmath261 gev / n , giving a diffusion mean free path @xmath27 pc ( @xmath28 is the particle velocity ) . on the `` microscopic level '' the spatial and momentum diffusion of cosmic rays results from the particle scattering on random mhd waves and discontinuities . in the linear approximation , the alfven ( with the dispersion relation @xmath29 ) , the fast magnetosonic ( @xmath30 ) , and the slow magnetosonic ( @xmath31 ) waves can propagate in a low @xmath32 plasma , @xmath33 @xmath34 , where @xmath35 is the alfven velocity and @xmath36 @xmath37 is the collisionless sound velocity ( for mhd conditions it is a factor of a square root of the gas adiabatic index larger ) , where @xmath38 , @xmath39 and @xmath40 are the magnetic field strength , the pressure , and the mass density of the interstellar gas respectively . in addition to waves , the static entropy variations can exist in the interstellar medium . in a collisionless plasma , the slow magnetosonic waves are not heavily damped only if the plasma is non - isothermal and the electron temperature considerably exceeds the ion temperature . the alfven velocity is @xmath41 cm / s , where @xmath42 is the number density of hydrogen atoms . the value of the sound velocity is @xmath43 cm / s in neutral gas with temperature @xmath44 ; @xmath36 is larger in the ionized gas because of the free electron contribution to gas pressure @xmath39 . the approximation of low @xmath32 plasma is valid in the dominant part of the interstellar medium at @xmath45 @xmath46 g since typically @xmath47 @xmath48 and @xmath49 k in hot hii regions , @xmath50 @xmath48 and @xmath51 k in warm intercloud gas , @xmath52 @xmath48 and @xmath53 k in clouds of atomic hydrogen , @xmath54 @xmath48 and @xmath55 k in molecular clouds and thus @xmath56 everywhere . the effective `` collision integral '' for energetic charged particles moving in small amplitude random fields @xmath57 can be taken from the standard quasi - linear theory of plasma turbulence , see e.g. @xcite . the `` collisional integral '' averaged over fast gyro - rotation of particles about the magnetic field @xmath58 contains the spectral densities of the effective frequencies of particle collisions with plasmons in the form @xmath59 , \label{eq2}\end{aligned}\ ] ] @xmath60 ( as presented by * ? ? ? the waves are assumed symmetric about the magnetic field direction and to have zero average helicity . the index @xmath61 characterizes the type of waves including the direction of their propagation , @xmath46 is the cosine of pitch angle , the energy densities of random magnetic fields for perpendicular and parallel to the average field @xmath58 components are @xmath62 and @xmath63 respectively , @xmath64 is the bessel function . it is explicitly taken into account that the scattering at @xmath65 occurs only on the fast magnetosonic waves propagating almost perpendicular to the magnetic field . the larmor radius is @xmath66 cm , where the particle magnetic rigidity @xmath67 is measured in gv , and the average magnetic field is measured in @xmath68 . the wave - particle interaction is of resonant character so that an energetic particle is predominantly scattered by the irregularities of magnetic field @xmath69 that have the projection of wave vector on the magnetic field direction equal to @xmath70 . the integers @xmath71 correspond to the cyclotron resonances of different orders . the efficiency of particle scattering depends on the polarization of the waves and on their distribution in @xmath72-space . the first - order resonance @xmath73 is the most important for the isotropic and also for the one - dimensional distribution of random mhd waves along the average magnetic field , @xmath74 . in some cases for calculation of scattering at small @xmath46 and for calculation of perpendicular diffusion the broadening of resonances and magnetic mirroring effects should be taken into account . the evolution of the particle distribution function on time - scales @xmath75 and distances @xmath76 can be described in the diffusion approximation with the following expressions for the spatial diffusion coefficient along the magnetic field @xmath77 and the diffusion coefficient in momentum @xmath78 : @xmath79 @xmath80^{2}}{v^{2}}-\left ( { \textstyle\sum_{\alpha , s,\mathbf{k } } } \nu_{\mu}^{\alpha}(\mathbf{k},s , p)\frac{v_{f}^{\alpha}(\mathbf{k})}{v}\right ) ^{2}\left ( { \textstyle\sum_{\alpha , s,\mathbf{k } } } \nu_{\mu}^{\alpha}(\mathbf{k},s , p)\right ) ^{-1}\right ] \!,\label{eq4 } \ ] ] see @xcite where the analogous equations were derived for one - dimensional turbulence with @xmath81 . here @xmath82 , the summations contain integrals over @xmath72-space , and the terms which contain @xmath83 should be corrected in eqs . ( [ eq3 ] ) , ( [ eq4 ] ) compared to eq . ( [ eq2 ] ) : multiplied by @xmath84 in @xmath85 , and multiplied by @xmath86 in other terms . one can check that the interaction of energetic particles with slow magnetosonic waves is relatively weak ( as @xmath87 ) and can be ignored . locally , the cosmic ray diffusion is anisotropic and occurs along the local magnetic field because the particles are strongly magnetized , @xmath88 . the isotropization is accounted for by the presence of strong large - scale ( @xmath26100 pc ) fluctuations of the galactic magnetic field . the problem is not trivial even in the case of relatively weak large scale random fields , since the field is almost static and the strictly one - dimensional diffusion along the magnetic field lines does not lead to non - zero diffusion perpendicular to @xmath58 , see @xcite , @xcite , @xcite . the eqs . ( [ eq3 ] ) and ( [ eq4 ] ) are too cumbersome for our present application . based on eq . ( [ eq3 ] ) and with reference to the detailed treatment of cosmic ray diffusion by @xcite and @xcite , we use below the following simplified equation for the diffusion coefficient : @xmath89 , \label{eq5}\ ] ] where @xmath90 is the resonant wave number , and @xmath91 is the spectral energy density of waves normalized as @xmath92 . the random field at the resonance scale is assumed to be weak , @xmath93 . the cosmic ray diffusion coefficient is equal to @xmath94 for particles with @xmath95 under the assumption of a power law spectrum of turbulence @xmath96 @xmath97 . we introduced here the principle wave number of the turbulence @xmath98 and the amplitude of random field @xmath99 at this scale . the diffusion has scaling @xmath100 . the diffusion in momentum is described by the following equation which is a simple approximation of eq . ( [ eq4 ] ) : @xmath101 eqs . ( [ eq5])-([eq7 ] ) give estimates of particle diffusion in position and momentum needed in eq . ( [ diffeq ] ) . they reflect the most essential features of cosmic - ray transport : the frequency of particle scattering on random magnetic field is determined by the energy density of this field at the resonance wave number @xmath102 ; the acceleration is produced by the waves moving with typical velocity @xmath103 and is stochastic in nature . ( [ eq5 ] ) implies equal intensities of waves moving along the magnetic field in opposite directions , the imbalanced part of the total wave energy density should not be taken into account when calculating @xmath104 , see @xcite for details . ( [ eq5])-([eq7 ] ) are also valid for small - amplitude nonlinear waves including weak shocks . the equations can be used for the isotropic distribution of alfven and fast magnetosonic waves and they give correct order of magnitude estimates for the wave distribution concentrated around the direction of average magnetic field . it should be pointed out however that the isotropization of the diffusion tensor does not occur in the case of a pure parallel propagation of waves ( @xmath74 ) . another special case is @xmath105d turbulence with perpendicular propagation of waves ( @xmath106 ) . in this case , the scattering occurs only for magnetosonic waves through the resonance @xmath65 which leads to a very large diffusion coefficient , about a factor @xmath107 larger than given by eqs . ( [ eq5 ] ) , ( [ eq6 ] ) . ( [ eq6 ] ) shows that the level of interstellar turbulence that is needed to account for the diffusion of gev cosmic rays is very small : @xmath108 at @xmath109 cm ( if @xmath110 ) . an extension to smaller wave numbers gives @xmath111 where @xmath112 is in @xmath8 . the description of mhd turbulence is a complicated and not completely solved problem even in the case of small - amplitude random fields . comprehensive reviews of mhd turbulence have been given by @xcite , and by @xcite , @xcite , and @xcite with application to interstellar turbulence . the classic problem is the determination of the wave spectrum in the presence of sources at small wave numbers @xmath113 and the strong dissipation at much larger wave numbers ( in some cases the cascade is inverse ) . note that the spectrum of interstellar mhd turbulence determines the transport coefficients eqs . ( [ eq5 ] ) , ( [ eq7 ] ) . according to the kolmogorov - obukhov hypothesis @xcite , the resulting spectrum at intermediate @xmath112 , i. e. in the inertial range , is characterized by a constant energy flux to higher wave numbers . the hypothesis was originally suggested for the description of developed hydrodynamic turbulence in incompressible fluids . the kolmogorov spectrum is of the form @xmath114 . the spectrum of weak acoustic turbulence @xmath115 was found by @xcite . this result was criticized by @xcite . they argued that the developed shocks produce an additional dissipation of acoustic wave energy . the shock - dominated turbulence ( also called the burgers turbulence ) is characterized by the spectrum @xmath116 irrespective of the nature of dissipation at the shock front . the presence of magnetic field in mhd turbulence complicates the issue because the turbulence becomes anisotropic and new types of waves arise in a magnetized medium . generally , all gradients are larger perpendicular to the field and all perturbations are elongated along the magnetic field direction even with isotropic excitation . @xcite and @xcite gave the first phenomenological theory of mhd turbulence and obtained the spectrum @xmath115 . they assumed that small - scale fluctuations are isotropic , which is contradictory because wave interactions break the isotropy in the presence of an external magnetic field . direct observations of mhd turbulence in the solar wind plasma where @xmath117 have shown the existence of a kolmogorov - type turbulence spectrum that contains waves moving both along the magnetic field and in almost perpendicular directions @xcite . such a spectrum was obtained over several decades of wave numbers in solar wind radio propagation studies @xcite . numerical simulations of incompressible ( @xmath118 ) mhd turbulence favored the kolmogorov spectrum @xcite . over the past decade , there has been a renewed interest in understanding of magnetohydrodynamic turbulence as it applies to interstellar magnetic field and density fluctuations ( @xcite , see also earlier work by @xcite ) . @xcite exploited anisotropy in mhd turbulence and obtained kolmogorov - like spectrum for the energy density of alfven waves . the `` elementary interactions '' between alfven waves satisfies the three - wave resonance conditions . however there is no exact relation between wave number and frequency in this case of strong turbulence . the main part of the energy density in this turbulence is concentrated perpendicular to the local magnetic field wave vectors @xmath119 , while the parallel wave numbers are small : @xmath120 ^{1/2}k_{\perp}$ ] . the cascade is anisotropic with energy confined within the cone @xmath121 . numerical simulations have confirmed this concept ( @xcite ) . although the formalism has been developed for incompressible mhd turbulence , @xcite argued that the compressibility does not essentially alter the results on the alfven wave spectrum . the distribution of slow magnetosonic waves passively follows that of alfven waves . the fast magnetosonic waves have an independent nonlinear cascade which is isotropic and has a kraichnan - type spectrum @xmath115 . these conclusions were supported by numerical simulations by @xcite . the description of weak mhd turbulence in low @xmath32 plasma outlined above is probably not complete and needs further analysis before it is accepted as a standard model of interstellar turbulence . first , there is still the discrepancy between theoretical results of different authors . thus considering the scattering of alfven waves and fast magnetosonic waves on slow magnetosonic waves , @xcite found the kraichnan - type spectra for all these types of waves with their preferentially parallel propagation , which disagrees with @xcite . second , consideration of processes in turbulent collisionless plasmas at the kinetic level involves additional nonlinear processes of induced wave scattering on thermal ions @xcite that may change the spectra @xcite . third , the real turbulence can be strongly intermittent , imbalanced , etc . ( e.g. * ? ? ? * ) , which may also affect the interstellar mhd spectrum . information on the extended interstellar turbulence spectrum has been obtained from radio scintillation and refraction observations ( sensitive to fluctuations of thermal electron density ) , measurements of the differential faraday rotation angles from distant sources ( mainly produced by fluctuations in the interstellar magnetic field ) , and the observations of random motions in the interstellar gas . these data are consistent with the assumption that a single close - to - kolmogorov spectrum extends from scales @xmath122 to @xmath123 cm @xcite , see @xcite and references therein . in the absence of an easily manageable and commonly accepted exact equation for the energy density @xmath91 , we employ below the simplest steady state phenomenological equation that represents the concept of a wave cascade in the inertial range of wave numbers : @xmath124 ( @xcite ; this approach goes back to @xcite and @xcite ) . here the approximation of a characteristic time @xmath125 is used to account for the non - linear wave interactions which provide the transfer of energy in k - space ) when a diffusive flux of wave energy in @xmath112-space is assumed and when the equation is written for a power spectrum @xmath126 not averaged over direction ( here @xmath127 ) . ] . formally , the kolmogorov - type spectrum @xmath114 follows from eq . ( [ eq8 ] ) if @xmath128 @xmath129 ^{-1}$ ] , where the constant @xmath130 as can be estimated from the simulations of verma et al ( 1996 ) . the kraichnan spectrum @xmath114 is obtained from eq . ( [ eq8 ] ) if @xmath131 @xmath132 ^{-1}$ ] , where @xmath133 . the kolmogorov - type nonlinear rate @xmath134 is relatively high since it is proportional to the amplitude of weak random field whereas the kraichnan rate @xmath135 is proportional to the amplitude squared . our preliminary estimates @xcite showed that the mhd cascade with the kolmogorov rate is not significantly affected by damping on cosmic rays even if the waves propagate along the magnetic field which makes the wave - particle interactions the most efficient . in addition , if the concept of @xcite works for interstellar turbulence , then the kolmogorov rate refers to the alfven waves which are distributed almost perpendicular to an external magnetic field and thus do not produce any significant scattering of cosmic rays , see section 2 . so , we do not consider the damping of cascades with the kolmogorov rate in what follows . for a cascade with the kraichnan rate @xmath135 which describes an isotropic or close to parallel distribution of mhd waves ( according to * ? ? * it refers only to fast magnetosonic waves but not to alfven waves ) , the equation for wave energy density can be written as follows : @xmath136 @xmath137 . here @xmath138 is the wave attenuation rate on cosmic - ray particles ( @xmath139 ) and on thermal particles ( @xmath140 ) , @xmath141 characterizes the source strength . the main sources of interstellar turbulence are supernova explosions , winds of massive stars , superbubbles , and differential rotation of the galactic disk . these sources produce strong random magnetic fields and probably initiate nonlinear wave cascades at the scale @xmath142 pc . below we ignore the dissipation on thermal particles ( see discussion below ) and set @xmath143 to study purely the effect of damping on cosmic rays . we actually deal with a small part of the possible global wave spectrum at wave numbers @xmath144 @xmath8 which resonantly scatter cosmic rays with energies less than about @xmath145 gev / n . the existence of a single interstellar mhd spectrum over @xmath146 orders of magnitude is an unsolved problem in itself and is beyond the scope of the present work . the quantities @xmath147 and @xmath148 are used here only for purposes of normalization . the equation for wave amplitude attenuation on cosmic rays is @xcite : @xmath149 where @xmath150 . the solution of eqs . ( [ eq9 ] ) , ( [ eq10 ] ) allows us to find the wave spectrum : @xmath151 , \label{eq11}\ ] ] @xmath152 . here the spectral wave density at the principal scale is determined by the source strength : @xmath153 . the second term in square brackets is increasing with @xmath112 . the wave damping on cosmic rays decreases the wave energy density and can even terminate the cascade if the expression in square brackets reduces to zero at some @xmath154 . the wave energy density should be set to zero at @xmath155 in this case . now with the use of eqs . ( [ eq5 ] ) , ( [ eq11 ] ) one can determine the cosmic ray diffusion coefficient , which is @xmath156@xmath157 @xmath158 . here @xmath159\varpropto v(p / z)^{1/2}$ ] is the diffusion coefficient calculated for the kraichnan - type spectrum without considering wave damping . the second term in the denominator of eq . ( [ eq12 ] ) describes the modification of the diffusion coefficient due to wave damping . the effect becomes stronger to smaller @xmath11 . the diffusion coefficient should be formally set to infinity at @xmath160 if the square bracket in eq . ( [ eq11 ] ) goes to zero at some @xmath161 . the constant @xmath162 characterizes the strength of the effect for a given cosmic ray spectrum @xmath163 . as the most abundant species , the cosmic ray protons mainly determine the wave dissipation . thus , with good precision only the proton component with @xmath164 needs to be taken into account to calculate @xmath165 by the simultaneous solution of eqs . ( [ diffeq ] ) and ( [ eq12 ] ) . the diffusion mean free path for other nuclei of charge @xmath166 is @xmath167 if it is @xmath168 for protons . let us estimate the effect of wave damping . assuming that the cosmic - ray energy density is about @xmath6 ev / cm@xmath5 , the diffusion coefficient @xmath170 cm@xmath4/s at @xmath6 gev , @xmath171 km / s , @xmath172 @xmath46 g , and @xmath173 , the second term in the denominator in eq . ( [ eq12 ] ) is about unity at gev energies and falls at higher energies . we conclude that the kraichnan - type cascade is significantly affected by damping on cosmic rays , and this should lead to the modification of cosmic - ray transport at energies less than about @xmath174 gev / n . to demonstrate the effect of wave damping , let us consider a simple case of one - dimensional diffusion with source distribution @xmath175 ( corresponding to an infinitely thin disk of cosmic - ray sources located at the galactic mid - plane @xmath176 ) and a flat cosmic - ray halo of height @xmath177 , see @xcite . the source spectrum at @xmath178 gv is @xmath179 , where @xmath180 approximately . considering energetic protons , we ignore energy losses and nuclear fragmentation ( @xmath181 , @xmath182 ) and assume that a galactic wind is absent ( @xmath183 ) in eq . ( [ diffeq ] ) . let us also assume that stochastic reacceleration does not significantly change the energies of cosmic - ray particles during the time of their diffusive leakage from the galaxy and set @xmath184 in eq . ( [ diffeq ] ) . the solution of eq . ( [ diffeq ] ) in the galactic disk is then @xmath185 the escape length ( the `` grammage '' ) , which determines the production of secondaries during the cosmic - ray leakage from the galaxy , is equal to @xmath186 , where @xmath187 is the gas surface density of the galactic disk . the simultaneous solution of eqs . ( [ eq12 ] ) and ( [ eq13 ] ) allows us to find @xmath163 and @xmath165 . introducing the function @xmath188 $ ] , which is the cosmic ray spectrum for an unmodified kraichnan spectrum , and substituting @xmath18 from eq . ( [ eq13 ] ) into eq . ( [ eq12 ] ) , one finds after two differentiations the following equation for the ratio @xmath189 as a function of @xmath190 : @xmath191}{9x}\varphi(x ) , \label{eq14}\ ] ] with the constraint @xmath192 . note that the same equation is valid for the ratio @xmath193 . the function @xmath194 in r.h.s . of eq . ( [ eq14 ] ) can be approximated by a power law function proportional to @xmath195 , @xmath196 . thus for @xmath197 , one has @xmath198 for ultrarelativistic ( @xmath199 ) and @xmath200 for nonrelativistic ( @xmath201 ) protons respectively . this allows us to find the solution of eq . ( [ eq14 ] ) in an explicit form . the diffusion coefficient @xmath202 is then : @xmath203 j_{1/(b-2)}[2w(p)]},\;w(p)=\frac{2}{3(b-2)}\sqrt{gp^{3/2}\psi_{0}(p ) } , \label{eq15}\ ] ] where @xmath204 is the gamma function , @xmath205 is the bessel function of the first kind . with @xmath11 decreasing from infinity and the corresponding increase of @xmath206 from zero , the bessel function goes to zero at some @xmath207 ( where @xmath208 at @xmath198 , and @xmath209 at @xmath200 ) . this means that @xmath165 becomes infinite at @xmath161 because of complete termination of the cascade ; @xmath210 at @xmath211 . the expansion of the bessel function for small and large arguments results in the following approximations for eq . ( [ eq15 ] ) : @xmath212/\gamma\left [ ( 2b-3)/(b-2)\right]\right\ } w^{2}(p)}\rm{\textrm{\ at } } w(p)\ll1 , \label{eq16}\ ] ] and @xmath213}}{\gamma\left [ ( b-1)/(b-2)\right ] \sin\left [ \frac{\pi}{4}(3b-4)/(b-2)-2w(p)\right ] } \rm{\textrm{\ at } } 1\lesssim w(p)<w_{\ast}. \label{eq17}\ ] ] the last approximate expression for @xmath165 has a singularity at @xmath214 ; @xmath215 at @xmath198 , and @xmath216 at @xmath200 which is close to the values obtained from eq . ( [ eq15 ] ) . the corresponding @xmath217 can be found from the equation @xmath218^{2}g^{-1}$ ] . ( [ eq15])-([eq17 ] ) exhibit the rapid transition from unmodified to infinite diffusion . [ dxx ] demonstrates that eq . ( [ eq15 ] ) derived in the considerably simplified model of cosmic - ray propagation ( dash - dot ) agrees well with the shape of the diffusion coefficient found in the numerical simulations ( solid ) described in the next section . the results of this section can be used for approximate estimates of cosmic ray diffusion coefficient without invoking the full - scale calculations based on the galprop code . modified by interactions with cosmic rays in drd model.[waves],title="fig:",width=316 ] modified by interactions with cosmic rays in drd model.[waves],title="fig:",width=326 ] 1 the numerical calculations have been made using the galprop code for cosmic - ray propagation . galprop is a flexible numerical code written in c++ which incorporates _ as much realistic _ astrophysical input _ as possible _ together with latest theoretical developments . the model is designed to perform cosmic - ray propagation calculations for nuclei ( isotopes of h through ni ) , antiprotons , electrons and positrons , and computes @xmath219-rays and synchrotron emission in the same framework . the current version of the model includes substantial optimizations in comparison to the older model and a full 3-dimensional spatial grid for all cosmic - ray species . it explicitly solves the full nuclear reaction network on a spatially resolved grid . galprop models have been described in detail elsewhere @xcite . for the present calculation we use three - dimensional galactic models with cylindrical symmetry : @xmath220 spatial variables plus momentum . galprop solves the transport equation eq . ( [ diffeq ] ) for all cosmic - ray species using a crank - nicholson implicit second - order scheme . the spatial boundary conditions assume free particle escape . the diffusion coefficient as a function of momentum and the reacceleration or convection parameters are determined by boron - to - carbon ( b / c ) ratio data . the spatial diffusion coefficient is taken as @xmath221 , @xmath222 , @xmath223 , if necessary with a break ( @xmath224 below rigidity @xmath225 , @xmath226 above rigidity @xmath225 ) . the injection spectrum of nucleons is assumed to be a power law in momentum , @xmath227 and may also have breaks . to account for stochastic reacceleration in the interstellar medium , the momentum - space diffusion coefficient @xmath228 is related to the spatial coefficient @xmath229 via the alfvn speed @xmath230 , see eq . ( [ eq7 ] ) . the interstellar hydrogen distribution @xcite uses h i and co surveys and information on the ionized component ; the helium fraction of the gas is taken as 0.11 by number . the h@xmath231 and h i gas number densities in the galactic plane are defined in the form of tables , which are interpolated linearly . the conversion factor is taken as @xmath232@xmath233 mols.@xmath234/(k km s@xmath235 ) @xcite . the extension of the gas distribution to an arbitrary height above the plane is made using analytical approximations . the halo size is assumed to be @xmath236 kpc . the distribution of cosmic - ray sources is chosen to reproduce the cosmic - ray distribution determined by analysis of egret @xmath219-ray data @xcite : @xmath237where @xmath238 is a normalization constant , @xmath239 kpc , @xmath240 and @xmath241 are parameters . we note that the adapted source distribution - ray emissivity and the distribution of snr , believed to be the sources of cosmic rays , can be plausibly solved by adopting a conversion factor @xmath232 which increases with @xmath242 ( see a discussion in * ? ? ? ] is flatter than the distribution of snr @xcite and pulsars @xcite . data below 200 mev / nucleon : ace @xcite , ulysses @xcite , voyager @xcite ; high energy data : heao-3 @xcite , for other references see @xcite.[bc],title="fig:",width=288 ] mv ) . data below 200 mev / nucleon : ace @xcite , ulysses @xcite , voyager @xcite ; high energy data : heao-3 @xcite , for other references see @xcite.[bc],title="fig:",width=288 ] mv ) . data below 200 mev / nucleon : ace @xcite , ulysses @xcite , voyager @xcite ; high energy data : heao-3 @xcite , for other references see @xcite.[bc],width=288 ] 1 the code includes cross - section measurements and energy dependent fitting functions @xcite . the nuclear reaction network is built using the nuclear data sheets . the isotopic cross section database is built using the extensive t16 los alamos compilation of the cross sections @xcite and modern nuclear codes cem2k and laqgsm @xcite . the most important isotopic production cross sections ( @xmath243h , @xmath244h , @xmath244he , li , be , b , al , cl , sc , ti , v , mn ) are calculated using our fits to major production channels ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? other cross sections are calculated using phenomenological approximations by @xcite ( code wnewtr.for versions of 1993 and 2003 ) and/or silberberg and tsao @xcite ( code yieldx_011000.for version of 2000 ) renormalized to the data where it exists . for @xmath245 inelastic cross section we adapted the parametrization by @xcite ( code crosec , * ? ? ? * ; * ? ? ? the details of proton and antiproton cross sections are given in @xcite . secondary positron and electron production is computed using the formalism described in @xcite , which includes a reevaluation of the secondary @xmath246- and @xmath247-meson decay calculations . energy losses for nucleons and leptons by ionization , coulomb scattering , bremsstrahlung , inverse compton scattering , and synchrotron are included in the calculations . the heliospheric modulation is treated using the force - field approximation @xcite . the reaction network is solved starting at the heaviest nuclei ( i.e. , @xmath248ni ) . the propagation equation is solved , computing all the resulting secondary source functions , and then proceeds to the nuclei with @xmath249 . the procedure is repeated down to @xmath250 . in this way all secondary , tertiary etc . reactions are automatically accounted for . to be completely accurate for all isotopes , e.g. , for some rare cases of @xmath251-decay , the whole loop is repeated twice . for the present work , a new iterative procedure which includes a self - consistent determination of the cosmic - ray diffusion coefficient was developed . in the first step , cosmic - ray propagation is calculated using the undisturbed diffusion coefficient @xmath252 . in the second , we use the propagated proton spectrum at every spatial grid point to re - calculate the modified diffusion coefficient according to eq . ( [ eq12 ] ) . in the third step , cosmic - ray propagation is calculated using the new diffusion coefficient . steps 2 and 3 are repeated until convergence is obtained . the convergence is fast and the procedure requires only few iterations . in this way , using the secondary to primary ratio b / c , one can derive the damping constant @xmath253 which is considered as an adjustable parameter . lccccccc plain diffusion ( pd ) & 2.30/2.15 & 2.40 & 40 & @xmath254 & @xmath255 & & 44_999726 + diffusive + reacceleration ( dr ) & 1.80/2.40 & 1.60/2.50 & 4 & @xmath256 & 0.34 & 36 & 44_599278 + diffusive + reacceleration with + damping ( drd ) & 2.40/2.24 & 2.70 & 40 & @xmath257 & 0.50 & 22 & 44_999714kr + mv ) . data : ace @xcite , heao-3 @xcite , for other references see @xcite ( symbols are changed).[carbon],title="fig:",width=288 ] mv ) . data : ace @xcite , heao-3 @xcite , for other references see @xcite ( symbols are changed).[carbon],title="fig:",width=288 ] mv ) . data : ace @xcite , heao-3 @xcite , for other references see @xcite ( symbols are changed).[carbon],width=288 ] 1 , title="fig:",width=288],title="fig:",width=288 ] , width=288 ] 1 , title="fig:",width=288],title="fig:",width=288 ] , width=288 ] 1 , title="fig:",width=288],title="fig:",width=288 ] , width=288 ] 1 , title="fig:",width=259],title="fig:",width=259 ] , width=259 ] 1 the new self - consistent model of diffusive reacceleration with damping ( drd model ) is compared with two reference models which were earlier found to give good fits to cosmic ray data : the plain diffusion model ( pd model ) with an ad - hoc break in the diffusion coefficient and the model with distributed reacceleration ( dr model ) and power - law diffusion with no breaks @xcite . the parameters of these models found from the fit to the observed spectra of primary and secondary nuclei are summarized in table [ table1 ] . note that the parameters of the drd model given in table [ table1 ] are the `` seed '' values used in the iteration procedure . the diffusion coefficient after iteration calculations is shown in fig . [ dxx ] . the diffusion coefficients in all three models are presented in fig . the dr model diffusion coefficient has a weak energy dependence with a single index 1/3 as dictated by the assumed kolmogorov spectrum of interstellar turbulence . the pd model requires a break in the diffusion coefficient and an additional factor of @xmath258 to be able to match the b / c ratio at low energies . the drd model diffusion coefficient lies between these at high energies and has a sharp increase at 1 gv . of course , the actual mean free path length can not be infinite . we put an upper limit of @xmath2615 pc based on the estimate of streaming instability effect that arises at low rigidities and leads to the generation of additional turbulence which limits cosmic ray escape from the galaxy , see discussion below in section 6 . the selfconsistent spectrum of waves calculated in the drd model is shown in fig . [ waves ] . the spectrum bends downward at wavenumbers @xmath259 @xmath8 due to the wave damping on cosmic rays . the b / c ratio and carbon spectra ( primary ) after modulation look almost identical in all three models ( figs . [ bc ] , [ carbon ] ) . the modulated proton spectrum ( fig . [ protons ] ) matches the data in all three cases after some tuning , while the interstellar spectra are different . the most dramatic difference is the reduction of proton flux in the drd model at low energies , consistent with the steep increase of the diffusion coefficient . the antiproton spectrum ( fig . [ pbars ] ) appears to be very sensitive to the model assumptions and might help to discriminate between the models . the dr model produces too few antiprotons , a well known effect @xcite . the pd and drd models are both consistent with antiproton measurements in the heliosphere , but predict very different spectra in the interstellar medium ; the interstellar antiproton flux at low energies ( @xmath260600 mev ) in drd model is an order of magnitude lower than in two other models . in both reacceleration models ( dr and drd ) , the majority of low - energy antiprotons come from inelastic scattering ( so - called `` tertiary '' antiprotons ) . figs . [ pos ] and [ ele ] show secondary positrons and primary plus secondary electrons as calculated in all three models . the spectra are similar in the pd and dr models , while drd spectra exhibit lower intensities at low energies . this may be an observable effect since the models predict different synchrotron emission spectra ( electrons ) . damping on cosmic rays may terminate the slow kraichnan - type cascade in the interstellar medium at @xmath261 @xmath8 . our estimates were made for the level of mhd turbulence which produces the empirical value of cosmic - ray diffusion coefficient . this finding suggests a possible explanation for the peaks in secondary / primary nuclei ratios at about 1 gev / n observed in cosmic rays : the amplitude of short waves is small because of damping and thus the low energy particles rapidly exit the galaxy without producing many secondaries . there is no other obvious reasons for a sharp cut off in the wave spectrum . if the concept of mhd turbulence by @xcite works for interstellar turbulence , the mhd waves we are dealing with in this context are the fast magnetosonic waves . the alfven waves propagate predominantly perpendicular to the magnetic field and because of this they do not significantly scatter cosmic rays . it also explains why radio scintillation observations show no sign of the termination of electron density fluctuations at wave numbers between @xmath262 to @xmath263 @xmath8 . according to @xcite these fluctuations are produced by the slow magnetosonic waves with @xmath264 which are almost not damped on cosmic rays . an alternative explanation is that the wave damping on cosmic rays and the radio scintillations mainly occur in separate regions of the interstellar medium ( see below in the discussion on the `` sandwich '' model of cosmic ray propagation in the galaxy ) . some minor contribution to the observed scintillations is possible from fast magnetosonic waves interacting with energetic particles , and in this respect it is of interest that the observations may need an enhancement in the power on large `` refractive '' scales @xmath265 cm relative to the power on small `` diffractive '' scales @xmath266 cm @xcite . this may indicate the cutoff of the spectrum of fast magnetosonic waves due to cosmic ray action . while the mere fact of a wave spectrum steepening under the action of damping on cosmic rays can be described by a simple eq . ( [ eq9 ] ) with some characteristic time for nonlinear wave interactions @xmath267 $ ] , the exact form of the function @xmath91 at large @xmath112 where the damping is significant depends critically on the form of the equation for waves . it involves in particular to the vanishing of @xmath91 at some @xmath268 ( @xmath26@xmath269 @xmath8 ) and the corresponding singularity of @xmath165 at some @xmath217 ( @xmath261 gv ) found in our calculations . less significant in this sense is our approximation of the resonant wave number in eq . ( [ eq5 ] ) , which does not include the particle pitch angle in an explicit form . ( if it were included , the term with @xmath162 in eq . [ [ eq9 ] ] would have a third integration over the pitch angle . ) we note however that this effect was included in the derivation of eq . [ [ eq10 ] ] for the attenuation rate . in the context of the approximations adopted in the present work , the problem of cosmic - ray transport at @xmath160 arises . the free streaming of cosmic rays from the galaxy leads to an instability and to the growth of waves which scatter particles and thus slow down the streaming , see e.g. @xcite . we shall consider the processes at low energies in a separate work . here we note only that the streaming instability develops above the galactic disk at @xmath270@xmath271 cm@xmath4/s ( if the wave damping is absent ) and leads to the diffusive - convective transport of cosmic rays . the given estimate of the diffusion coefficient at 1 gv follows from the condition of cosmic ray streaming instability @xmath272 where @xmath273 is the bulk velocity of cosmic ray gas . the bulk velocity is @xmath274 where the cosmic ray anisotropy perpendicular to galactic disk is @xmath275 . the alfven velocity in galactic halo is about @xmath276 cm / s . it is important that the magnetic rigidity @xmath6 gv corresponds to a kinetic energy @xmath277 gev for protons , and @xmath278 gev / n for nuclei with charge to mass ratio @xmath279 . the galactic spectrum of cosmic - ray protons and nuclei at such low energies can not be derived from direct observations at the earth because of strong modulation in the solar wind . dissipation other than on cosmic rays has been neglected in the present work , though it may completely destroy the mhd cascade or considerably change its angular distribution and thus affect @xmath165 in a large part of the interstellar medium , see @xcite , @xcite , @xcite . the region of the cosmic ray halo is the most `` safe '' in this sense @xcite . in particular , dissipation on ion - neutral collisions may destroy mhd turbulence in the galactic disk but not in the halo where neutrals are absent . we then come to the `` sandwich '' model for cosmic - ray propagation , with different diffusion coefficients in the galactic disk , @xmath280 , and in the halo , @xmath281 ( it is assumed here that some scattering is present in the galactic disk and the diffusion approximation works ) . in this case , the mean matter thickness traversed by cosmic rays does not depend on diffusion inside the disk even at zero gas density in the halo : @xmath282 @xcite . the energy dependence of secondary to primary ratios in cosmic rays is determined by the diffusion coefficient in the cosmic ray halo where the model developed in the present paper is applied . the estimate based on the empirical value of the diffusion coefficient for gev particles ( see section 1 ) gives the level of turbulence at the principal scale @xmath283 @xmath284 for a kraichnan - type spectrum @xmath115 , and @xmath285 for a kolmogorov - type spectrum @xmath114 , if @xmath286 @xmath8 . at the same time , the data on faraday rotation angles favor the kolmogorov spectrum with @xmath285 and @xmath286 @xmath8 . the cascades of alfven waves ( with the scaling @xmath287 ) and the fast magnetosonic waves ( @xmath288 ) are independent in the @xcite model of mhd turbulence , and the amplitude of alfven wave cascade may dominate at the principle scale . also , in the `` sandwich '' model described in a previous paragraph , the turbulence , which determines the confinement of cosmic rays in the galaxy , is distributed in the halo of size @xmath289 kpc whereas the observations of interstellar turbulence refer to the galactic disk and the adjacent region , where it can be much stronger . on the whole , the empirical diffusion model for cosmic rays with energies from @xmath122 to @xmath290 ev implies the presence of random magnetic field with an extended power law spectrum of fluctuations @xmath291 , @xmath292 at wave numbers from @xmath293 to @xmath294 @xmath8 . the existence of such a turbulence spectrum in the interstellar medium seems confirmed by various astronomical observations . it should be emphasized that this does not prove the existence of a spectral cascade or spectral transfer throughout this enormous wavenumber range , however tempting that conclusion may be . among other puzzles , an oddity is the absence of a spectral feature on spatial scales where ion - neutral collisional processes should be most pronounced . the two special cases of turbulence spectrum with @xmath295 and @xmath296 which correspond to the kolmogorov and the kraichnan spectra respectively are used in popular versions of the diffusion model as described in section 5 . in both cases , as the present work has demonstrated , one can get a satisfactory fit to the data on energy spectra of secondary and primary nuclei if account is taken for wave damping on cosmic rays for a slow kraichnan cascade . the fit can be obtained either in the dr model with reacceleration on the kolmogorov spectrum with no significant effect of cosmic ray damping ( because the kolmogorov cascade is fast ) or in the drd model with relatively weak reacceleration on the kraichnan spectrum which is significantly modified by cosmic ray damping . some problems still remain to be solved . the main difficulties with the kolmogorov spectrum are , firstly , the contradiction with the leading theory of mhd turbulence where this spectrum is associated with perpendicular propagating alfven waves which almost do not scatter cosmic - ray particles ; and secondly , the low flux of antiprotons characteristic of models with relatively strong reacceleration . a major problem of concern for diffusion on a kraichnan spectrum is the relatively strong dependence of diffusion on energy which leads to an unacceptably large anisotropy of cosmic rays especially above @xmath3 ev , see @xcite , @xcite . let us emphasize again that the models of cosmic ray propagation discussed in the present paper assume that the mhd turbulence required for cosmic - ray scattering is produced by some external sources . an alternative is the galactic wind model by @xcite and @xcite . in this model , the cosmic - ray pressure drives a wind with a frozen - in regular magnetic field which is shaped into huge spirals ( the radius of the galactic wind cavity is about @xmath297 kpc ) . the mhd turbulence in the wind is created by the streaming instability of cosmic rays moving predominantly along the regular magnetic field lines outward from the galactic disk . the level of turbulence is regulated by the nonlinear landau damping on thermal ions . the model explains well the cosmic - ray data up to ultra - high energies @xmath26@xmath290 ev with the exception of the observed low anisotropy ( about the same difficulty with anisotropy as occurs in the diffusion model with a given kraichnan spectrum ) . as regards interpretation of cosmic - ray observations , a preference can not yet be given to any of the models of cosmic - ray transport in the galaxy discussed above . v. s. p. and f. c. j. acknowledge partial support from nasa astrophysics theory program ( atp ) grants . v. s. p. and v. n. z. were supported by the rfbr grant at izmiran . i. v. m. acknowledges partial support from nasa astrophysics theory program ( atp ) grant and nasa astronomy and physics research and analysis program ( apra ) grant . alcaraz , j. , et al . 2000a , phys . b , 484 , 10 alcaraz , j. , et al . 2000b , phys . b , 490 , 27 armstrong , j. w. , cordes , j. m. , & rickett , b. j. 1981 , , 291 , 561 armstrong , j. w. , rickett , b. j. , & spangler , s. r. 1995 , , 443 , 209 asakimori , k. , et al . 1998 , , 502 , 278 asaoka , y. , et al . 2002 , , 88 , 051101 barashenkov , v. s. 1993 , cross - sections of interactions of particles and nuclei with nuclei ( dubna : joint institute for nuclear research ) barashenkov v. s. , & polanski a. 1994 , electronic guide for nuclear cross - sections , comm . jinr e2 - 94 - 417 ( dubna : joint institute for nuclear research ) basini , g. , et al . 1999 , in proc . 26th int . cosmic ray conf . ( salt lake city ) , 3 , 77 berezinskii , v. s. , bulanov , s. v. , dogiel , v. a. , ginzburg , v. l. , & ptuskin , v. s. 1990 , astrophysics of cosmic rays ( amsterdam : north holland ) boezio , m. , et al . 1999 , , 518 , 457 boezio , m. , et al . 2000 , , 532 , 653 boezio , m. , et al . 2001 , , 561 , 787 case , g. l. , & bhattacharya , d. 1998 , apj , 504 , 761 casse , f. , lemoine , m. , & pelletier , g. , 2001 , , 65 , # 023002 cesarsky , c. j. 1980 , , 18 , 289 cho , j. , & vishniac , e. t. 2000 , , 539 , 273 cho , j. , & lazarian , a. 2002 , , 88 , # 245001 chandrasekhar , s. 1948 , phys . , 75 , 896 chashei , i. v. , & shishov , v. i. 1985 , geomagnetism and aeronomy , 25 , 1 chuvilgin , l. g. , & ptuskin , v. s. 1993 , , 279 , 278 cranmer , s. r. , & van ballegooijen , a. a. 2003 , apj , 594 , 573 davis , a. j. , et al . 2000 , in aip conf . 528 , acceleration and transport of energetic particles observed in the heliosphere ( ace-2000 ) , eds . r. a. mewaldt et al . ( new york : aip ) , 421 davis , a. j. , et al . 2001 , in proc . . cosmic ray conf . ( hamburg ) , 3971 diehl , r. , parizot , e. , kallenbach , r. , & pvon steiger , r. , eds . 2001 , the astrophysics of galactic cosmic rays ( dordrecht : kluwer acad . publ . ) ; also : , 99 , nos . 1 - 4 ( 2001 ) duvernois , m. a. , simpson , j. a. , & thayer , m. r. 1996 , , 316 , 555 duvernois , m. a. , et al . 2001 , , 559 , 296 elmergreen , b. g. , & scalo , j. 2004 , , 42 , 211 engelmann , j. j. et al . 1990 , , 233 , 96 galtier , s. , nazarenko , s.v . , newell , a. c. , & pouquet , a. 2000 , j. plasma phys . , 63 , 447 giacolone , j. , & jokipii , j. r. 1999 , , 520 , 204 ginzburg , v. l. , & ptuskin , v. s. 1976 , , 48 , 161 gleeson , l. j. , & axford , w. i. 1968 , , 154 , 1011 goldreich , p. , & sridhar , s. 1995 , , 438 , 763 goldreich , p. , & sridhar , s. 1997 , , 485 , 680 grimani , c. , et al . 2002 , , 392 , 287 heisenberg , w. 1948 , proc . r. soc . london a , 195 , 402 iroshnikov , p. 1963 , sov . , 7 , 566 ivanenko , i. p. , et al . 1993 , in proc . 23rd int . cosmic ray conf . ( calgary ) , 2 , 17 jones , f. c. 1979 , , 229 , 747 jones , f. c. , lukasiak , a. , ptuskin , v. , & webber , w. 2001 , , 547 , 264 kadomtsev , b. b. , & petviashvili , v. i. 1973 , dokl . akad . nauk sssr , 20 , 794 kennel , c. f. , & engelmann , f. 1966 , phys . fluids , 9 , 2377 kobayashi , t. , nishimura , j. , komori , y. , shirai , t. , tateyama , n. , taira , t. , yoshida , k. , & yuda , t. 1999 , in proc . . cosmic ray conf . ( salt lake city ) , 3 , 61 kolmogorov , a. n. 1941 , dokl . nauk sssr , 30 , 299 kraichnan , r. 1965 , phys . fluids , 8 , 1385 kuznetsov , e. a. 2001 , jetp , 93 , 1052 lambert , h. c. , & rickett , b. j. 2000 , , 531 , 883 lee , m. a. , & jokipii , j. r. 1976 , , 206 , 735 lithwick , y. , & goldreich , p. 2001 , , 562 , 279 lithwick , y. , & goldreich , p. 2003 , , 582 , 1220l livshits , m. a. , & tsytovich , v. n. 1970 , nucl . fusion , 10 , 241 lorimer , d. r. 2004 , in young neutron stars and their environments , iau symp . 218 , f. camilo and b. m. gaensler ( eds . ) ( astro - ph/0308501 ) lukasiak , a. , mcdonald , f. b. , & webber , w. r. 1999 , in proc . 26th int . cosmic ray conf . ( salt lake city ) , 3 , 41 mashnik , s. g. , sierk , a. j. , van riper , k. a. , & wilson , w. b. 1998 , in proc . 4th workshop on simulating accelerator radiation environments , ed . gabriel , t. a. ( oak ridge : ornl ) , p.151 ( nucl - th/9812071 ) mashnik , s. g. , gudima , k. k. , moskalenko , i. v. , prael , r. e. , & sierk , a. j. 2004 , adv . space res . , 34 , 1288 mcivor , i. 1977 , , 179 , 13 menn , w. et al . 2000 , , 533 , 281 moskalenko , i. v. , & mashnik , s. g. 2003 , in proc . 28th int . cosmic ray conf . ( tsukuba ) , p.1969 moskalenko , i. v. , & strong , a. w. 1998 , , 493 , 694 moskalenko , i. v. , mashnik , s. g. , & strong , a. w. 2001 , in proc . 27th int . cosmic ray conf . ( hamburg ) , p.1836 moskalenko , i. v. , strong , a. w. , ormes , j. f. , & potgieter , m. s. 2002 , , 565 , 280 moskalenko , i. v. , jones , f. c. , mashnik , s. g. , ptuskin , v. s. , & strong , a. w. 2003a , in proc . 28th int . cosmic ray conf . ( tsukuba ) , 1925 moskalenko , i. v. , strong , a. w. , mashnik , s. g. , & ormes , j. f. 2003b , , 586 , 1050 ng , c. s. , & bhattacharjee , a. 1997 , phys . plasmas , 4 , 605 norman , c. a. , & ferrara , a. 1996 , , 467 , 280 obukhov , a. m. 1941 , dokl . nauk sssr , 32 , 22 orito , s. , et al . 2000 , , 84 , 1078 ptuskin , v. s. 1981 , , 76 , 265 ptuskin v. s. , 1989 , in `` problems of theoretical physics and astrophysics '' , eds . l. v. keldysh and v. ya . fainberg ( moscow : nauka ) , p.463 ptuskin , v. s. , jones , f. c. , moskalenko , i. v. , & zirakashvili , v. n. 2003a , in proc . 28th int . cosmic ray conf . ( tsukuba ) , 1929 ptuskin , v. s. , jones , f. c. , seo , e. s. , & sina , r. 2003b , in proc . 28th int . cosmic ray conf . ( tsukuba ) , 1933 ptuskin , v. s. , vlk , h. j. , zirakashvili , v. n. , & breitschwerdt , d. 1997 , , 321 , 434 sanuki , t. , et al . 2000 , , 545 , 1135 saur , j. , & bieber , j. w. 1999 , jgr , 104 , 9975 scalo , j. , & elmergreen , b. g. 2004 , , 42 , 275 seo , e. s. , & ptuskin , v. s. 1994 , , 431 , 705 seo , e. s. , ormes , j. f. , streitmatter , r. e. , stochaj , s. j. , jones , w. v. , stephens , s. a. , & bowen , t. 1991 , , 378 , 763 shebalin , j. v. , matthaeus , w. h. , & montgomery , d. 1983 , j. plasma phys , 29 , 525 silberberg , r. , tsao , c. h. , & barghouty , a. f. 1998 , , 501 , 911 simon , m. , heinrich , w. , & mattics , k. d. 1986 , , 300 , 32 stephens , s. a. , & streitmatter , r. a. 1998 , , 505 , 266 strong , a. w. , & mattox , j. r. 1996 , , 308 , l21 strong , a. w. , & moskalenko , i. v. 1998 , , 509 , 212 strong , a. w. , & moskalenko , i. v. 2001 , adv . space res . , 27 , 717 strong a.w . , moskalenko i.v . , reimer o. , digel s. , diehl r. 2004 , , 422 , l47 tidman , d. a. 1966 , , 144 , 615 toptygin , i. n. 1985 , cosmic rays in interplanetary magnetic fields ( dordrecht : d. reidel publ . co. ) tu , c - y . 1988 , jgr , 89 , 9695 verma , m. k. 2004 , , 401 , 229 verma , m. k. , roberts , d. a. , goldstein , m. l. , ghosh , s. , & stribling , w. t. 1996 , jgr , 101 , 21619 webber , w. r. , kish , j. c. , & schrier , d. a. 1990 , , 41 , 566 woo , r. , & armstrong , j.w . 1979 , jgr , 84 , 7288 yan , h. , & lazarian , a. 2004 , , 614 , 757 zakharov , v. e. , & sagdeev , r. z. 1970 , dokl . nauk sssr , 192 , 297 zhou , y. , matthaeus , w. h. , & dmitruk , p. 2004 , , 76 , 1015 zirakashvili , v. n. , breitschwerdt , d. , ptuskin , v. s. , & vlk , h. j. 1996 , , 311 , 113
we describe two implementations of the optimal error correction algorithm known as the maximum likelihood decoder ( mld ) for the 2d surface code with a noiseless syndrome extraction . first , we show how to implement mld exactly in time , where is the number of code qubits . our implementation uses a reduction from mld to simulation of matchgate quantum circuits . this reduction however requires a special noise model with independent bit - flip and phase - flip errors . secondly , we show how to implement mld approximately for more general noise models using matrix product states ( mps ) . our implementation has running time where is a parameter that controls the approximation precision . the key step of our algorithm , borrowed from the dmrg method , is a subroutine for contracting a tensor network on the two - dimensional grid . the subroutine uses mps with a bond dimension to approximate the sequence of tensors arising in the course of contraction . we benchmark the mps - based decoder against the standard minimum weight matching decoder observing a significant reduction of the logical error probability for .
1405.4883
the surface code @xcite is one of the simplest and most studied quantum error correcting codes . it can be realized on a two - dimensional grid of qubits such that the codespace is defined by simple four - qubit parity check operators acting on nearest - neighbor qubits . recent years have witnessed a surge of interest in the surface code as a promising architecture for a scalable quantum computing @xcite . experimental advances in manufacturing of multi - qubit devices @xcite give us hope that a small - scale quantum memory based on the surface code may become a reality soon . given high operational costs of a quantum hardware compared with the classical one , it is crucial to put enough efforts in optimizing algorithmic , or software aspects of error correction . in the present paper we focus on optimizing the decoding algorithm that takes as input measured syndromes of the parity checks and computes a recovery operation returning a corrupted state of the memory back to the codespace . as the name suggests , the maximum likelihood decoder ( mld ) is an algorithm that finds a recovery operation maximizing the probability of a successful error correction conditioned on the observed error syndrome . by definition , mld is the optimal error correction algorithm for a fixed quantum code and a fixed noise model . the first rigorous definition of mld for the surface codes was proposed by dennis et al @xcite . an important observation made in @xcite was that the computational problem associated with mld can be reduced to computing the partition function of a classical ising - like hamiltonian on the two - dimensional lattice . this observation has generated a vast body of work exploring connections between mld and the statistical physics of disordered ising - like hamiltonians , see for instance @xcite . the insights made in @xcite have also guided the search for efficient implementations of mld . although an exact and efficient algorithm for mld remains an elusive goal , several approximate polynomial - time algorithms have been discovered , most notably the renormalization group decoder due to duclos - cianci and poulin @xcite , and the markov chain monte carlo method due to hutter , wootton , and loss @xcite . in the case of concatenated codes an efficient exact algorithm for mld based on the message passing algorithm was proposed by poulin @xcite . by comparing mld with the level - by - level decoder commonly used for concatenated codes , ref . @xcite found that mld offers a significant advantage with almost two - fold increase of the error threshold for the depolarizing noise and a significant reduction of the logical error probability . here we propose an alternative method of implementing mld in the case of the surface code for two simple noise models known as the bit - flip noise and the depolarizing noise . our method combines the ideas of dennis et al @xcite and the standard classical - to - quantum mapping from classical 2d spin systems in the thermal equilibrium to quantum 1d spin chains . it enables us to reduce the computational problem associated with mld to simulating a particular type of quantum dynamics for a chain of qubits . in the case of the bit - flip noise , mld can be reduced to simulating a quantum circuit with a special type of two - qubit nearest - neighbor gates known as matchgates . it was shown by valiant @xcite that quantum circuits composed of matchgates can be efficiently simulated by classical means . matchgate circuits and their generalizations give rise to efficient holographic algorithms for certain combinatorial problems @xcite and efficient tensor network contraction methods @xcite . matchgate - based algorithms have been used to simulate quantum dynamics in systems of fermionic modes with quadratic interactions @xcite and study statistics of dimer coverings in classical lattice models @xcite . here we demonstrate that matchgates also have applications for quantum error correction . our simulation algorithm based on fermionic gaussian states @xcite provides an exact implementation of mld with the running time @xmath0 , where @xmath1 is the number of code qubits . the same algorithm can also be applied to a noise model with independent bit - flip and phase - flip errors . we note that a similar but technically different algorithm has been used by merz and chalker in the numerical study of the random - bond 2d ising model @xcite . in the case of the depolarizing noise , mld can be reduced to simulating the dynamics generated by matrix product operators with a small bond dimension . to perform the simulation efficiently we conjecture that all intermediate states generated by this dynamics are weakly entangled . this enables us to employ a vast body of efficient classical algorithms for simulating weakly entangled quantum spin chains based on matrix product states ( mps ) , see @xcite . our approximate implementation of mld for the depolarizing noise has running time @xmath2 where @xmath3 is a parameter that controls the approximation precision ( the bond dimension of the mps ) . although we do not have any rigorous arguments in support of the weak entanglement conjecture , it reflects the physical intuition that the classical 2d spin system associated with mld has a finite correlation length for error rates below the threshold @xcite . accordingly , one should expect that the classical - to - quantum mapping can not generate highly entangled states since the latter require long - range correlations . furthermore , we have justified the conjecture numerically by applying the mps - based decoder to the bit - flip noise . we observed that the logical error probabilities of the exact mld and the mps - based decoder with a relatively small bond dimension @xmath5 are virtually indistinguishable . likewise , in the case of the depolarizing noise we observed that the logical error probability exhibits a fast convergence as a function of @xmath3 suggesting that the mps - based decoder with @xmath5 implements nearly exact mld . finally , we benchmark the exact and the approximate implementations of mld against the commonly studied minimum weight matching ( mwm ) decoder @xcite . the benchmarking was performed for a fixed code distance @xmath6 and a wide range of error rates . in the case of the bit - flip noise we observed that the mwm decoder approximates the logical error probability of mld within a factor of two . the observed difference between mld and the mwm decoder can be attributed to the fact that the latter ignores the error degeneracy @xcite . since the observed difference is relatively small , we conclude that ignoring the error degeneracy does not have a significant impact on the decoder s performance for the studied noise model . in the case of the depolarizing noise we observed that the mps - based decoder is far superior than the mwm decoder offering more than two orders of magnitude reduction of the logical error probability even for small values of @xmath3 . this can be attributed to the fact that the mwm decoder often fails to find the minimum weight error consistent with the syndrome since it ignores correlations between @xmath7 and @xmath8 errors @xcite . the rest of the paper is organized as follows . we formally define the maximum likelihood decoder , the studied noise models , and the surface code in sections [ sect : mld],[sect : noise ] , and [ sect : scodes ] respectively . our exact implementation of mld for the bit - flip noise is described in section [ sect : xnoise ] . the approximate implementation of mld based on matrix product states is presented in section [ sect : xyznoise ] . a comparison between the exact mld , the approximate mld with various bond dimensions @xmath3 , and the minimum weight matching decoder is presented in section [ sect : numerics ] that describes our numerical results . in this section we formally define mld . we consider a quantum memory composed of @xmath1 physical qubits . let @xmath9 be the full @xmath1-qubit hilbert space and @xmath10 be the group of @xmath1-qubit pauli operators . by definition , any element of @xmath10 has a form @xmath11 , where @xmath12 are single - qubit pauli operators and @xmath13 is an overall phase factor . a quantum code of stabilizer type is defined by an abelian _ stabilizer group _ @xmath14 such that @xmath15 . quantum codewords are @xmath1-qubit states invariant under the action of any element of @xmath16 . such states define a _ codespace _ @xmath17 the encoding step amounts to initializing the memory in some ( unknown ) state @xmath18 supported on the codespace @xmath19 . we shall consider a stochastic pauli noise described by a linear map @xmath20 where @xmath21 is some normalized probability distribution on the pauli group . since the initial state @xmath18 is supported on the codespace @xmath19 , one has @xmath22 for any @xmath23 . by the same token , @xmath24 whenever @xmath25 . given a pauli operator @xmath26 , a subset @xmath27 is called a _ coset _ of @xmath16 . clearly , @xmath10 is a disjoint union of cosets @xmath28 , where @xmath29 is some fixed representative of @xmath30 . the above shows that errors in the same coset have the same action on the codespace . thus @xmath31 where the sum ranges over all cosets of @xmath16 and @xmath32 for simplicity , here we assumed that all coset representatives @xmath29 are hermitian operators . we shall refer to the quantity @xmath33 as a _ coset probability_. at the decoding step one attempts to guess the coset of the stabilizer group that contains the actual error based on a partial information about the error known as a syndrome . more precisely , let @xmath34 be some fixed set of generators of @xmath16 . since the generators @xmath35 pairwise commute , they can be diagonalized simultaneously . a configuration of eigenvalues @xmath36 can be described by a _ syndrome _ @xmath37 such that @xmath38 for all @xmath39 . assuming that the generators @xmath35 are independent , there are @xmath40 possible syndromes . the full hilbert space can be decomposed into a direct sum of syndrome subspaces @xmath41 where @xmath42 . note that the codespace @xmath19 corresponds to the zero syndrome . a pauli operator @xmath26 is said to have a syndrome @xmath43 iff @xmath44 for all @xmath39 . equivalently , @xmath45 has a syndrome @xmath43 iff @xmath46 . for each syndrome @xmath43 let us choose some fixed pauli operator @xmath47 with the syndrome @xmath43 . one can easily check that the set of all pauli operators with a syndrome @xmath43 coincides with the coset @xmath48 , where @xmath49 is a group known as the centralizer of @xmath16 . note that @xmath50 . thus each coset of @xmath51 can be partitioned into a disjoint union of several cosets of @xmath16 . in the present paper we only consider stabilizer codes with a single logical qubit . let @xmath52 be the logical pauli operators on the encoded qubit . then each coset of @xmath51 consists of four disjoint cosets of @xmath16 , namely , @xmath53 where @xmath54 @xmath55 the decoding step starts by a _ syndrome measurement _ that projects the corrupted state @xmath56 onto one of the syndrome subspaces @xmath57 . the above arguments show that @xmath58 has support on @xmath57 iff @xmath59 . thus the syndrome measurement reveals the coset of @xmath51 that contains the error @xmath45 , whereas our goal is to determine which coset of @xmath16 contains @xmath45 . using eqs . ( [ eq : final1],[eq : cg1]-[eq : cg3 ] ) one can write the post - measurement ( unnormalized ) state as @xmath60 where @xmath43 is the observed syndrome . here we assumed for simplicity that @xmath47 and the logical operators @xmath61 are hermitian . this shows that the effective noise model conditioned on the syndrome can be described by applying one of the four pauli errors @xmath62 , and @xmath63 with probabilities @xmath64 , and @xmath65 respectively . clearly , the best possible error correction algorithm for this effective noise model is to choose a recovery operator as the most likely of the four errors . equivalently , we should choose a recovery operator as any pauli operator that belongs to the most likely of the four cosets @xmath66 which we denote @xmath67 . these steps can be summarized as follows . the final step of the decoding is to apply the optimal recovery operator @xmath68 . it results in a state @xmath69 . we conclude that mld correctly identifies the coset of @xmath16 that contains the actual error and maps the corrupted state @xmath56 back to the encoded state @xmath18 with a probability @xmath70 in what follows we shall always ignore overall phase factors of pauli operators . such phase factors are irrelevant for our purposes since they do not change the outcome of error correction . we shall consider a stochastic i.i.d . pauli noise @xmath71 where @xmath72 and @xmath73 is called an _ error rate_. two commonly studied noise models are the classical bit - flip noise where only @xmath7-type errors are allowed ( the x - noise ) and the depolarizing noise where all types of errors are equally likely . the formal definitions are given below . @xmath74 the corresponding probability distributions on the pauli group are @xmath75 for the depolarizing noise and @xmath76 for the x - noise . here @xmath77 denotes the hamming weight of @xmath45 , that is , the number of qubits on which @xmath45 acts nontrivialy , while @xmath78 denotes the subgroup generated by single - qubit pauli @xmath7 operators . one may also consider a noise model with independent bit - flip and phase - flip errors , that is , @xmath79 and @xmath80 . since there are no correlations between the two types of errors , one can perform error correction independently for bit - flip and phase - flip errors . furthermore , since correcting phase - flip errors is equivalent to correcting bit - flip errors on the surface code lattice rotated by @xmath81 , it suffices to consider the @xmath7-noise model only . we consider the surface code on a square lattice of size @xmath82 with open boundary conditions . the boundaries parallel to the horizontal ( vertical ) axis are smooth ( rough ) . the surface code lattice with @xmath83 is shown on fig . [ fig : d3 ] . for the chosen geometry the surface code encodes one logical qubit into @xmath84 physical qubits with the minimum distance @xmath85 . we shall always consider odd values of @xmath85 such that the code corrects any combination of @xmath86 single qubit errors . let @xmath87 and @xmath88 be the stabilizers of the surface code associated with a site @xmath89 and a plaquette @xmath90 respectively . we have @xmath91 , where the product runs over all edges @xmath92 making up the boundary of @xmath90 . likewise , @xmath93 , where the product runs over all edges @xmath92 incident to @xmath89 . let @xmath94 and @xmath95 be the subgroups of the pauli group @xmath10 generated by all site stabilizers and all plaquette stabilizers respectively . finally , let @xmath96 be the full stabilizer group . logical pauli operators @xmath97 are shown on fig . [ fig : logical ] , while @xmath98 . surface code . solid dots , stars , and diamonds indicate locations of qubits , site stabilizers , and plaquette stabilizers respectively . stabilizers located near the boundary act only on three qubits . the distance-@xmath85 surface code has @xmath99 qubits on horizontal edges , @xmath100 qubits on vertical edges , and @xmath101 stabilizers of each type . [ fig : d3 ] , height=226 ] ( left ) and @xmath102 ( right ) . [ fig : logical ] , width=264 ] by a slight abuse of notations , below we shall often identify a pauli operator @xmath45 with the subset of edges in the lattice on which @xmath45 acts non - trivially . in this section we consider the x - noise and describe an exact implementation of mld . we begin by specializing mld to the x - noise ( section [ sect : mld4x ] ) and describing our algorithm ( section [ sect : algorithm1 ] ) . a reader interested only in the question of _ how _ the algorithm works can skip the remaining sections explaining _ why _ it works and proving its correctness . specifically , section [ sect : reduction ] shows how to express the coset probability as a matrix element of a matchgate quantum circuit . our derivation partially follows the one of refs . . an efficient method of simulating matchgate circuits based on fermionic gaussian states is described in section [ sect : gaussian ] . the material of this section mostly follows ref . @xcite . let @xmath43 be the input syndrome and @xmath103 be some fixed pauli error consistent with @xmath43 . we can always choose @xmath104 , that is , such that @xmath47 acts on any qubit by @xmath105 or @xmath7 . indeed , since only @xmath7-type errors can appear with a non - zero probability , the syndromes of all plaquette stabilizers must be zero . let @xmath106 be the syndrome of a site stabilizer @xmath87 . we choose the desired error @xmath47 by connecting each site @xmath89 with a non - zero syndrome @xmath106 to the left boundary by a horizontal string of @xmath7 errors and adding all such strings modulo two . note that @xmath47 can be constructed in time @xmath107 . let @xmath21 be the probability distribution on the pauli group describing the x - noise , see section [ sect : noise ] . to implement the ml decoder it suffices to compute the four coset probabilities @xmath64 , and @xmath65 as defined in section [ sect : mld ] . note that @xmath108 since any element of these two cosets acts by pauli @xmath8 on at least @xmath85 qubits . choose any logical operator @xmath109 and let @xmath110 . from now on we shall assume that @xmath45 is fixed . since @xmath8-type errors are not allowed , one has @xmath111 . thus it suffices to compute the coset probability @xmath112 . in this section we describe an algorithm that takes as input an @xmath7-type pauli operator @xmath45 and outputs the coset probability @xmath112 . the algorithm has running time @xmath0 . let us begin by introducing some notations . the sets of all horizontal and vertical edges of the surface code lattice will be denoted @xmath113 and @xmath114 respectively . for the code of distance @xmath85 one has @xmath115 and @xmath116 . we partition the set @xmath113 into columns of edges such that @xmath117 where @xmath118 denotes the @xmath119-th leftmost column of horizontal edges , see fig . [ fig : layers ] . edges of every column @xmath118 will be labeled by integers @xmath120 starting from the top edge . likewise , @xmath121 where @xmath122 denotes the @xmath119-th leftmost column of vertical edges , see fig . [ fig : layers ] . edges of every column @xmath122 will be labeled by integers @xmath123 starting from the top edge . we shall refer to @xmath118 and @xmath122 as horizontal and vertical columns respectively . and vertical columns @xmath124 . every edge is identified with the respective code qubit ( solid dot ) . [ fig : layers ] , height=188 ] for each edge @xmath92 of the surface code lattice define a weight @xmath125 recall that @xmath126 is the error rate . for any integer @xmath127 and a vector @xmath128 let @xmath129 be the anti - symmetric matrix of size @xmath130 that contains @xmath131 above the main diagonal and @xmath132 below the main diagonal . for example , if @xmath133 then @xmath134.\ ] ] let @xmath135 be the diagonal matrix of size @xmath136 that contains @xmath131 on the main diagonal . define also a standard antisymmetric matrix @xmath137 } & & & \\ & & \ddots & & \\ & & & { \left [ { \begin{array}}{cc } 0 & 1\\ -1 & 0 \\ { \end{array}}\right ] } & \\ -1 & & & & 0 \\ { \end{array}}\right]\ ] ] such that @xmath138 has size @xmath139 . the matrix @xmath138 contains @xmath140 blocks of size @xmath141 on the main diagonal and two non - zero elements @xmath142 . all remaining elements of @xmath138 are zero . let @xmath105 be the identity matrix of size @xmath139 . the first step of our algorithm is to compute the probability of the input error @xmath143 and the coefficients @xmath144 defined in eq . ( [ eq : wj ] ) . this step takes time @xmath107 . at each subsequent step of the algorithm we maintain a pair @xmath145 , where @xmath146 is an antisymmetric real matrix of size @xmath139 and @xmath147 is a real number . the algorithm calls two functions simulatehorizontal@xmath148 and simulatevertical@xmath148 that update the pair @xmath145 by applying a simple combination of matrix inversions and matrix multiplications . if implemented naively , each matrix inversion and each matrix multiplication takes time @xmath149 . likewise , computing each determinant takes time @xmath149 . simple counting then shows that the overall running time of the algorithm is @xmath150 . suggestions on improving stability of the algorithm against rounding errors can be found in section [ sect : numerics ] . consider any stabilizer @xmath151 . a simple algebra shows that @xmath152 where @xmath144 are the weights defined in eq . ( [ eq : wj ] ) . thus @xmath153 where @xmath154 is the list of coefficients @xmath144 and @xmath155 since the factor @xmath156 is easy to compute , below we concentrate on computing @xmath157 . we shall express @xmath157 as a matrix element of a certain quantum circuit acting on @xmath85 qubits . the circuit will be composed of single - qubit and two - qubit gates @xmath158 \quad \mbox{and}\quad g'(w)\equiv \left [ { \begin{array}}{cccc } 1 & 0 & 0 & w \\ 0 & 1 & w & 0 \\ 0 & w & 1 & 0\\ w & 0 & 0 & 1\\ { \end{array}}\right]\ ] ] where @xmath159 is a real parameter . we note that @xmath160 and @xmath161 are not unitary gates . let @xmath162 be the hilbert space of @xmath85 qubits . for each horizontal column @xmath118 and each vertical column @xmath122 defined at fig . [ fig : layers ] define linear operators @xmath163 acting on @xmath164 such that @xmath165 and @xmath166 here the subscripts indicate the qubits acted upon by each gate and @xmath167 denotes the @xmath168-th edge of the respective columns @xmath118 and @xmath122 counting from the top to the bottom . finally , define a state @xmath169 is the set of all @xmath85-bit binary strings with the even hamming weight . [ lem : mgates ] one has @xmath170 where @xmath171 is a quantum circuit on @xmath85 qubits shown at fig . [ fig : mgates ] . -noise model is equivalent to computing the matrix element @xmath172 , where @xmath173 is a quantum circuit on @xmath85 qubits shown above and @xmath174 is the superposition of all even - weight @xmath85-bit strings . the above example is for @xmath83 . each gate depends on a parameter @xmath144 defined in eq . ( [ eq : wj ] ) . [ fig : mgates ] , height=226 ] the gates @xmath160 and @xmath161 defined in eq . ( [ eq : gates ] ) are examples of the so - called matchgates discovered by valiant @xcite . it was shown in @xcite that quantum circuits composed of matchgates can be efficiently simulated by classical means . in the next section we describe an alternative algorithm for computing the quantity @xmath175 based on fermionic gaussian states with a running time @xmath0 . ( for comparison , the original algorithm of ref . @xcite would have running time @xmath176 since it requires computing the pfaffian of a matrix of size @xmath107 . ) in the rest of this section we prove lemma [ lem : mgates ] . [ prop : trivial ] for any subset @xmath177 such that @xmath178 is even for all @xmath179 there exists a unique @xmath151 such that @xmath180 . recall that a subset of edges @xmath68 is called a cycle iff any site has even number of incident edges from @xmath68 . let us first show that for any @xmath177 there exists exactly one cycle @xmath68 such that @xmath180 . indeed , consider any vertical column @xmath122 . it comprises a set of sites @xmath181 and a set of edges @xmath182 ( listed in the order from the top to the bottom ) . since @xmath183 and @xmath184 , the cycle condition at @xmath185 uniquely determines @xmath186 . once @xmath186 is determined , the cycle condition @xmath187 uniquely determines @xmath188 . continuing in this fashion uniquely determines @xmath189 . since @xmath122 can be any vertical column , we conclude that @xmath68 is uniquely determined by @xmath190 . it remains to note that @xmath191 coincides with the set of cycles that have even intersection with any column @xmath118 . let @xmath192 be the pauli operator constructed in proposition [ prop : trivial ] . then @xmath193 where the sum ranges over all subsets @xmath190 such that @xmath178 is even for all @xmath119 . let @xmath194 . we can regard @xmath195 as a binary @xmath85-bit string such that @xmath196 iff the @xmath168-th edge of @xmath118 belongs to @xmath190 . let @xmath197 be the basis vector corresponding to @xmath195 . since @xmath198 , we have @xmath199 where @xmath200 are the edges comprising the column @xmath118 listed in the order from the top to the bottom and @xmath160 is the single - qubit gate defined in eq . ( [ eq : gates ] ) . consider now some vertical column @xmath122 . let @xmath182 be the edges comprising @xmath122 listed in the order from the top to the bottom . we claim that @xmath201 where @xmath161 is the two - qubit gate defined in eq . ( [ eq : gates ] ) and @xmath202 denotes the gate @xmath161 applied to the pair of qubits @xmath203 . one can easily check eq . ( [ eq : zloops4 ] ) by noting that @xmath204 and following the arguments given in proof of proposition [ prop : trivial ] to reconstruct @xmath205 from @xmath195 and @xmath206 . let @xmath207 be the set of all @xmath85-bit strings with even hamming weight . combining eqs . ( [ eq : zloops2],[eq : zloops3],[eq : zloops4 ] ) one arrives at @xmath208 where @xmath209 and @xmath210 are the linear operators on @xmath164 defined in eqs . ( [ eq : hath],[eq : hatv ] ) . let @xmath211 be the subspace spanned by vectors @xmath212 with @xmath213 . note that the operators @xmath209 and @xmath210 preserve @xmath214 since the gates @xmath160 and @xmath161 preserve the hamming weight modulo two . the above observations imply that @xmath215 , which completes the proof of lemma [ lem : mgates ] . let @xmath164 be the hilbert space of @xmath85 qubits . for each @xmath216 define a _ majorana operator _ @xmath217 acting on @xmath164 such that @xmath218 the majorana operators obey the well - known commutation rules @xmath219 we shall often use a formula @xmath220 a _ covariance matrix _ of a pure ( unnormalized ) state @xmath221 is a @xmath139 matrix @xmath146 with matrix elements @xmath222 from eq . ( [ eq : cr ] ) one can easily check that @xmath146 is a real anti - symmetric matrix . consider as an example the state @xmath174 defined in eq . ( [ eq : psi ] ) . let us compute its covariance matrix @xmath146 . one can easily check that @xmath174 is a stabilizer state with the stabilizer group @xmath223 applying eq . ( [ eq : useful ] ) one can get an alternative set of generators that are quadratic in majorana operators , @xmath224 this shows that @xmath225 for all @xmath226 and @xmath227 . furthermore , @xmath228 whenever @xmath229 anti - commutes with at least one of the generators defined in eq . ( [ eq : gens ] ) . combining the above observations one can easily check that @xmath230 , where @xmath138 is the standard anti - symmetric matrix defined in eq . ( [ eq : m0 ] ) . a state @xmath221 is said to obey the wick s theorem iff the expectation value of any even tuple of majorana operators on @xmath231 can be computed from its covariance matrix @xmath146 using the formula @xmath232 where @xmath233 is the norm of @xmath231 , @xmath234 is the @xmath235 submatrix of @xmath146 formed by the rows and columns @xmath236 , and @xmath237 is the pfaffian @xcite . recall that the pfaffian of an anti - symmetric matrix @xmath238 of size @xmath235 is defined as @xmath239 where @xmath240 stands for the anti - symmetrization over all @xmath241 permutations of indexes . for example , @xmath242 for @xmath243 and @xmath244 for @xmath245 . a state @xmath221 is called a ( fermionic ) _ gaussian state _ iff it obeys the wick s theorem and , in addition , all odd tuples of majorana operators have zero expectation value on @xmath231 . by definition , a gaussian state @xmath231 is fully specified by the pair @xmath145 , where @xmath146 is the covariance matrix of @xmath231 and @xmath246 is the norm . below we shall identify a gaussian state and the corresponding pair @xmath145 . we shall need the following well - known facts , see for instance ref . @xcite . a state @xmath231 is gaussian iff its covariance matrix obeys @xmath247 . one can easily check that standard anti - symmetric matrix @xmath138 defined in eq . ( [ eq : m0 ] ) satisfies @xmath248 . this shows that @xmath174 is a gaussian state with the covariance matrix @xmath138 and the norm @xmath249 . let @xmath250 and @xmath251 be gaussian states of @xmath85 qubits . then @xmath252 be a ( complex ) anti - symmetric matrix of size @xmath139 . consider an operator @xmath253 then @xmath254 maps gaussian states to gaussian states . the last fact will be very important for us since the operators @xmath209 and @xmath210 constructed in section [ sect : reduction ] have the form eq . ( [ eq : w ] ) . indeed , consider the gates @xmath255 and @xmath256 where @xmath257 are defined in eq . ( [ eq : gates ] ) and the subscripts indicate which qubits are acted upon by the gate . one can easily check that @xmath258 , where @xmath259 is defined through @xmath260 . likewise , @xmath261 . from eq . ( [ eq : useful ] ) and eq . ( [ eq : hath ] ) one gets @xmath262 where @xmath263 is defined through @xmath264 likewise , eq . ( [ eq : hatv ] ) implies @xmath265 since @xmath174 is a gaussian state , fact 3 implies that all intermediate states obtained from @xmath174 by applying the operators @xmath209 and @xmath210 are gaussian . therefore @xmath266 can be efficiently computed if we have a rule describing how the covariance matrix and the norm of a gaussian state change upon application of @xmath209 and @xmath210 . the desired rule can be obtained using a fermionic version of the jamiolkowski duality between states and linear maps introduced in @xcite . let us first define a fermionic version of the maximally entangled state for a bipartite system of @xmath267 qubits . let @xmath268 be the majorana operators defined for a system of @xmath269 qubits according to eq . ( [ eq : majorana ] ) . define a @xmath269-qubit state normalized @xmath270 such that @xmath270 has a stabilizer group @xmath271 one can easily check that @xmath270 has a covariance matrix @xmath272,\ ] ] where each block has dimensions @xmath139 . fact 1 implies that @xmath270 is a gaussian state . let @xmath273 be the operator defined in eq . ( [ eq : w ] ) . define a @xmath269-qubit state @xmath274 note that @xmath275 is a gaussian state due to fact 3 . let @xmath276\ ] ] be the covariance matrix of @xmath275 . here @xmath277 are some matrices of size @xmath278 . let @xmath279 . we shall need the following fact proved in @xcite . let @xmath250 be a gaussian state . then @xmath280 , where @xmath281 and @xmath282 it remains to compute @xmath283 for the two special cases @xmath284 and @xmath285 . we shall perform the calculation for @xmath284 since both cases are quite similar . first we note that @xmath254 is a product of operators acting on disjoint pairs of majorana modes @xmath286 . accordingly , @xmath275 is a product of states involving disjoint @xmath287-tuples of majorana modes @xmath288 . it suffices to compute the covariance matrix and the norm for each of those @xmath287-tuples . equivalently , it suffices to do the calculation for @xmath289 . in this case @xmath290 is the single - qubit operator defined in eq . ( [ eq : gates ] ) . by definition , @xmath270 is a two - qubit state with stabilizers @xmath291 and @xmath292 . it can be written explicitly as @xmath293 hence @xmath294 this state has norm @xmath295 to compute the covariance matrix @xmath296 we shall use a shorthand notation @xmath297 by definition , @xmath298 a straightforward calculation shows that the only non - zero elements ( with @xmath299 ) of @xmath296 are @xmath300 @xmath301 @xmath302 and @xmath303 thus @xmath304,\ ] ] where @xmath305 \quad \mbox{and } \quad b=\left [ { \begin{array}}{cc } s & 0 \\ 0 & s \\ { \end{array}}\right].\ ] ] for an arbitrary @xmath85 we just need to take a direct sum of @xmath85 matrices @xmath296 as above and take the product of @xmath85 normalizing coefficients @xmath306 defined above . this yields @xmath307 whereas @xmath296 is given by eq . ( [ eq : mw ] ) where @xmath308 and @xmath309 with @xmath310 combining the above analysis and fact 4 we infer that the function simulatehorizontal@xmath148 defined in section [ sect : algorithm1 ] describes how the covariance matrix and the norm of a gaussian state change under application of the operator @xmath209 . a similar calculation shows that the function simulatevertical@xmath148 describes how the covariance matrix and the norm of a gaussian state change under application of the operator @xmath210 . the very last step of algorithm 1 correspond to computing the overlap between @xmath174 and the final state @xmath311 using eq . ( [ eq : overlap ] ) . this completes the proof of correctness of algorithm 1 . in this section we describe an approximate algorithm for computing the coset probabilities . it is applicable to a general stochastic i.i.d . pauli noise including the depolarizing noise . we assume some level of familiarity with matrix product states and tensor networks , see @xcite or @xcite for a thorough review . for the sake of completeness we summarize some basic facts about matrix product states in section [ sect : truncate ] . let @xmath312 be one of the cosets @xmath313 defined in section [ sect : mld ] . our goal is to compute the coset probability @xmath33 . let @xmath314 be any probability distribution on the single - qubit pauli group . for example , @xmath315 for the depolarizing noise with a rate @xmath126 . by definition , @xmath316 where the product ranges over all edges of the surface code lattice . let us parameterize @xmath317 by binary variables @xmath318 associated with sites @xmath89 and plaquettes @xmath90 such that @xmath319 here we used a convention @xmath320 and @xmath321 . let @xmath92 be some edge of the surface code lattice with endpoints @xmath322 and adjacent plaquettes @xmath323 , see fig . [ fig : edge ] . let @xmath324 be the restriction of @xmath68 onto the qubit @xmath92 . clearly , @xmath324 depends only on the bits @xmath325 and @xmath326 . thus we can write @xmath327 where @xmath328 is a function of just four binary variables @xmath329 . for horizontal edges located at the left or the right boundary of the lattice the variable @xmath330 or @xmath331 respectively is missing . likewise , for horizontal edges located at the top or the bottom boundary the variable @xmath332 or @xmath333 respectively is missing . we arrive at @xmath334 where the sums range over binary strings @xmath335 corresponding to all possible configurations of variables @xmath336 and @xmath337 onto the edge @xmath92 depends only on the variables @xmath330 , @xmath331 and @xmath332 , @xmath333 . [ fig : edge ] , height=75 ] the righthand side of eq . ( [ eq : coset2 ] ) coincides with the contraction value of a properly defined tensor network on a two - dimensional grid . to define this tensor network , consider the extended surface code lattice shown on fig . [ fig : network ] . the extended lattice has three types of nodes which we call @xmath43-nodes , @xmath338-nodes , and @xmath339-nodes . each @xmath43-node represents a location of a stabilizer ( either a site stabilizer @xmath87 or plaquette stabilizer @xmath88 ) while @xmath338-nodes and @xmath339-nodes represent code qubits located on horizontal and vertical edges of the original surface code lattice respectively . we shall refer to edges of the extended lattice as _ links _ to distinguish them from edges of the original surface code lattice . . locations of stabilizers are represented by @xmath43-nodes . code qubits located on horizontal and vertical edges of the original lattice are represented by @xmath338-nodes and @xmath339-nodes respectively . in general , the extended lattice has dimensions @xmath340 . [ fig : network ] , height=151 ] consider any configuration of variables @xmath341 and the corresponding term @xmath342 in eq . ( [ eq : coset2 ] ) . for each site stabilizer @xmath87 let us copy the corresponding variable @xmath343 to all links incident to the @xmath43-node @xmath89 . likewise , for each plaquette stabilizer @xmath88 let us copy the corresponding variable @xmath344 to all links incident to the @xmath43-node @xmath90 . we obtain a labeling of the links by binary variables @xmath345 with the property that all links incident to any @xmath43-node have the same label . let us call such a link labeling _ valid_. by definition , @xmath342 is a product of terms @xmath346 associated with @xmath338-nodes and @xmath339-nodes @xmath92 of the extended lattice . since @xmath347 and @xmath259 are uniquely determined by the link labeling @xmath345 , we can also write @xmath348 as a function of @xmath349 , that is , @xmath350 . this shows that @xmath351 where the product is over all @xmath338-nodes and @xmath339-nodes and the sum ranges over all valid link labelings . we can now extend the sum in eq . ( [ eq : coset3 ] ) to _ all _ link labelings @xmath349 by adding extra terms @xmath352 associated with @xmath43-nodes @xmath92 such that @xmath353 iff all links incident to @xmath92 have the same label and @xmath354 otherwise . we arrive at @xmath355 now the product ranges over all nodes of the extended lattice and the sum ranges over all link labelings . furthermore , by construction , each term @xmath356 depends only on the labels of links incident to the node @xmath92 . the expression in the righthand side of eq . ( [ eq : coset4 ] ) is known as a contraction value of the tensor network defined by the collection of tensors @xmath356 . tensor networks are usually represented by diagrams like the one shown on fig . [ fig : network ] such that each box on the diagram carries a tensor with several indexes . indexes of a tensor are associated with the links emanating from the corresponding box . diagrams representing the tensors @xmath356 are shown on eqs . ( [ eq : snode],[eq : hnode],[eq : vnode ] ) . all tensor indexes @xmath357 on these diagrams take values @xmath358 . for tensors located at the boundary some of the indexes may be missing . note that the order of arguments of @xmath324 is interchanged in eqs . ( [ eq : hnode],[eq : vnode ] ) . this is simply because the qubits located on horizontal edges ( @xmath338-nodes ) have site stabilizers on the left and on the right whereas qubits located on vertical edges ( @xmath339-nodes ) have site stabilizers on the top and on the bottom . @xmath359{snode.pdf } } \quad = \left\ { { \begin{array}}{rcl } 1 & \mbox{if } & i = j = k = l\\ 0 & & \mbox{otherwise } \\ { \end{array}}\right.\ ] ] @xmath360{snode.pdf}}= \pi_1(f_e g_e(j , l;i , k))\ ] ] @xmath361{snode.pdf}}= \pi_1(f_e g_e(i , k ; j , l))\ ] ] let @xmath362 and @xmath363 be the set of matrix product states and matrix product operators defined on a chain of @xmath364 qubits and having the bond dimension @xmath3 . in this section we shall identify a matrix product state ( operator ) with the corresponding tensor network . consider a partition of the extended surface code lattice into columns shown on fig . [ fig : slices ] . each column @xmath122 and each internal column @xmath118 defines a matrix product operator @xmath365 and @xmath366 respectively . the first and the last columns @xmath367 define matrix product states @xmath368 . here we identify horizontal links of the lattice with physical indexes of mpo and mps , while vertical links correspond to virtual indexes . by definition , contracting a consecutive pair of columns is equivalent to taking the product of the corresponding mpos . thus eq . ( [ eq : coset4 ] ) can be rewritten as @xmath369 and ` vertical ' columns @xmath124 ( here @xmath83 ) . [ fig : slices ] , height=170 ] to approximate the righthand side of eq . ( [ eq : network2 ] ) we shall employ the algorithm proposed by murg , verstraete and cirac @xcite . the approximation accuracy of the algorithm is controlled by an integer parameter @xmath370 such that the algorithm becomes exact if @xmath3 is exponentially large in @xmath85 . at each step of the algorithm we maintain a state @xmath371 . such a state can be described by a list of @xmath364 tensors of dimension @xmath372 which requires @xmath373 real parameters . we begin by initializing @xmath374 . note that @xmath375 . each step of the algorithm updates @xmath231 according to @xmath376 ( even steps ) or @xmath377 ( odd steps ) . this update is realized simply by taking the product of tensors of @xmath231 with the respective tensors of @xmath209 or @xmath210 which takes time @xmath373 . since @xmath209 and @xmath210 map @xmath362 to @xmath378 , extra measures have to be taken to reduce the bond dimension after each update . to this end we apply the truncation algorithm described in section 4.5 of ref . we shall use a function truncate@xmath379 that takes as input a state @xmath380 and returns a state @xmath371 approximating @xmath381 . such an approximation is obtained by computing the schmidt decomposition of @xmath381 across each bipartite cut of the chain and retaining only the @xmath3 largest schmidt coefficients . a detailed implementation of the function truncate@xmath379 is described in the next section . the last step of the algorithm is to compute the inner product between the final state @xmath371 and @xmath382 . this can be done in time @xmath383 by applying the standard contraction method for mps . as we explain in the next section , each call to the function truncate@xmath379 involves @xmath364 qr - decompositions and svd - decompositions on matrices of size @xmath384 and @xmath385 respectively , which takes time @xmath383 . since we need one truncation for each column of the lattice , the overall running time of the algorithm is @xmath386 . the above steps can be summarized as follows . @xmath387 in this section we describe implementation of the function truncate@xmath379 in algorithm 2 . our implementation closely follows section 4.5 of ref . @xcite . for the sake of completeness , we begin by summarizing the necessary facts about matrix product states . below we use a notation @xmath388 for the number of qubits per column of the lattice . a matrix product state @xmath389 describing a chain of @xmath390 qubits is defined by a list of @xmath391 matrices @xmath392 , where @xmath393 is a qubit index ( site of the chain ) . any amplitude of @xmath231 in the standard basis is expressed as a product of @xmath390 matrices @xmath394 we shall use a shorthand notation @xmath395 for the pair of matrices @xmath396 at some particular qubit @xmath43 . likewise @xmath397 will stand for the full matrix product state . the @xmath390-qubit state defined in eq . ( [ eq : mps ] ) will be denoted @xmath398 . let @xmath399 and @xmath400 be the number of rows and columns respectively in @xmath401 ( we shall always assume @xmath402 and @xmath403 have the same dimensions ) . since we want the the product of matrices in eq . ( [ eq : mps ] ) to be a @xmath404 matrix ( a complex number ) , dimensions of the matrices must satisfy @xmath405 a matrix product state is said to have a _ bond dimension _ @xmath3 iff @xmath406 and @xmath407 for all qubits @xmath43 . let @xmath362 be the set of all matrix product states @xmath397 on @xmath390 qubits with the bond dimension @xmath3 . we shall say that @xmath395 has a _ left canonical form _ ( lcf ) or _ right canonical form _ ( rcf ) iff @xmath408 or @xmath409 respectively . here @xmath410 denotes the identity matrix of size @xmath411 . the importance of lcf and rcf comes from the following lemma . here and below we use a notation @xmath412 for the column vector @xmath413{^{\mathsf{t}}}$ ] with ` @xmath414 ' at the @xmath168-th coordinate . [ lem : lcf ] suppose @xmath395 has lcf for @xmath415 . for each @xmath416 define a state @xmath417 with amplitudes @xmath418 then @xmath419 form an orthonormal family of vectors , i.e. , @xmath420 for all @xmath421 . indeed , using the definition of @xmath422 the inner product @xmath423 can be written as @xmath424 where the sum runs over @xmath425 . the lcf at qubit @xmath414 implies @xmath426 . hence @xmath427 is equal to @xmath428 where the sum runs over @xmath429 . applying the same argument to the remaining qubits one arrives at @xmath430 . exactly the same arguments show that if @xmath395 has rcf for all @xmath431 then states @xmath432 with amplitudes @xmath433 form an orthonormal family for @xmath434 . the first step of the function truncate is transforming all matrices @xmath395 to lcf . we shall describe this step by a function leftcanonical(@xmath397 ) that takes as input a matrix product state @xmath435 and returns a pair @xmath436 , where @xmath437 is a scalar and @xmath438 is a matrix product state such that @xmath439 and @xmath440 has lcf at every qubit . we shall define leftcanonical(@xmath397 ) by the following algorithm . let us explain the qr - decomposition step in the above algorithm and prove its correctness . consider any qubit @xmath43 and represent @xmath395 as a block matrix @xmath441.\ ] ] note that @xmath395 has @xmath442 rows and @xmath400 columns . let @xmath443 . applying the ` economic ' qr - decomposition to @xmath395 one gets @xmath444 where @xmath445 has dimensions @xmath446 , @xmath447 has dimensions @xmath448 , and columns of @xmath445 form an orthonormal family of vectors , that is , @xmath449 . finally , @xmath447 is an upper triangular matrix ( this property will not be important for us ) . let us write @xmath450,\ ] ] where @xmath451 have dimensions @xmath452 . the property @xmath449 is equivalent to @xmath453 . hence @xmath454 defined in the above algorithm has lcf . note that dimensions of @xmath455 may or may not be equal to the ones of @xmath401 . let @xmath456 be the updated version of @xmath457 defined in the algorithm . obviously @xmath458 for any @xmath459 . thus @xmath460 is equal to @xmath461 for all @xmath462 and for all @xmath463 . the last step of the algorithm ( @xmath464 ) applies a qr - decomposition to a column vector @xmath465 , possibly updated by the previous step of the algorithm . hence @xmath445 is a unit - norm column vector of size @xmath466 while @xmath447 is a scalar which determines normalization of the overall state . this proves that @xmath439 and @xmath440 has lcf at every qubit . suppose the input matrix product state @xmath397 has bond dimension @xmath3 . then the computational cost of each qr - decomposition is @xmath467 . therefore , the function leftcanonical(@xmath397 ) can be computed in time @xmath468 . since no step of the algorithm increases dimensions of the matrices , the final matrix product state @xmath440 also has bond dimension @xmath3 . we are now ready to describe the function truncate . choose any integer @xmath469 and partition the chain as @xmath470 , where @xmath471 consider a matrix product state @xmath397 such that @xmath395 has lcf for all @xmath472 and rcf for all @xmath473 . suppose also that the matrices @xmath401 have dimensions at most @xmath3 for all @xmath473 and at most @xmath474 for all @xmath472 . we assume that @xmath475 ( we shall be interested in the case @xmath476 ) . using the orthonormal families of states @xmath422 and @xmath477 defined in eqs . ( [ eq : phi],[eq : theta ] ) one can write @xmath398 as latexmath:[\[\label{eq } \psi(a)=\sum_{\alpha=1}^{r(m ) } \sum_{\beta=1}^{c(m ) } \sum_{x=0,1 } a_x(m)_{\alpha,\beta } the schmidt decomposition of @xmath398 with respect to the partition @xmath479 and truncate this decomposition by retaining only the @xmath3 largest schmidt coefficients . to this end consider the singular value decomposition ( svd ) of @xmath480 , namely , @xmath481= u s v^\dag,\ ] ] where the matrices @xmath482 have dimensions @xmath483 with @xmath484 the matrix @xmath485 is diagonal such that @xmath486 is the @xmath168-th largest singular value of @xmath480 . the matrices @xmath487 and @xmath114 are isometries , that is , @xmath488 let us represent @xmath114 as a block matrix @xmath489,\ ] ] where @xmath490 and @xmath491 have dimensions @xmath492 . using the above svd one can rewrite @xmath398 as @xmath493 where @xmath494 and @xmath495 are orthonormal family of @xmath1 states defined as @xmath496 we conclude that eq . ( [ eq : svd1 ] ) defines the schmidt decomposition of @xmath398 with respect to the partition @xmath479 , while @xmath486 are the schmidt coefficients . the best rank-@xmath3 approximation to @xmath398 which we denote @xmath497 is obtained from eq . ( [ eq : svd1 ] ) by retaining @xmath3 largest schmidt coefficients , that is , @xmath498 decompose matrices @xmath482 into blocks such that @xmath499 , \ ; s=\left [ { \begin{array}}{cc } s ' & 0 \\ 0 & s '' \\ { \end{array}}\right ] , \ ; v'=\left [ { \begin{array}}{c|c } v ' & v '' \\ { \end{array}}\right].\ ] ] by definition , @xmath500 have dimensions @xmath501 furthermore , @xmath502 is a square diagonal matrix that contains @xmath3 largest singular values of @xmath480 , while @xmath503 and @xmath504 are isometries , that is , @xmath505 and @xmath506 . we conclude that @xmath507 , where @xmath508 for @xmath473 and for @xmath509 , @xmath510 the fact that @xmath504 is an isometry implies that @xmath511 has rcf , so we can apply the above procedure again with @xmath512 and @xmath513 . starting from @xmath514 and moving towards the left boundary of the chain one can reduce the bond dimension from @xmath474 to @xmath3 . the above truncation algorithm can be summarized as follows . here we decomposed @xmath504 into blocks @xmath515 and @xmath516 similar to eq . ( [ eq : vblock ] ) . we have studied the following combinations of noise models and decoders : for each of the above combinations we estimated the probability of a logical error the decoding outcome in which the recovery operator differs from the actual error by a logical pauli operator ( we do not differentiate between @xmath517 , or @xmath102 logical errors ) . the performance of each decoder was measured in terms of its error threshold and its _ badness _ parameter the ratio between the logical error probabilities of a given decoder and the best available decoder for the considered noise model . thus badness @xmath518 for any decoder with smaller values indicating better decoders . the exact ml decoder and mps decoders were implemented as described in section [ sect : xnoise ] and section [ sect : xyznoise ] respectively . the mwm decoder was implemented by a reduction from the minimum weight perfect matching problem to the maximum weight matching problem as described in ref . @xcite . is between @xmath519 and @xmath520 , which is in a good agreement with the estimate @xmath521 of ref . @xcite which calculated the phase transition point in the respective spin model . each curve has data points at error rates @xmath522 . to compute the logical error probability , at least @xmath523 failed error correction trials have been accumulated for each datapoint . [ fig : xth ] , height=245 ] let us first discuss our results for the x - noise . the threshold error rate @xmath524 of the ml decoder coincides with the critical density of anti - ferromagnetic bonds in the random - bond ising model on the nishimori line @xcite . the latter has been estimated numerically by mertz and chalker @xcite who found @xmath525 . our data shown at fig . [ fig : xth ] suggest that @xmath526 which is in a good agreement with the estimate of ref . @xcite . for comparison , the mwm decoder is known to have the threshold @xmath527 , see @xcite . the curves representing the exact mld and mps decoders with @xmath528 are too close to be distinguishable on the main plot . the red curve represents the standard minimum weight matching decoder . the inset shows badness of various decoders as a function of the error rate . we define the badness as the ratio between logical error probabilities of a given decoder and the optimal decoder ( mld ) . each curve has data points at error rates @xmath529 . to compute the logical error probability , at least @xmath530 failed error correction trials have been accumulated for each datapoint . [ fig : xcomp ] , height=302 ] the performance of different decoders for a fixed code distance @xmath6 and a wide range of error rates is shown at fig . [ fig : xcomp ] . we observed that the mwm decoder remains nearly optimal for all simulated error rates with the badness parameter @xmath531 , even though for these error rates the logical error probability changes by several orders of magnitude . the slight difference between mld and the mwm decoder can be explained by the fact that the latter ignores the error degeneracy @xcite . the data shown on fig . [ fig : xcomp ] suggests that for x - noise ignoring the error degeneracy does not have a significant impact on the performance , even for large error rates and large code distances . perhaps more surprisingly , fig . [ fig : xcomp ] demonstrates that the mps decoder with a relatively small bond dimension @xmath528 is virtually indistinguishable from the optimal one in terms of the logical error probability . this serves as a numerical proof of correctness for the mps decoder . we observed numerically that the exact mld algorithm described in section [ sect : xnoise ] becomes very sensitive to rounding errors in the regime of large code distances and small error rates . one way to suppress rounding errors is to enforce an orthogonality condition @xmath532 on the covariance matrix @xmath146 in algorithm 1 . the orthogonality condition is satisfied automatically if all arithmetic operations are perfect ( because @xmath146 represents a covariance matrix of a pure gaussian state , see section [ sect : gaussian ] for details ) . in practice , we observed that the orthogonality can be quickly lost if no special measures are taken . a simple and computationally cheap solution of the above problem is to compute the qr - decomposition @xmath533 , where @xmath445 is an orthogonal matrix and @xmath447 is an upper - triangular matrix . note that @xmath532 is possible only if @xmath447 is a diagonal matrix with entries @xmath534 on the diagonal . this form of @xmath447 can be easily enforced by setting all off - diagonal entries of @xmath447 to zero and replacing each diagonal entry @xmath535 by the sign of @xmath535 . let @xmath536 be the resulting diagonal matrix . we found that replacing @xmath146 by @xmath537 after each call to the functions simulatehorizontal and simulatevertical in algorithm 1 makes the algorithm more stable against rounding errors . let us now discuss the depolarizing noise . in this case we only have an approximate implementation of mld with no direct means of estimating the approximation precision . hence the first natural question is whether the mps decoder with a fixed bond dimension @xmath3 has a non - zero error threshold @xmath524 . our data suggests ( although not conclusively ) that the answer is ` yes ' . most importantly , we observed an exponential decay of the logical error probability as a function of the code distance @xmath85 for a fixed error rate , see fig . [ fig : xyzdecay ] , where we used @xmath5 . assuming that the observed decay does not saturate for larger @xmath85 , the data shown at fig . [ fig : xyzdecay ] gives a lower bound @xmath538 . the logical error probability as a function of the error rate for a fixed @xmath85 is shown on fig . [ fig : xyzth ] which also exhibits a typical threshold - like behavior and suggests that @xmath539 . previously studied approximate versions of mld such as the renormalization group decoder @xcite and the markov chain decoder @xcite , as well the mwm decoder @xcite have error thresholds between @xmath540 and @xmath541 . the threshold of the exact ml decoder corresponding to the phase transition point in the disordered eight - vertex ising model is known to be @xmath542 , see ref . @xcite . since the correlation length of the ising model diverges at the phase transition point , we expect that the mps decoder can only achieve this optimal threshold if the bond dimension @xmath3 is a growing function of the code distance @xmath85 . as a function of the code distance @xmath85 for a fixed error rate @xmath543 . to compute the logical error probability , at least @xmath530 failed error correction trials have been accumulated for each datapoint . [ fig : xyzdecay ] , height=245 ] as a function of the error rate @xmath126 . assuming a non - zero error threshold @xmath524 , the data suggest that @xmath539 . to compute the logical error probability , at least @xmath523 failed error correction trials have been accumulated for each datapoint . [ fig : xyzth ] , height=245 ] the performance of different decoders for a fixed code distance @xmath6 and a wide range of error rates is shown at fig . [ fig : xyzcomp ] . in a striking contrast with the analogous x - noise data , we observed that the mwm decoder becomes highly non - optimal in the regime of small error rates with the badness parameter above @xmath544 . this can be attributed to the fact that mwm decoder often fails to find the minimum weight error consistent with the syndrome since it ignores correlations between @xmath7 and @xmath8 errors @xcite . we also observed that the logical error probability of mps decoders converges very quickly as one increases the bond dimension . the data shown on fig . [ fig : xyzcomp ] indicates that the mps decoder with @xmath5 is nearly optimal for all error rates and all code distances @xmath545 . while the logical error probability is the most natural figure of merit , one may also ask how well the mps - based algorithm with a small bond dimension @xmath3 approximates the coset probabilities for some fixed syndrome . for simplicity , we considered the trivial syndrome , that is , the cosets @xmath546 , and @xmath547 . we observed a very fast convergence for the most likely coset and a poor convergence for the remaining cosets , see tables [ table : cosetx],[table : cosetxyz ] . since the only goal of the decoder is to identify the most likely coset , the slower convergence for some of unlikely cosets might not be a serious drawback . .x - noise : probabilities of the two cosets computed by the mps algorithm . the simulation parameters are @xmath548 and @xmath6 . the exact values of the coset probabilities are @xmath549 and @xmath550 . [ cols="^,^,^",options="header " , ] is shown . the red curve represents the minimum weight matching decoder . the inset shows badness of various decoders as a function of the error rate . we define the badness as the ratio between logical error probabilities of a given decoder and the best decoder ( mps decoder with @xmath551 ) . each curve has data points at error rates @xmath552 . to compute the logical error probability , at least @xmath530 failed error correction trials have been accumulated for each datapoint . [ fig : xyzcomp ] , height=302 ] the mps decoder offers a lot of possibilities for improvement . one rather obvious improvement ( employed in the above simulations ) is to use a single run of algorithm 2 to compute two different coset probabilities . indeed , suppose we choose the logical operator @xmath102 supported in the right - most column of the lattice denoted @xmath553 on fig . [ fig : slices ] . then the tensor networks constructed for the cosets @xmath554 and @xmath555 are exactly the same except for the column @xmath553 . since we contract the network column by column starting from the left - most column @xmath556 , the difference between the two cosets manifests itself only in the very last step of algorithm 2 ( computing the inner product @xmath557 ) . since this step takes a negligible time compared with the rest of the algorithm , it makes sense to compute both probabilities @xmath558 and @xmath559 by performing a single network contraction . the same observation applies to the probabilities @xmath560 and @xmath561 . we also expect that a choice of the standard error @xmath47 consistent with the syndrome @xmath43 may affect the convergence of the algorithm . while we have chosen @xmath47 by connecting each syndrome to the left / top boundary , it may be advantageous to choose @xmath47 as a small - weight error , for example , using the mwm decoder . finally , a challenging open problem is how to extend the mps decoder to noisy syndrome extraction . a naive extension would require a contraction of a 3d tensor network . we anticipate that this problem can be attacked using recently developed algorithms for simulating 2d quantum systems based on projected entangled pairs states ( peps ) , see @xcite . it should be emphasized that the mps - based decoder is applicable to any noise model that can be described by a stochastic i.i.d . pauli noise . in contrast , the decoder based on matchgates is only applicable to noise models with independent bit - flip and phase - flip errors .
analysis of the observations of the sol2001 - 12 - 26 event related to ground - level - event gle63 , including microwave spectra and images from the _ nobeyama radioheliograph _ at 17 and 34ghz , _ siberian solar radio telescope _ at 5.7ghz , and _ transition region and coronal explorer _ in 1600 has led to the following results . a flare ribbon overlapped with the sunspot umbra , which is typical of large particle events . atypical were : i ) long duration of the flare of more than one hour ; ii ) moderate intensity of a microwave burst , aboutsfu ; iii ) low peak frequency of the gyrosynchrotron spectrum , around 6ghz ; and its insensitivity to the flux increase by more than one order of magnitude . this was accompanied by a nearly constant ratio of the flux emitted by the volume in the high - frequency part of the spectrum to its elevated low - frequency part determined by the area of the source . with the self - similarity of the spectrum , a similarity was observed between the moving microwave sources and the brightest parts of the flare ribbons in 1600 . comparison of the 17ghz and 1600 images has confirmed that the microwave sources were associated with multiple flare loops , whose footpoints appeared in ultraviolet as intermittent bright kernels . to understand the properties of the event , we simulated its microwave emission using a system of several homogeneous gyrosynchrotron sources above the ribbons . the scatter between the spectra and sizes of the individual sources is determined by the inhomogeneity of the magnetic field within the ribbons . the microwave flux is mainly governed by the magnetic flux passing through the ribbons and the sources . an apparent simplicity of microwave structures is caused by a poorer spatial resolution and dynamic range of the microwave imaging . the results indicate that microwave manifestations of accelerated electrons correspond to the structures observed in thermal emissions , as well - known models predict .
1611.08349
we continue a study ( @xcite ; article i ) on the 26 december 2001 event ( sol2001 - 12 - 26 ) . this solar eruptive - flare event produced a strong flux of solar energetic particles ( seps , mainly protons ) near earth and a 63rd ground - level enhancement of cosmic - ray intensity ( gle63 ) . gles represent the highest - energy extremity of seps ( see , _ e.g. _ , @xcite ; and references therein ) . gles are rare events ; seventy - two gles only have been recorded since 1942 . the rare occurrence of gles hampers understanding their origins and finding consistent patterns that might govern their appearance and properties . unlike electrons , whose signatures are manifold throughout the whole electromagnetic range , accelerated protons and heavier ions can only be detected on the sun from nuclear @xmath1-ray emission lines appearing in their interactions with dense material ( see , _ e.g. _ , @xcite ) . solar @xmath1-ray observations have been very limited in the past . no @xmath1-ray images were available before 2002 . due to the observational limitations , considerations of the origins of seps and , especially , gles mainly refer to the hypotheses proposed several years ago ( see , _ e.g. _ , @xcite for a review ) . on the other hand , studies based on recent observations @xcite indicate that some of these hypotheses might need refinement . gles are typically produced by major eruptive events associated with big flares ( mostly of the goes x - class ) , fast coronal mass ejections ( cmes ) , and strong microwave bursts . the common association with different solar energetic phenomena @xcite hampers the identification of the origins of seps and gles . all of these circumstances show how important the analysis of a solar source of each gle is . with a general correspondence between the magnitudes of seps , cme speeds , and flare manifestations , there are few outliers from the correlations @xcite . strong fluxes of high - energy protons were observed near earth after these events associated with moderate microwave bursts . the correlation between the proton fluences , on the one hand , and microwave and soft x - ray ( sxr ) fluences , on the other hand , is considerably higher . this is difficult to understand , if seps are exceptionally shock - accelerated . a flare - related contribution could also be significant in the events , whose proton productivity was enhanced for some unclear reasons . one of these events was the sol2001 - 12 - 26 eruptive - flare event responsible for gle63 . this solar event has not been comprehensively studied because of its limited observations . we are not aware of either low - coronal observations of an eruption or hard x - ray ( hxr ) data . on the other hand , the event was observed in microwaves by the _ siberian solar radio telescope _ ( ssrt : @xcite ) at 5.7ghz ; the _ nobeyama radioheliograph _ ( norh : @xcite ) at 17 and 34ghz , and in ultraviolet ( uv ) by the _ transition region and coronal explorer _ ( trace : @xcite ) . the ssrt observations have been studied in article i. its conclusions are : 1 . most likely , gle63 was caused by the m7.1 solar event in active region ( ar ) 9742 ( n08w54 ) . contribution from a concurrent far - side event is unlikely . the flare was much longer than other gle - related flares and consisted of two parts . the first , possibly eruptive , flare and a moderate microwave burst started at 04:30 and reached an m1.6 level . the main flare , up to m7.1 , with a much stronger burst started at 05:04 , when a cme was launched . 3 . the main flare involved strong magnetic fields presumably associated with a sunspot in the west part of ar 9742 . two non - thermal sources observed at 5.7ghz initially approached each other nearly along the magnetic neutral line , and then moved away from it like expanding ribbons , as if they were associated with the legs of the flare arcade . it was difficult to confirm and expand these indications in article i based solely on the ssrt data because of their insufficient spatial resolution and coalignment accuracy . to verify and elaborate these results , observations of the flare arcade or ribbons in a different spectral range , where they are clearly visible , should be compared with microwave data of a higher spatial resolution . of special interest is a conjectured localization of the non - thermal microwave sources in the legs of the flare arcade . this possibility does not contradict a commonly accepted view on the flare process ; however , hxr and microwaves almost always show a few simple non - thermal sources . ribbon - like hxr structures have been observed in exceptional events @xcite . the simplicity and confinement of non - thermal sources in impulsive flares suggested involvement of one to two loops @xcite . later observational studies extended this view up to some long - duration flares @xcite . a concept of a single microwave - emitting loop became dominant . one of the major purposes of our companion articles is to find the possible causes of the contrast between the rich proton outcome of this solar event and its moderate manifestations in microwaves . we consider _ a priori _ the contributions from the acceleration processes both in the flare and by the shock wave to be possible @xcite . accordingly , we examine the 26 december 2001 flare in this article ( article ii ) and eruptive event in article iii ( grechnev _ et al . _ , 2017 , in preparation ) . here we endeavor to shed further light on the listed issues related to the flare itself and its microwave emission , analyzing the observations carried out by trace in 1600 and by norh at 17 and 34ghz . trace observed the flare mostly in the 1600 channel , which we use . the 1600 images are similar to those observed in the h@xmath2 line , but they do not show filaments . some images were produced less often in other uv channels , 1550 and 1700 , and in white light . the whole set that we analyze consists of 668 images observed in 1600 from 04:23 to 05:22 ( all times hereafter refer to utc ) . the imaging interval between 63 images from 04:23 to 04:55 was around 30 seconds , and then it decreased to about 2 seconds from 05:04 to 05:22 . the exposure time was 0.43 seconds before 04:30 , then 0.26 seconds until 05:00 , and it changed cyclically between 0.020 , 0.031 , and 0.052 seconds during the main flare . a movie in the electronic supplementary material presents the whole flare observed in 1600 . we processed the trace images as follows : the background level was found for each image as a highest - probability value in its histogram . this individual level was subtracted from each image , which was converted to a reference exposure time of 0.43 seconds . the offset between the 1600 and white - light channels was corrected from their coalignment . comparison with white - light images produced by the _ michelson doppler imager _ ( mdi : @xcite ) onboard soho revealed a roll angle of @xmath3 in the trace images , which was compensated , and refined their absolute pointing coordinates . all of the images were transformed to 05:04 to compensate for the solar rotation . bright defects produced by high - energy particles were reduced in calculations by using a minimum of two images separated by a small interval @xcite . remaining issues of the processing described here are insignificant . the major flare configurations are presented by the trace images averaged during each of the two flare parts in figures [ f - flare_regions]a and [ f - flare_regions]b . the images are shown within a limited brightness range and scaled by a power - law function with a @xmath1 of 0.7 . a dark region near the centers of the panels corresponds to a sunspot . the orange contour traces the major magnetic - polarity inversion ( neutral ) line at the photospheric level calculated from a soho / mdi magnetogram observed on 26 december at 04:51 . the averaged images and movie present complex flare configurations . the first flare started from the appearance of two extended thin ribbons at both sides of the neutral line ( a remote northeast brightening is beyond the field of view of figure [ f - flare_regions ] ) . the east ribbon evolved in the first flare within the red contour . three structures are magnetically conjugate to the east ribbon . a long thin ribbon r1a was active early in the event , and then it smoothly faded . we do not consider this region . intermittent brightenings within the dark - blue contour were active during the first flare , and then became less important . a small bright region started to grow near the sunspot ( light - blue contour ) . . two distinct flare parts are separated by the dashed line . the gray bars in panel a represent the observation interval of a slowly drifting radio burst and the cme onset time extrapolated to the position of ar 9742 . the black curve in panel c represents the background - subtracted total flux at 17ghz . , scaledwidth=70.0% ] in the main flare , the east ribbon broadened and extended southwest ( figure [ f - flare_regions]b ) . the west ribbon seed broadened northwest and became the major west ribbon . a jet ( green contour ) appeared after 05:06 in a funnel - like configuration with a ring base , along which brightenings ran in the movie . such funnels appearing above magnetic islands inside opposite - polarity regions contain coronal null points @xcite . the magnetic structure of a small flux - rope erupting inside a funnel can not survive at a null point @xcite and released plasma flows out as a jet @xcite . the collision of an erupting flux - rope with a separatrix surface can produce a shock wave @xcite . figure [ f - flare_light_curves ] shows the flare light curves and milestones of the event . an inflection in the goes channels in figure [ f - flare_light_curves]a separates the two parts of the flare . the separation time is close to the cme onset time of 05:0605:10 extrapolated to @xmath4 in the cme catalog ( cdaw.gsfc.nasa.gov/cme_list/ ; @xcite ) based on the data from the soho s _ large angle and spectroscopic coronagraph _ ( lasco : @xcite ) . the heliocentric distance of the flare site was @xmath5 . with an average cme speed of 1446kms@xmath6 , its onset time at the flare site should be 65 seconds earlier ( the light - gray bar in figure [ f - flare_light_curves]a ) . the association of the cme with the main flare was previously found by @xcite . a slowly - drifting type ii and/or type iv burst was associated with the event . in the learmonth spectrum up to 180 mhz shown by @xcite , the burst is detectable from about 04:57 , too early for the fast cme . the burst can be followed back until 04:50 up to 300 mhz in the _ hiraiso radio spectrograph _ ( hiras ) spectrum ( 2001122605.gif ) at sunbase.nict.go.jp/solar/denpa/hirasdb/events/2001/ ( the dark - gray bar in figure [ f - flare_light_curves]a ) . this burst could only be caused by expanding ejecta or a wave , which started well before the fast cme from ar 9742 . no other cme was detected around that time . most likely , this burst was due to a slower eruption preceding the fast cme . figures [ f - flare_light_curves]b and [ f - flare_light_curves]c present the light curves for the first and main flare , respectively , calculated from the trace 1600images as total over the major flare regions denoted in figures [ f - flare_regions]a and [ f - flare_regions]b . the bars represent the imaging intervals . the black curve in figure [ f - flare_light_curves]c shows a microwave burst recorded by the _ nobeyama radio polarimeters _ ( norp : @xcite ) at 17ghz . the flare has two distinct parts . the first flare started at about 04:30 and lasted @xmath7 minutes before the main flare . the 17ghz burst did not exceed 200sfu . the main flare started at about 05:04 with a sharp increase of the uv emissions from the both major ribbons and the 17ghz burst . the temporal profiles of the west ribbon and microwave burst are similar but not identical . $ ] g ( black n - polarity , white s - polarity ) . the crosses in panels a , b , and d mark the base of the jet . ] figure [ f - trace_mag_wl ] compares the configuration observed in 1600 near the flare peak with mdi magnetograms and a white - light trace image produced at the main flare onset . the field of view presents the total length of the east ribbon with its remote extensions into a northeast s - polarity sunspot and a southwest region of weak magnetic fields . the color contours correspond to figure [ f - flare_regions ] . the red white - dashed contour denotes the east ribbon in the first flare . the pink contour corresponds to the n - sunspot umbra in figure [ f - trace_mag_wl]d . in the main flare , the west ribbon overlapped with the sunspot umbra in figure [ f - trace_mag_wl]a , as article i assumed . the magnetogram of ar 9742 located not far from the limb is affected by projection effects on magnetic field inclined to the line of sight . to find realistic fields , we consider the magnetograms observed just before the main flare ( figure [ f - trace_mag_wl]b ) and three days before ( figure [ f - trace_mag_wl]c ) . the magnetograms are scaled by a power - law function with a @xmath1 of 0.8 for each polarity separately ( bright positive , dark negative ) . ar 9742 evolved over three days . the major magnetic - field distribution persisted . the changes are mostly related to weaker - field regions and the shape of the neutral line . the s - polarity field under the future east ribbon concentrated . the flare - related south n - sunspot became slightly displaced . the magnetograms exhibit three kinds of distortions . i ) unlike the magnetogram in figure [ f - trace_mag_wl]c , the west parts of the sunspots in figure [ f - trace_mag_wl]b appear with a spurious inverted polarity ( this is a common feature in magnetograms close to the limb ) . ii ) the field strengths in figure [ f - trace_mag_wl]b are reduced _ vs. _ figure [ f - trace_mag_wl]c both in sunspots and plage regions , and a secant correction seems to be justified for all of the regions . iii ) a hook - like shape of the 2000 g contour in the south sunspot indicates a saturation - like distortion occurring in mdi magnetograms . this distortion is also present in the 26 december magnetogram . with a maximum field strength of 2807 g observed there on 23 december , the real strength in the south sunspot could reach @xmath8 g _ vs. _ 1164 g in the 26 december magnetogram . the cross in figure [ f - trace_mag_wl]a denotes the base center of the funnel - like configuration , where the jet occurred . figure [ f - trace_mag_wl]b shows there an island of an enhanced magnetic field up to @xmath9 g . the magnetograms of the preceding days indicate that , most likely , this was a negative - polarity island inside a positive environment . ] , and a power - law index of microwave - emitting electrons [ @xmath10 are specified for each spectrum . the gray curve represents a classical gs spectrum corresponding to the high - frequency observations at the peak of the burst.,scaledwidth=85.0% ] the microwave burst was observed at a number of fixed frequencies by the norp and learmonth radiometers . these data allow us to analyze the spectrum of the burst . we do not use the norp data at 80ghz , which look unreliable . to cross - calibrate the norp and learmonth data , the background - subtracted flux at each frequency was integrated from 04:20 to 05:30 . the logarithmic spectrum was fitted with a fifth - order polynomial , and the deviations from the fit were used as cross - calibration coefficients , which ranged between 0.87 and 1.21 . figures [ f - norp_lear]a[f - norp_lear]d present total flux temporal profiles at four frequencies . the gyrosynchrotron ( gs ) temporal profile extends to a very low frequency of 610mhz . a spiky component is due to plasma emission , and late - phase enhancements at 610mhz and 2ghz are probably due to a type iv burst from a different source . the spectra in figure [ f - norp_lear]e were calculated with an integration time of 50 seconds for the four instances denoted in the upper panels . the spectra are similar , although the ratio of the peak fluxes between the solid curve and the dotted one reached 18 . the peak frequency estimated from a sixth - order polynomial fit was 6.1ghz@xmath11 . the low - frequency branch was elevated relative to the slope of 2.9 for a classical gs spectrum ( gray curve ) and had an actual slope of @xmath12 at frequencies @xmath13ghz . the elevation could be due to inhomogeneity of the emitting source , _ i.e. _ increasing at lower frequencies contribution from higher layers , where magnetic field is weaker ( see , _ e.g. _ , @xcite ; @xcite and references therein ) . the high - frequency slope [ @xmath2 ] corresponds to a power - law index @xmath15 of the electron number spectrum , according to @xcite . .,scaledwidth=75.0% ] figures [ f - norh_timeprof]a and [ f - norh_timeprof]b show the total flux and maximum brightness temperatures of microwave sources measured at 17 and 34ghz from norh images and at 5.7ghz from ssrt images ( see article i ) . figure [ f - norh_timeprof]c presents the total area of the sources at 17 and 34ghz . the only way to increase the flux of a simple gs source while keeping its spectrum shape is to increase its area @xcite with an unchanged brightness temperature . figure [ f - norh_timeprof ] presents an opposite relation . when the total flux and brightness temperature strongly increased between the b and c instances , the change in the total area of the sources was minor . figure [ f - norh_timeprof]c also indicates that each of the microwave sources might not be well resolved . this complicates the situation . to get indications of microwave - emitting structures , we consider bright kernels in 1600 as footpoints of the loops , with whose legs microwave sources could be associated . figures [ f - norh_trace]a[f - norh_trace]d show the 1600images overlaid by contours of the 17ghz images ( see also the movie with contours at @xmath16t_{\mathrm{b}\ , \max}$ ] over each 17ghz image ) . figures [ f - norh_trace]e[f - norh_trace]h reveal bright kernels in 1600 emphasized by subtracting the images observed two minutes before , along with the outermost 17ghz contours . with a position of the flare site at n08w54 , the microwave sources in the low corona must be offset west - northwest from the upper - chromosphere 1600 images . the offset uncertainty can reach @xmath17 , while the coalignment accuracy is presumably within @xmath18 among the 17ghz images and within @xmath19 among the uv images . . contour levels are at @xmath20t_{\mathrm{b}\ , \max}$ ] specified in each panel . the thick contour is at a half - height level . the ellipses represent the half - height contours of the norh beam . bottom : two - minute running - difference 1600 images and 0.0625-level contours of the 17ghz images . solar rotation was compensated in all images to 05:04 . ] two microwave sources in figures [ f - norh_trace]a[f - norh_trace]d reside above two ribbons located in opposite magnetic polarities . such sources are usually interpreted as the conjugate legs of a single loop . however , different clusters of bright kernels in figures [ f - norh_trace]e[f - norh_trace]h correspond to different sets of loops dominating at each of the four times . comparison of the thick contours and the norh beam ( both at a half - height level ) shows that these sets of loops were unresolved by norh . if the two 17ghz sources were located in the legs of multiple loops , then correspondence is expected between the relative positions of the ribbons , on the one hand , and those of the microwave sources , on the other hand . ) . the accuracy of the results denoted by the open circles can be reduced due to overlap between the images of the sources produced by the ssrt.,scaledwidth=75.0% ] indications of this correspondence were found in article i from the relative motions of the two sources observed in ssrt images at 5.7 ghz . here we elaborate this result using the measurements from the 1600 and 17ghz images . for the measurements at 17ghz we used the same technique as in article i. for each ribbon we used the centroid of its current image in 1600 within the 10%-contour of the ribbon averaged over the whole flare . figure [ f - kernels_timeprof ] shows the distance between the microwave sources and their position angle relative to the major orientation of the neutral line ( @xmath21 from the west ) measured from the 5.7 and 17ghz images and the same parameters of the ribbons in 1600 . these relative measurements do not depend on any coalignment accuracy . the east 17ghz source was absent before 04:46:50 , and our analysis of the trace data ends at 05:22:04 . the vertical bars in figure [ f - kernels_timeprof]a show that the variations of the distances are well under the beam size of both ssrt and norh . this fact does not contradict the instrumental resolution , because we measure the centroid of a source rather than its structure , but it makes the results sensitive to imaging issues and changes in the shapes of the sources . to the west is a reference direction in the measurements of the shear angle in figure [ f - kernels_timeprof]b . see also the movie . ] the major tendencies observed in the three spectral ranges are similar from 04:45 until 05:15 . the microwave spectra in figure [ f - norp_lear]e are also similar in this interval . the decrease of the distance between the 17ghz sources in figure [ f - kernels_timeprof]a after 05:10 , dissimilar to the others , is probably due to an increasing emission from entire low loops filled with trapped higher - energy electrons . it is less pronounced in the lower - frequency ssrt images and absent in the 1600 images . it is difficult to separate this contribution from the main sources at 17ghz . the measurements from the ssrt data denoted by the open circles are complicated due to overlap of the sources ( article i ) . the difference between the measurements from the 5.7ghz and uv data in figure [ f - kernels_timeprof]b before 04:45 is due to two causes . the first , geometrical cause is the location of the uv ribbon centroid near a bend of the neutral line ( figures [ f - flare_regions ] and [ f - trace_mag_wl ] ) that distorts the measured shear angle . second , the magnetic field in this region was insufficient to produce a detectable gyromagnetic emission , and the 5.7ghz centroid was displaced to stronger magnetic fields . with the complications mentioned , the relative motions of the microwave sources and flare ribbons were similar . this correspondence allows us to compare the measured centers of the 17ghz sources and uv ribbons . we will not use here the positions of microwave sources measured from ssrt images due to their insufficient pointing accuracy . figure [ f - kernels ] presents the trajectories of the microwave sources ( blue ) and uv ribbons ( red ) . their increasing thickness or size indicates time , which is also specified at some positions . the gray - scale background is a negative variance image computed from all trace images obtained during the whole flare . this image represents all changes observed in 1600 according to their statistical contributions @xcite . the green dashed line represents the neutral line . the variance image clearly shows the major east ribbon , which extended south and broadened east , and the major west ribbon in the main flare . the initial west ribbon in the first flare ( r1 ) appears in figure [ f - kernels ] as three fragments . the red - dotted line represents the centroid of the east ribbon before 04:47 , when its microwave counterpart was absent . the east ribbon centroid moved nearly parallel to the neutral line from 04:36 to a hump , and then turned away from the neutral line . the of trajectory the east microwave source is mostly parallel to that of the east ribbon centroid . a later divergence is due to increasing contribution to microwaves from the upper part of the arcade . the offset of the microwave sources from the uv ribbons is due to their different heights . the west part of the flare site was more complex . intermittent brightenings on ribbon r1 during the first flare flipped the centroid between r1 and west - ribbon seed . the west microwave source was more stable due to a strong direct dependence on the magnetic field . in the main flare , both the west ribbon and microwave source drifted to the strongest - field region above the sunspot umbra . an excursion of the ribbon centroid around 05:16 was due to a brightening at the southwest edge of the sunspot that we do not consider . the agreement between the trajectories observed in 1600 and at 17ghz can not be a coalignment issue , because the west microwave source was localized from 04:47 to 05:22 within @xmath22 , while the east microwave source displaced systematically during this interval by @xmath23 ( excluding the later mismatching part of its trajectory ) . the trajectories of the ribbons and microwave sources correspond to each other , except for the mentioned deviations , whose causes are clear . the measurements of the shear angle between the uv ribbons were initially affected by a bending of the neutral line eastward . if one mentally straightens this part of the neutral line and the easternmost part of the east ribbon to keep the direction of @xmath21 everywhere , then the initial angle would substantially decrease . thus , the overall decrease of the shear throughout the event exceeded @xmath24 we measured . the correspondence between the uv ribbons at the bases of the arcade and microwave sources above the ribbons confirms association of the microwave sources with the legs of the arcade loops . non - thermal electrons precipitating in dense layers are ultimately responsible for the uv emission . on the other hand , both precipitating and trapped non - thermal electrons radiate gs emission in the low - corona magnetic loops . the uv and microwave emissions depend in opposite senses on the magnetic - field strength [ @xmath25 . with an estimated electron spectrum index of @xmath26 , the gs emissivity at optically thin frequencies has a direct dependence @xmath27 @xcite , whereas the well - known mirroring in strong magnetic fields hampers electron precipitation , which governs the uv emission . the real physical distinction between the structures emitting in these two spectral ranges and responsible processes might be emphasized by different instrumental characteristics of trace and norh . to verify this idea , the difference in the spatial resolution can be compensated by convolving the trace images ( @xmath19 resolution ) with a norh beam . its cross - section is an ellipse , whose parameters gradually change during the day . at 05:02 on 26 december , its half - height dimensions were @xmath28 ( we used the norh images synthesized at 17ghz by the enhanced - resolution fujiki software ) . figure [ f - trace_conv ] shows the results of our experiment for four different flare episodes . the left column presents the trace 1600 images , the right column presents nearly simultaneous norh images , and the middle column presents the trace images convolved with the norh beam , whose elliptic half - height contours are shown in the right panels . to facilitate comparison , the convolved images in the middle column are overlaid with contours of the microwave images . the structures in figures [ f - trace_conv]b and [ f - trace_conv]c are similar . a subtle counterpart of the east ribbon is also detectable at 17ghz . the environment in weaker fields is faint but detectable in microwaves , as the contours in figure [ f - trace_conv]b indicate . the major sources in figures [ f - trace_conv]e and [ f - trace_conv]f are also similar . a brighter microwave source , a , is located in stronger magnetic field . conversely , the image of the weaker - field east ribbon , b , is brighter in the convolved uv image , as expected . the presence of additional features c and d in figure [ f - trace_conv]h makes it dissimilar to figure [ f - trace_conv]i . feature c is the jet emanating from a configuration with a magnetic null point . the magnetic - field strength there steeply falls off upward , and a microwave counterpart is not expected . feature d is a southernmost end of the east ribbon located in a weak magnetic field . its absence in microwaves is not surprising , especially with a limited dynamic range of norh of about 300 . the images in figures [ f - trace_conv]k and [ f - trace_conv]l become less similar due to the upper part of the coronal arcade appearing at 17ghz , but invisible in the uv . the brightness temperature of the arcade in figure [ f - trace_conv]l exceeded 15 mk estimated for this time from goes data in article i ; the power - law index of its brightness temperature spectrum estimated from the images at 17 and 34ghz was around @xmath29 . therefore , thermal bremsstrahlung could only supply a minor contribution . most likely , the upper part of the arcade was dominated by trapped electrons with a harder spectrum , consistent with a general pattern established by @xcite , and in later studies . in summary , the microwave sources a and b were , most likely , associated with the legs of the arcade rooted in the ribbons . bright kernels in 1600 represented instantaneous loci of the electron precipitation . electrons trapped in numerous loops , whose footpoints were shown by the uv kernels previously , emitted a prolonged background gs emission . this long - lasting background reinforced the similarity between the images of the long - lived flare ribbons and microwave sources . this possible scenario should be manifested in the evolution of the microwave spectrum . in addition to the detailed gs spectra presented in figure [ f - norp_lear]e for four instances , here we examine the overall variations of the peak frequency [ @xmath30 and a power - law index of microwave - emitting electrons [ @xmath10 for the whole flare . the @xmath31 index of the electron - number spectrum can be calculated as @xmath32 @xcite ; @xmath2 should be estimated from optically thin data , _ e.g. _ , from the total flux norp data at 17 and 35ghz or from norh images at 17 and 34ghz . we used both methods , because norp data are characterized by a higher accuracy , while norh data provide a higher sensitivity . the contribution of thermal bremsstrahlung was subtracted . the result was smoothed over ten seconds . with corresponding styles.,scaledwidth=75.0% ] the peak frequency was estimated in a way similar to the technique used by @xcite and @xcite from norp and learmonth total flux data ( section [ s - microwave_observations ] ) . for each instance , a combined spectrum at 12 frequencies was averaged over 16 seconds , and a parabola was fitted to the five points of the log log spectrum closest to the peak . the result was smoothed over 30 seconds . figure [ f - delta ] shows the 17 and 35ghz fluxes along with an estimated thermal flux , and the calculated @xmath33 and @xmath31 . the intervals with doubtful estimates are rejected . the values of @xmath31 estimated from norh data are more reliable than those from norp data for weaker fluxes and conversely for stronger fluxes . the electron index [ @xmath10 in figure [ f - delta]b has an impulsive component superposed on a harder gradual background . the initial values of @xmath34 are too hard , probably due to underestimation of the thermal flux from the goes data . they are insensitive to plasma temperatures of @xmath35 mk , whose contribution to microwaves can be considerable . its possible role should lessen , as the gs emission increased . the gs emission from trapped electrons with a progressively hardened spectrum in the course of a continuous injection probably dominated late in the event . the trapping effect could also affect the initial part of figure [ f - delta]b , because the flare lasted about 20 minutes before 04:50 . comparison of figures [ f - delta]a and [ f - delta]b shows that the impulsive component corresponds to enhancements in the time - profiles . most likely , freshly injected electrons had a softer spectrum with @xmath36 , which did not change considerably throughout the flare . a probable power - law index of microwave - emitting electrons was mostly @xmath37 . the peak frequency in figure [ f - delta]c varies in a range of 57ghz . the values of @xmath33 ( and @xmath31 in figure [ f - delta]b ) at the four marked instances are close to those estimated in figure [ f - norp_lear]e in a slightly different way . the evolution of @xmath33 suggests interplay of parameters of emitting electrons and magnetic - field strength ; however , the range of @xmath33 seems to be atypically narrow and low for the observed microwave fluxes . a possible key to these challenges could be a distributed microwave - emitting system . the average @xmath33 of individual sources was around 6ghz , and their total number elevated the peak flux of each one up to the observed values . the scatter between the parameters of the sources could be a reason for the broadening of the microwave spectrum and its gradual shape . analysis of the microwave and uv images has shown that each of the two microwave sources observed at 5.7 , 17 , and 34ghz was associated with one of the two ribbons located in opposite - polarity magnetic fields . the appearance of each microwave source corresponds to the whole related ribbon , as if it had been viewed by norh . both the ribbons and microwave sources exhibited nearly identical systematic motions . until the peak of the microwave burst , the two flare centers approached each other nearly along the neutral line , so that the distance and shear between them considerably decreased . after that , the motion occurred away from the neutral line , corresponding to the usual expansion of the ribbons . the expansion was measured from the ssrt data until 06:30 in article i. the correspondence of each microwave source to the flare ribbon , their nearly identical motions , and surprisingly persistent shape of the microwave spectrum , in spite of 18-fold flux variations , indicate that microwaves were emitted by the conjugate legs of multiple loops constituting the flare arcade . the relation of the magnetic - field strengths under the ribbons points to a strongly asymmetric configuration . magnetic - flux conservation in flare loops requires larger areas of the sources above the weaker - field east ribbon relative to their conjugate counterparts above the sunspot - associated west ribbon . to verify these considerations , we will attempt to reproduce the observed spectra by means of a simple model , using the magnetic fields actually measured . the microwave flux density [ @xmath38 at optically thin frequencies [ @xmath39 is controlled by the instantaneous total number of emitting high - energy electrons [ @xmath40 . self - similarity of the spectra in a wide range of microwave fluxes in figure [ f - norp_lear]e shows that the @xmath41 relation in this event also applied at lower frequencies [ @xmath42 . accelerated electrons are produced in a reconnection process ; therefore , @xmath43 is associated with a power of flare energy release . let us find which observable parameters of a flare indicate this association . the energy flux density entering the flare current sheet from one of its sides is governed by the poynting vector , @xmath44/{4\pi}= -[[\textbf{\textit{v}}\times\textbf{\textit{b}}]\times\textbf{\textit{b}}]/{4\pi}= [ \textbf{\textit{v}}b^{2}-\textbf{\textit{b}}(\textbf{\textit{b}}\cdot \textbf{\textit{v}})]/{4\pi}= { \textbf{\textit{v}}b^{2}}/{4\pi}$ ] [ erg@xmath45s@xmath6 ] . here @xmath47/c$ ] ; @xmath48 is a vertical magnetic field , _ i.e. _ @xmath49 ; @xmath50 is a horizontal velocity of the plasma inflow into the vertical current sheet , _ i.e. _ , @xmath51 . the total power released in the current sheet dimensioned @xmath52 by @xmath53 is @xmath54[ergs@xmath6 ] . here @xmath55 , and @xmath56 is the input rate of the magnetic flux . let @xmath57 be the lifetime of a point - like uv kernel in the footpoint of a thin magnetic tube during reconnection in the current sheet and afterward . a multitude of kernels constitutes an instant uv - emitting stripe corresponding to a narrow moving flare ribbon in the standard model . then , @xmath58 , which is magnetic flux within the ribbon stripe at time [ @xmath59 . presumably , @xmath57 is proportional to the lifetime of accelerated electrons in the magnetic tube , and @xmath60 is proportional to the magnetic flux across a gs source at time [ @xmath59 . a particular value of @xmath57 is not important , if it is much shorter than the burst . if the flare process operated self - similarly throughout the burst , then the ratio of energy released in the current sheet during @xmath57 , @xmath61 , to the total energy of electrons produced at the same time , @xmath62 , was constant , _ i.e. _ @xmath63 . this relation is correct , as long as the vertical size [ @xmath53 ] of the current sheet is constant , and magnetic field [ @xmath25 in its vicinity is uniform . the latter assumption is justified by rapid disappearance with an increasing height of small magnetic features , which reflect strong inhomogeneity of the magnetic field on the photosphere . with a power - law energy distribution of gs - emitting electrons [ @xmath65 ( @xmath66 ) ] their energy density is @xmath67 \epsilon_{0 } n \equiv \overline{\epsilon}n$][erg@xmath68 , @xmath69 . here @xmath70 is the number of electrons per @xmath71 in a unit interval of energy [ @xmath72 , @xmath73 is the total number of gs - emitting electrons in 1@xmath71 , and @xmath74 is an average energy of a single electron . since @xmath75 , from the condition @xmath76 we get a final equation for the total number of gs - emitting electrons , @xmath77 . note that @xmath78 is related here to the vicinity of the current sheet . in turn , @xmath60 is an instant magnetic flux within one of the ribbon stripes , @xmath79 . the relation between the emission flux and total number of emitting electrons , @xmath80 , is transformed to the form @xmath81 . this relation , similar evolutions of microwave sources observed by two radio heliographs and flare ribbons observed in uv , and similarity between microwave images and convolved uv images motivated our usage of a model source system referring to the ribbons to simulate gs emission in this event . the spectrum of the non - thermal microwave emission in the 26 december 2001 event has two conspicuous features : persistent shape with weak changes in the peak frequency under large flux variations and an enhanced low - frequency part . @xcite found the peak - frequency variations to be small in about one third of the events . the authors related this behavior to gs self - absorption around the peak of the burst and to the tsytovich razin suppression in its early rise and late decay . @xcite reported the latter effect indeed ; however , its importance at the rise of a long - duration flare is difficult to reconcile with chromospheric evaporation . it is quantified by the neupert effect @xcite , _ i.e. _ similarity between the soft x - ray flux ( directly dependent on the plasma density ) and antiderivative of the microwave burst . the plasma density is initially low , reducing the tsytovich razin effect at this stage . at the decay of our burst , _ e.g. _ at 05:29 , the net total area at 34ghz was @xmath82@xmath83 ( figure [ f - norh_timeprof]c ) , emission measure estimated from goes data @xmath84@xmath85 ( article i ) , and plasma density @xmath86@xmath85 . with a magnetic field strength of @xmath87 g estimated in article i at 05:20 , the razin frequency was @xmath88ghz ( figure [ f - delta]c ) . thus , the ideas of @xcite are unlikely to help us , because the magnetic fields of @xmath89 g they considered are too weak for our sunspot - associated flare . to account for the low - frequency increase in a gs spectrum , inhomogeneity of the source and superposition of multiple sources have been proposed ( _ e.g. _ @xcite ) . the major inhomogeneity in these models is related to the magnetic field in a flare loop of a varying cross - section . this undoubted inhomogeneity affects the shape of the spectrum , especially its optically thick part @xcite . it is difficult to understand why the spectrum from a single inhomogeneous loop had a constant shape , while indications of multiple sources are certain . we are not aware of inhomogeneous multi - loop models . to verify our interpretations , we are forced to use a tentative simplified modeling of gs emission from a set of several homogeneous sources . we are interested in general features of this system and need a very simple analytic description of gs emission , which the @xcite approximations present . their reduced accuracy at the lowest and highest harmonics of the gyrofrequency is not crucial for the task of the model to understand properties of our event . our model contains a considerable number of homogeneous gs sources , each with a different magnetic - field strength and volume . their number depends on the width and length of the brightest parts of the uv ribbons . the model should also demonstrate the direct dependence of the total flux and spectrum of the microwave emission on the total magnetic flux and its distribution over each of the ribbons . the model does not consider the influence of the ambient plasma on generation and propagation of the gs emission , _ i.e. _ the tsytovich razin effect and free free absorption , whose importance in our event is unlikely . the loop system constitutes an arcade rooted in the ribbons . each ribbon in our event is extended and inhomogeneous in brightness and width . we relate a set of microwave sources to a brightest , broadest stripe of each ribbon . its width [ @xmath90 corresponds to a typical transversal size of a loop , whose end is rooted in this ribbon . the width of a loop varies according to the magnetic field strength [ @xmath25 along it , being equal to @xmath91 , with superscript `` 0 '' related to the first ribbon . for a narrow ribbon stripe , the number of emitting loops [ @xmath92 should be about its length to width ratio . if the loops do not overlap , then their total flux [ @xmath38 is the sum of the fluxes emitted by all of the loops . each @xmath93th loop is represented by two homogeneous cubic sources in its legs above both ribbons , corresponding to the observations at 5.7 and 17ghz . magnetic fluxes in conjugate cubes are equal to each other , @xmath94 . the ratio of their sizes [ @xmath95 is determined by the ratio of the magnetic - field strengths in the east source [ @xmath96 and the west one [ @xmath97 , so that @xmath98 . it is convenient to use a set of @xmath99 loops , each of which encloses equal magnetic flux @xmath100 . this assumption ensures the balance of magnetic fluxes in conjugate legs of any loop , irrespective of its location , and facilitates partition into @xmath99 cubic sources . two methods of partition are possible . in the first method , the total magnetic flux @xmath101 is divided into @xmath99 equal parts on the magnetogram within each ribbon in a fixed direction . the widths of the pieces can be different , while their magnetic fluxes are equal to each other . each @xmath93th pair corresponds to a loop . the loops do not overlap , and the procedure to find @xmath102 and @xmath103 seems to be physically justified . a rather formal second method considers the histograms @xmath104 of the magnetic - field distribution within each ribbon , where @xmath105 is the number of pixels where the magnetic - field strength is equal to @xmath78 . the area under the histogram is divided into @xmath99 equal parts corresponding to equal magnetic fluxes @xmath106 , which is easy to calculate . the cubic sources obtained in this way are different , and their paired link is lost . on the other hand , the scatter of the size and magnetic - field strength is maintained , as in the first method of partition . we use the second method , which is simpler to implement . the spectral flux density @xmath107 from each @xmath93th unpolarized source is @xmath108 ( \nu^2/c^2)\ , a_i / r^2 $ ] , where @xmath109 is boltzmann s constant , @xmath110 the source area , @xmath111 au , and @xmath112 the optical thickness . the effective temperature [ @xmath113 and absorption coefficient [ @xmath114 are calculated following @xcite . in their figure 3 , the log log plots of @xmath115 and @xmath116 deviate at low @xmath117 from the quasi - linear parts into opposite directions that reduces the errors @xcite . the deviations are less for @xmath118 in our case . the total flux spectrum is a sum of @xmath119 spectra from all sources . the number density of microwave - emitting electrons [ @xmath120 and their power - law index @xmath121 are identical for all sources . the viewing angles of the sources in the legs of the loops above the east and west ribbons @xmath122 and @xmath123 are different , while their half - sum is the longitude of the flare site . the optimal number of the paired sources [ @xmath92 was adjusted iteratively to meet three conditions : i ) the sum of @xmath119 sources provides a gradual spectrum with a single peak , ii ) the value of @xmath99 is about the ribbon length to width ratio , and iii ) the model acceptably fits the observed spectrum . with an optimal number [ @xmath92 , the field strengths [ @xmath124 estimated from the photospheric magnetogram should be corrected to the coronal values . it is possible to use a constant scaling factor @xmath125 , so that magnetic field strengths in coronal sources are @xmath126 . to estimate @xmath125 , we refer to @xcite , who found an average magnetic field of @xmath127 g using a homogeneous gs source model and a scaling law between @xmath128 and the total area of a source . an intuitive option to calculate @xmath125 as a ratio of 400 g to an average field strength measured from the magnetogram within the ribbons leads to a biased estimate because of a nonlinear dependence of the microwave flux on magnetic field . with any number of the sources , the flux at optically thin frequencies [ @xmath129 is controlled by the total number of emitting electrons and their emissivity . thus , @xmath130 with @xmath131 ; @xmath132 is the viewing angle @xcite . the constancy of the optically thin total flux emitted by the same electron population with any number of the sources , up to a single large one ( subscript `` s '' ) , results in an equality @xmath133 . using the constancy of the total number of emitting electrons , @xmath134 , we obtain an average magnetic - field strength in an equivalent single source @xmath135 . finally , we have estimated @xmath136 g for the sources above each ribbon separately and obtained @xmath137 . in this approach , the magnetic flux [ @xmath138 retains , and the change from @xmath102 to @xmath139 results in a corresponding change in the size of each @xmath93th source from @xmath103 to @xmath140 . figure [ f - model_spectrum]a presents the results of the modeling by the gray line for the first flare ( episode 1 ) and the black line for the rise of the main flare ( episode 2 ) . these two episodes were soon after fresh injections , when emission of trapped electrons , which we do not consider , can be neglected , and an assumption of a constant @xmath141 is justified . the symbols denote the observed fluxes . an enhancement at 2ghz was due to a contribution from plasma emission . the average magnetic flux over the ribbon stripe [ @xmath142 used in the modeling is specified for each episode . the average field strength above the west ribbon stripe [ @xmath143 additionally affects the microwave flux at optically thin frequencies . the number density of electrons with energies @xmath144kev was @xmath145@xmath85 in both episodes . the total number of electrons was @xmath146 in episode 1 and @xmath147 in episode 2 . in section [ s - reconnection_power ] we obtained @xmath148 assuming the magnetic field to be uniform at both sides of the neutral line that is not realistic . nevertheless , the model takes account of a real inhomogeneity of the photospheric magnetic field and acceptably matches the real spectrum . the relation between the magnetic flux and microwaves can be generalized to a variable magnetic field , considering the intrinsic dependence of the microwave flux on the magnetic field strength , @xmath149 . the ratio of the optically thin microwave fluxes in the two episodes in figure [ f - model_spectrum]a is 8.9 , while the ratio of the magnetic fluxes within the ribbon stripes is 4.4 . with @xmath150 , the expected microwave flux ratio is @xmath151 , which agrees with the model result of 8.9 within @xmath152 . this seems to be acceptable with our simplified approach . figure [ f - model_spectrum]b demonstrates the influence on the spectrum of the number of loops for episode 1 . the increase of the number of loops elevates the low - frequency branch of the spectrum . the effect is similar to that of the source inhomogeneity . the ratio of the 16-loop spectrum to the single - loop one at 2ghz is 5.1 in this case . the lowest - frequency slope , 2.9 , corresponds to the classical gs spectrum . figure [ f - model_spectrum]c illustrates the role of asymmetry in the magnetic configuration with the same magnetic flux reconnection rate . the solid line corresponds to the real situation in episode 2 . the dashed line represents the spectrum for a hypothetical situation , when both microwave - emitting regions were located above identical ribbons corresponding to the actual west ribbon . the dotted line represents a similar experiment with two east ribbons . the strongest effect of asymmetry at optically thin frequencies is illustrated by two extremities . the high - frequency flux from a highly asymmetric configuration is determined by a single source , and the flux is doubled in a symmetric configuration ( two identical sources ) . the effect of asymmetry varies between a factor of one and two . the same occurs for the opposite asymmetry . thus , with the same magnetic - flux reconnection rate , the high - frequency flux can vary within a factor of four . the actual ratios of the optically thin fluxes in figure [ f - model_spectrum]c are 1.26 between the dashed and solid lines and 1.40 between the solid and dotted lines . the asymmetry in this event progressively increased . the west ribbon expanded into the sunspot umbra , the east ribbon developed into weaker - field regions . to balance magnetic flux , an increasing high - frequency emission from the stronger - field west regions must be accompanied by an increasing area of the weaker - field east region that elevated the low - frequency part of the spectrum . the spectrum shape remained nearly constant , in spite of large changes in the microwave flux . the relevance of a homogeneous source and simplified expressions by @xcite were discussed by @xcite . as they showed , the usage of a scaling law between the average magnetic field and total source area makes the homogeneous model sufficient to estimate statistical characteristics of microwave bursts such as the peak flux and frequency and spectral index . each elementary loop in our model is replaced by two homogeneous sources of different size and magnetic field . a set of 32 homogeneous sources reflects inhomogeneity of the ribbons . the scaling factor to shift from their magnetic fields to those in microwave sources is based on the results of @xcite . our model acceptably reproduces the spectra also around the peak and at lower frequencies . modeling of the circular polarization of the gs emission is complicated by the near - the - limb location of the flare site . since the west microwave source is visible through quasi - transversal magnetic fields associated with the arcade , polarization reversal is expected in a wide frequency range . thus , the polarization of each elementary source is a result of interplay between the optical thickness and polarization reversal issues , both frequency - dependent . this makes the analysis of the polarization of the west source and the total emission too complex . the polarized emission of the east source can only be extracted in norh 17ghz and ssrt 5.7ghz images . for the degree of polarization of each optically thin elementary source we used a corrected formula from @xcite , otherwise we assigned an opposite polarization of 15% . the results for episode 1 are ( all negative ; first observed , second model ) : @xmath153\%$ ] at 5.7ghz , @xmath154\%$ ] at 17ghz ; for episode 2 : @xmath155\%$ ] at 5.7ghz , @xmath156\%$ ] at 17ghz . the model reproduces the actual degree of polarization within a factor of two with a correct sign . the outcome of our simple modeling can be summarized as follows . 1 . the observed properties of the gs spectrum are consistent with the emission from a distributed multi - loop system , _ i.e. _ the flare arcade . the asymmetry of the magnetic configuration is important . the magnetic - flux balance requires larger areas at the weaker - field side that elevated the low - frequency part of the spectrum , shifting the peak frequency left . inhomogeneities of the individual sources could strengthen this effect . 3 . to reproduce the observed spectrum with realistic magnetic fields , a large increase of the magnetic flux is required in the main flare ( the average field strength changed insignificantly ) . this result is consistent with a temporal correlation between the magnetic - flux change rate and hxr emission found by @xcite , who measured magnetic flux within expanding parts of the ribbons . replacement of a distributed multi - loop system by a single loop is generally not equivalent and can result in a different shape of the spectrum , peak frequency , and their behaviors during the burst . this leads to the next item . modeling of a microwave - emitting loop initiated by @xcite has been developed into a powerful tool @xcite . the models use real data on magnetic field and take account of its inhomogeneity , anisotropy and spatial distribution of electrons . a next - step challenge is a realistic multi - loop model , at least , for simplified conditions . elements of the scheme presented here might be helpful in its development . as known from observations in the h@xmath2 line , flare kernels and ribbons initially approach each other along the neutral line , and then they move away from it . the expansion of the ribbons was explained by the two - dimensional ( 2d ) standard flare model . the motions along the neutral line have not been clearly visualized . various motions observed later in hxr were summarized by @xcite . the authors interpreted them in terms of the 2d model ; their cartoons implied a questionable rotation of the current sheet around the vertical axis . the motions along the neutral line have reasonably been considered as an intrinsically 3d effect . its scheme was presented by @xcite . the authors supposed contraction of reconnected loops to be a necessary element of the unshearing process , whereas the two phenomena seem to be different results of the flare reconnection not directly related with each other . @xcite reported microwave observations of the motions along the neutral line . the authors noticed that an m2.6 flare developed along the arcade visible in the h@xmath2 and extreme - uv images and that , at least , several loops were involved in the process . however , they considered a single loop for each instance . in the discussion of the unshearing motions they mostly follow @xcite . the motions of the microwave sources in our event corresponded to those of the brightest parts of the ribbons , which we call the ribbon stripes for brevity . figure [ f - scheme ] explains the relative displacement of the stripes . the intrinsic 3d geometry implies the presence in an extended current sheet of a zone of most efficient energy release , where the reconnection process is most similar to the 2d model . the formation of major stripes is associated with this zone . figures [ f - scheme]a and [ f - scheme]b present evolution of three magnetic loops , which reconnected in this zone . initially , these loops belonged to a certain layer of magnetic arcades above the pre - eruptive filament . the dotted line connects the bases of the central loop to show the shear . figure [ f - scheme]c shows the central loops of different layers . the higher a loop , the smaller is the shear of its bases . the loops evolve similarly to the central loop in figures [ f - scheme]a and [ f - scheme]b . the major stripes corresponding to these loops displace as the arrows show , reflecting the decrease of the shear with the increasing height of the pre - eruptive arcade . the major motions initially occurred nearly along the neutral line and away from it during the main flare phase and afterwards . this behavior ( not shown in figure [ f - scheme ] ) is associated with a non - uniform decrease of the shear away from the neutral line . a similar behavior is exhibited by helical lines of a cylindric nonlinear force - free magnetic field ( @xmath157 ) when @xmath158 with @xmath159 . non - thermal emissions are generated by energetic electrons and carry the most direct information about acceleration processes in flares . as noted in section [ s - introduction ] , non - thermal sources observed in hxr and microwaves are usually simple and confined , which favors their identification with one or two flaring loops . this view drawn from microwave observations of almost all impulsive flares has been generalized to some major long - duration flares @xcite . still stronger confinement is typical of hxr sources ; complex structures like extended ribbons were observed in exceptional events @xcite . by contrast , observations of flares in thermal emissions ( h@xmath2 line , uv , extreme uv , sxr ) typically show complex multi - component structures of a larger extent . this dissimilarity has led to different views on flaring structures drawn from different observations . on the other hand , synergy between the structures observed in thermal and non - thermal emissions has been conjectured previously , including studies in which some of us participated @xcite . the causes of the differences between non - thermal sources and configurations visible in thermal emissions were unclear . a key idea by @xcite explaining this difference has not been commonly perceived . the authors pointed out the limitations on the sensitivity and dynamic range ( typically @xmath160 ) in the hxr imaging . indeed , images in hxr and @xmath1-rays have been provided by the imagers of the fourier - synthesis type with an intrinsically limited coverage of the @xmath161-plane . these are the _ hard x - ray telescope _ ( hxt : @xcite ) onboard _ yohkoh _ and the _ reuven ramaty high - energy solar spectroscopic imager _ ( rhessi : @xcite ) . sources weaker than 10% of the brightest one are not detectable in their images @xcite . @xcite also found that hxr sources accompanied the h@xmath2 kernels in strongest magnetic fields only ; `` the hxr sources indicate where large energy release has occurred , while the h@xmath2 kernels show the precipitation sites of nonthermal electrons with higher spatial resolution '' . flares have been imaged in microwaves also almost exclusively with fourier - synthesis interferometers ( mainly norh ) . they provide a better coverage of the @xmath161-plane than hxr imagers . on the other hand , optically thin gs emission strongly depends on the magnetic field @xmath162 with @xmath163 . with a dynamic range of norh of @xmath164 @xcite , its opportunities to observe weaker non - thermal structures seem to be comparable to hxr imagers . also , microwave telescopes have a poorer spatial resolution than hxr imagers . for these reasons , the strongest non - thermal sources are only expected in hxr and microwave images without weaker structures due to instrumental limitations . configurations , in which accelerated electrons are manifested , and those visible in thermal emissions , must actually be closely associated with each other . both thermal and non - thermal emissions in our event originated in basically the same configuration . dissimilarities between the structures visible in microwaves and uv were mainly due to different spatial resolution and dynamic range of the instruments and different dependencies of the emissions on the magnetic field strength . @xcite also concluded that , at least , some of the single - loop configurations in norh images corresponded to multi - loop arcades observed with telescopes of a higher spatial resolution . an appropriate proxy for a configuration responsible for non - thermal emissions may be a structure observed in extreme ultraviolet or in soft x - rays . this expectation corresponds to widely accepted model concepts of eruptive flares ( processes in confined flares are unlikely different in nature see , _ e.g. _ @xcite ) . this should be thoroughly verified . if this is correct , then considerations of simple configurations are justified , when they appear so in thermal emissions . inevitable simplifications should be recognized to avoid inadequate conclusions . we have studied the 26 december 2001 eruptive flare , combining the trace 1600 and norh 17 and 34ghz images , the results obtained in article i from the ssrt 5.7ghz images , and different data . the analysis has shown that the first flare and main flare were most likely associated with separate eruptions . the first eruption presumably occurred in ar 9742 around 04:40 and produced ejecta , which did not appear in the lasco field of view but manifested in a slowly drifting radio burst . a related moderate two - ribbon flare involved medium magnetic fields , produced a moderate microwave burst up to a few hundreds sfu at 56ghz , and reached a goes importance of m1.6 . this flare lasted half an hour and had not fully decayed when another eruption occurred in ar 9742 . the second eruption occurred around 05:04 and produced a fast cme . the flare passed into the major two - ribbon flare and reached a strength of m7.1 . the east ribbon observed in the first flare lengthened and broadened farther into regions of moderate and weak magnetic fields . the west ribbon entered the strongest magnetic fields above the sunspot umbra . the flare magnetic configuration was increasingly asymmetric . the microwave burst strengthened up to 4000sfu at 67ghz and 780sfu at 35ghz , and lasted about 15 minutes ( fwhm ) . furthermore , trace images reveal a jet - like eruption around 05:09 . its light curve in 1600 is a spike with a fwhm duration as short as three minutes . the jet is not detectable in microwaves and will be analyzed in article iii . the first flare and the following main one were most likely caused by the first and second eruptions , respectively . the eruptions stretched closed magnetic configurations and thus could facilitate escape of particles accelerated in the active region . sharp eruptions might have produced shock waves , which also could accelerate heavy particles . article iii will consider these possibilities . a conspicuous morphologic manifestation of a large particle event is flaring above the sunspot umbra @xcite . this feature indicates involvement in flare processes of the strongest magnetic fluxes , whose reconnection rate corresponds to flare energy release and governs particle acceleration . the flare ribbons in the events analyzed previously overlapped with the umbras of opposite - polarity sunspots . the sol2005 - 01 - 20 ( gle69 ) and sol2006 - 12 - 13 ( gle70 ) events studied in detail exhibited large variations of the peak frequency , whose maximum exceeded 25ghz , and peak fluxes at 35ghz exceeded @xmath0sfu . while the major flare in the sol2001 - 12 - 26 event looks similar to these events , its morphologic difference is involvement in the flare of a single sunspot . the microwave flux directly reflects the magnetic flux reconnection rate , being proportional to its instant value multiplied by a factor of @xmath165 at optically thin frequencies . the major difference between the moderate first flare and much stronger main flare was in the reconnected magnetic flux , while parameters of the acceleration process manifesting in the number density and power - law index of accelerated electrons remained almost unchanged . with the same magnetic - flux reconnection rate ( and presumably the same production of accelerated particles ) , the microwave response strongly depends on magnetic fields , including symmetry of the configuration . if it is asymmetric , then the microwave spectrum is broader , the peak frequency is lower , and its variations are small . the asymmetry of the configurations can cause an additional scatter within a factor of four in the correlations between high - frequency microwave bursts and near - earth proton fluxes . an indication of asymmetry is overlap of the flare ribbon(s ) with the umbra of a single , or two , or no sunspot . both flare parts were typical arcade flares , whose development is described by the 3d model . its intrinsic phenomena are the motions of the arcade legs visible in microwaves with their bases visible as the ribbons . their initial approach nearly along the neutral line reflects consecutive involvement in reconnection of structures corresponding to the pre - eruptive magnetic field vector distribution . it is similar to that one in a nonlinear force - free flux rope , _ i.e. _ from strongly sheared low structures to less sheared those located at increasing distances from the axis . a later expansion of the ribbons is accounted for by the 2d model . in spite of a seeming single - loop configuration visible in microwave images , the correspondence between the positions and motions of the uv ribbons and non - thermal microwave sources evidences that accelerated electrons emitted microwaves from the multi - loop arcade rooted in the uv ribbons . this conclusion is supported by a simple modeling of the microwave spectrum . configurations with more than one loop appear to be common in various flares : from small , spiky events @xcite to large , long - duration events ( @xcite and the present article ) . our analysis has demonstrated that dissimilarities between the structures visible in non - thermal and thermal emissions are due to different instrumental characteristics and different dependencies of the emissions on the magnetic field . in accordance with well - known models , a proxy of a configuration responsible for non - thermal emissions could be a structure observed in thermal emissions . we thank n.v . nitta for discussions and the reviewer for useful remarks . we thank the instrumental teams managing ssrt , norh , and norp ; trace ( nasa ) ; soho / mdi and lasco ( esa and nasa ) , goes ; usaf rstn network ; nict ( japan ) ; and the cme catalog at the cdaw data center ( nasa and catholic university of america ) . we appreciate the memories of t.a . treskov , one of the major developers of the ssrt , and n.n . kardapolova , who managed ssrt observations for many years . this study was supported by the russian state contract no . a. kochanov was supported by the russian foundation of basic research under grants 15 - 32 - 20504 mol - a - ved and 15 - 02 - 03717 . v. kiselev was supported by the marie curie pirses - ga-2011 - 295272 radiosun project . the authors claim that they have no conflicts of interest . grechnev , v.v . , lesovoi , s.v . , smolkov , g.y . , krissinel , b.b . , zandanov , v.g . , altyntsev , a.t . , kardapolova , n.n . , sergeev , r.y . , uralov , a.m. , maksimov , v.p . , lubyshev , b.i . : 2003 , _ solar phys . _ * 216 * , 239 . grechnev , v.v . , kurt , v.g . , chertok , i.m . , uralov , a.m. , nakajima , h. , altyntsev , a.t . , belov , a.v . , yushkov , b.y . , kuznetsov , s.n . , kashapova , l.k . , meshalkina , n.s . , prestage , n.p . : 2008 , _ solar phys._**252 * * , 149 .
we report on 34 m slit spectroscopy of 13 seyfert 2 nuclei . the 3.3 m polycyclic aromatic hydrocarbon ( pah ) emission is used to estimate the magnitudes of compact nuclear starbursts ( on scales less than a few 100 pc ) and to resolve the controversy over their energetic importance in seyfert 2 nuclei . for three selected seyfert 2 nuclei that have been well studied in the uv , the magnitudes of the compact nuclear starbursts estimated from the 3.3 m pah emission ( with no extinction correction ) are in satisfactory quantitative agreement with those based on the uv after extinction correction . this agreement indicates that the flux attenuation of compact nuclear starburst emission due to dust extinction is insignificant at 34 m , and thus allows us to use the observed 3.3 m pah luminosity to estimate the magnitudes of the compact nuclear starbursts in seyfert 2 nuclei . based directly on our 34 m slit spectra , the following two main conclusions are drawn : ( 1 ) except in one case , the observed nuclear 34 m emission is dominated by agn and not by starbursts , and ( 2 ) compact nuclear starbursts are detected in 6 out of 13 seyfert 2 nuclei , but can not dominate the energetics of the galactic infrared dust emission in the majority of the observed seyfert 2 galaxies . for several sources for which infrared space observatory spectra taken with larger apertures and/or soft x - ray data are available , these data are combined with our 34 m slit spectra , and it is suggested that ( 3 ) extended ( kpc scale ) star - formation activity is energetically more important than compact nuclear starbursts , and contributes significantly to the infrared luminosities of seyfert 2 galaxies , although the agn is still an important contributor to the luminosities , and ( 4 ) the bulk of the energetically significant extended star - formation activity is of starburst type rather than quiescent normal disk star - formation ; the extended starbursts are responsible for the superwind - driven soft x - ray emission from seyfert 2 galaxies . finally , a correlation between the luminosities of agns and compact nuclear starbursts is implied ; more powerful agns tend to be related to more powerful compact nuclear starbursts .
astro-ph0112299
according to the unification paradigm for seyfert galaxies , seyfert 1s , which show broad optical emission lines , and 2s , which do not , are intrinsically the same objects , but the nuclei of the latter class are obscured by dust along our line of sight in dusty molecular tori @xcite . it is widely accepted that the ultimate energy source of seyfert _ nuclei _ ( although not of their extended host galaxies ) is the release of gravitational energy caused by mass accretion onto a central supermassive blackhole ( so - called agn activity ) . however , it has recently been argued that strong signatures of starbursts are detected in the nuclear uv optical spectra of seyfert 2 galaxies ( heckman et al . 1997 ; gonzalez delgado et al . 1998 ; storchi - bergmann et al . 2000 ; gonzalez delgado , heckman & leitherer 2001 ) , and that the intrinsic extinction - corrected luminosities of compact nuclear starbursts ( hereafter ` compact ' is used to mean a size scale of less than a few 100 pc ) could be comparable to the luminosities of the agn @xcite . in contrast , @xcite investigated the co indices in the near - infrared @xmath1-band spectra of seyfert 2 nuclei , and found no evidence for the presence of strong compact nuclear starbursts . although these authors observed different samples of seyfert 2 galaxies , it is nevertheless striking that they draw such contradictory conclusions on the energetic role of compact nuclear starbursts in seyfert 2 nuclei . according to @xcite , compact nuclear starbursts are natural byproducts of agns dusty molecular tori . in this case , since direct uv optical emission from the agn is highly attenuated in seyfert 2 nuclei , the less strongly obscured compact nuclear starburst emission will inevitably make a relatively strong contribution to the observed uv optical fluxes in these objects . thus , the detection of signatures of compact nuclear starbursts in the observed uv optical spectra of seyfert 2 nuclei @xcite would not be surprising . however , to obtain a deeper understanding of the nature of seyfert 2 nuclei , it is certainly more important to quantify the energetic importance of compact nuclear starbursts than simply to investigate their presence . @xcite estimated the magnitudes of compact nuclear starbursts , based on their extinction - corrected uv data , for four uv - bright seyfert 2 galaxies , while for many other seyfert 2 galaxies that have been studied optically , no quantitative discussions of the absolute magnitudes of compact nuclear starbursts have been made @xcite . even the uv - based quantitative discussion could contain large uncertainties , because of the susceptibility of uv emission to dust extinction ; the extinction correction factor for uv data may vary drastically depending on the assumed amount of dust and its spatial distribution , and has often been claimed to be quantitatively uncertain in particular applications @xcite . the uncertainties in the dust extinction correction are much less in the near - infrared @xmath1-band data taken by ivanov et al . however , the spectral signatures of starbursts are weak , so that careful subtraction of old stellar and agn emission is required to estimate the magnitudes of compact nuclear starbursts @xcite . at 34 @xmath0 m , we possess powerful diagnostic tools to distinguish between starburst and agn activity , to detect weak starbursts , and to estimate the magnitudes of compact nuclear starbursts in seyfert 2 galaxies with fewer quantitative uncertainties . as shown by imanishi & dudley ( 2000 , their figure 1 ) , * if a seyfert 2 nucleus is powered by starbursts , strong polycyclic aromatic hydrocarbon ( pah ) emission will be detected at 3.3 @xmath0 m regardless of dust extinction . * if it is powered by obscured agn activity , a carbonaceous dust absorption feature will be detected at 3.4 @xmath0 m . * if it is powered by weakly obscured agn activity , the 34 @xmath0 m spectrum will be nearly featureless . * if it is a composite of starburst and agn activity , the absolute luminosity and the equivalent width of the 3.3 @xmath0 m pah emission feature will be smaller than in starburst galaxies , so that these values can be used to quantitatively estimate the energetic importance of starburst activity . since the 3.3 @xmath0 m pah emission is intrinsically strong @xcite , even the signatures of weak starbursts are detectable ; even if only @xmath210 % of the observed nuclear 34 @xmath0 m flux originates in starbursts , the 3.3 @xmath0 m pah emission peak is @xmath220 % higher than the continuum level , and thus is clearly recognizable in normal s / n @xmath2 1520 spectra . additionally , the effects of dust extinction are smaller at 34 @xmath0 m than at shorter wavelengths , which makes the uncertainties in the dust extinction correction factor much smaller . this advantage is demonstrated quantitatively below . suppose that the dust extinction in starbursts is found to be a@xmath3 = 24 mag , with an uncertainty of a factor of 2 in a@xmath3 . since the dust extinction in the uv ( @xmath22000 ) is @xmath22.5 @xmath4 a@xmath3 @xcite , the corresponding dust extinction correction factor in the uv is 10010000 , differing by a factor of 100 depending on the adopted a@xmath3 . the difference becomes even larger if we go to shorter - wavelength parts of the uv region or if the actual dust extinction is larger . in contrast , since the dust extinction at 34 @xmath0 m is @xmath20.05 @xmath4 a@xmath3 @xcite , the extinction correction factor at 34 @xmath0 m in the case of a@xmath3 = 24 mag is 1.11.2 . thus , the extinction correction is negligible at 34 @xmath0 m , and the uncertainty in the correction factor is only at the 10% level . the high detectability of weak starbursts and the small uncertainties in dust extinction correction at 34 @xmath0 m combine to make 34 @xmath0 m observations a very powerful tool to quantitatively address the issue of the energetics of compact nuclear starbursts in seyfert 2 nuclei . this paper reports the results of 34 @xmath0 m spectroscopy of seyfert 2 nuclei and their implications for the energetics of these objects . throughout the paper , @xmath5 @xmath6 75 km s@xmath7 mpc@xmath7 , @xmath8 = 0.3 , and @xmath9 = 0.7 are adopted . the seyfert 2 nuclei studied by @xcite and @xcite are selected in order to be able to compare our diagnostic results at 34 @xmath0 m directly with those previously obtained in the uv , optical , and @xmath1-band . our particular interests are ( 1 ) to assess the quantitative reliability of the extinction correction factor based on the uv data @xcite , ( 2 ) to make quantitative the optically based qualitative arguments on the magnitudes of compact nuclear starbursts @xcite , and ( 3 ) to find weak compact nuclear starbursts whose presence may have been missed with @xmath1-band diagnostics @xcite . the observed sources and their properties are summarized in table [ tbl-1 ] . due to our observing date ( table 2 ) , the observed seyfert 2 nuclei have right ascensions ( r.a . ) of between 7 and 16 hr . all eight seyfert 2 nuclei in this r.a . range in table 1 of @xcite are observed . @xcite explicitly presented the strength of the corrected co indices , after subtracting agn emission , in their table 5 . in total , there are eight seyfert 2 nuclei ( seyfert type 1.52.0 ) in the above r.a . range in the table . of these , ngc 5929 was in the sample of @xcite , leaving seven independent seyfert 2 nuclei . we attempted to observe all seven of these objects during our observing run , but could not observe ugc 6100 . we started observations on mrk 270 and mrk 461 , but based on the first few frames , found that these objects were too faint to obtain spectra of sufficient quality in a reasonable integration time ; we therefore discontinued the observations . thus , four out of the seven seyfert 2 nuclei in table 5 of ivanov et al . were observed . additionally , a 34 @xmath0 m spectrum of the very bright seyfert 2 nucleus ngc 1068 ( r.a . = 2 hr ) , originally presented by @xcite , is used because ngc 1068 is also in the list of @xcite . in summary , 13 seyfert 2 nuclei ( 12 sources with r.a . = 716 hr and ngc 1068 ) were observed . for the seyfert 2 galaxies studied in the optical by @xcite , our selection should not be biased with respect to the presence or absence of compact nuclear starbursts . for the seyfert 2 galaxies studied in the @xmath1-band by @xcite , however , some bias might be present , since seyfert 2 galaxies that are faint at 34 @xmath0 m have been excluded . nevertheless , our 34 @xmath0 m study of these seyfert 2 nuclei can provide useful information on the issue of the energetics of compact nuclear starbursts in seyfert 2 nuclei . an observing log is tabulated in table [ tbl-2 ] . details of the observations of ngc 1068 were described by @xcite , and are not repeated here . the nsfcam grism mode @xcite at irtf on mauna kea , hawaii , was used to obtain 34 @xmath0 m spectra of all the seyfert 2 galaxies except mrk 273 , mrk 463 , and ngc 1068 . the _ hkl _ grism and _ l _ blocker were used with the 4 pixel wide slit (= 1@xmath10 ) . the resulting spectral resolution was @xmath2150 at 3.5 @xmath0 m . mrk 273 and mrk 463 were observed with the cgs4 ( mountain et al . 1990 ) at ukirt on mauna kea , hawaii . the 40 l mm@xmath7 grating with 2 pixel wide slit (= 1@xmath10 ) was used . the resulting spectral resolution was @xmath2750 at 3.5 @xmath0 m . sky conditions were photometric throughout the observations . seeing sizes were 0@xmath1160@xmath119 ( full width at half maximum ) for both the irtf and ukirt observing runs . spectra were obtained toward the flux peak at 34 @xmath0 m . for sources observed with nsfcam , the position angles of the slit were set along the north - south direction . consequently , for mrk 266 , which has two nuclei lying on a southwest to northeast axis , only the southwest seyfert 2 nucleus ( mrk 266sw ; mazzarella & boroson 1993 ) was observed . for mrk 273 and mrk 463 , which were observed with cgs4 , the position angles were set at 35@xmath12 and 90@xmath12 east of north , respectively , so as to observe simultaneously emission from both the double nuclei with a separation of less than 5 arcsec @xcite . a standard telescope nodding technique with a throw of 12 arcsec was employed along the slit to subtract background emission . since 34 @xmath0 m emission from seyfert 2 galaxies is usually dominated by compact nuclear emission @xcite , this throw is believed to be sufficiently large . offset guide stars were used whenever available to achieve high telescope tracking accuracy . for mrk 463 , the 34 @xmath0 m emission was dominated by the eastern nucleus ( mrk 463e ) so that the obtained spectrum is almost equivalent to that of mrk 463e only . for mrk 273 , the double nuclei were not clearly resolvable and so emission from both nuclei was combined to produce a single spectrum . f- to g - type standard stars and one b - type star ( table [ tbl-2 ] ) were observed with almost the same airmass as individual seyfert 2 nuclei , to correct for the transmission of the earth s atmosphere . the @xmath13-band ( 3.5 @xmath0 m ) magnitudes of standard stars were estimated from their @xmath14-band ( 0.6 @xmath0 m ) magnitudes , by adopting the @xmath15 colors appropriate to the stellar types of individual standard stars ( tokunaga 2000 ) . standard data analysis procedures were employed within iraf . first , bad pixels were replaced with the interpolated values of the surrounding pixels . next , bias was subtracted from the obtained frames and the frames were divided by a flat image . the spectra of the targets and the standard stars were then extracted . spectra were extracted by integrating signals over 410 arcsec along the slit ( table [ tbl-3 ] ) , depending on the spatial extent of the actual 34 @xmath0 m signals . for nsfcam data , wavelength calibration was performed using the wavelength - dependent transmission of the earth s atmosphere , while an argon lamp was used for the cgs4 data . the spectra of seyfert 2 nuclei were divided by those of the standard stars , and multiplied by the spectra of blackbodies with temperatures corresponding to individual standard stars ( table [ tbl-2 ] ) . after flux calibration based on the adopted standard star fluxes , the final spectra were produced . nsfcam data at @xmath23.35 @xmath0 m for some sources are contaminated by an uncorrectable ghost image , and so these data points were removed . flux - calibrated spectra are shown in figure [ fig1 ] . the spectra of faint sources have been binned by a few spectral elements . the 34 @xmath0 m spectra of ngc 1068 and mrk 273 , previously presented by @xcite and @xcite respectively , are shown again here . the possible energy sources for seyfert 2 galaxies are agn , compact ( less than a few 100 pc ) nuclear starbursts @xcite , and extended ( kpc scale ) star - formation in the host galaxies . the employed slit width of 1@xmath112 ( except for ngc 1068 ) corresponds to the physical scale of 7090 pc and 1.1 kpc for the nearest ( ngc 3227 and ngc 5033 ; @xmath16 0.004 ) and farthest ( mrk 34 and mrk 463 ; _ z _ = 0.051 ) sources , respectively . therefore , except in ngc 3227 and ngc 5033 , the bulk of the compact ( less than a few 100 pc ) nuclear starburst emission is covered by our observations . although extended ( 410 arcsec ) emission along the slit direction is also included in our spectroscopy , the fraction of the extended emission inside this thin strip is negligible compared to the whole . therefore , the emission directly investigated based on our slit spectra is that due to agn and compact nuclear starburst activity , and the contamination of the emission from extended ( kpc scale ) star - formation activity is negligible . the slit widths employed for the uv , optical , and @xmath1-band observations are 1@xmath1121@xmath117 @xcite , so that these spectra also reflect the properties of the agn and compact nuclear starburst emission . thus , our 34 @xmath0 m slit spectra can be directly compared with uv , optical , and @xmath1-band spectra to investigate the nature of agn and compact nuclear starbursts . the bulk of the agn and compact nuclear starburst emission is , in principle , detectable in the slit . however , some flux from the compact emission could be lost in our slit spectra if the telescope tracking accuracy were insufficient . 34 @xmath0 m emission from seyfert 2 galaxies is generally dominated by compact nuclear emission @xcite . thus , table [ tbl-3 ] compares our spectro - photometric magnitudes with photometry at 34 @xmath0 m made with apertures of 39 arcsec in the literature , in order to estimate what fraction of compact nuclear 34 @xmath0 m emission may have been missed . ( we note , however , that time variability of the nuclear 34 @xmath0 m emission may affect the comparison . ) in table [ tbl-3 ] , our slit spectrum gives a magnitude 1.2 mag fainter than the aperture photometry for ngc 5033 . this is the closest source in our sample , so that it is quite plausible that a significant fraction of the 34 @xmath0 m continuum emission has been missed in this object . however , since ngc 5033 is not used in the systematic discussion ( @xmath17 5 ) , this missing flux for ngc 5033 will not have a significant effect on our main conclusions . with the exception of ngc 5033 , the magnitude difference is less than 0.8 mag , or a factor of less than @xmath22 in all cases . this agreement within a factor of less than 2 implies that our slit spectroscopy detects at least half of the compact nuclear emission ; given that aperture photometry may contain a contribution from extended emission , the actual slit loss for the compact nuclear emission is even smaller . therefore , as far as the compact nuclear emission is concerned , the absolute 3.3 @xmath0 m pah emission fluxes measured with our slit spectra should be certain within a factor of less than 2 . to estimate the flux , luminosity , and rest - frame equivalent width of the 3.3 @xmath0 m pah emission , the spectral profile of type-1 sources @xcite is adopted as a template of the 3.3 @xmath0 m pah emission . this profile can fit well the observed 3.3 @xmath0 m pah emission in some starbursts with spectra of high spectral resolution and high signal - to - noise ratios at @xmath23.3 @xmath0 m ( e.g. , ngc 253 and arp 220 in imanishi & dudley 2000 ; mrk 266sw in fig.1 ) . the rest - frame peak wavelength of the 3.3 @xmath0 m pah emission is assumed to be 3.29 @xmath0 m , and only the normalization is treated as a free parameter . the best linear fit for the data points which are affected by neither emission nor absorption is used as a continuum level . several different continuum levels are adopted , and the uncertainties of the pah fluxes ( luminosities , equivalent widths ) resulting from the continuum ambiguity are also taken into account , particularly for sources with small pah equivalent widths . the estimated values are summarized in table [ tbl-4 ] . if dust grains in the obscuring material of seyfert 2 nuclei are covered with an ice mantle , strong h@xmath18o ice absorption should be detected @xcite . this absorption feature has a peak wavelength of 3.08 @xmath0 m and is spectrally broad , extending from 2.8 to 3.6 @xmath0 m ( e.g. , smith , sellgren , & tokunaga 1989 ) . the estimates of the 3.3 @xmath0 m pah fluxes could be highly uncertain if the absorption is strong . however , most of our spectra cover 3.08 @xmath0 m in the rest - frame , and yet no sign of strong h@xmath18o ice absorption is found . the average spectrum of seyfert 2 galaxies @xcite show no clear h@xmath18o absorption feature either . since the absorption strength is much weaker at @xmath193.25 @xmath0 m than at 3.08 @xmath0 m @xcite , possible presence of currently undetectable , weak h@xmath18o absorption in seyfert 2 nuclei is unlikely to affect the estimates of the 3.3 @xmath0 m pah emission significantly . the fraction of the observed nuclear 34 @xmath0 m fluxes in our slit spectra that originates in starbursts can be estimated from the rest - frame equivalent widths of the 3.3 @xmath0 m pah emission feature ( ew@xmath20 ) . the ew@xmath20 value decreases as the agn contribution increases ( @xmath17 1 ) . the equivalent widths for starburst - dominated galaxies are @xmath2120 nm @xcite . in table [ tbl-4 ] , only mrk 266sw shows an ew@xmath20 value similar to those of starburst galaxies ; we can conclude from this that only the observed nuclear 34 @xmath0 m flux of mrk 266sw is dominated by starbursts . mrk 78 , mrk 273 , mrk 477 , ngc 3227 , and ngc 5135 all show detectable , moderately strong 3.3 @xmath0 m pah emission features , but their ew@xmath20 values are a factor 39 smaller than those of starburst galaxies . for these five sources , starbursts contribute some fraction , but less than half , of the observed nuclear 34 @xmath0 m fluxes , and the bulk of the observed 34 @xmath0 m fluxes originate in agn activity . for the remaining seven seyfert 2 nuclei ( ic 3639 , mrk 34 , mrk 463 , mrk 686 , ngc 1068 , ngc 5033 , and ngc 5929 ) , the ew@xmath20 is more than a factor of 7 smaller than that expected for starburst - dominated galaxies , indicating that the predominant fraction ( larger than 80% ) of their 34 @xmath0 m emission comes from agn activity . _ our first conclusion is that , except in the case of mrk 266sw , the observed compact nuclear 34 @xmath0 m emission is dominated by the agn and not by starbursts_. this result has implications for infrared studies of agns . @xmath13- ( 3.5 @xmath0 m ) and @xmath21-band ( 4.8 @xmath0 m ) emission from agns is usually dominated by compact emission , and photometric data at these bands are used to discuss dust obscuration toward agns , on the assumption that this compact emission comes predominantly from the agn and not from compact nuclear starbursts @xcite . our results provide supporting evidence for this assumption . the 3.4 @xmath0 m carbonaceous dust absorption feature is detected in mrk 463 and ngc 1068 . the observed 3.4 @xmath0 m absorption optical depths are @xmath22(observed ) @xmath6 [email protected] and [email protected] for mrk 463 and ngc 1068 , respectively . in agns , the 34 @xmath0 m continuum emission is dominated by dust in thermal equilibrium with a temperature of 8001000k , close to the dust sublimation temperature , located at the innermost part of the dusty torus @xcite . the dust extinction toward the 34 @xmath0 m continuum emitting region is thus almost the same as that toward the agn itself . assuming a galactic dust model ( @xmath22/a@xmath3 @xmath6 0.0040.007 ; pendleton et al . 1994 ) , the dust extinction toward the agns is estimated to be 611 and 1730 mag for mrk 463 and ngc 1068 , respectively . mrk 463 shows a broad emission component in its near - infrared hydrogen recombination lines @xcite so that dust obscuration toward the agn has been suggested to be relatively modest in this object . our result is consistent with this suggestion . for the remaining 11 seyfert 2 nuclei , no detectable 3.4 @xmath0 m absorption feature is found . the 3.4 @xmath0 m absorption feature may be suppressed if the bulk of the dust grains in the obscuring material is covered with an ice mantle @xcite . however , no strong 3.08 @xmath0 m ice absorption feature is found in their 34 @xmath0 m spectra . the dust absorption feature is smeared out if less obscured starburst emission contributes significantly to the observed nuclear 34 @xmath0 m flux @xcite . the six sources with detectable pah emission ( mrk 78 , mrk 266sw , mrk 273 , mrk 477 , ngc 3227 , and ngc 5135 ) could be this case . for the five pah - undetected seyfert 2 nuclei ( ic 3639 , mrk 34 , mrk 686 , ngc 5033 , and ngc 5929 ) , the 3.4 @xmath0 m absorption feature could remain undetectable if dust obscuration toward these agns is weak ( @xmath17 1 ) . the r - value , defined as @xmath24 is a good measure for dust extinction toward agns ( murayama , mouri , & taniguchi 2000 ) . it becomes smaller with higher dust extinction toward agns , and the boundary value between seyfert 1 and 2 nuclei is estimated to be r = @xmath250.6 @xcite . based on the 3.5 @xmath0 m ( table [ tbl-3 ] ) and _ iras _ 25 @xmath0 m ( table [ tbl-1 ] ) fluxes , the r - values for ic 3639 , mrk 34 , mrk 686 , ngc 5033 , and ngc 5929 , are estimated to be @xmath251.2 , @xmath250.8 , @xmath250.3 , @xmath250.6 , and @xmath251.2 , respectively . the r - values of mrk 686 and ngc 5033 are in the range of seyfert 1 nuclei , and thus weak dust obscuration is indicated , explaining the non - detection of the 3.4 @xmath0 m absorption . however , ic 3639 , mrk 34 , and ngc 5929 show small r - values . the non - detection could be caused by the effects of time variability or contamination from extended star - formation to the observed 25 @xmath0 m flux taken with large apertures @xcite . the bulk of the uv optical emission from the agns in seyfert 2 galaxies and starbursts @xcite is absorbed by dust and re - emitted as dust thermal emission in the infrared . infrared ( 81000 @xmath0 m ) luminosities are thus used to estimate the magnitudes of agn and starburst activity throughout this paper . infrared ( 81000 @xmath0 m ) luminosities are obviously better for evaluating these kinds of activity than far - infrared ( 40500 @xmath0 m ) luminosities because not all activity shows dust emission peaking in the far - infrared . in starburst - dominated galaxies , the 3.3 @xmath0 m pah to far - infrared ( 40500 @xmath0 m ) luminosity ratios ( l@xmath20/l@xmath26 ) are @xmath21 @xmath4 10@xmath27 @xcite . the 3.3 @xmath0 m pah to infrared ( 81000 @xmath0 m ) luminosity ratio ( l@xmath28/l@xmath29 ) is tentatively assumed to be also 1 @xmath4 10@xmath27 for the compact nuclear starbursts in seyfert 2 nuclei . the scatter of the pah - to - infrared luminosity ratio is likely to be a factor of 23 toward both higher and lower values around a typical value @xcite . for ic 3639 , mrk 477 , and ngc 5135 , @xcite explicitly estimated the extinction - corrected luminosities of compact nuclear starbursts , based on their uv data . the respective luminosities are 1.1 @xmath4 10@xmath30 , 1.4 @xmath4 10@xmath31 , and 4.2 @xmath4 10@xmath30 ergs s@xmath7 . the infrared luminosities of the compact nuclear starbursts estimated from the observed ( extinction - uncorrected ) 3.3 @xmath0 m pah emission , based on the assumption of l@xmath20/l@xmath29 @xmath2 1 @xmath4 10@xmath27 , are @xmath321.2 @xmath4 10@xmath30 , 1.5 @xmath4 10@xmath31 , and 5.8 @xmath4 10@xmath30 ergs s@xmath7 , respectively , for ic 3639 , mrk 477 , and ngc 5135 . it is noteworthy that the compact nuclear starburst luminosities estimated from uv data after extinction correction and from the 3.3 @xmath0 m pah emission with _ no _ extinction correction are in satisfactory agreement for all three sources . this agreement indicates that ( 1 ) discussions of the magnitudes of compact nuclear starbursts based on uv data after extinction correction are quantitatively reliable , ( 2 ) the extinction correction factor is not significant at 34 @xmath0 m , and ( 3 ) the assumption of l@xmath20/l@xmath29 @xmath2 1 @xmath4 10@xmath27 is reasonable . we have been concerned by the possibility that , if compact nuclear starbursts close to the central agns are directly exposed to x - ray emission from agns , pah emission from the compact nuclear starbursts might be suppressed due to the destruction of pahs @xcite . however , the agreement suggests that this is not the case . if the bulk of the compact nuclear starbursts in seyfert 2 nuclei occur at the outer reaches of the dusty tori around agns @xcite , pahs are shielded from the energetic radiation from the agns and thus are not destroyed , explaining the strong pah emission from compact nuclear starbursts in seyfert 2 nuclei . based on these findings , the magnitudes of compact nuclear starbursts can be estimated from the observed 3.3 @xmath0 m pah emission luminosities with some confidence . if compact nuclear starbursts energetically dominated the whole galactic infrared dust emission luminosities of seyfert 2 galaxies , the observed l@xmath20/l@xmath29 ratios measured from our slit spectra should be similar to those of starburst galaxies ( @xmath21 @xmath4 10@xmath27 ) . as shown in table [ tbl-4 ] , however , none of the observed seyfert 2 galaxies show such a large value of l@xmath20/l@xmath29 . the differences are by a factor of 3 ( mrk 477 ) to larger than 10 , and in most cases much larger than a factor of 23 , the possible flux loss in our slit spectra and the intrinsic scatter of the l@xmath20/l@xmath33 ratios for starbursts . therefore , _ our second conclusion is that compact nuclear starbursts can not be the dominant energy source of the whole galactic infrared dust emission luminosities in the majority of the observed seyfert 2 galaxies_. for the remaining ten seyfert 2 galaxies other than ic 3639 , mrk 477 , and ngc 5135 , our quantitative estimates of the magnitudes of the compact nuclear starbursts are compared with qualitative arguments based on the optical and @xmath1-band data @xcite . @xcite found signatures of compact nuclear starbursts ( young stars ) in the optical spectrum of mrk 273 . for mrk 273 , 3.3 @xmath0 m pah emission is detected ; ew@xmath20 and l@xmath20/l@xmath29 are a factor of @xmath23 and @xmath210 smaller , respectively , than those of starburst galaxies . the l@xmath20/l@xmath29 ratio indicates that the compact nuclear starbursts explain @xmath21/10 of the total galactic infrared luminosity . thus , compact nuclear starbursts are certainly present , supporting the optical results . for mrk 78 and ngc 1068 , @xcite found no signatures of compact nuclear starbursts , and detected only intermediate age stars in their optical spectra . for ngc 1068 , only stringent upper limits are found for the ew@xmath20 and l@xmath20/l@xmath29 , both of which are more than an order of magnitude smaller than those of starburst galaxies . for mrk 78 , however , 3.3 @xmath0 m pah emission is marginally detected . for mrk 463 , @xcite claimed that compact nuclear starbursts might be present , subject to further confirmation . no detectable 3.3 @xmath0 m pah emission is found . the upper limits for both ew@xmath20 and l@xmath20/l@xmath29 are more than an order of magnitude smaller than starburst galaxies . compact nuclear starbursts in mrk 463 are energetically insignificant , if they exist at all . for mrk 34 , @xcite found no signs of compact nuclear starbursts in the optical spectrum , with only old population stars being detected . no 3.3 @xmath0 m pah emission is detected either , supporting the conclusions based on the optical data . both @xcite and @xcite studied ngc 5929 and found no evidence for strong compact nuclear starbursts . no clear 3.3 @xmath0 m pah emission is detected . the remaining sources , mrk 266sw , mrk 686 , ngc 3227 , and ngc 5033 , were studied only by @xcite . they found that no strong compact nuclear starbursts are required in terms of the co indices after correction for agn emission . no 3.3 @xmath0 m pah emission is detected in mrk 686 and ngc 5033 , supporting their argument . however , in mrk 266sw , the nuclear 34 @xmath0 m emission must be dominated by starbursts , although these detected compact nuclear starbursts can explain only @xmath210% of the total galactic infrared luminosity . 3.3 @xmath0 m pah emission was also clearly detected in ngc 3227 , and compact nuclear starbursts can explain @xmath220% of the whole galactic infrared luminosity . in mrk 266sw and ngc 3227 , compact nuclear starbursts are certainly present , and they explain a non - negligible fraction of the total galactic infrared luminosities . the signatures of compact nuclear starbursts may have been missed during the complicated procedures required to find their weak signatures in the @xmath1-band @xcite . since the observed nuclear 34 @xmath0 m emission is dominated by agn activity ( @xmath17 5.1 ) , it would be expected that agn activity also contributes much more than compact nuclear starbursts to the net galactic infrared ( 81000 @xmath0 m ) luminosities of seyfert 2 galaxies . however , the agn - driven infrared luminosities of these objects are difficult to estimate because they are highly dependent on the spatial distribution of dust in the dusty torus , which is poorly constrained observationally . thus , we estimate the energetic importance of extended star - formation in the host galaxies , which is the remaining probable power source of seyfert 2 galaxies other than compact nuclear starbursts and agns , by using the pah emission . pah emission can be produced both with starbursts and normal quiescent star - formation in a similar way @xcite . to investigate the magnitude of extended ( kpc scale ) star - formation activity , _ infrared space observatory ( iso ) _ spectra taken with large apertures are useful . @xcite summarized the _ iso _ results ; they found that ( 1 ) the 7.7 @xmath0 m pah to far - infrared luminosity ratios in seyfert galaxies are similar to those of starburst galaxies , and that ( 2 ) the bulk of the pah emission is spatially extended . therefore , they argued that the bulk of the far - infrared emission from seyfert galaxies originates in extended star - formation activity . of the 13 seyfert 2 galaxies , @xcite have presented _ iso _ 2.511 @xmath0 m spectra taken with 24 @xmath4 24 arcsec@xmath34 apertures , and quote pah fluxes for four sources ( mrk 266 , ngc 3227 , ngc 5033 , and ngc 5929 ) . for these sources , the energetic importance of extended star - formation activity can be investigated by comparing the measured pah fluxes in our slit spectra with those in the _ iso _ spectra . the effects of dust extinction are similar at 38 @xmath0 m @xcite . the 7.7 @xmath0 m pah emission is the strongest pah emission feature . however , its flux estimates in the _ iso _ spectra may be highly uncertain @xcite , due to the presence of strong , spectrally broad 9.7 @xmath0 m silicate dust absorption feature and insufficient wavelength coverage longward of these emission and absorption features , which make a continuum determination difficult @xcite . quantitatively reliable flux estimates of the 3.3 @xmath0 m pah emission in the _ iso _ spectra are also difficult due to the scatter in the data points at 34 @xmath0 m ( clavel et al . 2000 , their figure 8) . consequently , the 6.2 @xmath0 m pah emission , which is isolated and moderately strong , is most suitable to investigate the extended star - formation activity based on the _ iso _ spectra @xcite . the 6.2 @xmath0 m pah to infrared luminosity ratios ( l@xmath35/l@xmath29 ) for starbursts are estimated to be @xmath2 6 @xmath4 10@xmath27 , with a scatter of a factor of 23 toward both higher and lower values @xcite , where it is assumed that l@xmath29 @xmath2 l@xmath26 for starbursts . the respective l@xmath35/l@xmath29 ratios are 4 @xmath4 10@xmath27 , 4 @xmath4 10@xmath27 , 3 @xmath4 10 @xmath27 , and 1 @xmath4 10@xmath27 , for mrk 266 , ngc 3227 , ngc 5033 , and ngc 5929 @xcite , roughly half of ( mrk 266 , ngc 3227 , and ngc 5033 ) or more than six times smaller than ( ngc 5929 ) the typical value for systems dominated by starbursts . since the observed l@xmath20/l@xmath29 ratios indicate that the compact nuclear starbursts energetically fall short of the total infrared luminosities by a factor of larger than 6 for these four sources ( table [ tbl-4 ] ) , it can be said that extended star - formation ( but not compact nuclear starbursts ) contributes significantly to the infrared luminosities of these seyfert 2 galaxies . this statement was made by @xcite for seyfert galaxies as a whole based on the 7.7 @xmath0 m pah emission . based on the 6.2 @xmath0 m pah emission , here , it has been confirmed that this is true also for the four individual seyfert 2 galaxies , and probably for seyfert 2 galaxies as a whole . the individual l@xmath35/l@xmath29 ratios for mrk 266 , ngc 3227 , and ngc 5033 are within the scattered range of the ratios for starbursts . however , the overall trend of lower values implies that agn activity also makes an important contribution to the infrared luminosities of seyfert 2 galaxies . while both starbursts and quiescent disk star - formation in normal galaxies can produce pah emission ( @xmath17 5.5 ) , strong soft x - ray emission driven by superwind is observed only if the star - formation rate per unit area exceeds a certain threshold ( 10@xmath7 m@xmath36 yr@xmath7 kpc@xmath37 ; heckman 2000 ) . starbursts surpass this threshold , while quiescent normal disk star - formation does not @xcite . these different characteristics can be used to understand the properties of energetically significant extended star - formation activity . of the 13 seyfert 2 galaxies , @xcite have estimated superwind - driven soft x - ray luminosities for eight sources ( ic 3639 , mrk 78 , mrk 266 , mrk 273 , mrk 463 , mrk 477 , ngc 1068 , and ngc 5135 ) , and found that , as a whole , their soft x - ray to far - infrared luminosity ratios are as high as those of starburst galaxies . therefore , starburst activity is a significant contributor to the far - infrared emission and also to the infrared dust emission . since compact nuclear starbursts detected in our slit spectra were found to be energetically insignificant ( @xmath17 5.3 ) , the energetically significant starbursts must be extended . _ our third conclusion is that the bulk of the energetically significant extended ( kpc scale ) star - formation activity is of starburst - type and not quiescent normal disk star - formation . it is the extended ( kpc scale ) starbursts , rather than the compact ( less than a few 100 pc ) nuclear starbursts , that are responsible for the superwind - driven soft x - ray emission . _ if starbursts are inevitable phenomena in seyfert 2 galaxies , their extended starburst activity is energetically more important than their compact nuclear starbursts . this is actually the case for the four seyfert 2 galaxies studied in detail in the uv @xcite and for the famous , well - studied seyfert 2 galaxy ngc 1068 @xcite . we test the hypothesis that more powerful agns might be related to more powerful compact nuclear starbursts @xcite . figure [ fig2 ] compares the 12 @xmath0 m luminosity with the 3.3 @xmath0 m pah emission luminosity . the total galactic 12 @xmath0 m luminosity is regarded as a measure of agn power @xcite . the 3.3 @xmath0 m pah luminosities , measured with our slit spectroscopy , reflect the magnitudes of compact nuclear starbursts . the picture of @xcite would be supported if we were to find a positive correlation between 12 @xmath0 m luminosity and 3.3 @xmath0 m pah emission luminosity . in fig . [ fig2 ] , this trend appears to be present , although the scatter is moderately large . we apply the generalized kendall s rank correlation statistic @xcite to the data points in fig . [ fig2 ] , and estimate the probability that a correlation is not present to be 0.11 , by using the software available at the web page : http://www.astro.psu.edu / statcodes/. thus , provided that the 12 @xmath0 m luminosity is a good measure of agn power , _ it is found that the luminosities of agns and compact nuclear starbursts in seyfert 2 galaxies are correlated , and more powerful agns tend to contain more powerful compact nuclear starbursts . _ we performed 34 @xmath0 m spectroscopy of 13 seyfert 2 nuclei that were previously studied in the uv , optical , and near - infrared @xmath1-band . making use of the 3.3 @xmath0 m pah emission feature detected in our slit spectra , the following results were obtained . 1 . our pah - based , extinction - uncorrected estimates of the luminosities of compact ( less than a few 100 pc ) nuclear starbursts were found to agree well with those based on extinction - corrected uv data in three selected seyfert 2 galaxies ( ic 3639 , mrk 477 , and ngc 5135 ) . this agreement indicates that compact nuclear starburst emission is not significantly attenuated due to dust extinction at 34 @xmath0 m , so that the observed 3.3 @xmath0 m pah emission luminosity is a powerful diagnostic of the magnitudes of compact nuclear starbursts . 2 . for the remaining ten seyfert 2 galaxies , our pah - based diagnostic results were compared with qualitative arguments based on the optical and @xmath1-band spectra . for mrk 273 , signs of compact nuclear starbursts were found , supporting the qualitative optically based argument . for mrk 463 , possible compact nuclear starbursts implied from the optical data were not confirmed at 34 @xmath0 m . for mrk 78 and ngc 1068 , both of which show signatures of only intermediate age stars in the optical , 3.3 @xmath0 m pah emission was marginally detected in mrk 78 . optical and/or @xmath1-band spectra show no evidence for compact nuclear starbursts in mrk 34 , mrk 266sw , mrk 686 , ngc 3227 , ngc 5033 , and ngc 5929 . at 34 @xmath0 m , signatures of compact nuclear starbursts were found in mrk 266sw and ngc 3227 . quantitative estimates of the magnitudes of nuclear starbursts were made based on the observed 3.3 @xmath0 m pah emission luminosities for the 13 seyfert 2 galaxies , and found that ( 1 ) nuclear 34 @xmath0 m emission is dominated by agn activity rather than by starburst activity in all seyfert 2 galaxies but one ( mrk 266sw ) , and ( 2 ) compact nuclear starbursts can explain only a small fraction ( @xmath321/101/3 ) of the infrared dust emission luminosities . the 3.4 @xmath0 m carbonaceous dust absorption feature is detected only in the two seyfert 2 nuclei , mrk 463 and ngc 1068 . when our observations were combined with large - aperture _ iso _ spectra for the four sources ( mrk 266 , ngc 3227 , ngc 5033 , and ngc 5929 ) , it was confirmed that extended ( kpc scale ) star - formation activity is energetically more important than compact nuclear starbursts , and contributes significantly to the infrared dust emission luminosities of seyfert 2 galaxies . however , the agns are likely to be still important contributors to the luminosities . 6 . making use of soft x - ray data for the eight sources ( ic 3639 , mrk 78 , mrk 266 , mrk 273 , mrk 463 , mrk 477 , ngc 1068 , and ngc 5135 ) , it was suggested that the energetically significant extended star - formation activity is of starburst - type and not quiescent normal disk star - formation . these extended starbursts are responsible for the superwind - driven soft x - ray emission from seyfert 2 galaxies . we find evidence supporting the scenario in which more powerful agns are related to more powerful compact nuclear starbursts , provided that 12 @xmath0 m luminosity is a good measure of agn power . we thank s. k. leggett , j. davies , t. wold , and t. carroll for their support during the ukirt observing runs , and j. rayner and b. golisch for their support before and during the irtf observing run . we are grateful to drs . k. aoki , t. nakajima , and y. p. wang for their useful comments on this manuscript , and dr . t. t. takeuchi for invaluable discussions about statistical tests . the anonymous referee and the scientific editor , dr . s. willner , gave invaluable comments , which improved this paper significantly . spectra of mrk 273 and mrk 463 were obtained while mi was at the university of hawaii . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . alonso - herrero , a. , simpson , c. , ward , m. j. , & wilson , a. s. 1998 , apj , 495 , 196 alonso - herrero , a. , quillen , a. c. , simpson , c. , efstathiou , a. & ward , m. j. 2001 , aj , 121 , 1369 antonucci , r. 1993 , ara&a , 31 , 473 baker , a. j. , lutz , d. , genzel , r. , tacconi , l. j. , & lehnert , m. d. 2001 , a&a , 372 , l37 calzetti , d. , armus , l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , & storchi - bergmann , t. 2000 , apj , 533 , 682 cid fernandes , r. jr . , & terlevich , r. 1995 , mnras , 272 , 423 clavel , j. et al . 2000 , a&a , 357 , 839 dudley , c. c. , 1999 , mnras , 307 , 553 fischer , j. 2000 , in iso beyond the peaks , ed . a. salama , m. f. kessler , k. , leech , & b. schulz ( esa sp-456 ; noordwijk : esa ) , 239 ( astro - ph/0009395 ) genzel , r. , & cesarsky , c. j. 2000 , ara&a , 38 , 761 glass , i. s. , & moorwood , a. f. m. 1985 , mnras , 214 , 429 gonzalez delgado , r. m. , heckman , t. , leitherer , c. , meurer , g. , krolik , j. , wilson , a. s. , kinney , a. , & koratkar , a. 1998 , apj , 505 , 174 gonzalez delgado , r. m. , heckman , t. , & leitherer , c. 2001 , apj , 546 , 845 goodrich , r. w. , veilleux , s. , & hill , g. j. 1994 , apj , 422 , 521 heckman , t. m. 1999 , astro - ph/9912029 heckman , t. m. 2000 , astro - ph/0009075 heckman , t. m. , gonzalez delgado , r. , leitherer , c. , meurer , g. r. , krolik , j. , wilson , a. s. , koratkar , a. , & kinney , a. 1997 , apj , 482 , 114 helou , g. , lu , n. y. , werner , m. w. , malhotra , s. , & silbermann , n. 2000 , apj , 532 , l21 imanishi , m. 2000 , mnras , 319 , 331 imanishi , m. 2001 , aj , 121 , 1927 imanishi , m. , & dudley , c. c. 2000 , apj , 545 , 701 imanishi , m. , terada , h. , sugiyama , k. , motohara , k. , goto , m. , & maihara , t. 1997 , pasj , 49 , 69 isobe , t. , feigelson , e. d. , & nelson , p. i. 1986 , apj , 306 , 490 ivanov , v. d. , rieke , g. h. , groppi , c. e. , alonso - herrero , a. , rieke , m. j. , & engelbracht , c. w. 2000 , apj , 545 , 190 kennicutt , r. 1998 , apj , 498 , 541 laureijs , r. j. et al . 2000 , a&a , 359 , 900 lawrence , a. , ward , m. , elvis , m. , fabbiano , g. , willner , s. p. , carleton , n. p. , & longmore , a. 1985 , apj , 291 , 117 le floch , e. , mirabel , i. f. , laurent , o. , charmandaris , v. , gallais , p. , sauvage , m. , vigroux , l. , & cesarsky , c. 2001 , a&a , 367 , 487 levenson , n. a. , weaver , k. a. , & heckman , t. m. 2001 , apj , 550 , 230 lutz , d. et al . 1996 , a&a , 315 , l269 marco , o. , & alloin , d. 2000 , a&a , 353 , 465 mazzarella , j. m. , & boroson , t. a. 1993 , apjs , 85 , 27 mazzarella , j. m. , gaume , r. a. , soifer , b. t. , graham , j. r. , neugebauer , g. , & matthews , k. 1991 , aj , 102 , 1241 mennella , v. , munoz caro , g. m. , ruiterkamp , r. , schutte , w. a. , greenberg , j. m. , brucato , j. r. , & colangeli , l. 2001 , a&a , 367 , 355 moorwood , a. f. m. 1986 , a&a , 166 , 4 mountain , c. m. , robertson , d. j. , lee , t. j. , & wade , r. 1990 , proc . spie , 1235 , 25 mouri , h. , kawara , k. , taniguchi , y. , & nishida , m. 1990 , apj , 356 , l39 murayama , t. , mouri , h. , & taniguchi , y. 2000 , apj , 528 , 179 pendleton , y. j. , sandford , s. a. , allamandola , l. j. , tielens , a. g. g. m. , & sellgren , k. 1994 , apj , 437 , 683 rieke , g. h. 1978 , apj , 226 , 550 rieke , g. h. , & lebofsky , m. j. 1985 , apj , 288 , 618 sanders , d. b. , & mirabel , i. f. 1996 , ara&a , 34 , 749 savage , b. d. , & mathis , j. s. 1979 , ara&a , 17 , 73 scoville , n. z. et al . 2000 , aj , 119 , 991 shure , m. a. , toomey , d. w. , rayner , j. t. , onaka , p. , & denault , a. j. 1994 , proc . spie , 2198 , 614 simpson , c. 1998 , apj , 509 , 653 smith , r. g. , sellgren , k. , & tokunaga , a. t. 1989 , apj , 344 , 413 spoon , h. w. w. , koornneef , j. , moorwood , a. f. m. , lutz , d. , & tielens , a. g. g. m. 2000 , a&a , 357 , 898 storchi - bergmann , t. , raimann , d. , bica , e. l. d. , & fraquelli , h. a. 2000 , apj , 544 , 747 surace , j. a. , & sanders , d. b. 1999 , apj , 512 , 162 surace , j. a. , sanders , d. b. , & evans , a. s. 2000 , apj , 529 , 170 tokunaga , a. t. , sellgren , k. , smith , r. g. , nagata , t. , sakata , a. , & nakada , y. 1991 , apj , 380 , 452 tokunaga , a. t. 2000 , in allen s astrophysical quantities , ed . a. n. cox ( 4th ed : aip press : springer ) , chapter 7 , p.143 veilleux , s. , goodrich , r. w. , & hill , g. j. 1997 , apj , 477 , 631 voit , g. m. 1992 , mnras , 258 , 841 zhou , s. , wynn - williams , c. g. , & sanders , d. b. 1993 , apj , 409 , 149 lcrrrrccc ic 3639 & 0.011 & 0.64 & 2.26 & 7.52 & 10.7 & 44.25 & 44.33 & g + mrk 34 & 0.051 & 0.07 & 0.46 & 0.81 & 0.80 & 44.59 & 44.78 & g + mrk 78 & 0.037 & 0.13 & 0.56 & 1.11 & 1.13 & 44.44 & 44.63 & g + mrk 266 & 0.028 & 0.23 & 0.98 & 7.34 & 11.1 & 45.07 & 45.03 & i + mrk 273 & 0.038 & 0.24 & 2.28 & 21.7 & 21.4 & 45.75 & 45.69 & g + mrk 463 & 0.051 & 0.51 & 1.58 & 2.18 & 1.92 & 45.00 & 45.34 & g + mrk 477 & 0.038 & 0.13 & 0.51 & 1.31 & 1.85 & 44.58 & 44.70 & g + mrk 686 & 0.014 & @xmath320.11 & 0.13 & 0.57 & 1.79 & 43.49 & 43.4343.56 & i + ngc 1068 & 0.004 & 39.70 & 85.04 & 176.2 & 224.0 & 44.72 & 44.96 & g + ngc 3227 & 0.004 & 0.67 & 1.76 & 7.83 & 17.6 & 43.46 & 43.49 & i + ngc 5033 & 0.003 & 0.95 & 1.15 & 13.8 & 43.9 & 43.54 & 43.48 & i + ngc 5135 & 0.014 & 0.64 & 2.40 & 16.9 & 28.6 & 44.84 & 44.81 & g + ngc 5929 & 0.008 & 0.43 & 1.62 & 9.14 & 13.7 & 44.06 & 44.06 & g , i + llccccc ic 3639 & 2001 apr 8 & 1920 & hr 5212 & 4.8 & f7v & 6240 + mrk 34 & 2001 apr 9 & 2520 & hr 4112 & 3.4 & f8v & 6200 + mrk 78 & 2001 apr 9 & 1728 & hr 3028 & 4.8 & f6v & 6400 + mrk 266sw & 2001 apr 9 & 2160 & hr 4767 & 4.8 & f8v g0v & 6000 + mrk 273 & 2000 feb 20 & 2560 & hr 4761 & 4.8 & f68v & 6200 + mrk 463 & 2000 jun 13 & 800 & hr 5243 & 4.9 & f6v & 6400 + mrk 477 & 2001 apr 9 & 2592 & hr 5581 & 4.3 & f7v & 6240 + mrk 686 & 2001 apr 9 & 2400 & hr 5423 & 4.7 & g5v & 5700 + ngc 1068 & 1995 nov 15 & 640 & hr 8781 & 2.6 & b9v & 10700 + ngc 3227 & 2001 apr 8 & 1050 & hr 3650 & 4.6 & g9v & 5400 + ngc 5033 & 2001 apr 8 & 2280 & hr 5423 & 4.7 & g5v & 5700 + ngc 5135 & 2001 apr 8 & 2160 & hr 5212 & 4.8 & f7v & 6240 + ngc 5929 & 2001 apr 9 & 2064 & hr 5581 & 4.3 & f7v & 6240 + lccc ic 3639 & 10.2 ( 1@xmath112 @xmath4 5@xmath38 ) & 10.4 ( 6@xmath38 ) & g + mrk 34 & 11.3 ( 1@xmath112 @xmath4 4@xmath38 ) & 11.1 ( 8@xmath115 ) & r + mrk 78 & 11.0 ( 1@xmath112 @xmath4 5@xmath38 ) & 11.0 ( 5@xmath119 ) & r + mrk 266sw & 12.1 ( 1@xmath112 @xmath4 5@xmath38 ) & 11.5 ( 8@xmath116 ) & i + mrk 273 & 10.5 ( 1@xmath112 @xmath4 10@xmath38 ) & 10.5 ( 5@xmath38 ) & z + mrk 463 & 8.2 ( 1@xmath112 @xmath4 7@xmath38 ) & 8.2 ( 4@xmath38 ) & m + mrk 477 & 11.1 ( 1@xmath112 @xmath4 4@xmath38 ) & 11.8 ( 6@xmath38 ) & l + mrk 686 & 11.0 ( 1@xmath112 @xmath4 5@xmath38 ) & 11.2 ( 5@xmath114 ) & i + ngc 1068 & 4.8 ( 3@xmath118 @xmath4 3@xmath118 ) & 4.8 ( 3@xmath38 ) & ma + ngc 3227 & 9.5 ( 1@xmath112 @xmath4 4@xmath38 ) & 9.0 ( 8@xmath116 ) & i + ngc 5033 & 10.7 ( 1@xmath112 @xmath4 5@xmath38 ) & 9.5 ( 8@xmath116 ) & i + ngc 5135 & 10.1 ( 1@xmath112 @xmath4 6@xmath38 ) & 9.7 ( 6@xmath38 ) & g + ngc 5929 & 11.6 ( 1@xmath112 @xmath4 5@xmath38 ) & 10.8 ( 5@xmath114 ) & i + lcccc ic 3639 & @xmath324.9 & @xmath3211.3 & @xmath325.2 @xmath4 10@xmath37 & @xmath328.5 @xmath4 10@xmath37 + mrk 34 & @xmath320.33 & @xmath3217.5 & @xmath322.9 @xmath4 10@xmath37 & @xmath321.6 @xmath4 10@xmath37 + mrk 78 & [email protected] & 108@xmath2327 & 2.5 @xmath4 10@xmath7 & [email protected] @xmath4 10@xmath7 + mrk 266sw & [email protected] & 133@xmath237 & 1.2 @xmath4 10@xmath7 & [email protected] @xmath4 10@xmath7 + mrk 273 & [email protected] & 422@xmath2319 & 8.7 @xmath4 10@xmath37 & [email protected] @xmath4 10@xmath7 + mrk 463 & @xmath323.0 & @xmath32159 & @xmath327.4 @xmath4 10@xmath37 & @xmath328.1 @xmath4 10@xmath27 + mrk 477 & [email protected] & 151@xmath2321 & 3.0 @xmath4 10@xmath7 & [email protected] @xmath4 10@xmath7 + mrk 686 & @xmath323.3 & @xmath3212 & @xmath324.6 @xmath4 10@xmath7 & @xmath321.2 @xmath4 10@xmath7 + ngc 1068 & @xmath3286.7 & @xmath3226.7 & @xmath322.9 @xmath4 10@xmath37 & @xmath321.2 @xmath4 10@xmath37 + ngc 3227 & [email protected] & [email protected] & 1.7 @xmath4 10@xmath7 & [email protected] @xmath4 10@xmath7 + ngc 5033 & @xmath320 & @xmath320 & @xmath320 & @xmath320 + ngc 5135 & [email protected] & 58@xmath2311 & 9.1 @xmath4 10@xmath37 & [email protected] @xmath4 10@xmath7 + ngc 5929 & @xmath322.5 & @xmath323.0 & @xmath322.7 @xmath4 10@xmath37 & @xmath321.4 @xmath4 10@xmath7 +
this article describes a new , fully adaptive particle - multiple - mesh ( pm ) numerical simulation code developed primarily for cosmological applications . the code integrates the equations of motion of a set of particles subject to their mutual gravitational interaction and to an optional , arbitrary external field . the interactions between particles are computed using a hierarchy of nested grids constructed anew at each integration step to enhance the spatial resolution in high - density regions of interest . as the code is aimed at simulations of relatively small volumes of space ( not much larger than a single group of galaxies ) with independent control over the external tidal fields , significant effort has gone into supporting isolated boundary conditions at the top grid level . this makes our method also applicable to non - cosmological problems , at the cost of some complications which we discuss . we point out the implications of some differences between our approach and those of other authors of similar codes , in particular with respect to the handling of the interface between regions of different spatial resolution . we present a selection of tests performed to verify the correctness and performance of our implementation . the conclusion suggests possible further improvements in the areas of independent time steps and particle softening lengths .
astro-ph9607156
ideally , a good cosmological @xmath1-body simulation should resolve in some detail the internal structure of individual galaxies ( on mass scales as low as @xmath2 to @xmath3 and on length scales of a few kiloparsecs or less ) while at the same time representing the growth of density perturbations on the scale of clusters and even superclusters of galaxies . galaxy morphologies are observed to correlate with their membership in clusters . the virgo cluster is estimated to be an appreciable , if not dominant , source of tides on the local group 15 to 20 mpc away . and when studying the nonlinear growth of galaxy - sized perturbations , one should not discount the possible coupling to modes with wavelengths as large as 50 mpc ( @xcite ) . achieving the required dynamic range is a difficult task given the limitations of present - day computers . generally , numerical codes have been able to obtain a large dynamic range either in mass or in length , but not in both . particle - particle ( pp ) methods ( @xcite ) and tree codes ( appel 1981 , 1985 ; @xcite ; @xcite ) have essentially unlimited length resolution , bounded only by the need to soften two - body encounters when modeling a collisionless physical system , but their use for simulations with more than a few tens , respectively hundreds , of thousands of particles is currently restricted to special - purpose hardware ( such as grape chips ) or massively parallel computers ( @xcite ) . particle - mesh ( pm ) methods ( @xcite , hereafter he ; @xcite ) , by contrast , can integrate the orbits of millions of interacting particles even on a more modest single - processor mainframe or high - end workstation . the forces are computed by solving poisson s equation on a cartesian grid . barring complications such as a memory latency that increases with problem size , the computational cost per time step is linear in the number of particles and of order @xmath4 in the number of grid cells ( which is typically of the same order as the number of particles ) . the dynamic range in length , however , is limited by the maximum size of the grid , typically 256 or 512 cells on a side in three dimensions on a current single - processor computer . parallel processing can naturally push this limit towards higher values . one can circumvent this obstacle by decomposing the inter - particle forces into a long - range part , calculated by the pm technique , and a small - range part which can be evaluated in a number of different ways . for some time , a popular approach has been to compute the short - range forces by direct summation over all sufficiently close pairs of particles . this technique , known as the particle - particle , particle - mesh ( p@xmath5 m ) method ( he 8) , was originally applied to plasma simulations where the electrostatic repulsion between charges of the same sign makes it difficult for large density contrasts to develop . its use with gravitating systems suffers from the tendency for ever more particles to condense into small volumes , causing the cost of the pp summation to become prohibitive as the system evolves . increasing the total number of particles and resolving smaller scales ( on which density perturbations become nonlinear at earlier times , at least in the presently favored `` bottom - up '' scenario ) both exacerbate the problem . we are unaware of any p@xmath5 m simulations using more than about @xmath6 particles . one may be able to raise this limit by using a tree method to compute the short - range forces , as in xu s ( 1995 ) tpm , but in principle the difficulty remains . most other attempts to enhance the spatial resolution of pm codes rely on the introduction of local grid refinements . this idea lies at the root of couchman s ( 1991 ) adaptive p@xmath5 m ( ap@xmath5 m ) ( which however still uses direct summation where the number of particles is small enough ) , of numerous hierarchical pm codes ( @xcite ; @xcite ; @xcite ; @xcite , hereafter anc ; @xcite ; @xcite , hereafter jdc ; @xcite ) and of various related approaches ( @xcite ; @xcite ) . as pointed out by various authors , and most recently by suisalu and saar ( 1995b ) , the choice between pm , p@xmath5 m , and tree codes is not only a question of the spatial and mass resolution that can be achieved , but also of the behavior of each method with respect to two - body gravitational collisions . these are to be avoided when the physical system being modeled is essentially collisionless , as is the case for example for cosmological dark matter . pm codes usually prove to be less collisional , although it should be emphasized that this is at least in part a consequence of their spatial resolution being weaker than their mass resolution , and need not carry over to hierarchical pm methods unless precautions are taken to ensure that the mass granularity remains adequate at all times . the purpose of this article is to document in some detail our implementation of a dynamically adaptive multiple - mesh code and the tests we performed to validate it . so far , only a brief report of an earlier stage of development ( @xcite ) and a description of the methodology but not of the tests ( @xcite ) have appeared . our code , like that of jdc , is able to track the formation and subsequent motion of density concentrations by dynamically adjusting the number , nesting depth and location of the subgrids . in our case , the entire grid structure is chosen afresh on every step . since our first intended application was a cosmological problem the formation of the local group that requires the freedom to specify external tidal fields other than would be implied by periodic image charges , our code presents the relatively uncommon combination of an expanding system of coordinates with isolating ( more properly non - periodic ) boundary conditions . this introduces a number of complications not usually discussed in the literature on pm codes , and of which we shall give an account here . ( our method is also applicable to non - cosmological problems , for which isolating boundary conditions are an asset and the aforementioned complications do not arise . ) a future article ( @xcite ) will detail the results of our simulations of group formation . the general outline of this paper is as follows . section [ s : method ] describes in detail the method used . the tests and their results are presented in section [ s : tests ] . we conclude , in section [ s : conclusions ] , with our assessment of the strengths and limitations of this code . our code integrates the equations of motion for a set of gravitating particles . these equations can be derived from the lagrangian @xmath7 here the indices @xmath8 and @xmath9 span the set of particles in the system , @xmath10 is the mass of particle @xmath8 , @xmath11 its position at time @xmath12 , and @xmath13 describes the law of pairwise interaction between particles . ideally @xmath14 ( we choose units such that @xmath15 ) but the numerical method forces us to use an approximation which , due to adaptive grid refinement , can depend explicitly on time and on the individual coordinates @xmath16 and @xmath17 , not just on their difference . we could have included the term proportional to @xmath18 ( where @xmath18 is an arbitrary constant ) within the external potential @xmath19 , but for cosmological applications it is useful to show it explicitly . it is often convenient to apply a time - dependent rescaling of the coordinate system , with new coordinates @xmath20 defined by @xmath21 . an equivalent lagrangian is then @xmath22 for notational convenience , we define @xmath23 note that in principle @xmath24 is arbitrary , reflecting our freedom to choose the coordinate system . for cosmological applications one normally chooses @xmath24 to be the solution of friedman s equation for some cosmological model , in which case @xmath25 is the background density and @xmath26 the hubble constant for that model . later in this paper we shall impose the additional restriction @xmath27 , which holds for cosmological models in the matter - dominated era . the second term in the lagrangian ( [ q : lag - x ] ) can be regarded as involving an effective potential @xmath28 which satisfies @xmath29 . similarly , in the continuum limit and for a coulomb interaction ( @xmath30 ) , the third term satisfies @xmath31 , where the proper density @xmath32 is given by @xmath33 and @xmath34 ( respectively @xmath35 ) denotes differentiation with respect to the first ( respectively the second ) of the two position vectors on which @xmath36 depends . the usual derivation of the equations of motion from the lagrangian ( [ q : lag - x ] ) yields : @xmath37 for later convenience , we define @xmath38 this allows the equation of motion ( [ q : motion ] ) to be written more compactly as @xmath39 it is also useful to recast the equation in terms of a generalized time coordinate @xmath40 : @xmath41 with @xmath42 a common choice is the so - called conformal time , @xmath43 ; another ( @xcite ) is @xmath44 for some constant @xmath45 . naturally , any sufficiently differentiable monotonic function of @xmath12 is acceptable , and one is free to construct such a function _ ad hoc _ for each individual simulation . we avail ourselves of this freedom . we lay down a rectangular grid of @xmath46 by @xmath47 by @xmath48 nodes with uniform spacings @xmath49 , @xmath50 and @xmath51 . the values of @xmath46 , @xmath47 and @xmath48 must be acceptable to the fast fourier transform ( fft ) routines used . ( all our tests were done with @xmath52 a power of 2 , and with @xmath53 , as this is the most common situation in practice and the only one we needed for our applications . ) at every step , a density is assigned to each grid point using the cloud - in - cell ( cic ) algorithm . ffts are then used to solve for the potential on the same grid , and an acceleration is obtained for each particle by differentiating the grid potential , then using cic interpolation . we use a two - point finite - difference formula to compute the derivatives of the potential at grid nodes . this choice means that truncation errors in the force law scale with the square of the grid spacing ( he , 5 - 4 ) . the discrete green s function we use is obtained by sampling @xmath54 at grid points ( and fourier transforming the result ) . this differs from the more common approach in cosmological pm codes ( @xcite ) of basing the green s function on the seven - point finite difference approximation to the laplacian ; our choice has the merit of guaranteeing the truncation of the interaction law at large separations , as required for a correct implementation of isolated boundary conditions ( @xcite ) . unlike in the p@xmath5 m method , where the grid - based part of the force must possess a good degree of translational invariance to avoid introducing terms in the direct particle - particle contribution that depend explicitly on the position of each pair relative to the grid , we are not compelled to soften the interaction at small scales . to do so would cost us precious spatial resolution ; already with our choice of green s function the effective force turns out to be softened at separations @xmath55 grid cells . a consequence of our decision not to soften the force law any further is that the @xmath56-minimization procedure of he 8 - 3 - 3 , which requires the reference force to have no harmonics on scales smaller than the grid spacing , is not appropriate for us . the choice of the shape of the interaction law at small separations and of the charge assignment and force interpolation scheme is a matter of compromise between accuracy and computational cost . our choices could undoubtedly be improved upon , but we preferred to concentrate our efforts on the more innovative aspects of our method . isolated boundary conditions are implemented by conceptually doubling the size of the grid in each dimension , and padding the density array with an appropriate constant value ( normally zero ) outside the principal octant , which contains the particles ( he , 6 - 5 - 4 ) . if the green s function is truncated so that the interaction falls to zero at separations larger than the system ( before doubling ) , this completely suppresses any interaction between the system and its periodic images . the computed potential , however , will only match that of an isolated system within the principal octant . since we need to apply a gradient operator to the potential in order to compute the forces , there is a layer ( one cell thick for our approximation to the gradient ) in which we would not be able to compute them . we therefore allow no particles in this outer layer . ( we are free to increase the thickness of this layer . it has been convenient to do so in some of the tests discussed below , to achieve an integral ratio of cell counts in comparisons between different grid resolutions . ) we do not use james ( 1977 ) method to impose isolated boundary conditions without doubling the grid by calculating appropriate screening charges on the surface , since strictly speaking the procedure is only justified when a finite - difference approximation to the laplacian is used in solving poisson s equation . the fft technique does not satisfy this condition , and would require screening charges throughout the volume of the box . in cosmological applications , the space surrounding the system is not to be thought of as empty , but rather as containing a background of uniform density @xmath57 . ( departures from uniformity can be represented through an appropriate choice for the tidal potential @xmath19 . ) we must therefore add this background , convolved with the same ( cic ) charge assignment function that is used for the particles . its contribution to the density at any node on the boundary of the region where particles are allowed is then proportional to the number of cells adjacent to this node that lie outside the particle region : @xmath58 for nodes on a face , @xmath59 on an edge , @xmath60 at a vertex . we implement the @xmath61 term of the equation of motion ( equation [ q : motion ] ) by subtracting @xmath62 from the density at all points before solving for the potential . if @xmath62 coincides with the background cosmological density , the source function for the potential outside the principal octant is exactly zero . this is the most common case , and the one that presents the fewest conceptual and practical difficulties . with periodic boundary conditions it would also be the only possible case since the mean density outside the simulation box must always equal that inside it . with isolated conditions a mismatch is permissible , and could be used for example in applying an expanding or contracting grid to follow the evolution of an isolated system without a cosmological density background . the varying @xmath24 would then be adjusted to match the expansion or contraction of the simulated system , providing better resolution at lower cost during the collapse phase . a very significant difference between periodic and isolated boundary conditions is that the latter allow the exchange of mass , momentum , energy , angular momentum between the particles and the exterior . with periodic boundary conditions , there is effectively no exterior . particle flows are a significant concern in cosmological applications , where matter may both leave and enter the computational box during the simulation . in a typical cosmological model , the mass variance is of order unity in a sphere of radius @xmath63 mpc . one expects particles at the edge of a sphere of radius @xmath64 to be displaced with respect to the center of mass of the sphere by about @xmath65 , a significant length when compared to the size of a group or cluster of galaxies . a simulation of characteristic size @xmath64 may reasonably be expected to exhibit a twofold increase or decrease in the total mass within the box during a run . on smaller scales the variance is even larger , at least for the currently fashionable `` bottom - up '' scenarios of structure formation . exiting particles are easily handled by removing them from the simulation , but incoming particles have to be injected according to some prescription that fits the physical problem at hand . in the cosmological case , as long as the density perturbations grow linearly on the scale of the box , a reasonable prescription can be based on extrapolations from the linear theory . at late times , in the strongly non - linear regime , simply extrapolating from the linear solution causes large amounts of material to be injected which in reality would collapse into bound objects outside the box and remain outside . in other words , the zeldovich approximation is clearly inappropriate beyond the time at which caustics form . it would lead to a gross overestimate of the total mass flow into the box . an adequate model of the inflow in the nonlinear regime therefore requires an actual simulation of the mass flows in a larger region . this has become normal practice in simulations with tree codes , and hierarchical grid methods such as ours also lend themselves well to this approach . the main difficulty is that in order to keep the total number of particles manageable , one must essentially run a preliminary simulation simply to find out where the mass that flows into the region of interest originated and sample it with a finer granularity in the initial conditions . this may lead to problems of contamination by more massive `` background '' particles if the small - scale structure that is not resolved by the preliminary simulation turns out to have a significant impact on the dynamics . furthermore , the presence of particles of different masses in the system could lead to spurious mass segregation effects . for these reasons , we attempted to keep the overall size of the computational box as small as possible , handling the tidal fields on larger scales , as well as any mass flows ( as long as they remained moderate ) as externally imposed boundary conditions . this turned out not to be a particularly successful design choice , and we can not recommend it to others . injecting too many , or too few , particles can have a destabilizing effect . the first term on the right hand side of equation [ q : motion ] has , in the usual case @xmath66 , the effect of accelerating particles towards the boundary of the box . this is balanced by the second term , which represents the attraction between the gravitating particles . if more particles leave the box than are injected , the first term will tend to dominate and cause even more particles to be ejected ; conversely , if too many particles are added the material will tend to collapse towards the center . this may be taken to represent a physical effect : a void is expected to expand faster than the universal average , an over - density more slowly . but unless the algorithm for replenishing the box with particles is well thought out , a runaway instability may occur . in practice we have found it expedient to couple the injection of particles to the outflow so as to maintain a constant mass within the system . we then monitor the cumulative mass of particles so recycled and compare it with the total mass in the box . if the ratio becomes too large ( greater than 10% or so ) , we take it as an indication that one needs to simulate a larger volume of space . as the volume that needs to be simulated grows larger , our original motivation for using isolated boundary conditions becomes weaker . it turns out that the method we described in this paragraph works much better if the system is located in a region of lower density than the cosmic mean , as the mass fraction that undergoes reinjection is smaller in that case . this allows us to simulate a number of interesting systems , but is not suitable for statistical studies involving a random selection of initial conditions . momentum , angular momentum , and energy can be carried by the particles that flow in and out of the system and by direct interaction with the external potential @xmath67 and the uniform density we impose in the space that surrounds the particle region . the latter unfortunately has the cubic symmetry of the computational box rather than the more convenient spherical symmetry that would be required for an exact cancellation of the induced tides . consequently , these tides are always present except in the pure non - cosmological case with @xmath68 a constant , where the background density vanishes . whether this is a serious drawback depends on the physical problem being studied . if necessary , one can enlarge the computational box ( this may be required in any case to keep mass flows under check ) , include a compensating term in @xmath19 , or both . the idea of adding subgrids in regions where better spatial resolution is required is not a new one . past approaches have differed on whether to add mass resolution at the same time by splitting the particles into a larger number of less massive ones , on how to minimize errors at the interface between the finer and the coarser grid , and on whether the solution on the coarser grid should be modified to take into account the results from the finer grid . unlike villumsen ( 1989 ) and splinter ( 1996 ) , we do not automatically introduce a new set of less massive particles on each subgrid . we take the view that our initial mass granularity already matches the resolution we wish to achieve , and simply increase the force resolution ( by adding subgrids ) when and where the particle density is high enough to make this permissible and worthwhile . this allows the decision of where to place subgrids to be made on a step by step basis , and spares us the need to perform a first simulation without subgrids to find out where particles of smaller mass should be placed in the initial conditions . an additional advantage of having a single set of particles of equal mass is that we need not worry about mass segregation effects . a drawback is that maintaining equivalent resolution within a larger computational volume requires more particles . of course our code does support multiple particle masses , and as will become clear below the criteria for subgrid placement can be tuned to favor tracking the lower - mass particles ; we may therefore decide to experiment with particles of different masses in future . in our scheme the various grids are merely devices , introduced independently on each integration step , to compute inter - particle forces . in this respect we are closer in spirit to couchman s ap@xmath5 m than to most other multiple - grid approaches . ours is effectively ( in the terminology of anc ) a particle - multiple - mesh ( pm@xmath0 ) scheme . this distinction will have important consequences below . since there is only one set of particles that exist independently of any subgrids , the equivalent of a back - reaction from the subgrid solution to the parent grid is automatically included : it is mediated by the particles themselves . we typically decrease the grid spacing by a factor of two for each additional level of subgrids . each subgrid thus covers a substantial number of cells of its parent grid . other integral refinement factors are allowed by our formulation ( subject to the condition that the boundaries of any subgrid must coincide with cell boundaries of the parent grid ) but become increasingly inefficient in terms of volume coverage and we have not tried them in practice . they would also exacerbate any `` ringing '' at the interface between a subgrid and its parent grid ; since this is one of the most delicate aspects of any hierarchical grid scheme , it is probably not advisable to use refinement factors larger than two under any circumstances . a particle passing from a region of low resolution to one of higher resolution effectively undergoes a sudden change in its spatial extent . ( in pm codes , particles are best thought of as extending over about one grid cell or more , depending on the shape of the interaction law and on the charge assignment scheme . ) in its current state , our code takes no particular precautions against any transients that may result . our comparison tests between runs that use adaptive subgridding and runs with uniform resolution show that such transients are not important enough to produce significant discrepancies in the results . this is fortunate , since otherwise we would have had to associate a time - dependent smoothing length with every particle , which would complicate the method . that the smoothing length would have to be associated with the particle rather than with the location relative to a subgrid follows both from our choice to treat the particles as primary and the grids as auxiliary objects and from the fact that new subgrids may be added , removed or relocated to follow the flow at every step . our strategy for computing the forces between subgrid particles differs slightly from that of villumsen . his approach was to recalculate the potential on the entire parent grid after having set the density in the volume occupied by the subgrid to a constant value equal to the mean density within the subgrid , and use this solution as the tidal field on the subgrid particles from the rest of the system . the forces between subgrid particles were then computed in the usual way by solving for the potential on the subgrid . instead , we first compute forces for all particles on the parent grid , then generate the density array on the subgrid ( whose nodes must be aligned with those of the parent grid ) as well as on a coarser grid that has the same spacing as the parent grid and nodes in the same locations but covers only the volume of the subgrid . since coordinates on the subgrid are affected by the same expansion factor @xmath24 as on the parent grid , we must subtract the same @xmath62 . we solve for the forces on all subgrid particles from both these density arrays , and add the difference to the forces we had previously calculated on the parent grid . this avoids double counting and means in effect that pairwise forces between particles that both lie in the subgrid are always evaluated at the higher of the two resolutions . the coarse fft on the subgrid involves far fewer grid points , so our approach is more efficient than villumsen s . it is also more flexible , in that we only need to store the accumulated force for every particle and the density potential for a single subgrid at a time , independently of the number and nesting depth of subgrids . the cost of actually storing a force vector for every particle may seem high when one knows that in a traditional pm code with leapfrog time stepping the particle accelerations can be interpolated when the velocities are updated and need not be kept afterwards ; but saving them allows us to use them to adjust the time step and to compute subgrid corrections to the energy conservation test . on the other hand , we could have supported independent time steps had we chosen to save the potential on each subgrid instead . this , however , would have caused the storage requirements to scale with the subgrid nesting depth . let @xmath69 be the set of particles in the subgrid , @xmath70 its complement . schematically , the acceleration on particle @xmath8 is computed as follows . for @xmath71 , @xmath72 for @xmath73 , @xmath74 where @xmath75 is the approximation to the coulomb potential on the parent grid , @xmath76 the approximation on the subgrid , and @xmath77 that on the subset of the parent grid that covers the subgrid . ( in reality , the sums are evaluated by fft and include the @xmath62 term . ) @xmath78 to a very good degree , provided only that the grid nodes are in the same locations . ( this is an essential requirement : we have found that violating it has deleterious effects on the solution . ) the two rightmost terms in the display of equation ( [ q : acc - sgr-1 ] ) therefore cancel each other . note that we add together force contributions from the various grids rather than the potentials ; this allows the arbitrary additive constant in the potential to differ from grid to grid without affecting our solution . we impose isolated boundary conditions on the subgrid , with an external background density ( @xmath62 ) equal to that used on the parent grid . it may seem more appropriate to use the mean density within the subgrid , or better yet the mean density in the immediate vicinity of the subgrid ; but one should note that some ringing is to be expected no matter what the exact scheme adopted since the density at the boundary of the subgrid always has some short - wavelength component that is resolved on the subgrid but not on the parent grid . the usual approach in codes of this type is to try to limit the ringing by introducing around the subgrid a buffer region , in which particles contribute to the potential but are not affected by it , and/or a transition region in which the potentials of the parent grid and of the subgrid are blended linearly . the transition region implements the variable smoothing length we alluded to at the end of the previous subsection . as already noted , it is of limited usefulness in our case since it does nt guard against transients induced by the sudden activation of a new subgrid . the possible motivation for introducing a buffer region can be understood by considering the case of a particle @xmath79 located on the subgrid but near its edge , and subject to the forces of two nearby equidistant particles @xmath80 , also on the subgrid , and @xmath81 just outside it . in the exact solution , @xmath80 and @xmath81 exert forces of equal magnitude on @xmath79 . without a buffer region , the calculated force from @xmath80 is stronger . this can create spurious small - wavelength perturbations as the pair @xmath82 tends to collapse on its own rather than as a triplet @xmath83 . the introduction of a buffer region would allow the balance to be preserved between the attractions of @xmath80 and @xmath81 . our implementation of the buffer region is as follows . extending the notation introduced in the previous subsection , we have split @xmath70 into the buffer region @xmath84 and a distant exterior @xmath85 . equation [ q : acc - sgr-1 ] is modified in that the sum over @xmath86 becomes a sum over @xmath87 and the sums over @xmath88 become sums over @xmath89 . we normally set the thickness of this buffer region to be equivalent to three cells of the parent grid : the grid - based force does not differ significantly from the exact coulomb force at separations this large , so that @xmath90 . since the thickness of the buffer region is fixed in units of the spacing of the parent grid , large refinement factors result in the buffer region `` eating away '' most of the volume of the subgrid . this is one of the reasons why we only experimented with a refinement factor of 2 at each level . an essential feature of this buffer region is that it enables us to use multiple adjacent subgrids to cover extended structures that are worthy of higher resolution but do not fit within a single grid . this could occur either because of their size or because they are located too close to other high - density objects already covered by their own subgrids . ( we do not allow arbitrary overlap of subgrids , for both simplicity and efficiency . ) a problem that might arise when the boundary between two adjacent subgrids cuts through the middle of a high density structure is that the force between two nearby particles separated by the boundary would be calculated with the ( poorer ) resolution of the parent grid . thanks to the buffer region around each subgrid , this difficulty is avoided : the force exerted by either particle on the other will be computed at subgrid resolution by virtue of each particle lying in the buffer region of the other s subgrid . in the overlap region between adjacent subgrids , it matters little whether the border region is implemented as described above or in the following alternative way , by subjecting particles in @xmath84 to the subgrid forces without including them in the subgrid potential . this corresponds to applying equation [ q : acc - pgr-1 ] to particles @xmath91 only , and equation [ q : acc - sgr-1 ] to particles @xmath92 . our first implementation used the latter approach , and comparison tests show only minute differences between the results of both variants . a difficulty arises wherever the buffer region has finite width and does not overlap with an adjacent subgrid . then the balancing of the forces by @xmath80 and @xmath81 over @xmath79 comes at the cost of a violation of newton s second law : the force exerted by @xmath79 on @xmath81 , being computed on the parent grid , does not balance the force by @xmath81 on @xmath79 . the net result is that @xmath81 will be accelerated outwards , away from the subgrid . ( if the alternative implementation of the buffer region is adopted , the net effect has opposite sign and tends to push the particles inwards . ) the consequences for the orbit of a bound clump can be spectacular , as the following worst - case example , illustrated by figure [ f : tf4 g ] , shows . we launch a 4096-particle realization of a truncated isothermal sphere ( with a half - mass radius of 0.02 units , corresponding roughly to half the cell width on the top grid ) with a mean velocity of 0.1 units towards a region covered by a fixed subgrid ( the inner bounds of which are indicated by dotted lines in the figure ) with refinement factor of two and a buffer region of width 0 ( solid curve ) or 2 ( dashed curve ) parent grid cells . for this example we adopted the alternative implementation of the buffer region , which causes the particles to be accelerated inwards . the initial velocity is along the @xmath93 axis , perpendicular to the faces of the subgrid . the figure shows the @xmath93 coordinate of the center of mass of the sphere as a function of time @xmath12 . when the buffer region is suppressed ( width 0 , solid curve ) the clump enters and exits the subgrid without changes to its bulk velocity . ( the apparent turnaround occurs when the clump reaches the edge of the computational volume and some particles leave the box . the solid curve reflects the mean velocity of only those particles that remain in the box . ) the presence of a finite buffer region , by contrast , causes the clump to accelerate as it enters the subgrid , and to bounce back as it tries to exit on the other side . the reason it does nt simply lose the momentum it gained when entering the subgrid is that the momentum change depends on the difference between the forces calculated on the subgrid and on the parent grid , and that the passage through the more highly resolved subgrid has allowed the clump core to relax to a more sharply peaked density profile , increasing the force mismatch . the main features of this behavior are quite generic : a more carefully constructed clump with a larger number of particles that was allowed to settle into a numerical equilibrium before being launched through the subgrid underwent similar , if slightly milder , accelerations and rebounds . clearly it is better for us to suppress the buffer region where there is no adjacent subgrid . the violation of momentum conservation that a finite border region implies can have a drastic influence on the orbit of a bound clump . this would not be much of a concern if we could guarantee that bound clumps are always entirely covered by subgrids at the same resolution ; the problem does not arise for smooth distributions of matter where short - range forces do nt predominate . ( in section [ s : gunn - gott - test ] we demonstrate this fact through a test of smooth infall onto a localized seed mass perturbation . ) in principle , an adaptive algorithm for subgrid placement based on the density should tend to ensure this . however , one may want to place additional restrictions on which regions are followed at higher resolution , in which case the guarantee does not hold and we must take the precaution of suppressing the buffer region in the absence of an adjacent subgrid . note that the problem is less severe when expanding coordinates are used , as in cosmological applications , since the peculiar velocity acquired while crossing a subgrid interface will be damped away . why has this same difficulty not been recognized by other authors , notably anc and splinter ( 1996 ) ? the reason may be traced to a fundamental difference in philosophy between our code and theirs . the very idea of a particle on a subgrid interacting symmetrically with a particle on the parent grid is a consequence of our decision to regard the particles , rather than the density field on the grids , as primary . in a scheme where every particle is associated with only one grid , and particularly when the dynamics of the parent grid particles are unaffected by the subgrid ( the so - called `` one - way interface '' schemes ) , momentum conservation should be examined separately on each grid : on the parent grid , no violation is expected , while on the subgrid the effects of the parent grid particles are treated as an external field and the buffer and transition regions are used to smoothly bring any particles exiting the subgrid under the control of the parent grid field alone , as test particles , rather than reflecting them back into the subgrid . thus , the apparent contradiction between these authors use of a buffer region two cells wide and our restriction of the buffer region to the sole case of adjacent subgrids does not signal an error either on our part or on theirs , but is merely a manifestation of our different view of the respective roles of particles and grids . a distinguishing feature of our code is that the activation of subgrids is entirely automated : the user need only set a few parameters ( maximum depth , and various optional thresholds ) , after which the code decides at every step where to place subgrids . tuning the criteria for subgrid placement is not an easy task ; our current choices can undoubtedly be improved upon . here we shall describe some of the criteria we have implemented . not all of them have proved very useful in practice . our subgrids have fixed shape and size , and may not overlap . consequently , we found it simplest to introduce a uniform tiling of subgrids covering the entire volume of the parent grid . each subgrid may be either active or inactive . a subgrid is activated whenever our requirements for subgridding are satisfied within its volume . we are free to choose the origin of the tiling independently for every parent grid at every step . we make use of this freedom to reduce the number of active subgrids . typically we try eight different possible origins , and adopt one that requires the activation of fewest subgrids . the first criterion for subgrid activation is that the particle number density be sufficiently high in at least part of the covered region : there is no point in the grid ever being much finer than the smallest distance between particles . to activate a subgrid , we require that among the corresponding cells of the parent grid either ( a ) at least one cell contains more than @xmath94 particles ; or ( b ) one cell contains at least @xmath95 particles and its 26 immediate neighbors together contain a further @xmath96 particles or more . @xmath94 , @xmath95 and @xmath97 are specified as input parameters to the code . the main reason for considering cubes of @xmath98 cells as well as single cells is that one needs to detect the collapse of a high - density region before it is too advanced : the additional resolution is already required during the later stages of collapse . we do nt try to detect larger , lower - contrast potentially collapsing structures on larger scales since for separations larger than about three cells the forces are computed accurately on the parent grid . one way of choosing @xmath95 and @xmath97 is the following . let @xmath99 be a measure of the mean number of particles per cell in some larger region , such as the entire candidate subgrid . ( this measure need not be unbiased ; in fact , if the mean particle number density is very low we found it necessary to artificially increase @xmath99 by setting it to the value 1 . otherwise our subgrid placement criterion would be satisfied far too easily in low - density regions . ) if the particle counts were distributed according to poisson statistics , one would expect a group of @xmath100 cells to contain on average @xmath101 particles , with a standard deviation @xmath102 around that mean . by setting @xmath103 and @xmath104 , where @xmath105 is a tunable parameter , we try to ensure that only particularly significant deviations from homogeneity trigger subgrid activation . we have found that this approach works reasonably well during the early stages of nonlinear evolution , where it causes higher resolution to be applied to the regions we are most interested in . however , as the simulation progresses we have needed to increase @xmath105 in order to limit the proliferation of subgrids , as our goal was merely to achieve high resolution in a few large peaks near the center of the computational box . clearly the criteria can and should be tuned according to the needs of individual applications ; no single choice is optimal for all cases . we follow the usual practice in pm codes of advancing the positions and velocities of the particles in leapfrog fashion : @xmath106 an alternative formula applies on a starting step , when both the positions and the velocities are known at the same time @xmath107 : @xmath108 likewise , one can synchronize positions and velocities at the final step by advancing the velocities from @xmath109 to @xmath110 then updating the positions as @xmath111 in large simulations , it is particularly important to avoid taking more time steps than required to obtain accurate results . the time step should therefore be allowed to vary . in the standard leapfrog formulation , the cancellation of second - order terms in equation ( [ q : lf - v ] ) depends on @xmath112 being evaluated at the midpoint of the velocity step . the cancellation of the second - order term in the formula to update the positions is less important since the @xmath113 are known ; however , it contributes to lowering the operation count for the method . various approaches can be used to change the time step in the middle of a simulation . one is to use a different midpoint in equation ( [ q : lf - v ] ) and calculate the corresponding @xmath114 . the other , which is the one we adopted , is to keep @xmath115 constant by an appropriate choice of the function @xmath40 . we construct this function and its first and second derivatives , which are needed to evaluate @xmath112 and @xmath84 according to equations ( [ q : a ] ) and ( [ q : b ] ) , step by step as the simulation progresses . ( in practice , we do not allow @xmath116 to vary too quickly , since that tends to spoil the results . we shrink @xmath117 by at most 25% on each step , and let it grow even more slowly . situations in which this forces a larger @xmath117 than desired have fortunately been very rare in our runs . ) our preferred criterion for determining the time step is the courant condition : @xmath118 where the index @xmath9 spans the set of all particles , @xmath119 is the grid spacing of the finest grid used to compute the forces on particle @xmath9 , and @xmath120 is a constant coefficient . @xmath121 should be taken as being the mean velocity over the time step , and is a function of both the velocity before the step is taken and of the acceleration . the latter can be important if the initial velocity is uncharacteristically small , for example when all the particles are initially at rest . the equation for the maximum acceptable @xmath117 for each particle is actually a quartic . we do not solve it exactly , preferring to estimate a close lower bound on the solution for the particles with the tightest @xmath117 constraints . this way of taking the accelerations into account fulfills the same function as other authors use of the maximum density on a subgrid to estimate a local dynamical time . in some cosmological simulations with a large number of particles , we found that adopting the smallest @xmath117 found in this fashion was leading to time steps much shorter than warranted by our physical understanding of the dynamical time scales involved ( based on the maximum density ) . this turned out to be due to a small number of high velocity particles : the tail of the velocity distribution produced by the violent relaxation of a newly collapsed object , which is naturally better sampled as the number of particles in the simulation increases . since these particles represent a small fraction of the mass in the typically high - density regions in which they are found , their motion has almost no impact on the mean gravitational field , which evolves on much longer time scales . this led us to modify the time step criterion for these simulations by pretending that the velocity of each particle before the time step is negligible and basing the choice of @xmath117 solely on the accelerations . this is not unreasonable for cosmological applications since in the linear regime the velocities and peculiar accelerations are related and in virialized clumps they are a better indication of the dynamical time scale than the velocity of the fastest particle ( which is not the same as the local velocity dispersion ) . moreover , expanding coordinates have the property that peculiar velocities are damped away and need to be continually regenerated by the forces . as a further precaution , we have analyzed the velocities of the particles at a late stage in a simulation run with this relaxed time step criterion and found , as intended , that the number of particles that violated the stricter courant condition was small . the relaxed criterion is known to fail , however , for highly ordered collapse ( such as that of a homogeneous sphere ) . in that case the largest acceleration does not provide a good estimate of the time scale over which the mean field evolves : based as it is on the more extended distribution at the beginning of the step , it systematically overestimates the maximum allowable time increment . accordingly , we only resort to the relaxed criterion , with some reluctance , for runs in expanding coordinates , with large numbers of particles , and where different mass shells collapse at different times . our tests on smaller cosmological runs have shown the results to be substantially identical , but with the relaxed condition requiring a much smaller number of steps . we use a single value of @xmath117 for all particles on all subgrids at any given step . a far better approach in principle , but much more complicated to implement , would be to adopt multiple time steps and to distinguish between the time steps used for advancing individual particles and those for updating the potentials on subgrids . independent time steps for the particles would require a way of estimating the accelerations at positions and times slightly different from those for which the potential has been calculated . for this , it may be necessary to calculate the time derivative of the subgrid potential as well as the potential itself . in addition to the constraints resulting from the rate of change of the subgrid density , the frequency with which the subgrid potential is recalculated can also be affected by particles flowing out of the region in which forces can be interpolated from the subgrid potential : in our scheme , such events change the mapping between particles and subgrids and require recalculation of all the subgrid potentials involved if the total momentum is to be conserved . these are our main reasons for adopting a single time step for all grids . the other main reason is that a single time step allows us to process subgrids one by one , accumulating the force contributions on each particle without needing to store the solutions for the potential on many nested subgrids simultaneously . almost any other time step scheme requires these subgrid potentials to be available simultaneously . the coefficients in the leapfrog formulae ( equations [ q : lf - v ] and [ q : lf - x ] ) can depend on time . for good accuracy , their variation in a time step must be small . this constraint is clearly unrelated to the grid spacing , and must be imposed separately . it is equivalent to the requirement that @xmath122 . conservation laws provide an important diagnostic of how accurately the equations of motion have been integrated in a given simulation . good conservation of physical invariants is not a sufficient condition for a good integration , but it is a necessary one . in the absence of coordinate expansion , subgrids , and external fields , a pm code such as ours would be expected to conserve total momentum algebraically , energy a little less well , and angular momentum only in the limit where the interparticle separation is much larger than the grid spacing ( he 76 ) . cosmic expansion causes the usual equation of conservation of the total energy , @xmath123 ( where @xmath124 is the kinetic , @xmath125 the potential energy ) , to be replaced with the layzer - irvine ( @xcite ; @xcite ) equation , which can be derived from the well - known property of the hamiltonian @xmath126 : @xmath127 computing the canonical momentum from equation [ q : lag - x ] , @xmath128 it follows in the usual way that @xmath129 we define the kinetic and potential energies as : @xmath130 then @xmath131 our exclusion of the external potential @xmath19 from the definition of @xmath125 is somewhat arbitrary . in practice , one wants to separate the terms associated with the mutual interactions of particles , which do measure the accuracy of the integrator , from the correction terms that merely account for known , external sources and sinks of energy . in applying equation [ q : dham - dt - law ] , one notes that in @xmath132 the last term does not vanish if @xmath133 depends explicitly on time ( as may occur in practice when a new subgrid is activated and the force resolution increases locally ) , or if @xmath134 is not constant . for simplicity , we shall assume @xmath135 from now on , since that is the only case of actual practical interest to us . straightforward algebra leads to the two equations : @xmath136 @xmath137 which although physically equivalent are evaluated numerically in slightly different ways when @xmath24 is not constant . the corresponding conserved quantities are : @xmath138 @xmath139 in proper - length coordinates ( @xmath140 ) , the first integral term ( @xmath141 or @xmath142 ) vanishes and both criteria reduce to the usual law of conservation of energy . the last two terms represent respectively the work done by external fields and time dependence in the interaction law . one may question the appropriateness of including the latter effect , since its origin lies in the numerical method used rather than in the underlying physics . should nt contributions from the @xmath143 term be counted as violations of energy conservation by the code ? in reply we point out that much of that term may result from fluctuations in the zero level of the gravitational potential on subgrids , which we did not attempt to suppress since they have no bearing on the computed forces . and in any case the @xmath143 contribution should be evaluated so that its magnitude relative to the other terms may be assessed . it is unrealistic to expect a pm code to conserve @xmath70 and @xmath144 arbitrarily well when gravitational instability causes the individual terms ( @xmath125 , @xmath145 ) to grow by orders of magnitude . like many other authors before us , we deem @xmath70 and @xmath144 adequately conserved if their variation is a small fraction ( typically 13% ) of the change in @xmath125 or @xmath146 , respectively . in our definition of @xmath125 ( equation [ q : w ] ) we excluded the self - energy terms @xmath147 . but the natural way of calculating the potential in a pm code effectively includes these terms ( the restriction @xmath148 can be ignored since the scheme avoids self - forces ) ; we subtract them explicitly as part of our energy conservation test . we adopted a momentum - conserving scheme for calculating the forces . the total momentum should therefore be unaffected by particle - particle interactions on any given grid , and ( given the precautions we have taken in our handling of the buffer region surrounding each subgrid ) be only weakly modified by interactions across the boundary between adjacent subgrids . however , it is still affected by all other interactions between particles and external fields , including the constant - density background outside the computational box . ( this field gives rise to forces only when isolated boundary conditions are used , which is the reason why it is nt normally discussed in the literature . ) the canonical momentum @xmath149 of particle @xmath8 satisfies @xmath150 summation over all particles and time integration yield a conserved quantity @xmath151 in the special case when ( i ) @xmath19 has no spatial dependence , ( ii ) @xmath152 or @xmath153 , and ( iii ) @xmath154 , the total momentum is constant . if one s aim is to measure the deviation of the results from those expected when @xmath154 ( such a deviation can occur in the presence of subgrids ) , one should naturally omit the corresponding term from equation [ q : pc ] . likewise , the conservation of angular momentum @xmath155 amounts to the constancy of @xmath156 angular momentum is only conserved if ( i ) @xmath157 and ( ii ) @xmath19 is a function of @xmath158 and @xmath12 alone . the exact coulomb force law satisfies the first condition , but grid - based approximations to it do nt . as in the analogous case for the momentum , if the aim is to measure deviations from the ideal behavior where ( i ) is satisfied , then one should compare the magnitude of the corresponding term to @xmath159 itself , and to the magnitude of the external torque term when present . our first tests are meant to see how accurately the code calculates the accelerations of particles given a known mass configuration . the simplest case is that of the forces due to a single point mass , which can be compared to the value predicted by coulomb s law . figure [ f : tf1b.dff ] illustrates this comparison . it shows the accelerations induced by a single particle at the center of a @xmath160 grid with isolated boundary conditions . a single @xmath160 subgrid , twice as fine as the top grid , was centered on the massive particle . two sets of points are readily distinguished , corresponding to test particles inside and outside the subgrid . as expected , the acceleration is systematically underestimated since the effective potential is softer than that of a point mass . also as expected , the subgrid extends the range over which the force is accurate to within a few percent ; the accuracy threshold ( a little over 3% in this test ) could be lowered by increasing the size of the subgrid , the spacing being equal . our second test checks that the accelerations computed on a set of adjacent subgrids agree with those one would obtain on a single , larger grid with the same spacing as the subgrids . we compare forces on individual particles randomly chosen to sample a uniform distribution . the long - range forces , which are adequately resolved on the parent grid , tend to cancel , so that the system is dominated by short - range interactions that will be most affected by the grid refinement . our reference forces are from a non - subgridded run with @xmath161 cells and @xmath161 particles . we compare these to those obtained from three different runs with only @xmath162 cells on the parent grid , in which all possible first - level subgrids are activated . the only difference between the three runs is in the width of the buffer region where adjacent subgrids overlap . the results are illustrated in figure [ f : df ] . the magnitude of the force difference is plotted against that of the reference force for a randomly selected 1% of the particles ( to avoid overcrowding the plot ) . panels ( a ) , ( b ) , and ( c ) correspond respectively to a buffer width of 0 , 1 , and 2 parent grid cells . the solid diagonal line represents @xmath163 ; the last panel also contains a dotted line where @xmath164 . the results show that when the buffer is sufficiently wide ( at least two cells ; we like to use three cells in production runs with larger grids ) , the forces computed by the subgrid method are mostly within 1% of the forces from a single grid of equivalent resolution . whenever the buffer region is suppressed , an appreciable fraction of the particles exhibits force errors larger than 10% . here it is essential to understand what we mean by `` errors '' . figure [ f : df ] only compares the forces computed using a set of abutting subgrids of equal resolution with those from a single , larger grid with the same spacing , and shows that some overlap is both required and sufficient to avoid loss of resolution at the boundary between such abutting subgrids . it does _ not _ compare forces computed at different grid spacings , and does not address the issue of convergence to a `` perfect '' solution in the limit of an infinitely fine grid . we know that at the interface between a subgrid and a coarser parent grid the accuracy of the forces within the outer two or three cell layers of the subgrid will be less than that of a uniform fine grid regardless of the thickness of the buffer region . ( it will , however , be no worse than on the parent grid alone . ) we compensate for this by making the subgrids cover a slightly larger volume than the region where the higher resolution is required . the collapse of a homogeneous self - gravitating pressure - less ellipsoid initially at rest can be described by a small set of coupled ordinary differential equations for the lengths of the principal axes , which can be integrated in a straightforward way ( @xcite ) . this solution provides a convenient test of our code . the test corresponds to that of collapse to a sheet or filament in a cosmological code with periodic boundary conditions , but is more appropriate to our use of isolated boundary conditions . in particular , the plane - wave test of efstathiou et al . ( 1985 ) is only easy to interpret and compare with a semi - analytic solution when periodic boundary conditions are used , and these lie outside the scope of the present article . another test , in section [ s : voidtest ] below , shows the behavior of our code when integrating the collapse of a moving sheetlike ridge in the presence of cosmological expansion . we followed the collapse of a homogeneous ellipsoid with a @xmath165 ratio between the lengths of the principal axes . the experiment was repeated for two different choices of the grid spacing in order to verify convergence towards the correct solution . for the higher resolution , we performed two runs , one using a single grid of @xmath166 cells and one with a top - level grid of only @xmath160 cells but up to two levels of subgrids . each subgrid also has only @xmath160 cells ; when necessary , multiple adjacent subgrids were automatically generated by the code to cover the entire ellipsoid . the number of particles is the same in all three runs , about @xmath167 . figure [ f : ts7-overview ] shows the short axis @xmath168 of the ellipsoid as a function of time @xmath12 . the solid curve represents the analytic solution , the dot - dash curve the solution computed on the coarse grid , while the remaining two dashed curves show the results of the two high - resolution runs . the horizontal dotted line corresponds to the grid spacing in the high resolution runs ; that of the low resolution grid is four times larger . figure [ f : ts7-blowup ] presents a blown - up view of the same results around the time of collapse of the ellipsoid to a spindle . one sees from these figures that increasing the resolution does indeed yield a better agreement between the @xmath1-body and analytic solutions . furthermore , the results with subgridding are nearly identical to those obtained with a single grid of equivalent resolution . in the high - resolution runs , the minimum radius of the spindle is comparable to the grid spacing . at the lower resolution , it is significantly less than the grid spacing . we did not investigate the reasons for this in detail , but it is likely that the convergence is also affected by the number of particles used to represent the ellipsoid . in these tests we have kept this number constant . the two runs without subgrids also used a small , constant value of the time step @xmath169 . collapse occurs after 1524 and 1644 steps respectively . the subgridded run , by contrast , used the courant condition to adjust the time step ; only 125 steps were required to reach the point of collapse . the good agreement between the high - resolution results obtained with both choices of time step confirms the validity of our implementation of adaptive time steps . we have also repeated the low - resolution run using our adaptive time step scheme ; on figure [ f : ts7-blowup ] the results would be indistinguishable from those obtained with a fixed time step . an interesting test of the behavior of mass flows into a subgrid is the case of accretion from a uniform - density background onto a point mass ( or any other compact , spherically symmetric density profile ) . a regular lattice of @xmath161 particles with zero initial peculiar velocity was laid down in a computational box of unit side to be evolved with isolated boundary conditions in an @xmath170 , @xmath171 cosmological model . we added at the center of the box a single particle of mass @xmath172 , where @xmath173 is the mean density of the other particles and of the external background , @xmath174 , and @xmath175 . we followed the collapse to a time @xmath176 , corresponding to an expansion factor @xmath177 . a scaling solution is available ( @xcite ; @xcite ; @xcite ; @xcite ) : @xmath178 . this simple law is expected to hold only for those shells that enclose a mass much larger than that of the initial seed and that have evolved for a few crossing times after their initial turnaround . the crossing time for a shell is of the same order as its turnaround time , and its proper radius is a factor @xmath179 times the turnaround radius . using a linear theory approximation ( in which the density perturbation grows in place ) until the turnaround at a mean enclosed overdensity @xmath180 , we find that for a given initial enclosed overdensity @xmath181 the turnaround occurs at @xmath182 , the subsequent virialization at @xmath183 , and the final comoving radius of the shell is @xmath184 , where @xmath185 is the initial comoving radius . requiring @xmath186 , only shells with @xmath187 are expected to show self - similar behavior , and then only for @xmath188 . one can also evaluate bertschinger s ( 1985 ) equation ( 2.7 ) with @xmath189 , @xmath190 , @xmath191 ( these values lead to @xmath192 ) and make direct use of his numerically determined similarity solution . a twist of our simulations is that shells with @xmath193 are prevented from falling in by the fact that they are not entirely included within the simulation volume . accordingly , the accretion will be starved for @xmath194 , and the last complete shell should virialize at @xmath195 . our main reason for performing this test was to compare the solutions obtained with various treatments of the border region around a subgrid . we performed four runs : one with a uniform @xmath196 grid , the other three with a @xmath160 grid and a nested @xmath160 subgrid with a linear refinement factor of 2 . ( after subtracting edge cells as discussed in [ s : bc ] , the computational box was covered respectively by @xmath161 and @xmath162 cells . ) in one run the border region was suppressed , as suggested by our tests with a bound clump crossing the boundary , while the other two both used a border width of two parent grid cells , once with the particles in the border region contributing to the subgrid potential without feeling its effects and once with the border particles being subjected to the subgrid potential without contributing to it . in the last two cases , the net effect of the force asymmetries on overdense regions results in an acceleration pointing respectively away from the subgrid and towards it . figure [ f : tf6d - rho ] shows the logarithmic density profiles for the runs we just described . the differences between the results of all these runs are minute ; we find it difficult to ascribe any significance to such small deviations . this is reassuring : it suggests that the exact way in which subgrid borders are treated does not unduly affect the profiles of collapsed peaks . the dotted line connects open triangles that correspond to the values given by bertschinger ( 1985 ) in his table 4 . at small radii , his solution tends towards the expected @xmath197 , and ours is in good agreement with his . the first few caustics can easily be identified . peebles ( 1987 ; see also @xcite , hereafter pmhj ) has proposed and used the following test for cosmological pm codes : integrating the evolution of a spherically symmetric under - dense region surrounding by a compensating over - dense shell . the interior of such a void expands faster than the universe as a whole , and in so doing compresses the surrounding shell , making its profile higher and narrower . the evolution can be computed analytically until the time at which the first density cusp appears . for peebles profile , @xmath198 with @xmath199 , this occurs at @xmath200 in a flat @xmath170 universe . ( here @xmath201 is the initial value of the expansion factor @xmath68 . ) we integrated such a void , realized with @xmath196 particles , on a @xmath160 grid and compared our results to the analytic prediction at @xmath202 . figure [ f : ts6aa ] can be directly compared to the corresponding figure 4 of pmhj . a significant difference between our code and theirs lies in their use of a staggered grid for the force ( @xcite ) . the staggered grid increases the spatial resolution twofold ( which is why these authors used it ) at the cost of inducing self - forces on the particles ( for which reason we did not follow their example ) . because of this difference , our results obtained with @xmath196 particles on a @xmath196 grid correspond to their results with the same number of particles on a @xmath160 grid . a comparison reveals differences of detail , particularly in the structure of the outer parts of the shell , but our results match the analytic solution as closely as theirs . we take this occasion to illustrate the energy conservation properties of the code . figure [ f : ts6aa - econs ] shows the evolution with time of the conserved quantities @xmath70 and @xmath144 of equations [ q : c ] and [ q : cprime ] ( top ) , as well as that of the ratio @xmath203 of the changes in @xmath70 and in the total potential energy @xmath125 from the start of the run ( bottom ) . apart from an initial transient due to rounding in the print - out from which the plot was made , an examination of the ratio shows the energy to be conserved to about 12% accuracy throughout the run . we have developed a useful and versatile tool for the simulation of collisionless systems of gravitating particles . our code is particularly suited to situations that call for both fine - grained mass resolution everywhere and high spatial resolution in selected regions of high density the exact locations of which need not be known in advance . our method is a development of the well - known particle - mesh technique . tests indicate that in comparable conditions our code performs substantially like similar codes described in the published literature . when subgrids are introduced to increase the spatial resolution locally , the results in the regions covered by the subgrids are in very good agreement with those of a conventional pm code with a single , finer grid . if the number of subgridded regions is sufficiently small , our approach requires less computing time and less memory than the equivalent single - grid approach . furthermore , by not increasing the resolution in regions where the particle number density is low , we avoid making the behavior of the code collisional , which would be undesirable for applications to collisionless systems . a significant advantage of pm ( and its variants p@xmath5 m , ap@xmath5 m , etc . ) for cosmological applications is that comoving coordinates and periodic boundary conditions can be supported in a natural way . however , periodic boundary conditions are not always a physically appropriate approximation . in particular , if one s interest is in systems not much smaller than the size of the computational cube , periodic boundary conditions introduce a much stronger coupling between the external tidal field and the internal dynamics of the system than one would expect to occur in the aperiodic real universe . periodic boundary conditions are only appropriate when the individual structures of interest are much smaller than the computational volume , in which case the tides due to periodic images are small ; but then the relatively small dynamic range of conventional pm methods severely limits the resolution that can be achieved . by introducing hierarchical subgrids , we are able to extend that dynamic range ; however , it is also tempting to construct the simulation in such a way that the system of interest fills as much of the computational box as possible . this is what led us to implement isolated boundary conditions . in principle , the additional cost of imposing isolated boundary conditions ( which is relatively small when hierarchical grids are used since isolated conditions have to be imposed on the subgrids in any case ) is offset by the greater freedom one has to apply an arbitrary time - dependent external tidal field and to adopt a system of expanding coordinates that matches the evolution of the simulated object without necessarily coinciding with the expansion of the global cosmological model . of course our method can also be used in a more conventional way , with periodic boundary conditions on the top grid ; in fact , numerous simplifications occur when this is done . in addition , our code should be well - suited to non - cosmological dynamical simulations , _ e.g. _ , of brief interactions between already formed galaxies , star clusters , and so on . our approach also has a few limitations . the most important is shared by all particle simulation methods : it is generally impossible to improve the spatial resolution without a corresponding refinement of the time resolution : more time steps need to be taken . other difficulties of note are the non - radial nature of forces at small separations ( which could be cured , at some cost in computing time , by softening the interaction law and increasing the depth of subgridding to compensate for this softening ) , the complexity and difficulty of tuning the subgrid placement algorithm , and the fact that subgrids of fixed shape will almost inevitably also cover some regions where the particle number density is low . the behavior of the code can become collisional in such regions ; this should not affect the computed structure of high - density condensations , but may well be a source of high - velocity particles that lead to shorter time steps according to the courant condition . an enhancement to our current code could be to refrain from applying the refined forces to such low - density cells . alternatively , it could prove useful to introduce individual time - varying softening lengths associated with individual particles , to represent the fact that particles in this method are to be thought of as possessing a finite extent that depends on the local number density . another desirable improvement is the adoption of different time steps at different levels of grid refinement ; the savings that can be achieved from this depend on the detailed structure of the tree of subgrids , and will be greatest for deep trees with most of their branches at the finer resolution levels . this work was partially supported by grants nsf - ast-86 - 57647 , nsf - ast-91 - 19475 , and nasa - nagw-2224 . some calculations were conducted using the resources of the cornell theory center , which receives major funding from the u.s . national science foundation and the state of new york , with additional support from the advanced research projects agency , the national center for research resources at the national institutes of health , ibm corporation , and other members of the center s corporate partnership program . we thank adrian melott , richard james , and the referee randall splinter for interesting and valuable remarks on an earlier version of this paper .
we present solutions for the accretion disks and boundary layers in pre - main - sequence stars undergoing fu orionis outbursts . these solutions differ from earlier disk solutions in that they include a self - consistent treatment of the boundary layer region . in a previous paper ( popham 1996 ) , we showed that these stars should stop accreting angular momentum once they spin up to modest rotation rates . here we show that for reasonable values of , these low angular momentum accretion rate solutions fit the spectra and line profiles observed in fu orionis objects better than solutions with high rates of angular momentum accretion . we find solutions which fit the observations of fu orionis and v1057 cygni . these solutions have mass accretion rates of 2 and , stellar masses of 0.7 and , and stellar radii of 5.75 and , respectively . they also have modest stellar rotation rates , comparable to the observed rotation rates of t tauri stars , and angular momentum accretion rates of zero . this supports our earlier suggestion that fu orionis outbursts may regulate the rotation rates of t tauri stars . 2i_k^2(r_in ) 2o_k^2(r_out ) 2s_k^2(r _ _ * )
astro-ph9606111
the fu orionis objects are a class of accreting pre - main - sequence stars which are undergoing dramatic outbursts in their brightness ( see hartmann , kenyon , & hartigan 1993 ; kenyon 1995 ; hartmann & kenyon 1996 for reviews ) . these outbursts are believed to result from a sudden , large increase in the mass accretion rate in an accretion disk around a t tauri star . line profiles in fu orionis systems are generally double - peaked , as expected from a rotating disk ( hartmann & kenyon 1985 , 1987 ) . the observed variation in linewidths as a function of temperature also suggests a disk origin ; line profiles in the optical are broader than those in the infrared . this is the trend expected for lines produced in a keplerian disk , where the rotational velocity and temperature both decrease with radius , so that smaller rotational linewidths are produced in cooler regions ( hartmann & kenyon 1987 ) . the spectral energy distributions of these systems are also quite broad , which is consistent with a disk with a range of effective temperatures at different radii . accretion disk models have been successful in reproducing observations of individual systems . kenyon , hartmann , & hewett ( 1988 , hereafter khh ) constructed steady thin disk models for fu orionis and v1057 cygni . they found that simple thin disk models were able to reproduce both line profiles and broad - band spectra of these systems . these models had mass accretion rates of about @xmath5 , stellar masses of @xmath6 , and stellar radii of @xmath7 . they had keplerian rotational velocities , and maximum disk temperatures of 7200 k for fu orionis and 6590 k for v1057 cygni . one major simplifying assumption made by the khh models is that the accretion disks in fu orionis systems can be approximated by a standard keplerian thin disk model . this assumption breaks down in the inner disk , where the transition from the disk to the accreting star occurs in the boundary layer region . early estimates for the temperature of the boundary layer region suggested that it would be much hotter than permitted by observations ( hartmann & kenyon 1985 ; kenyon 1989 ) . these estimates were made by assuming that the boundary layer luminosity is half of the total accretion luminosity , and that it is radiated from a small region of the inner disk , with a radial extent of a few percent of the stellar radius . these large luminosities radiated from such a small area would produce boundary layer temperatures approaching 30,000 k , which would mean that the boundary layer would emit large ultraviolet fluxes . iue observations of fu orionis objects by kenyon ( 1989 ) demonstrated that these large fluxes were not present . kenyon ( 1989 ) suggested three possible reasons why the boundary layer might not be as hot as predicted by these simple estimates . first , some of the boundary layer energy might be going into expanding the central star . second , the central star might be rotating rapidly due to accretion spinup . third , the disk might not be thin , and so the radial extent of the boundary layer might be much larger than a few percent of the stellar radius . recently , we have begun to calculate self - consistent solutions for the structure of boundary layers in fu orionis systems . these solutions suggest that all of the mechanisms mentioned by kenyon ( 1989 ) act to decrease the effective temperature of the boundary layer below the temperatures suggested by simple estimates . popham ( 1993 , hereafter pnhk ) obtained boundary layer solutions for pre - main - sequence stellar parameters and mass accretion rates @xmath8 ranging from @xmath9 . they showed that at the higher accretion rates , which correspond to fu orionis systems , the radial extent of the boundary layer region becomes quite large , and the boundary layer temperature drops much closer to that of the inner disk . thus as @xmath8 increases it becomes progressively more difficult to distinguish the contribution of the boundary layer region to the overall spectrum . pnhk also showed that in high-@xmath8 systems , the accreting material is quite hot when it reaches the stellar radius , with a temperature which is a substantial fraction of the virial temperature . this means that a substantial fraction of the accretion luminosity is advected into the star , decreasing the luminosity radiated by the boundary layer . pnhk calculated two models with @xmath10 , one with @xmath11 and the other with @xmath12 . the @xmath11 solution reached a peak @xmath13 k , with @xmath14 , while the @xmath15 solution had a lower peak @xmath16 k and @xmath17 . godon ( 1996 ) confirmed these results with a time - dependent model , finding a peak @xmath18 of 14,000 k and @xmath19 for @xmath20 and @xmath21 . recently , popham ( 1996 , hereafter paper i ) presented boundary layer solutions for fu orionis parameters , and examined the effect of fu orionis outbursts on the spin evolution of the underlying t tauri stars . event statistics suggest that the outbursts may dominate the mass accretion onto t tauri stars ( hartmann & kenyon 1996 ) ; if so , they should also dominate the angular momentum accretion . disk accretion generally adds angular momentum to the accreting star and spins it up ; however , when the accreting star nears breakup , the angular momentum accretion rate drops rapidly and becomes negative , so that the accreting star stops spinning up ( popham & narayan 1991 ; paczyski 1991 ) . paper i showed that for the disk parameters which characterize fu orionis systems , the accreting star stops spinning up once it reaches 20 - 40% of breakup speed ( i.e. , the keplerian rotation rate at the stellar surface ) . the star can then continue to accrete mass while maintaining an equilibrium rotation rate . paper i suggested that these low angular momentum accretion rate solutions , obtained during outbursts , play an important role in maintaining the observed slow rotation rates of t tauri stars . the boundary layer region should make an important contribution to the observed spectral energy distribution and spectral line profiles of fu orionis systems . in the boundary layer , the rotational velocity of the accreting material , which determines the linewidths , varies rapidly with radius . it drops from its disk value , generally close to keplerian , to the surface rotational velocity of the star , which is usually much slower . as the material loses its rotational kinetic energy , up to half of the accretion luminosity is released in the boundary layer region . since disk models which include the boundary layer will have quite different temperature and rotational velocity profiles from keplerian thin disk models , they will produce somewhat different spectra and line profiles . also , as shown in paper i , low angular momentum accretion rate solutions have different temperature and rotational velocity profiles from those with high angular momentum accretion rates . this raises several questions : 1 ) whether solutions which include the boundary layer region will be able to match the observations of fu orionis systems ; 2 ) which types of boundary layer solutions will match the observations , and 3 ) how the inclusion of the boundary layer will change the model parameters inferred from keplerian disk solutions . in this paper , we attempt to answer these questions by calculating boundary layer solutions for fu orionis systems , and comparing the spectra and line profiles produced by these solutions to the observations . in 2 , we briefly describe our boundary layer model and the procedures we use for calculating spectra and line profiles from our solutions . in 3 , we discuss the general characteristics of the spectra and profiles produced by our boundary layer solutions . we show how these characteristics depend on solution parameters such as the mass accretion rate , the stellar mass , radius , and rotation rate , the angular momentum accretion rate , and the viscosity parameter @xmath0 . we compare our solutions to observations of the spectra and line profiles of individual fu orionis systems in 4 , and discuss our results and their implications in 5 . 6 gives a summary . the model we use to calculate the structure of the boundary layer and accretion disk is identical to the one used in paper i and very similar to the model used in several earlier papers ( narayan & popham 1993 ; pnhk ; popham & narayan 1995 ) , where it is described in detail . the model uses the steady - state , axisymmetric slim disk equations originally developed by paczynski and collaborators for modeling disks around black holes ( paczynski & bisnovatyi - kogan 1981 ; muchotrzeb & paczynski 1982 ; abramowicz 1988 ) to describe the radial structure of the disk . the slim disk equations dispense with some of the assumptions of the standard thin disk model ( shakura & sunyaev 1973 ) , and include terms which are assumed to be small in the thin disk formulation . these include the radial pressure gradient and acceleration in the radial momentum equation , and radial energy transport by radiation and advection . also , while the thin disk equations assume that the angular momentum accretion rate takes on a standard value , the slim disk equations allow for different angular momentum accretion rates . the additional terms in the slim disk equations involve radial derivatives which make them differential equations requiring a numerical solution . we use approximate relations to describe the vertical structure of the disk . we solve the equations using a relaxation method , on a 1001-point radial grid which extends from @xmath22 to @xmath23 , with grid points concentrated in the inner boundary layer region . our solutions yield the effective temperature of the disk surface as a function of radius . from this , we construct spectra by simply adding up the contributions from each disk annulus . we use two different methods for calculating the spectrum of each annulus . the first is simply to assume that the disk surface emits as a blackbody at the local effective temperature . this has the advantages of being simple and independent of any assumptions about the vertical structure of the disk . the second method is to use a grid of stellar atmospheres , using the methods described by hartmann & kenyon ( 1985 ) . this should do a better job of reproducing features such as the balmer jump ; however , it is still only approximate , since the disk atmosphere probably differs somewhat from a stellar atmosphere . we assume that the only effect of inclination is to change the projected area of the disk as seen by an observer . thus , the observed flux from the disk is @xmath24 , where @xmath25 is the total disk luminosity , @xmath26 is the inclination angle , and @xmath27 is the distance . at the average value of @xmath28 , we have the standard expression @xmath29 . in order to calculate spectral line profiles from our models , we begin by calculating the line profile produced by a single annulus of the disk . we take the profile function for a rotating disk , @xmath30^{-1/2}$ ] , where @xmath31 is the deviation in velocity from line center , and @xmath32 is the projected rotational velocity at inclination angle @xmath26 . we convolve this profile function with a gaussian broadening function @xmath33 $ ] , where @xmath34 is the instrumental broadening due to the resolution of the spectrograph . next , we sum up the line profile contributions from all disk annuli . we weight the annuli according to their contribution to the blackbody flux at the wavelength of the spectral line . this is equivalent to assuming that the equivalent width of the line stays constant as the effective temperature varies . for the lines we consider , metal lines at 6170 and co lines at 2.2 @xmath35 , this is a reasonable assumption , except at high temperatures , where the species in question disappear . for this reason , we set the weight to zero when the effective temperature exceeds a cutoff value , which we take to be 8000 k for the 6170 lines and 5000 k for the 2.2 @xmath35 lines . our disk and boundary layer solutions have a number of free parameters which can affect their spectra and line profiles . these include the usual parameters used to characterize accreting systems : the mass accretion rate @xmath8 , and the stellar mass and radius @xmath36 and @xmath22 . as discussed in paper i , even when these parameters are specified , there are still a range of solutions available , which are characterized by different values of the stellar rotation rate @xmath37 and the disk height at the stellar radius @xmath38 , which then specify the angular momentum accretion rate @xmath39 . finally , one can specify the parameter @xmath0 , which characterizes the magnitude of the viscosity . with such a wide range of solution parameters available , we begin by examining the effects of each parameter on our model spectra and line profiles . we then compare these to the general observed characteristics of fu orionis systems in order to narrow down the parameter space in which solutions in reasonable agreement with the data will be found . later in 4 , we will attempt to find solutions which match the observations of individual systems ( fu orionis and v1057 cygni ) . the general observed characteristics of fu orionis systems , which we will use to try to limit the parameter space for our solutions , are as follows : the line profiles are generally double - peaked , both in the optical and the infrared ( hartmann & kenyon 1985 , 1987 ) . the line widths vary widely from system to system ; some are below the instrumental resolution of @xmath40 , while others have fwhm values in excess of @xmath41 . such a range is not surprising , since the linewidth depends on the projected rotational velocity @xmath42 . the linewidths in the optical are broader than those in the infrared ; linewidths measured at 6170 are @xmath43 larger than those measured at 2.2 @xmath44 . the dereddened broad - band spectra of fu orionis systems generally have their peak @xmath45 values in the red part of the spectrum ( khh ) . the spectra fall off quite rapidly at blue and ultraviolet wavelengths , with @xmath45 in the u band a factor of three or more smaller than the peak value . the best estimates for the distances of these systems give total luminosities of a few hundred @xmath46 ( bell & lin 1994 ; hartmann & kenyon 1996 ) . the three most fundamental parameters describing an accretion disk are the mass accretion rate @xmath8 and the mass @xmath36 and radius @xmath22 of the accreting star . in the thin disk formulation , these parameters can be combined to derive three standard quantities which characterize the spectra and linewidths produced by the disk : the accretion luminosity @xmath47 , the maximum disk temperature @xmath48 , and the keplerian rotational velocity at @xmath22 , @xmath49 . these quantities provide a good general guide to the variations in the spectra and line profiles that result from variations in @xmath8 , @xmath36 , and @xmath22 . the accretion luminosity characterizes the brightness ( the vertical position ) of the spectrum , the maximum temperature controls the peak wavelength ( the horizontal position ) of the spectrum , and the keplerian rotational velocity at @xmath22 characterizes the width of a line profile . for a standard thin accretion disk , the quantities @xmath50 , @xmath51 , and @xmath52 represent the exact values of the luminosity and the maximum effective temperature and rotational velocity of the disk . in our models , the situation is more complex , for two reasons . first , we include the boundary layer region in our solutions . the effects of variations in @xmath8 , @xmath36 and @xmath22 upon the boundary layer are more complex than their effects upon the disk , and the boundary layer is also affected by other parameters such as the rotation rate of the accreting star @xmath37 and the angular momentum accretion rate . second , our slim disk equations include important terms , described in 2 , which are not included in the standard thin disk model . thus , for our solutions , @xmath50 , @xmath51 , and @xmath52 are only characteristic values which provide a good approximate description of the outer parts of the disk , but not of the inner portions close to the star . the radiated luminosity of our solutions can differ substantially from @xmath50 , since energy can be advected into the star , and variations in @xmath37 change the boundary layer luminosity . the boundary layer region is usually hotter than @xmath51 , and the rotational velocity of the disk is often well below keplerian there . as an example of the changes in the boundary layer structure produced by variations in @xmath8 and @xmath22 , we show in figure 1 three solutions with the parameters used in paper i. the first solution has @xmath53 , @xmath54 , the second has @xmath55 , @xmath56 , and the third @xmath57 , @xmath58 . note that these choices of parameters keep the accretion luminosity @xmath59 approximately constant at @xmath60 , since the ratio @xmath61 stays nearly constant . the maximum disk temperature @xmath51 and the keplerian rotational velocity at the stellar radius @xmath52 both decrease as @xmath8 and @xmath22 increase . all three solutions have @xmath62 , and @xmath63 , and all three use the boundary condition @xmath64 , which will be discussed below . these are high angular momentum accretion rate solutions . throughout this work , we express the angular momentum accretion rate @xmath39 as @xmath65 , so that @xmath66 for a standard thin disk solution . the solutions in fig . 1 have @xmath67 , 0.9 , and 0.8 , respectively . figure 1a shows the angular velocity @xmath68 for these solutions . all three have a peak in @xmath68 ; inside the peak , @xmath68 drops to @xmath69 , while outside the peak , @xmath68 decreases at a roughly keplerian rate . the main feature we wish to emphasize is that the location of the peak changes from about @xmath70 to @xmath71 as @xmath8 increases ; thus the dynamical boundary layer width doubles from @xmath72 to @xmath73 as @xmath8 doubles . another important feature is that as @xmath8 increases , @xmath68 drops farther below @xmath74 , indicated by a dashed line . at @xmath57 , the peak value of @xmath68 is only half of @xmath74 at that radius . this is caused by the increase in pressure support at large values of @xmath8 . figure 1b shows the effective temperatures of these solutions . all three reach peak values just outside of @xmath22 , but the peak is much more pronounced at lower @xmath8 . this is not surprising , since the dissipation of energy is much more concentrated in the smaller boundary layer . as @xmath8 increases , the peak in @xmath18 due to the boundary layer dissipation becomes broader until it becomes difficult to distinguish it from the @xmath18 profile of the disk . this broadening and cooling of the boundary layer with increasing @xmath8 was pointed out in our earlier paper ( pnhk ) for @xmath8 ranging from @xmath75 to @xmath5 . 1 shows that at the upper end of this range , a factor of two increase in @xmath8 can make a qualitative difference in the appearance of the boundary layer region . the 6170 line profiles for these solutions are shown in figure 1c . these profiles reflect the variations in the rotational velocities in the inner portion of the disk . all three profiles are centrally peaked , due to the presence of the hot , slowly rotating material in the boundary layer region . at the lowest @xmath8 , the rapidly rotating inner disk produces a double - peaked component , which is filled in by the contribution from the more slowly rotating boundary layer , leaving broad shoulders on the combined line profile . as the boundary layer extends to larger radii at higher values of @xmath8 , the double - peaked component becomes even more difficult to distinguish ; it only appears as wings on the centrally - peaked boundary layer profile . the width of the profiles decreases dramatically as @xmath8 and @xmath22 increase , since the peak rotational velocity is much smaller . none of the profiles resemble the double - peaked profiles observed in fu orionis systems ; clearly the inclusion of the boundary layer makes a crucial difference in the overall line profile by reducing the rotational velocities over a wide region of the inner disk , where the disk is brightest . figure 1d shows blackbody spectra of these solutions . they have nearly the same luminosity , but the spectrum moves to longer wavelengths as @xmath8 increases : the peak @xmath45 moves from @xmath76 to @xmath77 . this change in peak wavelength is larger than would be expected from the change in the @xmath51 of the disk , due to the decrease in the boundary layer effective temperature . separate boundary layer and disk components are not distinguishable in the spectrum , as they are at lower values of @xmath8 ( pnhk ) . in paper i , we showed that a wide range of boundary layer solutions can be found for a single set of values of @xmath8 , @xmath36 , and @xmath22 , these solutions have different values of the stellar rotation rate @xmath37 and the disk height at the stellar radius @xmath38 , and they also have different values of the angular momentum accretion rate @xmath78 . note that only two of these three variables are independent , so we can specify a unique solution by choosing the values of two of them . we plotted the locations of these solutions in the @xmath79 plane as lines of constant @xmath78 . we found that solutions with high angular momentum accretion rates @xmath80 formed z - shaped tracks which fell in the low-@xmath37 or the low-@xmath38 regions of the plane , as illustrated in figure 2 for solutions with @xmath81 , @xmath55 , @xmath62 , @xmath56 , and @xmath82 . we were also able to find solutions with small or negative angular momentum accretion rates , @xmath83 . these solutions have moderate values of both @xmath37 and @xmath38 , as shown by the @xmath84 and @xmath85 tracks in figure 2 . how do the spectra and line profiles produced by these two types of solutions compare to the data ? we begin by considering high-@xmath78 solutions . figure 3 shows a set of solutions with @xmath81 which sit on the z - shaped track in figure 2 . they have different values of @xmath37 and @xmath38 . the low-@xmath38 solutions have a narrow boundary layer , as evidenced by the pronounced peaks in both @xmath68 and @xmath86 close to @xmath22 . the high-@xmath38 solutions have much broader boundary layers , and much flatter @xmath68 and @xmath18 profiles . these are reflected in very different line profiles and spectra ; the high-@xmath38 solutions have strongly centrally peaked narrow profiles as a result of their small rotational velocities throughout the wide boundary layer . as @xmath38 decreases , the rotational velocity in the inner disk increases , and the boundary layer gets smaller ; thus the profiles get wider , and the double - peaked disk component becomes more prominent , forming a triple - peaked profile at @xmath87 and finally a double - peaked profile at @xmath88 . the spectra produced by these solutions also vary substantially with @xmath38 ; as @xmath38 decreases , the smaller boundary layer reaches higher peak effective temperatures and produces bluer spectra . none of the high-@xmath78 solutions agrees with the observed spectra and line profiles of fu orionis systems . only the low-@xmath38 solutions produce reasonably double - peaked profiles , but their spectra are too blue . the @xmath87 solution peaks around @xmath89 ; the @xmath88 solution actually peaks in the red , but is still fairly bright in the blue and ultraviolet , unlike the observed systems . note that the @xmath90 solution has both a lower luminosity and redder peak wavelength than the @xmath87 solution , reversing the trend of the other solutions this points to a more serious problem , which was discussed in paper i : in the low-@xmath38 solutions , much of the boundary layer is located at @xmath91 . this can be seen by noting that @xmath92 is still quite large at @xmath22 , so that @xmath68 must continue to drop inside @xmath22 . some portion of the boundary layer flux is thus not being included in the spectrum . in order to avoid these problems , we need to specify an additional boundary condition . the need for an additional condition is also clear from the fact that we have needed to specify values for two of the three variables @xmath37 , @xmath38 , and @xmath78 . in nature , one expects that the stellar rotation rate @xmath37 is the only true variable , and that the disk height at the stellar surface @xmath38 and the angular momentum accretion rate @xmath78 will be determined by the accretion flow . an additional condition will allow us to specify only @xmath37 in our solutions , and the values of @xmath38 and @xmath78 will follow . as discussed in paper i , we have selected a somewhat arbitrary condition on the radial velocity at @xmath22 , @xmath93 . in practice , this keeps the value of @xmath38 nearly constant with varying @xmath37 . as we showed in paper i , @xmath78 drops fairly rapidly as @xmath37 increases , and becomes negative when @xmath37 reaches 20 to 40% of the breakup stellar rotation rate @xmath94 . with this additional condition , we are ready to examine spectra and line profiles for low-@xmath78 solutions . figure 4 shows five solutions with different values of @xmath95 , 0.9 , 0.84 , 0.6 , 0 , which lie along the track with @xmath93 . the stellar rotation rate @xmath37 increases as @xmath78 drops ; @xmath96 at @xmath97 . the high-@xmath78 solutions have @xmath68 reaching a maximum and then dropping down to @xmath37 , while in the low-@xmath78 solutions , @xmath68 continues to increase all the way in to @xmath22 . the high-@xmath78 solutions also have a more pronounced peak in @xmath18 , while the low-@xmath78 solutions have lower effective temperatures in the boundary layer , but higher in the disk . the high-@xmath78 solutions produce centrally - peaked line profiles similar to the ones in figs . 1 and 3 . on the other hand , the @xmath98 and @xmath84 solutions produce double - peaked profiles . the low-@xmath78 solutions also have redder spectra . the @xmath84 spectrum is more luminous than the others , but it peaks in the red , and drops off rapidly in the blue and ultraviolet due to the lack of a high-@xmath18 boundary layer region . thus , we find that low-@xmath78 solutions match the observations better than high-@xmath78 solutions . none of the high-@xmath78 solutions have spectra and line profiles which agree with observations ; they either produce centrally - peaked line profiles , or spectra which are bluer than the observed ones . the @xmath84 solution , with moderate values of both @xmath37 and @xmath38 , produces spectra and line profiles which are in much better qualitative agreement with the observations . note that this conclusion is based on solutions calculated with @xmath82 ; we discuss solutions for other values of @xmath0 below . why do low-@xmath78 solutions fit the observations better ? some insight into this can be gained by noting that khh were able to find good agreement with spectra and line profiles of the same systems , using standard thin disk models with the usual run of effective temperatures and rotational velocities . this leads one to wonder why both types of disk models , with or without boundary layers , can be successful in reproducing the observed characteristics of fu orionis systems . the answer lies in the fact that the boundary layer solutions which best fit the observations , our low-@xmath78 solutions , in some sense lack a boundary layer . conversely , more conventional high-@xmath78 solutions clearly do not agree with the observations . the low-@xmath78 solutions do not show the rapid drop in the angular velocity between the disk and the star which is usually associated with the boundary layer region . instead , @xmath68 increases slowly inward , so that the rotational velocity in the innermost region of the disk stays nearly constant over a wide range of radius . this produces double - peaked line profiles like the ones observed in fu orionis systems . also , @xmath37 is moderately large , generally around 20% of the breakup stellar rotation rate @xmath94 , so that there is no part of the inner disk which has very low rotational velocity . this inner portion of the disk produces most of the optical flux , so any region of it which has very low rotational velocity tends to produce a peak in the center of the line profile , eliminating the double - peaked structure . a second important characteristic of the low-@xmath78 solutions is that they lack the pronounced peak in the effective temperature which is characteristic of conventional high-@xmath78 boundary layer solutions . in high-@xmath78 solutions , this peak is produced by the rapid release of energy which accompanies the rapid drop in @xmath68 . this energy is released at fairly high effective temperatures , which produces blue and ultraviolet fluxes substantially in excess of those observed . for example , the high-@xmath78 , low-@xmath38 solutions in fig . 3 reach effective temperatures of @xmath99 10,000 k , whereas low-@xmath78 solutions for the same parameters only reach about 7,500 k. at the same time , low-@xmath78 solutions are fairly luminous , as illustrated by fig . 4 . although these solutions are cooler than high-@xmath78 solutions in the innermost region , they are hotter in the remainder of the disk . this is a result of the different temperature distribution of a disk with a different value of @xmath78 . for example , if we adopt the standard thin disk assumptions that @xmath100 and that the flux radiated from the disk surface is just the energy per unit area released by viscous dissipation , then the effective temperature of the disk is given by @xmath101^{1/4}.\ ] ] for @xmath102 , we recover the usual expression for @xmath18 , but for @xmath84 , the factor in square brackets is eliminated , producing higher effective temperatures . note that this expression only applies when the thin disk assumptions listed above are satisfied , so that it will not apply to the innermost portion of the disk , where @xmath68 deviates strongly from keplerian , and radial energy transport is important . however , at larger radii , the disk comes closer to satisfying these conditions , and @xmath84 solutions should produce higher effective temperatures than high-@xmath78 solutions . this is confirmed by fig . 3 , where the @xmath84 solution is cooler than the high-@xmath78 solutions at @xmath103 , but hotter at @xmath104 . solutions with @xmath105 are also `` equilibrium '' solutions which allow the star to accrete without spinning up or down ( see paper i and 5 for further discussion of this point ) . for these reasons , we concentrate on @xmath84 solutions in the remainder of this paper . one additional parameter that must be specified in our model is the viscosity parameter @xmath0 . most estimates of @xmath0 are based largely on the assumption that the outbursts observed in dwarf novae proceed on the viscous timescale . this produces fairly large values of @xmath106 ; however , application of the same idea to fu orionis outbursts suggests that much smaller values of @xmath0 are required to reproduce the long outbursts observed in these systems . bell & lin ( 1994 ) found that disk instability models for fu orionis systems required values of @xmath107 where hydrogen is neutral and @xmath108 where hydrogen is ionized . however , such small values of @xmath0 lead to problems with evolutionary timescales and gravitational instability in the disk ( bell 1995 ) . in this work , we have adopted a compromise value of @xmath63 for most of our models . in order to explore the effects of varying @xmath0 in our models , we have calculated a sequence of solutions with @xmath109 . these solutions all have the same values of @xmath110 , @xmath62 , @xmath56 , and @xmath84 , and we have varied the value of @xmath111 such that @xmath112 . these solutions are shown in figure 5 . the larger values of @xmath0 correspond to thinner disks , which produce faster stellar rotation rates and higher effective temperatures in the inner disk . if we were to continue to increase @xmath0 ( perhaps to unreasonably large values @xmath113 ) , eventually we would reach a nearly keplerian , thin disk solution . such a solution would have the star rotating near breakup , and would resemble the solutions found by khh . the smallest value , @xmath114 , produces a solution in which the rotation rate is very small throughout the inner disk . this results in a very narrow , centrally - peaked line profile . larger values of @xmath0 produce progressively wider double - peaked profiles . there is also a clear trend in the spectra as @xmath0 changes ; larger values of @xmath0 produce bluer spectra and higher total luminosities . the variation in luminosity is due to the advection of a larger fraction of the accretion luminosity into the star in the small-@xmath0 solutions , where the disk has become quite thick . we now use our disk and boundary layer solutions to fit the spectra and line profiles of the two best - observed fu orionis systems : fu orionis and v1057 cygni . both have resolved line profiles available in both the optical and near - infrared , along with broad - band photometry which illustrates the shape of the spectral energy distribution from 3600 out to 10 @xmath35 and beyond . these were two of the original three systems discussed by herbig ( 1977 ) in his seminal paper . the third system , v1515 cygni , has very narrow , unresolved line profiles which suggest that it is probably viewed pole - on , making it difficult to constrain models for this system . another well - observed system which is a probable member of the fu orionis class is z cma , which has very broad line profiles . unfortunately , this system has a binary companion which greatly alters the shape of the infrared spectrum . the other candidate systems have some of these data available , but not all of them , and for many of them their identification as fu orionis systems is still somewhat uncertain ( see hartmann & kenyon 1996 for a discussion of these objects ) . most of the observations of fu orionis and v1057 cygni used in this work are the ones which were used by khh in their comparison to keplerian thin disk models . these consist of high - resolution spectra , which are used to derive line profiles , and photometry to show the broad - band spectral energy distributions of these objects . the high - resolution spectra include echelle spectra of a 50-region around 6170 , and fourier transform spectra of the 2.2 @xmath35 region . we have used a more recent 2.2 @xmath35 spectrum of v1057 cyg with improved resolution of @xmath115 . unlike the earlier spectrum , the cross - correlation produced from this spectrum is clearly double - peaked . as described by khh , the photometric data include low - resolution optical spectrophotometry which has been binned into the standard photometric bands , along with standard optical and infrared photometry . the photometric data have been dereddened using the extinction law given by savage & mathis ( 1979 ) and assuming @xmath116 mag for fu orionis and @xmath117 mag for v1057 cygni . based on the results discussed in 3 , in attempting to fit the observations , we have fixed certain parameters : since low-@xmath78 solutions seem to fit the general characteristics of the observations , we have fixed @xmath84 . we also have fixed the boundary condition on @xmath64 . this together with the condition on @xmath78 fixes the position of the solution in the @xmath79 plane ( see fig . we also fix the value of @xmath118 at @xmath119 . this leaves the standard accretion parameters @xmath8 , @xmath36 , and @xmath22 , which we vary in order to fit the observations . as discussed in 3.1 , we can use these three parameters to calculate @xmath120 , @xmath121 , and @xmath122 , which relate more directly to the observations which we are trying to match . these parameters give us a good sense of how changes in @xmath8 , @xmath36 , and @xmath22 will affect our spectra and line profiles . for instance , if we want to increase the luminosity of a particular solution without changing the temperature or rotational velocity , the dependences given above suggest that we need to keep the ratios @xmath123 and @xmath124 constant . if we increase @xmath22 , we can increase @xmath36 by the same factor and increase @xmath8 as @xmath125 . this will increase @xmath50 as @xmath125 while keeping @xmath51 and @xmath52 constant . there are several other parameters which enter into our fits : the inclination angle , distance , and extinction of the individual fu orionis systems . the inclination angle @xmath26 and the distance @xmath27 affect the derived luminosity of the fu orionis systems , since we convert the luminosity of our solutions to flux using the relation @xmath126 , and then compare these fluxes to the observed ones . the inclination angle also affects the linewidths derived from our solutions , since the linewidth is set by the line - of - sight velocity of the disk material @xmath127 . the extinction @xmath128 affects both the shape and the luminosity of the spectra , since the observed photometric points are dereddened before they are used for comparison with the solutions . we adopt values published elsewhere for the distance ( hartmann & kenyon 1996 ) and extinction ( khh ) , and leave these fixed . we vary the inclination as part of our fitting procedure . our fits to the observations are all approximate , and are done by eye . in comparing the spectra produced by our solutions with photometric data , we weight some of the data points less heavily than others . at wavelengths longer than 10 @xmath35 , the data frequently show that the objects are much brighter than our models would predict . this is generally attributed to the presence of reprocessing of some of the disk luminosity by the outer disk or by a dusty envelope surrounding the disk ( khh ; kenyon & hartmann 1991 ) . thus we confine our comparison with the photometric data to wavelengths shorter than @xmath129 . at the blue end of the spectrum , the u - band point at 3600 is generally substantially fainter than the b - band point at 4400 . this rapid drop in the spectral energy distribution is probably due largely to the balmer jump , and it is therefore difficult to reproduce it with a blackbody spectrum . our spectra produced using a library of stellar spectra tend to be closer to the observed point . a number of other factors could also contribute to the rapid dropoff from b to u ; the extinction increases dramatically as one moves to shorter wavelengths in this portion of the spectrum , and so error in the dereddening applied to the observed spectrum will have their greatest effect here . also , inclination effects tend to depress the blue end of the spectrum . when we compare our line profiles to the data , we do not compare them directly to individual line profiles , which tend to be noisy and frequently blended with other nearby lines , but rather to the shape of the cross - correlation function . the cross - correlation function represents the mean line shape in the spectral region for which it is derived . we have assigned the zero point of our model line profiles to the zero point of the cross - correlations , and normalized the two by their peak heights . we quantitatively compare the overall shape of our line profile to the cross - correlation by determining the velocity widths of the two at half - maximum and at full - maximum . we also compare the ratio of the optical and infrared line widths produced by our solutions to the ratio seen in the data . this object has broad line profiles , with full - width at half - maximum of @xmath41 at 6170 and @xmath130 at 2.2 @xmath35 . it is also quite bright , with a peak value of @xmath131 in the v - band , assuming @xmath132 . adopting a distance of 500 pc , we find good fits to the line profiles and spectrum from a solution with @xmath133 , @xmath134 , @xmath135 , and @xmath136 . as mentioned above , the solution has @xmath84 and @xmath82 . it also has @xmath93 , giving a stellar rotation rate of @xmath137 and a disk height at the stellar radius @xmath138 . these values correspond to a stellar rotation period of 8.76 days and @xmath139 . the angular velocity @xmath68 and rotational velocity @xmath140 are shown in figure 6a . the rotational velocity at the stellar surface is only @xmath141 , and the maximum rotational velocity is @xmath142 at @xmath143 . figure 6b shows the effective temperature @xmath18 , which reaches a maximum value of 8212 k at @xmath144 . figure 6c shows the line profiles derived from this disk model at 6170 and 2.2 @xmath35 , respectively . both profiles are double - peaked , like the cross - correlations , and we find fwhm values of 98 and @xmath145 at the two wavelengths . note that the cross - correlation peaks are rather asymmetric ; this is probably due to the presence of a mass loss in a wind , as discussed by khh . figure 6d shows the fits of blackbody and stellar composite spectra to the photometry of fu orionis . both spectra fit the data points quite well from the b band to 10 @xmath35 , deviating from most data points by 10% or less . both are too bright in the u band , although the stellar composite spectrum comes substantially closer to the observed point . the luminosity emitted by the boundary layer and disk is @xmath146 . if this were a standard keplerian thin disk , the disk parameters @xmath8 , @xmath36 , and @xmath22 for this solution would give an accretion luminosity @xmath147 , a maximum disk temperature @xmath148 k , and a keplerian rotational velocity @xmath149 . these are substantially different from the actual luminosity , peak effective temperature , and peak rotational velocity found in our solution . these differences indicate the importance of including the boundary layer region in our disk model . this system has narrower line profiles than fu orionis , with fwhm values of @xmath150 at 6170 and @xmath142 at 2.2 @xmath35 . it also appears to be somewhat fainter ; after dereddening with @xmath117 , it has a peak @xmath151 , and the peak occurs in the r band , indicating that this system is slightly cooler than fu orionis . we adopt a distance of 600 pc . we find good fits to the spectrum and line profiles of v1057 cygni from a disk and boundary layer solution with @xmath20 , @xmath62 , and @xmath152 . this solution is less luminous and slightly cooler than the solution for fu orionis described above , but in most other respects the two solutions are quite similar . like the solution for fu orionis , this solution has @xmath84 , @xmath63 , and @xmath93 this gives a stellar rotation rate @xmath153 , which corresponds to a rotation period of 8.19 days . the disk height at @xmath22 is @xmath154 , so @xmath139 . the angular and rotational velocities of this solution are shown in figure 7a . the rotational velocity is @xmath155 at the stellar surface , and reaches a peak value of @xmath156 at @xmath157 . the effective temperature , shown in figure 7b , reaches a maximum value of 7111 k at @xmath158 . figure 7c shows the line profiles at 6170 and 2.2 @xmath35 for this solution . the profiles have fwhm values of 54 and @xmath159 , respectively , compared to the observed values of 57 and @xmath142 . the ratio of the optical to infrared linewidth is quite close to the observed ratio , where the khh thin disk model gave infrared linewidths which were about 25% smaller than those observed . the separation of the two peaks is 26 and @xmath160 in our disk solution , and 33 and @xmath161 in the data . the fits of our blackbody and stellar composite spectra to the photometric data for v1057 cygni are shown in figure 7d . the blackbody spectrum fits quite well in both the optical and near - infrared regions ; it is too bright in the u - band , as expected . the data show a drop in the i - band and an excess at 3.5 and 4.8 @xmath35 ; these points would be difficult to fit with any smooth curve which also fits the optical and near - infrared points . the stellar composite spectrum comes much closer to the u - band point but is slightly fainter than the data in the optical . it is interesting to compare the disk parameters @xmath8 , @xmath36 , and @xmath22 of our best - fit models for fu orionis and v1057 cygni to those found by khh using thin disk models . khh list @xmath36 and @xmath22 for each system as a function of the inclination angle @xmath26 ; these values of @xmath36 and @xmath22 keep the projected rotational velocity at the stellar surface @xmath162 constant at @xmath163 for fu orionis and @xmath164 for v1057 cygni . khh also keep the maximum disk temperature constant at 7200 k in fu orionis and 6590 k in v1057 cygni , so @xmath8 also varies with inclination . our models use an inclination of @xmath165 for fu orionis and @xmath166 for v1057 cygni . for these inclinations , khh found @xmath167 , @xmath168 , and @xmath169 for fu orionis , and @xmath170 , @xmath171 , and @xmath172 for v1057 cygni . our best - fit models give @xmath134 , @xmath173 , and @xmath133 for fu orionis , and @xmath62 , @xmath174 , and @xmath20 for v1057 cygni . thus we find stellar masses which are a factor of 23 larger , mass accretion rates which are a factor of 23 smaller , and slightly larger stellar radii . these differences are due to the fact that our models include the boundary layer region . thus , rotational velocities are substantially below keplerian in the inner disk , requiring larger stellar masses to produce the observed linewidths . to compensate for this increased stellar mass , the mass accretion rate must decrease by a similar factor in order to produce similar luminosities and temperatures . note that the khh models only included the disk luminosity , which is half of the total accretion luminosity in the standard thin disk model . our models include both the disk and boundary layer luminosities , but they do not radiate the full accretion luminosity , as mentioned in 4.3 and discussed below . finally , the presence of the boundary layer increases the inner disk temperatures , and so a slightly larger stellar radius is required to keep the spectra from becoming too blue . we have found disk and boundary layer solutions with significant departures from keplerian rotation that can reproduce the key spectral features of fu orionis and v1057 cygni . however , observations of these objects were originally modelled with reasonable success using standard keplerian disk theory ( hartmann & kenyon 1985 ; khh ) . here we discuss why it is difficult to compare the relative merits of the two types of solutions on the basis of observations alone . we then outline why we think the boundary layer solutions are nevertheless an important improvement in treating fu orionis accretion physics . both standard thin disk solutions and disk and boundary layer solutions agree quite well with observations . thus , comparisons with observations do not provide a definitive test of the relative merits of the two types of solutions for fu orionis objects . the disk and boundary layer solutions appear to match the differential rotation with wavelength somewhat better than the thin disk solutions of khh ; the thermal pressure support in the inner regions of the disk tends to reduce the ratio of the optical line widths to the near - infrared line widths , as observed . however , as pointed out by khh , there are significant uncertainties in strengths of the infrared co lines in the outer disk due to the possibility of dust formation , and with plausible models of dust contributions a thin disk in keplerian rotation can match the observed differential rotation with wavelength . similarly , differences in how well the two types of solutions match the observed spectral energy distributions , or small differences in the inferred masses , radii , and accretion rates , are not significant in view of the many uncertainties involved . specifically , there are uncertainties in the inclination angles @xmath26 , distances @xmath27 , and extinctions @xmath128 of fu orionis and v1057 cygni . we have used fairly standard inclinations of @xmath136 for fu orionis and @xmath175 for v1057 cygni . these inclinations allow solutions with similar rotational velocities to produce the rather different linewidths seen in these two systems . in fact , the ratio @xmath176 is almost identical to the ratio of the 6170 linewidths ( 57 and @xmath41 ) in these objects . nonetheless , other inclinations , in combination with different choices of the other solution parameters , might produce solutions which match the data . the distances to fu orionis and v1057 cygni are probably only accurate to 1020% , so that the luminosities inferred for these systems could be in error by 2040% . finally , the extinctions of these systems are not well known , and could represent an additional source of error in both the luminosities and the shape of the spectra . all of these considerations limit our ability to distinguish between different disk models . nonetheless , we feel that the disk and boundary layer model presented in this paper offers significant advantages over the standard thin disk model . first , it avoids some uncomfortable assumptions which are implicit in the thin disk model . the two major observational constraints on models of the inner disk in fu orionis systems are the apparent lack of hot boundary layer emission and the lack of slowly - rotating material in the observed line profiles . the absence of these features gives the impression that the boundary layer is not present is fu orionis systems . the standard thin disk model ( hartmann & kenyon 1985 ; khh ) simply did not include any boundary layer region . by omitting the boundary layer , this model made the implicit assumption that the accreting star is rotating at breakup speed . rapid stellar rotation seems unlikely in view of the generally slow rotations of t tauri stars , plus the recognition that fu orionis and v1057 cygni have not accreted enough angular momentum during their outbursts to spin up the entire central star as a solid body ( hartmann & kenyon 1996 ) . this leaves open the possibility that only the outer layers are spun up to keplerian angular velocities . a second implicit assumption of the early model is that the disk can be keplerian all the way in to the stellar surface , which implies that pressure gradients in the inner disk are insignificant . this also seems unlikely ; at the high mass accretion rates found in fu orionis systems , radial pressure gradients will support the accreting material and produce sub - keplerian angular velocities unless @xmath0 is very large . our disk and boundary layer model avoids these assumptions . we have found that by using the slim disk model , we can include the boundary layer region in our calculations of fu orionis disks and still find solutions which agree with the observations . these solutions demonstrate explicitly what the thin disk model implicitly assumed : fu orionis disks do not have a standard accretion disk boundary layer where the angular velocity of the accreting material drops rapidly over a short radial distance . unless @xmath0 is very large , the high accretion rates of fu orionis objects require quite optically thick disks which are very hot in their interiors , resulting in signficant thermal pressure support . thus the narrow , hot boundary layer typically expected in pre - main - sequence stars ( lynden - bell & pringle 1974 ) automatically disappears , for a wide range of stellar rotation rates . instead , the `` boundary layer '' is a broad region where @xmath68 changes gradually ; in our best - fitting solutions , @xmath68 increases gradually and becomes nearly constant as the accreting material approaches the star . as a result , like the early thin disk solutions , our solutions lack the high - temperature , slowly - rotating region expected from a standard boundary layer . but unlike the early solutions , they do not require that the star be rotating at breakup or that @xmath0 be very large . in fact , for @xmath63 , our best solutions have the star rotating at only @xmath177 of breakup speed . 5 shows that as @xmath0 increases , @xmath37 increases , reaching @xmath178 of breakup speed at @xmath179 . this suggests that the keplerian thin disk solutions of hartmann & kenyon ( 1985 ) and khh represent a special , extreme case of the solutions found in this paper , where the values of both @xmath0 and @xmath37 are very large . in addition to eliminating the assumptions of very large values of @xmath0 and @xmath37 which are implicit in the thin disk model , our disk and boundary layer model can offer insight into new areas of fu orionis accretion physics . the model explicitly includes the effects of the rotation rate of the star on the boundary layer region and on the angular momentum accretion rate . since radial advection of energy is included in the slim disk equations , our solutions directly give the rate at which energy is carried into the accreting star . these issues can only be studied when the boundary layer region and the additional physics of the slim disk equations are included in the model . they have important consequences for the evolution of pre - main - sequence stars , which will be discussed in more detail in the following sections . our best - fit solutions for fu orionis and v1057 cygni are equilibrium solutions with @xmath84 , i.e. the central star is neither gaining nor losing angular momentum , and have stellar rotation rates @xmath180 which correspond to rotation periods of 89 days . these periods are comparable to those observed in t tauri stars ( bouvier 1993 , 1995 ) . these low-@xmath78 solutions have important implications for the spin evolution of pre - main - sequence stars , because they support the scenario proposed in paper i , where fu orionis outbursts regulate the rotation rates of t tauri stars . in paper i we showed that the angular momentum accretion rate @xmath78 drops rapidly as the stellar rotation rate @xmath37 increases , and reaches @xmath84 when @xmath181 for fu orionis parameters . for smaller values of @xmath37 , @xmath182 and the star spins up , while for larger values of @xmath37 , @xmath183 and the star spins down . thus , fu orionis outbursts will move @xmath37 toward an equilibrium value where @xmath105 . if fu orionis outbursts dominate mass accretion onto t tauri stars , as seems likely from event statistics , then they should also dominate angular momentum accretion and control the spin evolution of these stars . when the t tauri star is between outbursts , it may spin up or down , but during each outburst it will return to the equilibrium rotation rate . this scenario therefore predicts that fu orionis systems should attain a low-@xmath78 equilibrium state during outbursts , and we find that low-@xmath78 solutions fit the observations . we are unable to make more precise predictions of the spin rates of t tauri stars because of uncertainties in the structure of the central star undergoing fu orionis accretion . as discussed in 5.4 below , it is likely that the advection of large amounts of thermal energy through the inner disk will cause the central star to expand from its equilibrium state . after an outburst ceases , we expect that the star will contract to a smaller radius , and will therefore spin up somewhat in the absence of any angular momentum loss . note that the disk may still remove large amounts of angular momentum from the star and spin the star down substantially during the course of the outburst ; however , at the end of the outburst , we expect that there will always be some spinup due to the contraction of the star . thus the equilibrium rotation rates during the fu orionis phase should correspond to faster t tauri rotation rates . the amount of spinup will depend upon the moment of inertia of the expanded layers of the star , which we are not able to estimate at present . observations of t tauri stars suggest that classical t tauri stars which have accretion disks rotate more slowly than weak - line t tauri stars which lack disks ( bouvier 1993 , 1995 ; edwards 1993 ; eaton , herbst , & hillenbrand 1995 ; choi & herbst 1996 ) . our results , combined with those of paper i , suggest a new way in which the disk can regulate the rotation rate of an t tauri star . previously suggested methods of regulating the stellar rotation rate have relied upon the interaction between the stellar magnetic field and the disk ( knigl 1991 ; cameron & campbell 1993 ; hartmann 1994 ; shu 1994 ) . our method does not depend on the stellar magnetic field , which is presumably not strong enough to disrupt the disk during fu orionis outbursts . however , it does require that outbursts occur , and that they dominate the angular momentum accretion onto the star . we noted in 4.3 that our solutions for fu orionis and v1057 cygni produce luminosities which are substantially smaller than the accretion luminosity . for fu orionis , we found that our solution gave @xmath184 , whereas @xmath185 , so that @xmath186 . for our v1057 cygni solution , @xmath187 and @xmath188 , so we again have @xmath189 . the remaining luminosity goes into heating the disk material . at the stellar surface , the midplane temperature of the disk in our solutions becomes quite large : @xmath190 k in the fu orionis solution , and @xmath191 k in the v1057 cygni solution . this means that the disk material carries energy into the star at the rate @xmath192 for fu orionis and @xmath193 for v1057 cygni . the high temperature of the disk material is due to the large vertical optical depth of the disk and boundary layer in these solutions . if the accreting stars in fu orionis objects are typical pre - main - sequence stars , they should have luminosities @xmath194 , so this advected energy represents a major perturbation to the star . another interesting feature of these solutions is that only a fraction of the luminosity comes from energy liberated by viscous dissipation . popham & narayan ( 1995 ) derived an expression for the total viscous dissipation rate of the disk and boundary layer @xmath195 where we have left out several small terms . the first term in this expression represents the gravitational potential energy and the rotational kinetic energy transferred between the star and the disk . for @xmath84 , the disk is removing a small amount of rotational kinetic energy from the star . both of our solutions have @xmath196 , so the first term is just @xmath197 . in a thin disk , the second term is very small , but in our fu orionis solutions , this term becomes quite large due to the importance of the radial pressure gradient in supporting the accreting material against gravity . in both solutions , we find @xmath198 , so that the viscous dissipation rate is only @xmath199 . nonetheless , we know that both solutions radiate @xmath200 ; the remaining @xmath201 comes from the entropy term in the energy equation @xmath202 the four terms in this equation represent the viscous dissipation , the vertical flux from the disk surface , the advected entropy , and the divergence of the radial flux . if we integrate this equation over the surface of the disk , we have @xmath203 = 0.\ ] ] our boundary conditions assume that the radial fluxes at the inner and outer edges of the disk are small ; at the inner edge we assume that the flux entering the disk is @xmath204 , where we assume @xmath205 k , so @xmath206 , and at the outer edge the radial flux is insignificant . if we neglect these radial flux terms , we can use @xmath207 and the expression for @xmath208 given above to write @xmath209 @xmath210 @xmath211 where the internal energy @xmath212 , @xmath213 , and @xmath214 . this accounts for the difference between the actual luminosities radiated by our solutions and the accretion luminosities for those disk parameters . our best - fit solutions for fu orionis and v1057 cygni have stellar radii of @xmath215 . these are substantially larger than the radii of t tauri stars , which are generally estimated at @xmath216 ( bouvier 1995 ) . since fu orionis outbursts are believed to occur in t tauri star accretion disks , this seems to imply that the star expands rapidly during the outburst , probably as a result of the rapid addition of high - temperature material ( khh ; hartmann , cassen , & kenyon 1996 ) . prialnik & livio ( 1985 ) calculated the effects of accretion onto a @xmath217 , @xmath218 fully - convective main - sequence star . they found that if the accretion carries a sufficient amount of energy into the star , it can cause the star to expand stably or unstably . for accretion rates comparable to those of fu orionis stars , their calculations indicate that the expansion will proceed unstably if the accretion energy is added to the star at a rate faster than about 10% of the accretion luminosity . the rapid expansion is due to the conversion of the normally convective star to a convectively - stable , radiative structure . as the discussion of the luminosities shows , our solutions add energy to the accreting star at an enormous rate - @xmath219 in our solutions for fu orionis and v1057 cygni . this rate of energy transfer will almost certainly produce rapid expansion of the star . the character of this expansion is not well understood . the accreting material comes from a disk , so it carries angular momentum as well as energy , and is added around the star s equator . thus the expansion is likely to be nonspherical . more sophisticated models of this process are needed , but the problem is clearly a difficult one . the large energy input into the star and the resulting stellar expansion should have important implications for the decline from fu orionis outbursts . the star will contract and release the energy it gained during the outburst , which is a substantial fraction of the total accretion energy of the outburst . thus we predict , qualitatively , that at the end of the mass accretion outburst in the disk , there may be a phase in which the central star is overly luminous , and fades over some ( rapid ) time set by the kelvin time of the perturbed portion of the stellar envelope , similar to the original explanation put forth by larson ( 1983 ) for fu orionis outbursts . by including the boundary layer region and the disk structure self - consistently , the solutions presented here represent a substantial advance in our understanding of disks in fu orionis objects and the spectra and line profiles they produce . nonetheless , there are additional explorations which could be made within the context of our assumptions , and improvements which could be made to our model . the solutions we have found do a good job of fitting the observed line profiles and spectra of fu orionis and v1057 cygni . nevertheless , this does not mean that they are the only solutions which would fit the observations of these systems . the number of solution parameters is too great to permit a full exploration of the entire parameter space . accordingly , as described in 4.2 , we have used the standard parameters @xmath59 , @xmath51 , and @xmath52 to guide our search . we have also kept certain parameters constant in order to simplify the fitting procedure . we keep @xmath84 for the reasons discussed in 3 , where we demonstrated that high-@xmath78 solutions fail to produce spectra and line profiles that agree with observations . other values of @xmath78 could produce reasonable solutions ; for instance , in paper i we proposed that fu orionis outbursts may spin down the accreting star , which would require solutions with negative values of @xmath78 . negative-@xmath78 solutions are similar in all respects to @xmath84 solutions ; the boundary layer region has @xmath68 increasing inward throughout and lacks a strong peak in effective temperature . as we demonstrated in paper i , @xmath78 drops rapidly with increasing @xmath37 , so that a @xmath85 solution has @xmath37 only slightly larger than a @xmath84 solution with the same parameters ( see fig . 2 ) . we also use @xmath63 for all of our solutions . as discussed in 3.3 , solutions with smaller values of @xmath0 have thicker disks , so that a larger fraction of the accretion luminosity is advected into the star . thus a smaller fraction is radiated , and the effective temperature is lower . another consequence of changing @xmath0 is that the rotational velocities change ; small values of @xmath0 result in small rotational velocities which would produce centrally peaked line profiles unlike those observed . this suggests that values of @xmath220 may produce acceptable solutions , but those with @xmath0 substantially below @xmath119 may not . this is particularly interesting because bell & lin ( 1994 ) and bell ( 1995 ) found that they needed @xmath221 to obtain satisfactory outbursts using the thermal instability mechanism . our model spectra could also be improved by treating the effects of inclination in more detail . we have implicitly assumed that the disk surface is flat , so that the inclination angle for all parts of the disk surface is the same . in fact , most of our solutions have @xmath222 nearly constant with @xmath223 , with @xmath17 . if we take @xmath224 as representing the surface of the disk , then the disk surface is inclined at @xmath225 to the disk midplane . this means that at inclination angles greater than @xmath226 , a portion of the disk surface will not be visible . even if @xmath227 , the side of the disk closer to the observer will effectively be seen at a larger inclination angle than the opposite side . ( note that when we refer to `` sides '' of the disk here , we are referring not to the top and bottom surfaces of the disk , but to regions of the same surface at different angles around the rotation axis . ) also , one side of the disk will be heated by radiation from the other side . we have also assumed that the disk spectrum does not vary with inclination angle , whereas in fact limb darkening will generally produce a decline in the blue end in the spectrum for systems viewed at large inclination angles . finally , there are improvements which could be made in the physical treatment of the disk and boundary layer . the assumption of a steady disk might be relaxed , although we feel this is unlikely to be a major issue , even though we are comparing our solutions to objects experiencing outbursts , because fu orionis and v1057 cygni have both remained fairly steady over recent years ( khh ; kenyon & hartmann 1991 ) , and we are only addressing the innermost regions of the disk . time - dependent models of boundary layers in disks around pre - main - sequence stars ( godon 1996 ) also seem to agree with our steady - state results ( pnhk ) . calculations of the structure of advection - dominated accretion disks ( narayan & yi 1995 ) have demonstrated that the slim disk equations used in our model provide a fairly good representation of the disk structure even for disks which are quite vertically thick ; still , a two - dimensional model for the boundary layer and disk might provide insights into fu orionis objects that we are unable to make using our current model ( e.g. , kley 1991 ) , especially when considering a more subtle matching of the disk to the inherently two - dimensional star . some consideration should also be given to understanding the effects of rapid accretion on the central star . presumably these include rapid expansion of the star , which could have important effects on the boundary layer region and the angular momentum accretion by the star . ultimately this will have to be a two - dimensional , time - dependent calculation as well , but perhaps some progress can be made with calculations similar to those performed by prialnik & livio ( 1985 ) for main - sequence convective stars . we have calculated solutions for the structure of the accretion disk in fu orionis objects which include a self - consistent treatment of the boundary layer region . we have also computed the line profiles and continuum spectra that would be observed from these solutions . we have explored the dependence of these observable characteristics on the solution parameters , including the mass accretion rate , the stellar mass , radius , and rotation rate , the angular momentum accretion rate , and the viscosity parameter . we find that our slim disk solutions can account for the absence of hot boundary layer emission without requiring that the central star rotate at breakup velocity . the differential rotation of the innermost disk departs from keplerian rotation and gives marginally better agreement with observations than a keplerian disk , but uncertainties in line formation preclude using the observations to make a definitive test . we find solutions at spin equilibrium ( @xmath228 ) for stellar angular velocities comparable to those observed in t tauri stars , and these solutions fit the observations of fu orionis and v1057 cygni . this result supports the proposal made in paper i that fu orionis outbursts play an important role in regulating the rotation rates of t tauri stars . in our solutions , large amounts of thermal energy are being advected into the central star during an fu orionis outburst . this should cause the star to expand , which would help to explain why the central stars of fu orionis objects appear to be twice as large as typical pre - main - sequence stars . the heating and expansion of the star should produce an observational signature : the fading of the central star after the end of rapid disk accretion . further theoretical explorations of the effects of rapid disk accretion on pre - main - sequence stars are needed . rp acknowledges the support of grants nasa nag5 - 2837 and nsf ast-9423209 at the center for astrophysics and grants nasa nagw-1583 , nsf ast 93 - 15133 , and nsf phy 91 - 00283 at the university of illinois . the research of lh and sk is supported in part by nasa grant nagw-2306 .
the partition function of the- or-symmetric euclidean barrett crane model can be understood as a sum over all quantized geometries of a given triangulation of a four - manifold . in the original formulation , the variables of the model are balanced representations of which describe the quantized areas of the triangles . we present an exact duality transformation for the full quantum theory and reformulate the model in terms of new variables which can be understood as variables conjugate to the quantized areas . the new variables are pairs of-values associated to the tetrahedra . these-variables parameterize the hyperplanes spanned by the tetrahedra ( locally embedded in ) , and the fact that there is a pair of variables for each tetrahedron can be viewed as a consequence of an-valued parallel transport along the edges dual to the tetrahedra . we reconstruct the parallel transport of which only the action of on is physically relevant and rewrite the barrett crane model as an lattice-theory living on the-complex dual to the triangulation subject to suitable constraints whose form we derive at the quantum level . our reformulation of the barrett crane model in terms of continuous variables is suitable for the application of various analytical and numerical techniques familiar from statistical mechanics .
gr-qc0112002
spin foam models have been proposed as candidates for a quantum theory of space - time geometry @xcite and therefore form candidates for a quantum theory of gravity . spin foam models also play a central role in the construction of invariants of piecewise - linear three- and four - manifolds @xcite . they arise as the models that are strong - weak dual to non - abelian lattice gauge theory @xcite and form a possible starting point for the generalization of gauge theories from lie groups to quantum groups @xcite . for reviews on the subject , see , for example @xcite . a spin foam @xcite whose symmetry group is a compact lie group @xmath6 , is an abstract oriented @xmath5-complex consisting of faces , edges and vertices , together with a colouring of the faces with representations of @xmath6 and a colouring of the edges with compatible intertwiners ( representation morphisms ) of @xmath6 . a spin foam model is a discrete model in the framework of statistical mechanics whose partition function is a sum over spin foams each of which is assigned a boltzmann weight . if the boltzmann weight factors into contributions for each vertex , edge and face , these factors are usually called vertex , edge and face amplitudes . barrett , crane and baez @xcite have developed a particular spin foam model whose partition function can be understood as the sum over all quantized euclidean geometries that can be assigned to a given triangulation of a four - manifold . the spin foam model is usually defined on the @xmath5-complex dual to that triangulation . several versions of this model are studied in the literature @xcite which differ in their boltzmann weights ( amplitudes ) , in particular in their edge amplitudes . some of the models are formulated on a fixed @xmath5-complex while others include a sum over @xmath5-complexes . in the following , we call all these models barrett crane models . in this article , we study a generalized version which encompasses all interesting choices of amplitudes , but we restrict ourselves to a fixed @xmath5-complex . the partition function of our model is a sum over all possible assignments of _ balanced _ irreducible representations of the lie group @xmath0 to the faces . the balanced representations are of the form @xmath7 where @xmath8 is an irreducible representation of @xmath9 . they are also called the _ simple _ representations . each edge is then assigned a unique intertwiner , the barrett crane intertwiner @xcite . we allow quite generic edge and face amplitudes , but the vertex amplitude is always the generalized @xmath10-symbol @xcite consisting of one barrett crane intertwiner for each edge attached to the vertex . we make use of the fact that the barrett crane intertwiner can be written in terms of an integral over the sphere @xmath2 @xcite and of the techniques familiar from the duality transformation for non - abelian lattice gauge theory @xcite and derive a dual expression for the partition function of the barrett crane model in which the fundamental variables are pairs of @xmath2-values associated to the edges . the partition sum is then given by an integral over @xmath2 for each of these variables . provided that the edge amplitude of the original model is of a particular form , the dual model has local interactions , its boltzmann weight factors into one contribution for each face , and it depends on the @xmath2-variables at the edges in the boundary of that face . this reformulation of the barrett crane model has the following geometric interpretation if the @xmath5-complex is chosen to be dual to a triangulation of a four - manifold . the @xmath2-variables are then associated to the tetrahedra . they parameterize the ( normalized ) normal vectors of the hyperplanes spanned by the tetrahedra where the tetrahedra are locally embedded in @xmath3 . the factorized boltzmann weight provides an interaction term for each triangle so that the interaction depends on the @xmath2-values associated to all tetrahedra that contain the triangle in their boundaries . however , there are two @xmath2-values associated to each tetrahedron one of which belongs to either one of the two four - simplices that contain the tetrahedron in their boundaries . this suggests that there exists an @xmath0-valued parallel transport along the tetrahedron which maps the first @xmath2-value to the second one . however , this parallel transport is defined only up to elements of @xmath0 that stabilize the first of the @xmath2-values . following this idea , we show that the barrett crane model can be obtained from @xmath0 lattice @xmath4-theory ( the @xmath0 ooguri model @xcite ) if we impose a constraint at the quantum level by averaging over this stabilizer in a suitable way . this means that of the @xmath0-valued parallel transport only the action of @xmath0 on @xmath2 is physically relevant . finally , we consider the boltzmann weight of the dual formulation of the barrett crane model and list possible face and edge amplitudes for which the partition function is well defined ( not just divergent ) . we outline how one can analyze this model in the context of statistical mechanics , for example in order to study its ground state and fluctuations around it . in addition , one can expect that the dual formulation of the barrett crane model is well suited for numerical studies . the present article is organized as follows . in section [ sect_prelim ] , we review some mathematical background material on the algebra of representation functions of a compact lie group , we construct the spherical functions of @xmath2 , and we introduce our notation for abstract @xmath5-complexes . the barrett crane model and @xmath0 lattice @xmath4-theory are introduced in section [ sect_bcmodel ] . in section [ sect_transform ] , we present the duality transformation for the barrett crane model in detail and show how this model is related to @xmath0 lattice @xmath4-theory subject to a suitable constraint . we also discuss its geometric interpretation and the possible choices of face and edge amplitudes . section [ sect_conclusion ] finally contains our concluding comments . note added : the barrett crane model is often understood as a path integral in the ` real time ' picture , using an integrand of the form @xmath11 for some action @xmath12 . however , baez and christensen @xcite have shown that for the common choices of edge amplitudes , the integrand of the partition function is always positive so that the path integral admits an interpretation in the framework of statistical mechanics in terms of probabilities . in this article , we use the language of statistical mechanics . in this section , we briefly summarize definitions and some basic statements related to the algebra of representation functions @xmath13 of @xmath6 where @xmath6 is a compact lie group . furthermore , we describe the spherical functions of @xmath2 and present our notation for the @xmath0 and @xmath14 actions on @xmath2 . more background material and most of the proofs can be found , for example , in @xcite . finally , we introduce the @xmath5-complexes that we employ in our definition of the barrett crane model and in the duality transformation . let @xmath6 be a compact lie group . we denote finite - dimensional complex vector spaces on which @xmath6 is represented by @xmath15 and by @xmath16 the corresponding group homomorphisms . since each finite - dimensional complex representation of @xmath6 is equivalent to a unitary representation , we select a set @xmath17 containing one unitary representation of @xmath6 for each equivalence class of finite - dimensional representations . the tensor product , the direct sum and taking the dual are supposed to be closed operations on this set . this amounts to a particular choice of representation isomorphisms @xmath18 , @xmath19 , which is implicit in our formulas . we furthermore denote by @xmath20 the subset of irreducible representations . for a representation @xmath21 , the dual representation is denoted by @xmath22 , and the dual vector space of @xmath15 by @xmath23 . the dual representation is given by @xmath24 , where @xmath25 @xmath26 for all @xmath27 . there exists a one - dimensional ` trivial ' representation of @xmath6 which is denoted by @xmath28}\cong\c$ ] . for the unitary representations @xmath15 , @xmath21 , we have standard ( sesquilinear ) scalar products @xmath29 and orthonormal bases @xmath30 . for each basis @xmath30 of @xmath15 , there exists a dual basis @xmath31 of @xmath23 so that @xmath32 , and @xmath23 possesses a scalar product such that @xmath33 , @xmath34 . the complex - valued functions @xmath35 where @xmath21 , @xmath36 and @xmath37 , are called the _ representation functions _ of @xmath6 . they form a commutative and associative unital algebra over @xmath38 , @xmath39 whose operations are given by @xmath40 where @xmath41 , @xmath36 , @xmath42 , @xmath43 , @xmath44 and @xmath45 . the zero element of @xmath13 is given by @xmath46}(g)=0 $ ] and its unit element by @xmath47}(g)=1 $ ] . the algebra @xmath13 is furthermore equipped with a hopf algebra structure employing the coproduct @xmath48 , the co - unit @xmath49 and the antipode @xmath50 which are defined by @xmath51 for @xmath21 , @xmath36 , @xmath37 and @xmath52 . here @xmath53 denotes the group unit . in the standard orthonormal bases , the representation functions are given by the coefficients of representation matrices , @xmath54 so that the coproduct corresponds to the matrix product , @xmath55 while the antipode refers to the inverse matrix , @xmath56 , and the co - unit describes the coefficients of the unit matrix , @xmath57 . furthermore , the antipode relates a representation to its dual , @xmath58 which , in the case of unitary @xmath59 , is just the conjugate representation because on the other hand @xmath60 . here the bar denotes complex conjugation . the structure of the algebra @xmath13 can be understood if @xmath13 is considered as a representation of @xmath61 by combined left and right translation of the function argument , @xmath62 it can then be decomposed into its irreducible components as a representation of @xmath61 . [ thm_peterweyl ] let @xmath6 be a compact lie group . there is an isomorphism @xmath63 of representations of @xmath61 . here the direct sum runs over the equivalence classes of finite - dimensional irreducible representations of @xmath6 . the direct summands @xmath64 are irreducible as representations of @xmath61 . the direct sum in is orthogonal with respect to the @xmath65-scalar product on @xmath13 which is formed using the haar measure of @xmath6 on the left hand side , and using the standard scalar products on the right hand side , namely @xmath66 where @xmath67 are irreducible . the haar measure is denoted by @xmath68 and normalized so that @xmath69 . each representation function @xmath70 can be decomposed according to , @xmath71 the analytical aspects of @xmath13 can now be stated . let @xmath6 be a compact lie group . then @xmath13 is dense in @xmath72 with respect to the @xmath65 norm . we use the peter weyl theorem in order to complete @xmath13 with respect to the @xmath65 norm to @xmath72 . functions @xmath73 then correspond to square summable series in . these series are therefore invariant under a reordering of summands , and their limits commute with group integrations . we make use of these invariances in the duality transformation . the _ characters _ of @xmath6 are the algebraic class functions , those functions @xmath70 that satisfy @xmath74 for all @xmath75 . for class functions @xmath70 , the peter weyl decomposition specializes to the _ character decomposition _ @xmath76 here @xmath77 denotes the character of the representation @xmath21 . for irreducible @xmath67 , the orthogonality relation implies , @xmath78 the formal @xmath79-element , @xmath80 satisfies for all @xmath70 , @xmath81 in order to present the partition function of lattice gauge theory both in the original formulation using group integrations and in the dual formulation @xcite as a spin foam model , we define a particular intertwiner @xcite based on the haar measure . let @xmath6 be a compact lie group and @xmath82 , @xmath83 , be finite - dimensional irreducible representations of @xmath6 . the _ haar intertwiner _ , [ eq_haarinter ] @xmath84is defined by its matrix elements , @xmath85 the haar intertwiner @xmath86 of satisfies for all @xmath87 , @xmath88 the first equation is just the definition while the second and third are consequences of the left - right translation invariance of the haar measure . in particular , @xmath86 forms a morphism of representations of @xmath6 . in the subsequent sections , the haar intertwiner will appear in rather complicated calculations . it is therefore convenient to introduce diagrams that visualize the structure of the indices in these formulas ( figure [ fig_diagrams ] ) . the diagrams are read from top to bottom . we draw directed lines which are labelled with irreducible representations @xmath89 of @xmath6 . if the arrow points down , the line denotes the identity map of @xmath15 , figure [ fig_diagrams](a ) . if the arrow points up , it refers to the identity map of the dual representation @xmath23 . a representation function @xmath90 is denoted by a box with an incoming and an outgoing line ( b ) , and a product of representation functions by boxes placed next to each other ( c ) . the haar intertwiner is visualized by the box labelled @xmath86 in ( d ) . we construct the spherical functions of @xmath2 from the identification @xmath91 , we can make use of the representations functions of @xmath9 . the finite - dimensional irreducible representations of @xmath9 are characterized by half - integers @xmath92 . we write @xmath93 , @xmath94 , for the irreducible representations . using the peter weyl decomposition , we can give a basis for @xmath95 , @xmath96 in particular , we obtain the ( algebraic ) spherical functions of @xmath2 from the identification @xmath91 as @xmath97 for @xmath92 and @xmath98 . we also wish to describe the transitive actions of @xmath0 and @xmath99 on @xmath2 for which the stabilizer of a point @xmath100 is @xmath101 or @xmath9 , respectively . writing @xmath102 complex matrices with the usual properties for the elements of @xmath9 , we obtain @xmath101 as the quotient @xmath103 where @xmath104 and @xmath105 where @xmath106 . the relevant actions are the following . [ prop_su2action ] the map @xmath107 is a transitive group action . the stabilizer of a point @xmath100 is given by @xmath108 and forms a subgroup of @xmath14 via the inclusion @xmath109 the action @xmath110 on @xmath2 is well defined for the quotient @xmath111 in which case the stabilizer is isomorphic to @xmath101 and given by the inclusion @xmath112 which we have written down for representatives . [ corr_decomposes3 ] the algebra of spherical functions @xmath113 forms a representation of @xmath14 by translation of the function argument , @xmath114 which also gives rise to a representation of @xmath111 . the decomposition of @xmath113 into irreducible representations of @xmath14 coincides with the peter weyl decomposition of @xmath9 , @xmath115 where the direct summand @xmath116 for a given @xmath117 has the basis @xmath118 in our notation , the spherical functions @xmath119 therefore transform as the coefficients of a vector in @xmath120 . furthermore , the haar measure of @xmath9 provides an integral over @xmath2 with the desired translation invariance under the @xmath0-action . [ prop_integrals ] let @xmath121 and @xmath122 . then the integral over @xmath2 , obtained from the haar measure of @xmath9 under the identification @xmath91 , satisfies @xmath123 there is a scalar product on @xmath113 , @xmath124 with respect to which the spherical functions are orthogonal , @xmath125 where @xmath126 and @xmath98 , @xmath127 . we also construct the algebras of functions on @xmath14 and @xmath0 . we denote the irreducible representations of @xmath14 by pairs @xmath128 of half - integers @xmath129 . they are of the form @xmath130 . a basis for the algebra @xmath131 is given by the functions @xmath132 where @xmath129 , @xmath133 . it is known that the functions @xmath134 are well defined on @xmath111 if and only if @xmath135 , if in @xmath130 either both factors are even- or both are odd - dimensional . therefore @xmath136 if and only if @xmath135 . usually , the spherical functions of @xmath2 are constructed as the functions on @xmath137 , as the functions on @xmath0 that are constant on cosets of @xmath101 @xcite . there exist representation functions @xmath138 that are constant on these cosets only if @xmath139 , in the class-1 representations of @xmath0 with respect to @xmath101 . these are the _ balanced _ representations @xmath140 . our spherical functions , however , transform as the coefficients of vectors in @xmath141 . recall that for @xmath9 , there exist isomorphisms of representations @xmath142 ( which we have not explicitly constructed ) which can be applied to the right tensor factor in order to relate our spherical functions to the standard construction . in order to define lattice gauge theory and the barrett crane model , we need essentially an abstract oriented @xmath5-complex . this is a collection of faces , edges and vertices together with a boundary relation specifying which edges are contained in the boundary of a given face , . for the partition function of lattice gauge theory , we require a formulation which explicitly describes the cyclic ordering of the edges in the boundary of a given face . this is achieved by the following definition of a _ @xmath5-complex with cyclic structure_. [ def_complexcyclic ] a _ finite @xmath5-complex with cyclic structure _ @xmath143 consists of finite sets @xmath8 ( _ vertices _ ) , @xmath144 ( _ edges _ ) and @xmath145 ( _ faces _ ) together with maps [ eq_twocomplex ] @xmath146 so that [ eq_boundary ] @xmath147 for all @xmath148 . the conditions state that the edges in the boundary of a face @xmath148 are in cyclic ordering from @xmath149 to @xmath150 where one encounters the edges with a relative orientation given by @xmath151 , see figure [ fig_boundary ] . observe that contains combinatorial information similar to the condition @xmath152 on the boundary operator @xmath153 in abelian simplicial homology . for the existence of the partition functions in the following sections , we require that the @xmath5-complexes have only a finite number of vertices , edges and faces . in the subsequent calculations , it is convenient to use the following abbreviations . let @xmath143 denote a finite combinatorial @xmath5-complex with cyclic structure . for a given edge @xmath154 , the sets @xmath155 contain all faces that have the edge @xmath156 in their boundary with positive ( @xmath157 ) or negative ( @xmath158 ) orientation , and we write @xmath159 . for a given face @xmath148 , the set @xmath160 denotes all vertices that belong to the boundary of the face @xmath161 . finally , the sets @xmath162 contain all edges in the boundary of the face @xmath161 that occur with positive ( @xmath157 ) or negative ( @xmath158 ) orientation , and @xmath163 . finally , for a given face @xmath148 , @xmath164 denotes the edge @xmath165 for which @xmath166 , and similarly , @xmath167 the edge @xmath165 for which @xmath168 . given any triangulation of a four - manifold , we can construct a @xmath5-complex with cyclic structure that is combinatorially dual to the triangulation . the vertices @xmath8 are the four - simplices of the triangulation , the edges @xmath144 the tetrahedra and the faces @xmath145 the triangles . an edge is contained in the boundary of a given face whenever the corresponding tetrahedron contains the triangle in its boundary , . a cyclic structure can then be constructed by induction . in particular , if the triangulation contains only finitely many four - simplices , we obtain a finite @xmath5-complex with cyclic structure . when we present the definition of the barrett crane model @xcite , we wish to emphasize its structural similarity to lattice gauge theory with pure gauge fields , in particular to @xmath0 lattice @xmath4-theory . therefore we recall the two formulations of lattice gauge theory , the original one using group integrations and the dual formulation @xcite as a spin foam model before we present the definition of the class of barrett crane models which we consider in this article . for more background on lattice gauge theory , we refer the reader to text books such as @xcite . the partition function of lattice gauge theory can be formulated on any @xmath5-complex with cyclic structure as follows . let @xmath6 be a compact lie group and @xmath143 be a finite @xmath5-complex with cyclic structure . the _ partition function _ of lattice gauge theory is defined as @xmath169 here @xmath170 is the local _ boltzmann weight _ which is required to be a positive and @xmath65-integrable class function of @xmath6 that satisfies @xmath171 . our notation in means that there is one integral @xmath68 for each edge @xmath154 . we use this notation very often for integrals and sums in the subsequent calculations . the partition function of lattice gauge theory is an integral over the group @xmath6 for each edge of the @xmath5-complex so that the set of configurations is @xmath172 . the boltzmann weight is calculated for each face where @xmath173 denotes the ordered product of group elements assigned to the edges in the boundary of the face . this is the place where the cyclic structure of the @xmath5-complex enters the definition . the value of the partition function is independent of the choice of cyclic structure . the starting point for the cyclic numbering of edges in the boundary of a face does not matter because the boltzmann weight is given by a class function and thus invariant under cyclic permutation of the factors of @xmath173 . reversal of the orientation is also a symmetry because it replaces @xmath173 by @xmath174 . for lattice yang mills theory , the boltzmann weight is given by @xmath175 in terms of the euclidean action @xmath176 . for lattice @xmath4-theory @xcite , however , we have @xmath177 . for a generic lie group @xmath6 and a generic @xmath5-complex , the partition function is therefore not well defined . the boltzmann weight of lattice gauge theory is invariant under local gauge transformations . let @xmath178 associate a group element to each vertex . the boltzmann weight @xmath179 is invariant under the _ local gauge transformations _ , @xmath180 in order to prove this invariance , one has to make use of the conditions and of the fact that @xmath181 is a class function . there exists a dual formulation of lattice gauge theory in which the partition function is no longer given by group integrals but rather by sums over representations and intertwiners @xcite , so that it forms a spin foam model . we present a definition of the dual model which coincides with an intermediate step in @xcite and directly corresponds to the diagrammatical formulation of @xcite . [ thm_dualpart ] let @xmath6 be a compact lie group and @xmath143 be a finite @xmath5-complex with cyclic structure . the partition function of lattice gauge theory is equal to @xmath182 where @xmath183 denotes the haar intertwiner for the following representations , @xmath184 we require here that the ordering of tensor factors parameterized by @xmath185 agrees with the ordering of indices in . the coefficients @xmath186 are the coefficients of the character expansion of the boltzmann weight , @xmath187 the dual form of the partition function is a sum over all colourings of faces @xmath148 with irreducible representations @xmath89 of @xmath6 . there is a haar intertwiner @xmath183 assigned to each edge @xmath154 , and under the sum we have a weight @xmath188 for each face . the sums over the indices @xmath189 for each face @xmath148 and each vertex @xmath190 in the boundary of the face , contract the indices of the haar intertwiners . this is illustrated in figure [ fig_transform](a ) . for lattice @xmath4-theory , we have @xmath191 , the sums do not converge . in particular , for @xmath192 and a @xmath5-complex that is dual to a given triangulation of a four - manifold , we obtain the ooguri model @xcite whose @xmath0 version we will consider later . for lattice yang mills theory , the transformation relating to is a strong - weak or high temperature - low temperature duality transformation @xcite . this is a consequence of the fact that the character expansion of the boltzmann weight has many similarities to fourier decomposition and maps a boltzmann weight with a wide peak to a function with a narrow peak and conversely . the barrett crane model has close similarities to the spin foam model dual to @xmath0 lattice @xmath4-theory . we obtain the barrett crane model starting from the partition function if we restrict the sum over irreducible representations to the balanced representations of @xmath0 . furthermore , we have to replace the haar intertwiner of by a pair of barrett crane intertwiners . following barrett , freidel and krasnov @xcite , the barrett crane intertwiner can be written as an integral over @xmath2 . [ def_bcinter ] let @xmath193 and @xmath194 , @xmath195 , denote balanced irreducible representations of @xmath0 whose right tensor factor has been dualized . the _ barrett crane intertwiner _ [ eq_bcinter ] @xmath196is defined in terms of the matrix elements , @xmath197 of the intertwiner @xmath198 here the identification @xmath199 was used as well as the fact that @xmath200 and @xmath201 . the index pairs @xmath202 therefore correspond to the tensor factors @xmath203 of the domain of @xmath204 while the @xmath205 belong to the image @xmath206 . the map @xmath204 defined by is a morphism of representations of @xmath0 . the action of @xmath0 is given by the action of @xmath14 in . if we write @xmath207 for the range of the integral , @xmath208 this action corresponds to @xmath9 left - right translation of all function arguments which is an invariance of the haar measure of @xmath9 . the barrett crane model can now be defined by an expression similar to the dual of lattice gauge theory ( theorem [ thm_dualpart ] ) in which the haar intertwiner is replaced by a pair of barrett crane intertwiners , one associated to each vertex of the given edge . these two barrett crane intertwiners are ` independent ' which corresponds to the first version of the model proposed in @xcite . let @xmath143 denote a finite @xmath5-complex with cyclic structure , let @xmath209 , @xmath92 , be a face amplitude and @xmath210 @xmath211 , an edge amplitude . the barrett crane model is defined by the partition function , @xmath212 here @xmath213 and @xmath214 , @xmath154 , denote the coefficients of the barrett crane intertwiners , [ eq_bcintermap]@xmath215 where @xmath216 are balanced irreducible representations of @xmath0 whose right factors have been dualized . the partition function can be described in words as follows . the faces @xmath148 are coloured in all possible ways with representations @xmath217 , @xmath92 , of @xmath0 . there is an amplitude @xmath218 for each face and an amplitude @xmath219 for each edge which can depend on the representations at all faces @xmath220 . each edge @xmath154 is labelled with a pair of barrett crane intertwiners @xmath221 and @xmath222 , each one associated to one vertex , and the vector indices ( here pairs @xmath223 for @xmath116 ) are contracted precisely as in the spin foam model dual to lattice gauge theory ( theorem [ thm_dualpart ] and figure [ fig_transform](a ) ) . figure [ fig_bcinter](a ) visualizes the replacement of the @xmath0 haar intertwiner @xmath86 by the pair of barrett crane intertwiners @xmath204 . the contraction of the indices in the barrett crane model is illustrated in figure [ fig_transform](b ) . in this figure , we denote the barrett crane intertwiners by the full dots , a simplification which we justify below . the face amplitude is usually chosen to be @xmath224 while there are various edge amplitudes proposed in the literature @xcite . we leave the amplitudes unspecified here and discuss the question of the convergence of the partition function later in section [ sect_bcregular ] . for all balanced irreducible representations @xmath225 of @xmath14 , we have identified the tensor factor corresponding to the right @xmath9 by its dual . our model still agrees with the original proposal @xcite since the corresponding isomorphism appears in each representation of @xmath0 , in each line in figure [ fig_transform](b ) , together with its inverse in such a way that they cancel . we have , however , obtained a considerable technical advantage with this construction as we can now use the spherical functions of section [ sect_s3 ] which agree with the representation functions of @xmath9 . in this section , we comment on a number of properties of the barrett crane intertwiner and translate some results of @xcite into our language . this intertwiner is visualized by the box diagrams labelled @xmath204 in figure [ fig_bcinter](a ) . each line carries a label @xmath92 and symbolizes the representation @xmath226 of @xmath14 if the arrow points down and the dual representation @xmath227 if the arrow points up . commutativity of the algebra @xmath113 gives rise to rules such as the one depicted in figure [ fig_bcinter](b ) . if we use @xmath228 in the integrand , we obtain rules as in figure [ fig_bcinter](c ) . the lines can therefore be freely moved around the diagram , and what matters is only whether the arrows point towards or away from the box . therefore , in figure [ fig_transform](b ) , it is sufficient to draw full dots with incoming or outgoing lines for the barrett crane intertwiners . note that each arrow can be reversed if at the same time the representation label @xmath117 is replaced by its dual @xmath229 . we can make use of the fact that @xmath113 forms an algebra for which the spherical functions provide a basis ( corollary [ corr_decomposes3 ] ) , therefore , @xmath230 for @xmath129 and @xmath133 , with uniquely determined coefficients @xmath231 . this is illustrated in figure [ fig_trivalent](a ) . rules such as the one depicted in diagram ( b ) follow from the commutativity of the algebra . linearity of the integral implies rules as in diagram ( c ) , and the orthogonality of spherical functions finally leads to diagram ( d ) . these rules can be used in order to decompose the barrett crane intertwiner into a tree diagram involving only three - valent vertices @xcite . for each line connecting two vertices , the corresponding indices are contracted , and we sum over all representations of the form @xmath226 for each internal edge . obviously , all possible decompositions into trees evaluate to the same intertwiner @xcite . this is the property that was originally required in @xcite . we emphasize that we here use the intertwiner in the form involving the spherical functions @xmath119 whose orthogonality relation involves a denominator of @xmath232 . the normalized basis vectors of @xmath113 are therefore @xmath233 . furthermore , we can employ the identification of the @xmath2-integral of spherical functions with the @xmath9-integral of representation functions ( proposition [ prop_integrals ] ) and express the barrett crane intertwiner as an @xmath9 integral , @xmath234 so that we find [ eq_bcsu2b ] @xmath235 where @xmath86 is the @xmath9 haar intertwiner , @xmath236 this relation is shown in figure [ fig_bcsu2 ] . in diagram ( a ) , the representations @xmath237 of @xmath0 are split into their tensor factors as representations of @xmath14 where we have dualized all right factors . diagram ( b ) visualizes . finally , we show that the barrett crane intertwiner of definition [ def_bcinter ] agrees with the unnormalized intertwiner in the terminology of @xcite . therefore we consider the intertwiners @xmath238 and @xmath239 for a four - fold tensor product and compute @xmath240 this calculation is illustrated in figure [ fig_normalize ] . a comparison shows that our barrett crane intertwiner agrees with the unnormalized one of @xcite . if we wish to use a different normalization and multiply our intertwiner by a factor , we can always absorb this factor into the edge amplitude @xmath241 . we emphasize that the barrett crane intertwiner for given representations is unique only up to normalization @xcite , but that the relative weight of different representations is not determined . we demonstrate in section [ sect_bcregular ] that the physical properties of the model are affected by the choice of normalization . we can employ the integral presentation of the barrett crane intertwiner in order to introduce @xmath2-variables into the barrett crane model . it is possible to derive a transformation similar to the reversed direction of the duality transformation for lattice gauge theory in order to promote the new @xmath2-values to the configuration variables of the partition function . first we make use of the integral presentation of the barrett crane intertwiner , @xmath242 we insert this expression into the partition function and move all integrals to the left , @xmath243.\end{aligned}\ ] ] here we have also employed the antipode in order to remove dual representations in favour of inverse group elements . keeping the similarity with lattice gauge theory in mind , we try to factor the integrand into one contribution for each face . therefore , we sort the products in the above expression by face rather than by edge . this step is almost entirely hidden in our notation , @xmath244.\end{aligned}\ ] ] the structure of the integrand becomes clearer if we re - introduce the cyclic ordering of edges in the boundary of each face , @xmath245.\end{aligned}\ ] ] recall that the cyclic structure of the edges around a face is described by the conditions . we notice that in the last product in , both indices of the second @xmath246 of each factor are contracted with the indices of the first @xmath246 of the following factor and so on . we therefore shift the factors of the product by a single @xmath246 . now the indices @xmath247 and @xmath248 occur only in a single factor , @xmath249,\end{aligned}\ ] ] where @xmath164 denotes the edge in the boundary of the face @xmath148 for which @xmath166 , . the last expression involves just a group character . we arrive at the following result . [ thm_bcdual ] let @xmath143 be a finite @xmath5-complex with cyclic structure , let @xmath250 , @xmath92 , be a face amplitude and @xmath219 denote an edge amplitude , @xmath251 . the partition function of the barrett crane model is equal to the following expression , @xmath252.\end{aligned}\]]if the edge amplitude factors into one contribution for each face attached to the edge , @xmath253 @xmath254 , this can be written as @xmath255 where @xmath256 the partition function describes a system in statistical mechanics whose variables are pairs of @xmath9 values @xmath257 attached to each edge @xmath154 . for a generic edge amplitude @xmath241 , the boltzmann weight under the partition sum forms a non - local expression . if the edge amplitude factorizes into one contribution per face , the boltzmann weight factorizes into a product over all faces . the factor @xmath258 for each face @xmath148 depends on the group elements at all edges in the boundary of @xmath161 , but only through the characters @xmath259 for each pair of group elements @xmath260 that are associated to neighbouring edges . although this interaction is considerably more complicated than just a nearest neighbour interaction , it is still local in the sense that there is one contribution for each face , and the interaction term for a given face relies only on data associated to the nearby edges . we discuss the properties of the boltzmann weight @xmath261 and the possible choices for the amplitudes @xmath262 and @xmath241 in section [ sect_bcregular ] . equation is an intermediate step in the transformation in which both types of variables are present , the old representation numbers and the new group variables . in this section , we restrict ourselves to the case in which the interactions are local , to the situation in which the edge amplitude factorizes according to . we consider a @xmath5-complex @xmath143 that is dual to a given triangulation of a four - manifold . for more general @xmath5-complexes dual to a generic cellular decomposition , the interpretation has to be generalized accordingly . using the identification @xmath207 , the partition function suggests the following geometric interpretation . the variables of the model are the pairs of @xmath2-values associated to the tetrahedra of the triangulation . an oriented tetrahedron , embedded in @xmath3 , spans an oriented hyperplane which can be parameterized by its ( normalized ) normal vector , by a point in @xmath263 . the partition function contains an interaction term @xmath261 for each triangle . the interaction depends on the @xmath2-values for all tetrahedra that contain the triangle in their boundaries . recall that the characters of @xmath9 are given by @xmath264 where @xmath265 denotes the polar angle on @xmath91 . the interaction therefore depends only on the relative angles between the normal vectors of each pair of neighbouring tetrahedra , the so - called _ dihedral angles_. two tetrahedra are neighbours with respect to a given triangle if they are contained in the boundary of the same @xmath266-simplex ( this is the same as saying that on the dual @xmath5-complex , two neighbouring edges contain a common vertex , but expressed in the language of the original triangulation ) . however , there is the complication that each tetrahedron carries two @xmath2-variables , @xmath267 and @xmath268 . for a given triangle , each of these values is involved in the interaction term : the first variable , @xmath267 , in association with one neighbour and the second variable , @xmath269 , with the other one , see figure [ fig_geometry](a ) . the tetrahedron dual to @xmath156 therefore contributes with different @xmath2-values @xmath267 and @xmath269 depending on the four - simplex in which the interaction takes place . it is thus plausible to conjecture that there is some parallel transport associated to the tetrahedra which contains the information of how @xmath269 is rotated to @xmath267 when we transit from one four - simplex along a tetrahedron to a neighbouring four - simplex . the obvious candidate for the parallel transport is the transitive @xmath0-action on @xmath2 for which the @xmath0 element is , however , only defined up to an element in the stabilizer of a given point on @xmath2 . in section [ sect_bcgauge ] , we reconstruct the @xmath0 parallel transport , and our intuition about the ambiguity due to the stabilizer is shown to be correct . the weight resembles the character decomposition of a function on @xmath14 where the character decomposition can be understood as a generalization of fourier expansion . because of the geometric picture , the @xmath2-variables have a natural interpretation as variables conjugate to the representations labelling the faces conjugate in the sense of the quantum mechanical uncertainty relation . if we suppose that the partition function can be interpreted as the trace over a suitable hilbert space , there are two pictures for the barrett crane model in analogy to position and momentum basis in quantum mechanics : in the original formulation , the partition function is a sum over all possible quantized areas associated to the triangles where the areas are determined by the balanced irreducible representation of @xmath0 @xcite . if we are in a quantum state with definite areas of the triangles , the directions of the tetrahedra in @xmath3 which are represented by the @xmath2-variables and the corresponding dihedral angles , would be maximally uncertain . in addition to this ` area basis ' we also have a ` direction basis ' which is given by the dual partition sum . the dual partition sum is the integral over all possible @xmath2-directions of the tetrahedra . in a quantum state with definite directions , however , we would expect a maximal uncertainty of the quantized areas assigned to the triangles . we have conjectured above that it is possible to find the parallel transports of an @xmath0 connection which , for each edge @xmath154 , map the first @xmath2-variable @xmath269 to the second one @xmath267 . this would mean that the barrett crane model has some properties of an @xmath0 gauge theory and therefore enjoys the corresponding local symmetry . a suitable notion of local @xmath0 symmetry is the invariance of the boltzmann weight under a transformation which takes an arbitrary @xmath0 element for each vertex @xmath270 and transforms each edge variable @xmath257 depending on the group elements that have been chosen for the two vertices . let @xmath271 define a generating function . here @xmath272 represents a class of @xmath111 . the barrett crane model is invariant under the local transformations , @xmath273 by proposition [ prop_su2action ] , this transformation involves the action of @xmath0 on @xmath2 for each of the @xmath2-variables at the given edge . for a given face @xmath148 and a vertex @xmath190 , the transformation reads @xmath274 so that @xmath275 is invariant . it is now possible to reconstruct @xmath0 parallel transports from the pairs of @xmath2-values . therefore we recall that the action of @xmath0 on @xmath2 given in proposition [ prop_su2action ] is transitive , for two points @xmath276 , there exists a @xmath277 such that @xmath278 of course , we can not see elements in the stabilizer of @xmath279 . therefore for any @xmath280 , the action is , @xmath281 we wish to start with an @xmath0 gauge theory and then impose the condition that elements from these stabilizers are irrelevant . in other words , we wish to describe an @xmath0 gauge theory of which only the action of the parallel transport on @xmath2 is physically relevant . technically , the key idea is to consider a representation function of @xmath0 whose argument is the @xmath0 element @xmath282 and to average over the stabilizer as parameterized in . the result of the averaging is indeed a product of two spherical functions on @xmath91 with arguments @xmath279 and @xmath283 , respectively . [ lemma_simpleproject ] let @xmath284 and @xmath277 . then @xmath285 if we integrate over @xmath0 , the right hand side of this equation factorizes into two @xmath207 integrals . the above lemma crucially makes use of the special relations among @xmath2 , @xmath9 and @xmath0 . in particular , in higher dimensional generalizations @xcite , we would still have @xmath286 , but the right hand side of would no longer factorize . the averaging over the stabilizer automatically singles out the balanced irreducible representations of @xmath0 for which @xmath287 . the map @xmath288 is a projector . in lemma [ lemma_simpleproject ] , we can replace @xmath0 by @xmath99 in which case the stabilizer is @xmath9 . the lemma is still satisfied . in particular , each balanced irreducible representation of @xmath1 is also a representation of @xmath0 . lemma [ lemma_simpleproject ] suggests that it should be possible to start with an @xmath0 lattice gauge theory and then to apply the averaging over the stabilizer in order to restrict the partition function to the balanced irreducible representations and at the same time split the @xmath0 integral of the haar intertwiner into the pair of independent @xmath2-integrals in the definition of the barrett crane intertwiner . in the remainder of this section , we prove that the barrett crane model is indeed obtained from lattice @xmath0 @xmath4-theory by applying lemma [ lemma_simpleproject ] in a suitable manner to the integrand of the partition function . we therefore confirm that the barrett crane model agrees with @xmath0 @xmath4-theory subject to constraints @xcite whose form we derive at the quantum level . we wish to apply the averaging procedure of lemma [ lemma_simpleproject ] to the @xmath0 haar intertwiner in order to obtain the barrett crane intertwiner together with the balancing ( _ simplicity _ ) condition @xmath287 . obviously , we need one averaging per edge and per face attached to that edge , namely one averaging for each representation function in the haar intertwiner . we therefore introduce the _ exploded configurations _ of lattice gauge theory which provide several copies of each edge variable so that we have separate variables attaining the same value for all faces attached to a given edge . let @xmath143 be a finite @xmath5-complex with cyclic structure . we define the _ explosion map _ @xmath289 so that @xmath290 . the elements of the set @xmath291 are called the _ exploded configurations_. the _ simplicity constraint _ , [ eq_simplicity ] @xmath292 is defined by specifying its image on a basis of the tensor product @xmath293 , namely for each @xmath294 , @xmath295 given any local boltzmann weight @xmath296 , the _ exploded boltzmann weight _ @xmath297 , is defined as , @xmath298 we have absorbed the edge amplitude in its factorized form into the definition of the simplicity constraint . the partition function of @xmath0 lattice gauge theory can , of course , be written in terms of the exploded configurations . let @xmath143 be a finite @xmath5-complex with cyclic structure . the partition function of @xmath0 lattice gauge theory on @xmath143 with boltzmann weight @xmath296 is equal to @xmath299 we can now state our theorem about the reformulation of the barrett crane model as an @xmath0 lattice gauge theory that is subject to the simplicity constraint . let @xmath143 be a finite @xmath5-complex with cyclic structure . the partition function of @xmath0 lattice gauge theory on the exploded configurations , subject to the simplicity constraint , @xmath300 agrees with the barrett crane model for face amplitudes , @xmath301 that are the coefficients of the @xmath0 character expansion of the boltzmann weight for the balanced irreducible representations . the edge amplitudes are of the factorized form . in particular , for @xmath0 lattice @xmath4-theory , @xmath177 , we obtain the barrett crane model with @xmath302 . for edge amplitudes with coefficients @xmath303 , the simplicity constraint forms a projector . first we consider the boltzmann weight @xmath304 on the exploded configurations and carry out the duality transformation for lattice gauge theory until the analogue of for the exploded configurations has been reached . we move all integrals from the haar intertwiners to the left of the expression and obtain the following integrand , @xmath305 applying the simplicity constraint , we find @xmath306 where we write @xmath307 for the @xmath0 variables of the exploded configurations . setting @xmath290 , we can integrate over @xmath0 for each edge and obtain two independent @xmath9 integrals over @xmath267 and @xmath269 per edge . the result agrees with . in a similar way , we could implement more general edge amplitudes into the simplicity constraint even if they do not factorize . however , the condition that the simplicity constraint forms a projector , singles out a particular edge amplitude . in this case the amplitude factorizes ( equation ) , and @xmath308 . in the entire discussion , we can replace @xmath0 by @xmath99 and at the same time @xmath101 by @xmath9 . passing to the covering does not change any result . the boltzmann weight in the dual formulation of the barrett crane model is obtained from the boltzmann weight of lattice gauge theory by applying the simplicity constraint . this means in particular that @xmath309 where @xmath310 denote representatives in @xmath111 . it is no coincidence that the derivation of the group theory picture in the present section closely resembles the reversed direction of the construction of oriti and williams @xcite . our result which was motivated by the idea that there is an @xmath0 parallel transport associated to the tetrahedra ( dual edges ) , turns out to agree with a ` trivial gluing ' in @xcite . in @xcite , a certain prescription for the gluing of four - simplices is employed in order to extend the geometrical constraints on a single four - simplex @xcite to the entire triangulation . different ways of imposing the constraints in the gluing procedure would give rise to different edge amplitudes @xcite . in our formulation , the edge amplitude @xmath311 in looks a bit artificial . for our choice of the constraint , only one amplitude seems to be natural , namely the case in which @xmath312 is a projector . in order to implement the constraints differently in the gluing procedure , one would have to change the constraint @xmath312 and perhaps also the combinatorics of the explosion map and would then recover a preferred edge amplitude from the requirement that @xmath312 forms a projector . the results of @xcite demonstrate that there exist alternative constraint maps @xmath312 that have a particular geometric interpretation and that give rise to different edge amplitudes . further work on these matters is in progress @xcite . the reformulation of the barrett crane model in terms of continuous variables presented in theorem [ thm_bcdual ] , offers a framework in which one can study the ground state of the model and small fluctuations around it , and which is also well suited for numerical studies . in this section , we sketch some central ideas and also comment on the question of which amplitudes @xmath313 and @xmath241 are admissible in the partition function . the amplitudes @xmath313 and @xmath241 are not fixed from geometric considerations @xcite . the simplest choice is , @xmath314 in this case the boltzmann weight behaves similarly to a @xmath79-function whose arguments are the products @xmath315 . if the two @xmath2-values of each pair agree , the weight diverges because @xmath316 . on the other hand , a simple numerical calculation suggests that the weight is very small ( grows with @xmath117 only much more slowly than the peak at @xmath317 ) whenever there is a pair of @xmath2-variables whose values do not agree . in order to obtain a first intuition , consider figure [ fig_geometry ] and assume that the boltzmann weight vanishes whenever two @xmath2-variables ( dots ) that form a pair in one of the characters ( pair of solid lines ) are not aligned . according to the structure of double lines in figure [ fig_geometry](b ) , there exists a vanishing weight unless all five @xmath2-values agree . the four - simplices would therefore be rigid , and the hyperplanes of all five tetrahedra in the boundary of a given four - simplex would have to be aligned . this is geometrically a very degenerate configuration in which the four - simplex has zero four - volume . in addition , there is no relation between the @xmath2-values assigned to different four - simplices . if we transit from one four - simplex along a tetrahedron to a neighbouring four - simplex and imagine that the change in @xmath2-value is given by a parallel transport along the ( edge dual to the ) tetrahedron , then there is no restriction on that parallel transport . however , these arguments are not quite correct because we did not precisely quantify the value of the total boltzmann weight if we multiply very large ( if @xmath317 ) and very small ( if some other @xmath318 ) numbers . we therefore have to regularize the weight function . this can be done , for example , by a non - trivial edge amplitude @xmath241 or by introducing a one parameter family of functions that tends to the degenerate face amplitude in a suitable ` limit ' . obvious choices of amplitudes are the following . let @xmath209 denote the face amplitudes , let @xmath319 define the factorized edge amplitude , and @xmath320 denote the boltzmann weight for a @xmath321-gon , expressed in terms of the dihedral angles @xmath322 on @xmath2 . for @xmath302 and @xmath323 or @xmath308 , the series does not converge if all @xmath324 . if @xmath302 and @xmath325 , then @xmath326 , @xmath327 , converges pointwise for all @xmath328 because the characters are bounded , @xmath329 for all @xmath265 . if @xmath330 and @xmath323 or @xmath308 , then @xmath326 converges pointwise for all @xmath328 and any @xmath331 . furthermore , a simple numerical calculation suggests that @xmath326 is strictly positive only in case ( 3 ) provided that @xmath308 , and that in the cases ( 2 ) and ( 3 ) , the function @xmath326 has a unique maximum when all @xmath332 . both choices ( 2 ) and ( 3 ) yield a non - trivial shape of the function @xmath326 . therefore the four - simplices will be not exactly rigid and degenerate , and there will be a non - trivial interaction between the four - simplices mediated by the boltzmann weights for each triangle . in order to obtain a well - defined partition function , it is sufficient that the integrand is an @xmath65 function of the @xmath9-variables . however , for numerical studies and also for an intuitive understanding , it is useful to have a pointwise convergent series so that we can always compare the function values for different angles @xmath333 . if we insist on a strictly positive and regular boltzmann weight function and on the projector property of the simplicity constraint , choice ( 3 ) with @xmath308 for which the @xmath313 are given by the character expansion coefficients of the @xmath0 heat kernel action ( for one of the casimir operators ) @xcite , deserves to be studied in greater detail . in this case , there exists a limit @xmath334 in which the boltzmann weight tends to the degenerate weight @xmath335 of the barrett crane model . choice ( 3 ) therefore defines a generalized barrett crane model with a bare ` inverse temperature ' @xmath336 or with a bare ` coupling constant ' @xmath337 . this regularized model provides a framework in which one could study the classical ground state , the configuration of maximal probability , and small fluctuations around it . the boltzmann weight has a non - trivial shape , and it makes sense to ask the question of whether the interaction terms for the triangles can be responsible for long range correlations . a perturbative analysis would correspond to an asymptotic expansion around @xmath334 which is precisely the interesting case . in the model with generic @xmath338 , the configurations of the spin foam formulation of the barrett crane model correspond to the coefficients of the strong coupling expansion , of an expansion for small @xmath338 , similarly to the situation in lattice gauge theory @xcite . at strong bare coupling , the configurations for small representations dominate . at this point , some warnings are appropriate . the reformulation of the barrett crane model might exhibit significant non - perturbative effects . for all choices of amplitudes @xmath313 and @xmath311 discussed above , the classical ground state , the state of maximal probability , is the configuration in which all tetrahedra of a given four - simplex are aligned . this configuration is geometrically degenerate and spans only a zero four - volume , so that one would need quantum effects in order to regularize the degeneracy and to provide a proper four - volume . this in turn means that the typical configurations one would find in the path integral , are ` far away ' from the classical ground state . confinement in pure @xmath339 gauge theory is a famous example for such a non - classical behaviour . in this case , perturbation theory can not be expected to describe the quantum system correctly because it is an expansion around the wrong vacuum . the new parameter @xmath338 and maybe the freedom to choose different edge amplitudes , also raise the question of whether and how we have to renormalize the model . from the above considerations , one might conjecture that the barrett crane model had to be renormalized along the lines of lattice gauge theory . the choice of a triangulation or of a dual @xmath5-complex for a given smooth manifold would correspond to a choice of coordinates , and the triangulation had to be ` fine ' enough , had to contain sufficiently many simplices , for the aspects we wish to study . an interesting question is certainly whether there exist long range correlations which would arise from local degrees of freedom of the model and correspond in some sense to the euclidean analogue of the graviton . long range is here referred to in a combinatorial sense , variables that are separated by many simplices should still be correlated . provided that such long range effective degrees of freedom exist , several scenarios are conceivable . first , we might have to tune or fine - tune @xmath338 or the edge amplitude in order to obtain an infinite or at least a very large correlation length . alternatively the model could be , at least for some range of parameters , in a coulomb like phase and exhibit correlations that decay with the reciprocal distance ( again in the combinatorial sense ) . this would provide a scenario in which there are always long range correlations and which is largely insensitive to the choice of parameters . after one has finally understood the spectrum of correlation lengths of the model in a combinatorial sense of ` length ' , one would have to reconstruct metric information from the data provided by the model so that , say in a suitable ` classical ' limit , the combinatorial correlations could be interpreted as correlation functions with respect to a metric distance . it is an interesting question how a fundamental length scale might arise in such an approach . finally , we mention that two versions of the barrett crane model , @xcite with normalized intertwiner , and @xcite , have edge amplitudes that do not factorize . in this case , the dual model has a non - local boltzmann weight , and the classical ground state is no longer obvious and might even have better properties . this is a second possibility , besides quantum effects , how one could avoid the degenerate ground state . a more detailed study of the possible edge amplitudes of the dual model is in preparation . we have presented a reformulation of the euclidean barrett crane model in terms of dual @xmath2-variables . the new variables have a geometric interpretation and can be understood as variables conjugate to the quantized areas . the fact that the dual model can be phrased in terms of continuous variables , prepares the ground for the application of a variety of analytical and numerical techniques . interesting are in particular the question of how the degenerate classical ground state can be regularized , maybe by quantum effects , maybe by a non - local boltzmann weight for the dual model or maybe only if we introduce a cosmological constant or couple gravity to matter . also important is the question of which correlations are relevant in the dual model . the possible choices of edge amplitudes and regularization parameters @xmath338 raise the question of whether and how one has to renormalize the theory . on the one hand , the model is not immediately triangulation independent , so that some refinement or sum over triangulations will be necessary . on the other hand , the result ( however this will be achieved ) is expected to be background independent . this requires an entirely new interpretation of the renormalization procedure as opposed to what is familiar from the context of lattice gauge theory and statistical mechanics . we are confident that the reformulation presented here offers a framework in which these questions can be approached . the author is grateful to emmanuel college , cambridge , for a research fellowship . i would like to thank john barrett , mtys kardi , alan macfarlane , robert oeckl , daniele oriti , nuno romo , ruth williams and toby wiseman for many valuable discussions , for comments on the manuscript and on relevant literature . 999 j. c. baez : spin , loops , knots and gauge fields . in _ knots and quantum gravity _ , ed . j. c. baez . oxford university press , oxford , 1994 . reisenberger : world sheet formulation of gauge theories and gravity . preprint ` gr - qc/9412035 ` . m. p. reisenberger and c. rovelli : ` sum over surfaces ' form of loop quantum gravity . * d 56 * ( 1997 ) 34903508 , ` gr - qc/9612035 ` . j. w. barrett and l. crane : relativistic spin networks and quantum gravity . _ j. math . phys . _ * 39 * , no . 6 ( 1998 ) 32963302 , ` gr - qc/9709028 ` . j. c. baez : spin foam models . _ class . _ * 15 * , no . 7 ( 1998 ) 18271858 , ` gr - qc/9709052 ` . v. g. turaev and o. y. viro : state sum invariants of @xmath340-manifolds and quantum @xmath341-symbols . _ topology _ * 31 * , no . 4 ( 1992 ) 865902 . l. crane , l. h. kauffmann and d. n. yetter : state - sum invariants of @xmath266-manifolds . _ j. knot th . _ * 6 * , no . 2 ( 1997 ) 177234 , ` hep - th/9409167 ` . r. oeckl and h. pfeiffer : the dual of pure non - abelian lattice gauge theory as a spin foam model . phys . _ * b 598 * , no . 1 - 2 ( 2001 ) 400426 , ` hep - th/0008095 ` . h. pfeiffer and r. oeckl : the dual of non - abelian lattice gauge theory . to be published in the proceedings of lattice 2001 ( 2002 ) . preprint ` hep - th/0110034 ` . h. pfeiffer : four - dimensional lattice gauge theory with ribbon categories and the crane yetter state sum . _ j. math . _ * 42 * , no . 11 ( 2001 ) 52725305 , ` hep - th/0106029 ` . r. oeckl : generalized lattice gauge theory , spin foams and state sum invariants ( 2001 ) . preprint ` hep - th/0110259 ` . j. c. baez : an introduction to spin foam models of @xmath4 theory and quantum gravity . in _ geometry and quantum physics _ , lecture notes in physics 543 . springer , berlin , 2000 , pp . 2593 , ` gr - qc/9905087 ` . d. oriti : space - time geometry from algebra : spin foam models for nonperturbative quantum gravity . * 64 * ( 2001 ) 17031756 , ` gr - qc/0106091 ` . j. c. baez and j. w. barrett : the quantum tetrahedron in 3 and 4 dimensions . phys . _ * 3 * , no . 4 ( 1999 ) 815850 , ` gr - qc/9903060 ` . r. de pietri , l. freidel , k. krasnov and c. rovelli : barrett - crane model from a boulatov - ooguri field theory over a homogeneous space . phys . _ * b 574 * , no . 3 ( 2000 ) 785806 , ` hep - th/9907154 ` . a. perez and c. rovelli : a spin foam model without bubble divergences . phys . _ * b 599 * , no . 1,2 ( 2001 ) 255282 , ` gr - qc/0006107 ` . d. oriti and r. m. williams : gluing @xmath266-simplices : a derivation of the barrett crane spin foam model for euclidean quantum gravity . * d 63 * ( 2001 ) 024022 , ` gr - qc/0010031 ` . j. w. barrett : the classical evaluation of relativistic spin networks . _ * 2 * , no . 3 ( 1998 ) 593600 , ` math/9803063 ` . l. freidel and k. krasnov : simple spin networks as feynman diagrams . _ j. math . phys . _ * 41 * , no . 4 ( 2000 ) 16811690 , ` hep - th/9903192 ` . h. ooguri : topological lattice models in four dimensions . lett . _ * a 7 * , no . 30 ( 1992 ) 27992810 . n. kawamoto , n. sato and y. uchida : @xmath266-dimensional @xmath4 gravity on the lattice . _ * b 574 * ( 2000 ) 809848 , ` hep - th/9911228 ` . j. c. baez and j. d. christensen : positivity of spin foam amplitudes ( 2001 ) . preprint ` gr - qc/0110044 ` . vilenkin and a. u. klimyk : representations of lie groups and special functions class i representations , special functions and integral transforms , volume 2 . kluwer academic publishers , doordrecht , 1993 . r. carter , g. segal and i. macdonald : lectures on lie groups and lie algebras . london mathematical society student texts 32 . cambridge university press , cambridge , 1995 . h. j. rothe : lattice gauge theories an introduction . world scientific , singapore , 1992 . i. montvay and g. mnster : quantum fields on a lattice . cambridge university press , cambridge , 1994 . reisenberger : on relativistic spin network vertices . _ j. math . _ * 40 * , no . 4 ( 1999 ) 20462054 , ` gr - qc/9809067 ` d. oriti : in preparation . l. freidel , k. krasnov and r. puzio : @xmath4 description of higher - dimensional gravity theories . _ adv . th . math . * 3 * ( 1999 ) 12891324 , ` hep - th/9901069 ` . r. de pietri and l. freidel : @xmath342 plebanski action and relativistic spin - foam model . _ class . * 16 * ( 1999 ) 21872196 , ` gr - qc/9804071 ` .
we present the physical properties of v404 lyr exhibiting eclipse timing variations and multiperiodic pulsations from all historical data including the and superwasp observations . detailed analyses of 2,922 minimum epochs showed that the orbital period has varied through a combination with an upward - opening parabola and two sinusoidal variations , with periods of=649 d and=2,154 d and semi - amplitudes of=193 s and=49 s , respectively . the secular period increase with a rate of.41 d yr could be interpreted as a combination of the secondary to primary mass transfer and angular momentum loss . the most reasonable explanation for both sinusoids is a pair of light - travel - time effects due to two circumbinary objects with projected masses of=0.47 m and=0.047 m . the third - body parameters are consistent with those calculated using the wilson - devinney binary code . for the orbital inclinations 43 , the fourth component has a mass within the hydrogen - burning limit of.07 m , which implies that it is a brown dwarf . a satisfactory model for the _ kepler _ light curves was obtained through applying a cool spot to the secondary component . the results demonstrate that the close eclipsing pair is in a semi - detached , but near - contact , configuration ; the primary fills approximately 93% of its limiting lobe and is larger than the lobe - filling secondary . multiple frequency analyses were applied to the light residuals after subtracting the synthetic eclipsing curve from the _ kepler _ data . this revealed that the primary component of v404 lyr is a dor type pulsating star , exhibiting seven pulsation frequencies in the range of 1.85.11 d with amplitudes of 1.38.72 mmag and pulsation constants of 0.24.27 d. the seven frequencies were clearly identified as high - order low - degree gravity - mode oscillations which might be excited through tidal interaction . only eight eclipsing binaries have been known to contain dor pulsating components and , therefore , v404 lyr will be an important test - bed for investigating these rare and interesting objects .
1405.5658
orbital periods of eclipsing binaries ( ebs ) are one of the most accurately measured physical parameters that are required in order to understand the dynamics of the systems and their stellar structures and evolutions . the period variability may be represented using an eclipse timing diagram , which displays the differences between the observed and calculated minimum epochs _ versus _ time . the orbital periods of many ebs have varied in some combination of secular and sinusoidal variations rather than in a monotonic manner ( kreiner et al . 2001 ; lee et al . 2009b , 2013a ) . the presence of a third body orbiting an eb causes a periodic variation of the eclipsing period due to the increasing and decreasing light - travel times ( ltt ) to the observer ( irwin 1952 , 1959 ) . if the historical database for light curves and eclipse times is large and sufficiently long to understand the binary s period change , circumbinary objects that are gravitationally bound to ebs can be detected . the circumbinary companions may have played an important function in the formation and evolution of close ebs , which would cause them to have short initial orbital periods and evolve into their present configurations through angular momentum loss due to magnetic braking ( pribulla & rucinski 2006 ) . ebs can precisely and directly measure the masses and radii for each component from time - series photometry and spectroscopy , while pulsating stars provide significant information about the interior structure of the stars through their pulsation features . some ebs contain @xmath16 sct or @xmath14 dor type pulsating components with multiple periods , which are very useful for asteroseismical studies . the binary stars that exhibit both eclipses and pulsations are very promising targets that could lead to unique constraints on theoretical models , and they have increased significantly in number by ground - based survey projects ( e.g. , kim et al . 2003 ) and space missions such as _ kepler _ and _ corot _ ( e.g. , southworth et al . 2011 ; maceroni et al . most of them have been found to be @xmath16 sct - type members of algol - type semi - detached ebs designated by the so - called oea star ( mkrtichian et al . in contrast , there have only been eight ebs reported with @xmath14 dor pulsating components ( zhang et al . 2013 ; maceroni et al . 2014 ) . in general , @xmath14 dor pulsators are main - sequence or sub - giant stars with spectral types ranging from a7 to f5 near the red edge of the @xmath16 sct instability strip in the h - r diagram . their observational properties are very similar to those of @xmath16 sct stars , except the significant difference in pulsation periods and pulsation constants ( handler & shobbrook 2002 ) . recently , stars with both @xmath16 sct and @xmath14 dor characteristics have been discovered , three of which exhibit a hybrid nature of those classes in ebs ( corot 100866999 , chapellier & mathias 2013 ; kic 4544587 , hambleton et al . 2013 ; kic 3858884 , maceroni et al . the possible existence of hybrid @xmath14 dor/@xmath16 sct stars is of great interest because they offer additional constraints on stellar structures . in the field of the old galactic cluster ngc 6791 , v404 lyr ( kic 3228863 , tyc 3121 - 1291 - 1 , 2mass j19190594 + 3822005 ) was classified as a @xmath17 lyr - type eb system with an orbital period of 0.7309 d by popova & kraicheva ( 1984 ) . however , from their incomplete ccd photometry in the @xmath18 bandpasses , csizmadia & sndor ( 2001 ) suggested that the binary star is not a member of the cluster . since then , somosvri & csizmadia ( 2006 ) made additional observations in order to complete the previous light curves but approximately 20 @xmath19 of them remained deficient . through analyzing the @xmath18 light curves , they presented that v404 lyr is an algol - type semi - detached binary with a mass ratio of @xmath20=0.485 , an orbital inclination of @[email protected] , and a temperature difference between the components of @xmath22=1,928 k. recently , the highly precise photometric observations of v404 lyr have been obtained in long cadence mode of the _ kepler _ satellite , which led to a sampling rate of 29.4244 min . detailed information regarding the _ kepler _ spacecraft and its performance can be found in borucki et al . ( 2010 ) and koch et al . the orbital period changes for v404 lyr were examined by rappaport et al . ( 2013 ) and conroy et al . ( 2014 ) using the _ kepler _ public data . the former found that the orbital period has varied due to a periodic oscillation with a cycle length of 668.4 d and showed that its variation would be produced by the existence of a physically linked third body with a mass function of @xmath23=0.017 m@xmath9 and an eccentricity of @xmath24=0.08 . the latter concluded that the cause of the period change is an ltt effect due to a third body with a period of 644.1 d and a semi - amplitude of 195 s in a circular orbit . in order to obtain a reasonable set of photometric solutions and to examine whether the previous suggestions are appropriate for the orbital period change , we have meticulously investigated the long - term photometric behavior of v404 lyr using all historical data . for this , we used 56,333 individual observations ( bjd 2,454,953@xmath152,456,391 ) from quarter 1 through quarter 16 of the _ kepler _ operations . our results from eclipse timing variation , light - curve synthesis , and frequency analysis for light residuals indicate that v404 lyr is most likely a multiple system with a @xmath14 dor - type pulsating component . from the _ kepler _ observations of v404 lyr , we determined 2,882 eclipse timings and their errors using the method presented by kwee & van woerden ( 1956 ) . these are given in the first column of table 1 , wherein 12 additional eclipses were obtained using the data from the superwasp ( wide angle search for planets ) public archive ( butters et al . in addition to these , 28 ccd timings have been collected from the literature ( agerer & hbscher 1996 , 1997 , 1998 , 1999 , 2000 ; csizmadia & sndor 2001 ; diethelm 2001 ; safr & zejda 2002 ; borkovits et al . 2003 ; locher 2005 ; nelson 2005 , 2007 ; br et al . 2006 ; hbscher et al . 2006 ; brt et al . the timings were transformed from hjd based on utc into tdb - based bjd using the online applets developed by eastman et al . ( 2010 ) . for ephemeris computations , weights were calculated as the inverse squares of the timing errors and were then scaled from the standard deviations ( @xmath25=0.0010 d ) of all timing residuals . previous researchers ( rappaport et al . 2013 ; conroy et al . 2014 ) have proposed that the eclipse timing variation of v404 lyr can be represented using an ltt caused by the presence of a third body orbiting the eclipsing pair . first of all , we fitted the times of minimum light to the single ltt ephemeris , as follows : @xmath26 where @xmath27 is the ltt due to a tertiary companion ( irwin 1952 , 1959 ) and includes five parameters ( @xmath28 , @xmath29 , @xmath30 , @xmath31 , @xmath32 ) . here , @xmath33 , @xmath29 , and @xmath30 are the orbital parameters of the eclipsing pair around the mass center of the triple system . the parameters @xmath31 and @xmath32 denote the keplerian mean motion of the mass center of the binary components and its epoch of periastron passage , respectively . the levenberg - marquart ( lm ) technique ( press et al . 1992 ) was applied to solve the equation for the ephemeris parameters . the result is plotted in figure 1 and the orbital parameters are summarized in the second column of table 2 , together with their related quantities . the absolute dimensions presented in the following section have been used for these and subsequent calculations . the residuals from the ltt ephemeris appear as @xmath34@xmath35 in the fourth column of table 1 . as shown in figure 1 , the single ltt ephemeris could not describe the previously published minima and the timing residuals in the lower panel indicate the existence of a ( or some ) further effect(s ) . after testing several other forms , such as a quadratic term _ plus _ a single - ltt ephemeris and a two - ltt ephemeris , we found that the eclipse timing variation is best fitted using a quadratic _ plus _ two - ltt ephemeris ( lee et al . 2009a , 2013a ) : @xmath36 the lm method was again applied to evaluate the thirteen parameters of the ephemeris which are listed in the third and fourth columns of table 2 . this ephemeris resulted in a much smaller @xmath37=0.996 than the single ltt ephemeris ( @xmath37=2.256 ) . the short - term periods ( @xmath1 ) and semi - amplitudes ( @xmath3 ) for the two ephemerides are in excellent agreement with each other . the @xmath34@xmath38 diagram constructed using the linear terms is presented at the top of figure 2 with the solid curve due to the sum of the non - linear terms and the dashed parabola due to the quadratic term of equation ( 2 ) . the second and third panels display the @xmath39 and @xmath40 orbits , respectively , and the bottom panel represents the residuals from the full ephemeris . these appear as @xmath34@xmath41 in the fifth column of table 1 . assuming the orbits of the circumbinary objects are coplanar with that ( @[email protected] ) of the eclipsing pair of v404 lyr , the masses of the third and fourth bodies are @xmath8=0.469 m@xmath9 and @xmath10=0.047 m@xmath9 , respectively , and the radii are calculated to be @xmath42=0.484 r@xmath9 and @xmath43=0.054 r@xmath9 from the empirical relation described by southworth ( 2009 ) . if the third component is on main sequence , its temperature and bolometric luminosity are @xmath44=3,445 k and @xmath45=0.030 @xmath46 , respectively , which would contribute only 0.5% to the total luminosity of the quadruple system . when the ltt periods are very short , the eclipse timing variations could be caused by the geometrical and perturbative effects of the circumbinary companions ( borkovits et al . 2011 , 2013 ; rappaport et al . we computed the semi - amplitudes of the dynamic perturbations ( physical delays ) on the motion of the eclipsing pair to be 0.00005 d and 0.000002 d , respectively , for the third and fourth components and found that their contributions are not significant . alternatively , the periodic oscillations in the eclipse timing diagram might be caused by a magnetic activity cycle for systems with a spectral type later than f5 ( applegate 1992 ; lanza et al . however , it is difficult for the applegate model to produce a perfectly smooth periodic component in the timing residuals . moreover , lanza ( 2006 ) presented that this mechanism is not adequate in explaining the period modulation of close binaries with a late - type secondary . thus , the periodic variations most likely arise from the ltt effects due to the existence of the two circumbinary companions that are gravitationally bound to the close eclipsing pair . these indicate that v404 lyr is a quadruple system . the positive coefficient of the quadratic term ( @xmath47 ) listed in table 2 indicates a secular period increase with a rate of @xmath51.41 @xmath6 d yr@xmath7 . the most common explanation of the parabolic variation is a mass transfer from the less massive secondary to the primary component if the secondary fills its inner roche lobe . our light - curve synthesis shows that the eclipsing pair of v404 lyr is a semi - detached binary with a secondary - filling configuration ( cf . section 3 ) . this indicates that the secondary to primary mass transfer can be possible in the system . under the assumption of a conservative mass transfer , the transfer rate is 5.42@xmath4810@xmath49 m@xmath9 yr@xmath7 . the observed value is approximately 40% smaller when compared with the predicted rate of 9.06@xmath4810@xmath49 m@xmath9 yr@xmath7 , which is calculated by assuming that the secondary transfers its present mass to the primary in a thermal time scale . thus , the possible explanation of the secular variation might be some combination of non - conservative mass transfer and angular momentum loss due to magnetic braking . in figure 3 , the _ kepler _ light curve of v404 lyr is plotted as normalized flux _ versus _ orbital phase . its shape resembles that of @xmath17 lyr and therefore indicates a significant temperature difference between the two components and a significant distortion of the photospheres . in order to derive the physical parameters of this system , the _ kepler _ data were analyzed using the 2007 version of the wilson - devinney binary code ( wilson & devinney 1971 , hereafter w - d ) . independently of eclipse timing diagrams , this version can extract third - body orbit parameters ( @xmath50 , @xmath51 , @xmath52 , @xmath53 , @xmath54 , @xmath55 ) that revolve around an eclipsing pair from whole light curves ( van hamme & wlson 2007 ) , where the primed notation refers to the third - body orbit in the w - d code . the parameters @xmath51 , @xmath52 , @xmath53 are the inclination , eccentricity , and argument of periastron of the third - body orbit , respectively . @xmath50=@xmath56 is the semimajor axis of the outer relative orbit , and @xmath55 is the time of the superior conjunction of the mass center of the close binary with respect to the third body . for the light - curve synthesis , the effective temperature of the hotter and presumably more massive primary star was initialized at @xmath57=6,561 k from the _ kepler _ input catalogue ( brown et al . the logarithmic bolometric ( @xmath58 , @xmath59 ) and monochromatic ( @xmath60 , @xmath61 ) limb - darkening coefficient were interpolated from the values of van hamme ( 1993 ) in concert with the model atmosphere option . the gravity - darkening exponents and bolometric albedoes were fixed at standard values ( @xmath62=0.32 and @xmath47=0.5 ) for stars with convective envelopes , as surmised from the components temperatures . furthermore , a synchronous rotation for both components and a circular orbit were adopted , and the detailed reflection effect and third light source ( @xmath63 ) were considered throughout the analyses . the values with parenthesized errors in table 3 signify adjusted parameters . in this paper , subscripts 1 and 2 refer to the primary and secondary stars eclipsed at min i ( at phase 0.0 ) and min ii , respectively . although the photometric studies of v404 lyr were performed by somosvri & csizmadia ( 2006 ) and pra et al . ( 2011 ) , their results are preliminary and there is still no spectroscopic mass ratio ( @xmath20 ) . in order to generally understand the geometrical structure and photometric parameters of the system , an extensive @xmath20-search procedure was conducted for various modes of the w - d code , without permitting third body or spot effects . in this procedure , we considered the orbital inclination ( @xmath21 ) , effective temperature ( @xmath64 ) , dimensionless surface potential ( @xmath65 ) , and luminosity ( @xmath66 ) . because the _ kepler _ data base is very large , we used 1,000 normal points that were calculated using bin widths of 0.001 in phase from all individual observations and were assigned weights equal to the number of observations per normal point . the @xmath20-search result exhibited acceptable photometric solutions for mode 5 only , which are semi - detached systems with the secondary filling its inner roche lobe . as seen in figure 4 , the weighted sum of the squared residuals ( @xmath67 ) reached a minimum value at @xmath20=0.38 . for subsequent calculations , all individual points were used and the mass ratio and third - body parameters were considered to be additional free variables . the results are listed in the second and third columns of table 3 and they appear as a dashed curve in the top panel of figure 3 . the light residuals from the solution are calculated to see the details of non - modeled lights and plotted in the middle panel of the figure , wherein it can be seen that the unspotted model does not describe the observed light curve around phases 0.39 and 0.61 . in short - period semi - detached binaries , the light discrepancy may result from either a mass transfer between the component stars or a magnetic dynamo in the systems with deep convective envelopes . therefore , we tested three possible spot models : a hot spot on the primary star due to the mass transfer and a cool spot on either binary component caused by magnetic activity . the best solution is given in the fourth and fifth columns of table 3 and is described using a solid curve in the top panel of figure 3 . the residuals from this spot model are plotted in the bottom panel of the figure . separate trials for other spot configurations were not as successful as the fitting provided in table 3 . from the table and displays , it is clearly seen that the photometric parameters without and with a spot are very close to each other and that the cool spot on the secondary component could almost entirely explain the light variation . the third - body orbit parameters computed from all _ kepler _ data are in good agreement with those from the historical eclipse timings within the uncertainties . in all procedures , we searched for a possible third light but the parameter remained indistinguishable from zero within its error . as discussed in the previous section , the light contribution of the third body to the multiple system is approximately 0.5 % , assuming the binary and third - body coplanar orbit ; thus , the absence of this evidence is not surprising . the differential correction ( dc ) program of the w - d code produces the errors computed from the covariance matrix using the standard method . the parameter errors should arise from a final run that includes all adjustable parameters together . however , some papers ( e.g. , maceroni & rucinski 1997 ) commented that the error estimates in the output from dc are unrealistically small due to the strong correlations between the relatively numerous parameters and the non - normal distribution of the measurement errors . following the procedure described by koo et al . ( 2014 ) , we estimated the errors for the fitted parameters . for this purpose , the _ kepler _ data were split into ten subsets and modeled individually with the w - d code . then , the error estimates were calculated as the standard deviations of the ten different values for each parameter . the errors in table 3 are the 1@xmath25-values adopted from the larger of the values from the ten datasets and from all data . the absolute dimensions of v404 lyr can be estimated from our photometric solutions and empirical relations between spectral type and mass . assuming that the primary component is a normal main - sequence star with a spectral type of f4 v from its temperature ( harmanec 1988 ) , we obtained the absolute parameters for the eclipsing pair given in the bottom of table 3 . the luminosity ( @xmath68 ) and bolometric magnitudes ( @xmath69 ) were computed by adopting @xmath70@xmath9=5,780 k and @xmath69@xmath9=+4.73 as solar values . for the absolute visual magnitudes ( @xmath71 ) , we used the bolometric corrections ( bcs ) appropriate for the temperature of each component from the relationship between @xmath72 and bc given by torres ( 2010 ) . with an apparent magnitude of @xmath73=+11.39 and an interstellar absorption of @xmath74=0.30 ( csizmadia & sndor 2001 ) , we calculated a nominal distance to the system of 465 pc . this is too small compared with the distance ( 4,188 pc ) of ngc 6791 ( basu et al . 2011 ) and , hence , v404 lyr is not a member of the open cluster . figure 5 plots the light curve residuals from our cool - spot model distributed in bjd instead of orbital phase as in figure 3 , wherein the lower panel presents a short section of the residuals . this figure clearly exhibits light variations with a semi - amplitude of about 30 mmag maximum . the amplitude changes from cycle to cycle , which indicates that multiple curves with different periods are superimposed . we performed multiple frequency analyses for the residuals using the period04 program ( lenz & breger 2005 ) . the binary components block each other s lights during eclipses , which can make complicated variations . in order to remove this unmanageable effect , we analyzed only out - of - eclipse residuals ( phases [email protected] and [email protected] ) after removing the times of the primary and secondary eclipses . the top panel of figure 6 displayed the amplitude spectra in the frequency range from 0 to the nyquist limit of 24.47 d@xmath7 . because only the out - of - eclipse phase data was used , the orbital frequency @xmath75 caused alias effects , i.e. 1.368 d@xmath7 multiplets deviated from the dominant peak , which is seen in the figure . the main signals lie in the frequency region between 0.5 and 5.0 d@xmath7 . after consecutive pre - whitening processes , a total of 40 frequencies were detected based on the criterion of the signal to noise amplitude ratio ( s / n ) larger than 4.0 ( breger et al . 1993 ) . at each pre - whitening procedure , we applied a multiperiodic least - squares fit to the light residuals using the equation of @xmath76 = @xmath77 + @xmath78 @xmath79(2@xmath80 ) . here , @xmath76 and @xmath77 denote the calculated magnitude and zero point , @xmath81 and @xmath82 are the amplitude and phase of the @xmath21th frequency , and @xmath83 is the time of each measurement . the amplitude spectra after pre - whitening the first six frequencies and then all 40 frequencies are presented in the middle and bottom panels of figure 6 , respectively . the results are listed in table 4 . the uncertainties in the table were derived according to kallinger et al . the synthetic curve computed from the 40-frequency fit is displayed in figure 5 . we examined the frequency variations with time through analyzing the out - of - eclipse residuals at intervals of approximately 100 d. the 12 subsets resulted in a little differences from each other and 9@xmath1516 frequencies were detected at each subset with the same criterion of s / [email protected] . figure 7 displayed the variability of the 12 frequencies detected most repeatedly . the eight frequencies of @xmath85@xmath15@xmath86 , @xmath87 , and @xmath88 are almost constant . on the contrary , the four frequencies of @xmath89@xmath15@xmath90 and @xmath91 are varied significantly , which might be partly affected by the changes in the spot parameters with time and/or by the irregularity in the mass transfer from the secondary component to the primary . these unequable phenomena may cause several other frequencies near the four frequencies that were detected in the full out - of - eclipse data in table 4 . in order to determine which component caused the seven dominant and constant frequencies near 2.0 d@xmath7 , we compared the light variations between the out - of - eclipse and primary / secondary eclipse data . a section of the light residuals from figure 5 is magnified in figure 8 , where the synthetic light curve calculated from the parameters in table 4 is well matched with both the out - of - eclipse and secondary eclipse data . however , the curve is poorly fitted with the primary eclipse data . in our light - curve representation for v404 lyr , the primary eclipse is a partial transit , while the secondary eclipse is a total occultation of the secondary star by the primary . therefore , the agreement in the secondary eclipse but disagreement in the primary eclipse implies that the primary star is the main light variation source . this discrepancy can be explained by the phase shift of non - radial pulsation during the eclipse as simulated by reed et al . ( 2005 ) . considering the physical parameters in table 3 , the primary component of v404 lyr is an early f - type main sequence star , located near the red edge of instability strip for @xmath14 dor type pulsating stars ( handler & shobbrook 2002 ) . the multiperiodic pulsations with low amplitudes less than 10 mmag can occur in this type of pulsators . we calculated the pulsation constants of seven frequencies near 2.0 d@xmath7 , @xmath85@xmath15@xmath86 and @xmath87 , from the equation of @xmath92 ( breger 2000 ) . the results are listed in the third column of table 5 . the @xmath93s of [email protected] d correspond to the gravity modes of @xmath14 dor pulsating stars . the ratios of the orbital frequency to the seven pulsation frequencies @xmath75:@xmath94 are very close to 2:3 , which implies that the @xmath14 dor type pulsations might be excited by the tidal interaction of the close secondary component . we attempted to identify the pulsation modes of the seven frequencies by applying the frequency ratio method ( frm ) , which was introduced by moya et al . according to the asymptotic approximation ( tassoul 1980 ) assuming adiabatic and non - rotation condition , the angular frequency of high - order ( @xmath31@xmath951 ) low - degree ( @xmath96@xmath973 ) gravity modes for typical @xmath14 dor stars can be simplified as follows : @xmath98 where @xmath99 is the integral of the brunt - visl frequency . the theoretical value of @xmath99 is approximated as a constant for gravity modes of typical @xmath14 dor stars ( moya et al . therefore , the frequency ratio is expressed as follows : @xmath100 we calculated the model frequency ratios from equation ( 4 ) for several pulsation mode sets ( @xmath31 , @xmath96 ) and found out the best fitted ratios with the observed ratios , as listed at table 5 . although the rotation velocity of v404 lyr has not been known , the primary component may be a fast rotator of approximately 120 km s@xmath7 assuming a synchronized rotation with the orbital motion . because the frm intrinsic error increases with rotational velocity ( surez et al . 2005 ) , the ratio error was set to have a slightly larger value of @xmath1010.02 ; rodriguez et al . ( 2006 ) set the error to be @xmath1010.012 for hd 218427 with a moderate rotation velocity of 72 km s@xmath7 . the four frequencies of @xmath102 , @xmath103 , @xmath85 , and @xmath104 are identified as @xmath96=2 mode for four consecutive radial orders of @xmath31=24 , 25 , 26 , and 27 , respectively . and the two frequencies @xmath105 and @xmath86 are @xmath96=1 mode for two orders of @xmath31=14 and 15 , respectively . the other @xmath87 with a lower amplitude than the former six frequencies may be excited by a less - detectable @xmath96=3 mode for @xmath31=39 . we obtained the observed frequency integral @xmath106=771.5 @xmath101 4.2 @xmath107hz from equation ( 3 ) , averaging the values for the seven frequencies . this is similar to the theoretical value @xmath108 near 750 @xmath107hz for a model of @xmath109 @xmath70=3.817 , 1.35 @xmath110 , and [ fe / h]=0.0 ( see figure 5 in the paper by moya et al . considering that the physical properties of v404 lyr , such as metal abundance and effective temperature , are uncertain , the difference of about 20 @xmath107hz may be acceptable . we have presented the long - term photometric behavior of v404 lyr which demonstrates total eclipses and multiperiodic pulsations , based on all available data as well as the @xmath0 data . our analyses of 2,922 eclipse timings , spanning more than 18 yrs , reveal that the orbital period of the system has varied due to two periodic variations superimposed on an upward - opening parabola , with cycle lengths of @xmath1=649 d and @xmath2=2,154 d and semi - amplitudes of @xmath3=193 s and @xmath4=49 s , respectively . the upward parabola indicates a continuous period increase , which can be plausibly explained by a combination of mass transfer from the secondary to the primary star and angular momentum loss . the most reasonable explanation for both cycles is a pair of ltt effects driven by the presence of a third and fourth component with minimum masses of @xmath8=0.47 m@xmath9 and @xmath10=0.047 m@xmath9 . as long as the inclination of the orbital plane of the fourth component is higher than 43@xmath12 , the mass of this object is below the theoretical threshold of @xmath130.07 m@xmath9 for a hydrogen - burning star . in eclipse timing diagrams , it is possible that the periodic variations with small amplitudes may be produced by the sporadic asymmetries of eclipse light curves due to stellar activity such as starspots and pulsations ( cf . lee et al . 2014 ) . as in the process described by lee et al . ( 2013b ) , we combined the _ kepler _ data at intervals of 10 orbital periods ( @xmath13 7.3 d ) and calculated 174 minimum epochs for those datasets using the w - d code through adjusting only the ephemeris epoch ( @xmath111 ) in the spot model of table 3 . the results are illustrated with the ` x ' symbols in figure 2 . it can be seen that the light - curve timings calculated from the w - d code agree well with our analysis of the eclipse timing variation . thus , we conclude that the orbital period of v404 lyr has varied due to two periodic oscillations plus an upward parabola . the @xmath0 light curves of v404 lyr were satisfactorily modeled using a cool spot on the secondary component and third - body parameters . the results represent the eclipsing pair as a semi - detached binary with @xmath20=0.382 , @xmath21=83@xmath12 , and @xmath22=1,193 k , in which the primary component fills its limiting lobe by approximately 93% . the cool spot may be produced by magnetic dynamo - related activity because the system is rotating rapidly and the components have a deep convective envelope . the third - body parameters from the w - d code are consistent with those calculated from the eclipse timings themselves . from the absolute parameters given in section 3 , it is possible to estimate the evolutionary state of the binary system in the mass - radius , mass - luminosity , and hertzsprung - russell ( hr ) diagrams ( hilditch et al . 1988 ; . ibanolu et al . 2006 ) . in these diagrams , the primary star lies in the main - sequence band , while the secondary is clearly beyond the terminal - age main sequence and its radius is more than two times oversized compared with the main - sequence stars of the same mass . furthermore , the locations of the two component stars fall amid those of previously - known near - contact binaries ( shaw 1990 , 1994 ) . thus , the eclipsing pair is a semi - detached and fo vir - type near - contact binary consisting of a detached main - sequence primary star and an evolved lobe - filling secondary component . through analyzing the light residuals from our w - d binary model , we found intrinsic variations of 40 frequencies between 0.60 and 4.11 d@xmath7 with amplitudes from 0.49 to 5.72 mmag . among these , seven frequencies near 2.0 d@xmath7 with large amplitudes and high stability for about 4 years transpired to originate from the pulsation of the primary component . @xmath16 sct pulsators are stars pulsating in low - order pressure modes with short periods of [email protected] d ( breger 2000 ) , whereas @xmath14 dor pulsators are a - f stars of luminosity class iv - v pulsating in high - order gravity modes with typical periods in the range 0.4@xmath153 d ( kaye et al . 1999 , henry et al . the pulsation periods of v404 lyr indicate that the primary component is a candidate for @xmath14 dor type pulsating stars , rather than @xmath16 sct variables . handler & shobbrook ( 2002 ) argued the relationship between the @xmath16 sct and @xmath14 dor variables and showed that the two pulsators are clearly separated by their pulsation constants : the @xmath16 sct stars have @xmath112 0.04 d and the @xmath14 dor stars @xmath113 0.23 d. as listed in the third column of table 5 , the pulsation constants of v404 lyr are in the range of [email protected] d. a recent study by zhang et al . ( 2013 ) indicates that the ratio of pulsation and orbital periods could function as a criterion to distinguish them in ebs , where @xmath1140.09 is the upper limit for @xmath16 sct stars . all of these results reveal that the primary component is a @xmath14 dor pulsator . the eclipsing pair of v404 lyr is a short - period semi - detached binary and its pulsating component has the largest filling factor ( 93% ) of the 78 known ebs with @xmath16 sct or @xmath14 dor pulsators ( zhang et al . 2013 ; maceroni et al . 2014 ; yang et al . 2014 ) . based on this , the pulsating characteristics of the primary star may be partly influenced by tidal interaction and secondary to primary mass transfer . the properties of a convective core in the @xmath14 dor star can be demonstrated through studying tidally induced and gravity mode pulsations . because it has evolved with mass accretion between the binary components , the pulsating primary star may have a different evolutionary history than single @xmath14 dor pulsators . high - resolution spectroscopy will assist in determining the absolute parameters of the multiple system and in understanding its evolutionary status better than is possible with photometry alone . this research has made use of the _ kepler _ and superwasp public archives . _ kepler _ was selected as the 10th mission of the discovery program . funding for the _ kepler _ mission is provided by the nasa science mission directorate . the wasp consortium comprises of the university of cambridge , keele university , university of leicester , the open university , the queen s university belfast , st . andrews university and the isaac newton group . funding for wasp comes from the consortium universities and from the uk s science and technology facilities council . we have used the simbad database maintained at cds , strasbourg , france . this work was supported by the kasi ( korea astronomy and space science institute ) grant 2014 - 1 - 400 - 06 . applegate , j. h. 1992 , apj , 385 , 621 agerer , f. , & hbscher , j. 1996 , inf . variable stars , 4383 , 1 agerer , f. , & hbscher , j. 1997 , inf . bull . variable stars , 4472 , 1 agerer , f. , & hbscher , j. 1998 , inf . bull . variable stars , 4562 , 1 agerer , f. , & hbscher , j. 1999 , inf . bull . variable stars , 4711 , 1 agerer , f. , & hbscher , j. 2000 , inf . bull . variable stars , 4912 , 1 basu , s. , et al . 2011 , apj , 729 , l10 br , i. b. , et al . 2006 , inf . bull . variable stars , 5684 , 1 borkovits , t. , et al . 2003 , inf . bull . variable stars , 5434 , 1 borkovits , t. , csizmadia , s. , forgcs - dajka , e. , & hegeds , t. 2011 , a&a , 528 , a53 borkovits , t. , et al . 2013 , mnras , 428 , 1656 borucki , w. j. , et al . 2010 , sci , 327 , 977 brt , l. , zejda , m. , & svoboda . stars , 74 , 1 breger , m. 2000 , in asp conf . 210 , delta scuti and related stars , eds . m. breger and m. montgomery ( san francisco : asp ) , 3 breger , m. , et al . 1993 , a&a , 271 , 482 brown , t. m. , latham , d. w. , everett , m. e. , & esquerdo , g. a. 2011 , aj , 142 , 112 butters , o. w. , et al . 2010 , a&a , 520 , l10 chapellier , e. , & mathias , p. 2013 , a&a , 556 , a87 csizmadia , s. , & sndor , z. 2001 , inf . variable stars , 5045 , 1 conroy , k. e. , pra , a. , stassun , k. g. , orosz , j. a. , fabrycky , d. c. , & welsh , w. f. 2014 , aj , 147 , 45 diethelm , r. 2001 , inf . variable stars , 5027 , 1 eastman , j. , siverd , r. , & gaudi , b. s. 2010 , pasp , 122 , 935 hambleton , k. m. , et al . 2013 , mnras , 434 , 925 handler , g. , & shobbrook , r. r. 2002 , mnras , 333 , 251 harmanec , p. 1988 , bull . czechoslovakia , 39 , 329 henry , g. w. , fekel , f. c. , & henry , s. m. 2005 , aj , 129 , 2815 hilditch , r. w. , king , d. j. , & mcfarlane , t. m. 1988 , mnras , 231 , 341 hbscher , j. , paschke , a. , & walter , f. 2006 , inf . stars , 5731 , 1 . ibanolu , c. , soydugan , f. , soydugan , e. , & derviolu , a. 2006 , mnras , 373 , 435 irwin , j. b. 1952 , apj , 116 , 211 irwin , j. b. 1959 , aj , 64 , 149 kallinger , t. , reegen , p. , & weiss , w. w. 2008 , a&a , 481 , 571 kaye , a. b. , handler , g. , krisciunas , k. , poretti , e. , & zerbi , f. m. 1999 , pasp , 111 , 840 kim , s .- l . , lee , j. w. , kwon , s .- g . , youn , j .- h . , mkrtichian , d. e. , & kim , c. 2003 , a&a , 405 , 231 koch , d. g. , et al . 2010 , apj , 713 , l79 koo , j .- , et al . 2014 , aj , 147 , 104 kreiner , j. m. , kim , c .- h . , & nha , i .- s . 2001 , an atlas of @xmath34@xmath115 diagrams of eclipsing binary stars ( krakow : wydawn . pedagogicznej ) kwee , k. k. , & van woerden , h. 1956 , bull . netherlands , 12 , 327 lanza , a. f. 2006 , mnras , 369 , 1773 lanza , a. f. , rodono , m. , & rosner , r. 1998 , mnras , 296 , 893 lee , j. w. , kim s .- l . , kim c .- h . , koch r. h. , lee c .- u . , , & park j .- h . 2009a , aj , 137 , 3181 lee , j. w. , kim , s .- l . , lee , c .- u . , & youn , j .- h . 2009b , pasp , 121 , 1366 lee , j. w. , hinse , t. c. , & park j .- h . 2013a , aj , 145 , 100 lee , j. w. , kim , s .- l . , lee , c .- u . , lee , b .- c . , park , b .- , & hinse , t. c. 2013b , apj , 763 , 74 lee , j. w. , park , j .- h . , hong , k. , kim , s .- l . , & lee , c .- u . , 2014 , aj , 147 , 91 lenz , p. , & breger , m. 2005 , comm . asteroseismology , 146 , 53 locher , k. 2005 , open eur . j. var . stars , 3 , 1 maceroni , c. , & rucinski s. m. , 1997 , pasp , 109 , 782 maceroni , c. , montalbn , j. , gandolfi , d. , pavlovski , k. , & rainer , m. 2013 , a&a , 552 , 60 maceroni , c. , et al . 2014 , a&a , 563 , 59 mkrtichian , d. e. , et al . 2004 , a&a , 419 , 1015 moya , a. , et al . 2005 , a&a , 432 , 189 nelson , r. h. 2005 , inf . stars , 5602 , 1 nelson , r. h. 2007 , inf . bull . stars , 5760 , 1 popova , m. , & kraicheva , z. 1984 , astrofiz . issled . izv . spets . obs . , 18 , 64 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes ( cambridge : cambridge univ . press ) , chapter 15 pribulla , t. , & rucinski , s. m. 2006 , aj , 131 , 2986 pra , a. , et al . 2011 , aj , 141 , 83 rappaport , s. , et al . 2013 , apj , 768 , 33 reed , m. d. , brondel , b. j. , & kawaler , s. d. 2005 , apj , 634 , 602 rodriguez , e. , et al . 2006 , a&a , 450 , 715 safr , j. , & zejda , m. 2002 , inf . bull . variable stars , 5263 , 1 shaw , j. s. 1990 , in active close binaries , ed . c. ibanoglu ( dordrecht : kluwer ) , 241 shaw , j. s. 1994 , mem . soc . astron . italiana , 65 , 95 somosvri , b. m. , & csizmadia , s. 2006 , in asp conf . ser . 349 , astrophysics of variable stars , eds . sterken and c. aerts ( san francisco : asp ) , 331 southworth , j. 2009 , mnras , 394 , 272 southworth , j. , et al . 2011 , mnras , 414 , 2413 surez , j. c. , et al . 2005 , a&a , 443 , 271 tassoul , m. 1980 , apjs , 43 , 469 torres , g. 2010 , aj , 140 , 1158 van hamme , w. 1993 , aj , 106 , 209 van hamme , w. , & wilson , r. e. 2007 , apj , 661 , 1129 wilson , r. e. , & devinney , e. j. 1971 , apj , 166 , 605 yang , y .- g . , wei , j .- y . , & 2014 , aj , 147 , 35 zhang , x. b. , luo , c. q. , & fu , j. n. 2013 , apj , 777 , 77 @xmath115 diagram of v404 lyr . in the upper panel , constructed with the linear terms of equation ( 1 ) , the continuous curve represents the ltt orbit . the residuals from this ephemeris are plotted in the lower panel . ] @xmath115 diagram of v404 lyr with respect to the linear terms of the quadratic _ plus _ two - ltt ephemeris . the full ephemeris is drawn as a solid curve and the dashed parabola results from the quadratic term of equation ( 2 ) . the second and third panels display the short- and long - term ltt orbits , respectively . the bottom panel represents the residuals from the complete ephemeris . in all panels , the blue ` x ' symbols refer to the minimum times obtained by means of analyzing 174 _ kepler _ datasets with the w - d code . ] lcrccc 2,454,230.63111 & @xmath1010.00031 & @xmath15989.0 & @xmath150.00103 & @xmath150.00092 & i + 2,454,249.63673 & @xmath1010.00077 & @xmath15963.0 & @xmath50.00042 & @xmath50.00057 & i + 2,454,252.55891 & @xmath1010.00046 & @xmath15959.0 & @xmath150.00110 & @xmath150.00096 & i + 2,454,256.57899 & @xmath1010.00070 & @xmath15953.5 & @xmath150.00113 & @xmath150.00098 & ii + 2,454,257.67461 & @xmath1010.00038 & @xmath15952.0 & @xmath150.00190 & @xmath150.00176 & i + 2,454,278.50706 & @xmath1010.00077 & @xmath15923.5 & @xmath150.00087 & @xmath150.00072 & ii + 2,454,279.60470 & @xmath1010.00048 & @xmath15922.0 & @xmath50.00037 & @xmath50.00053 & i + 2,454,282.52827 & @xmath1010.00044 & @xmath15918.0 & @xmath50.00024 & @xmath50.00039 & i + 2,454,283.62463 & @xmath1010.00049 & @xmath15916.5 & @xmath50.00021 & @xmath50.00036 & ii + 2,454,286.54651 & @xmath1010.00044 & @xmath15912.5 & @xmath150.00161 & @xmath150.00146 & ii + 2,454,287.64502 & @xmath1010.00032 & @xmath15911.0 & @xmath50.00050 & @xmath50.00066 & i + 2,454,291.66576 & @xmath1010.00050 & @xmath15905.5 & @xmath50.00115 & @xmath50.00130 & ii + 2,454,953.89883 & @xmath1010.00030 & @xmath50.5 & @xmath150.00020 & @xmath50.00057 & ii + 2,454,954.26383 & @xmath1010.00022 & @xmath51.0 & @xmath150.00066 & @xmath50.00011 & i + 2,454,954.63106 & @xmath1010.00079 & @xmath51.5 & @xmath50.00109 & @xmath50.00187 & ii + 2,454,954.99363 & @xmath1010.00070 & @xmath52.0 & @xmath150.00179 & @xmath150.00102 & i + 2,454,955.36131 & @xmath1010.00042 & @xmath52.5 & @xmath50.00042 & @xmath50.00119 & ii + 2,454,955.72618 & @xmath1010.00029 & @xmath53.0 & @xmath150.00017 & @xmath50.00060 & i + 2,454,956.09245 & @xmath1010.00070 & @xmath53.5 & @xmath50.00063 & @xmath50.00140 & ii + 2,454,956.45536 & @xmath1010.00073 & @xmath54.0 & @xmath150.00191 & @xmath150.00115 & i + lccccc @xmath111 & 2,454,953.53370(24 ) & & & bjd + @xmath116 & 0.73094341(21 ) & & & d + @xmath117 & 0.395(57 ) & & 0.387(30 ) & 0.098(31 ) & au + @xmath30 & 153.1(8.7 ) & & 43.6(4.5 ) & 191(17 ) & deg + @xmath29 & 0.08(27 ) & & 0.05(11 ) & 0.00(54 ) & + @xmath31 & 0.5543(98 ) & & 0.5549(44 ) & 0.167(12 ) & deg d@xmath7 + @xmath32 & 2,454,903(15 ) & & 2,454,703(8 ) & 2,454,301(101 ) & bjd + @xmath118 & 649(11 ) & & 648.8(5.1 ) & 2154(160 ) & d + @xmath119 & 197(28 ) & & 193(15 ) & 49(16 ) & s + @xmath120 & 0.0195(28 ) & & 0.0184(14 ) & 0.0000272(89 ) & m@xmath9 + @xmath121 & 0.479(38 ) & & 0.469(20 ) & 0.0468(75 ) & m@xmath9 + @xmath122 & 1.541(62 ) & & 1.545(32 ) & 4.91(42 ) & au + @xmath47 & & & & d + @xmath123/@xmath124 & & & & d yr@xmath7 + reduced @xmath125 & 2.256 & & & + lccccc @xmath111 ( bjd ) & & & + @xmath116 ( d ) & & & + @xmath20 & & & + @xmath21 ( deg ) & & & + @xmath64 ( k ) & 6,554(69 ) & 5,363(47 ) & & 6,555(67 ) & 5,362(46 ) + @xmath65 & 2.8406(6 ) & 2.6277 & & 2.8549(6 ) & 2.6400 + @xmath58 , @xmath59 & 0.638 , 0.243 & 0.645 , 0.186 & & 0.638 , 0.243 & 0.645 , 0.185 + @xmath60 , @xmath61 & 0.613 , 0.285 & 0.693 , 0.219 & & 0.613 , 0.285 & 0.693 , 0.219 + @xmath68/(@xmath126+@xmath127 ) & 0.8094(2 ) & 0.1906 & & 0.8072(2 ) & 0.1928 + @xmath128 ( pole ) & 0.4013(1 ) & 0.2778(1 ) & & 0.3999(1 ) & 0.2789(1 ) + @xmath128 ( point ) & 0.4605(2 ) & 0.4009(1 ) & & 0.4588(2 ) & 0.4024(1 ) + @xmath128 ( side ) & 0.4215(1 ) & 0.2895(1 ) & & 0.4199(1 ) & 0.2907(1 ) + @xmath128 ( back ) & 0.4380(1 ) & 0.3222(1 ) & & 0.4364(1 ) & 0.3234(1 ) + @xmath128 ( volume ) & 0.4207 & 0.2976 & & 0.4192 & 0.2989 + [ 1.0 mm ] + @xmath50(@xmath129 ) & & & + @xmath51 ( deg ) & & & + @xmath52 & & & + @xmath53 ( deg ) & & & + @xmath54 ( d ) & & & + @xmath55 ( bjd ) & & & + + colatitude ( deg ) & & & & & 60.7(5 ) + longitude ( deg ) & & & & & 352.9(2 ) + radius ( deg ) & & & & & 22.9(7 ) + @xmath64@xmath130/@xmath64@xmath131 & & & & & 0.872(9 ) + @xmath67 & & & + [ 1.0 mm ] + @xmath132(@xmath110 ) & 1.35 & 0.51 & & 1.35 & 0.52 + @xmath133(@xmath129 ) & 1.76 & 1.25 & & 1.76 & 1.26 + @xmath109 @xmath62 ( cgs ) & 4.08 & 3.95 & & 4.08 & 3.95 + @xmath134 ( g @xmath135 & 0.35 & 0.37 & & 0.35 & 0.37 + @xmath68(@xmath136 ) & 5.14 & 1.15 & & 5.12 & 1.17 + @xmath69 ( mag ) & 2.95 & 4.57 & & 2.96 & 4.56 + bc ( mag ) & @xmath50.01 & @xmath150.18 & & @xmath50.01 & @xmath150.18 + @xmath71 ( mag ) & 2.94 & 4.75 & & 2.95 & 4.74 + distance ( pc ) & & & + lccccc @xmath137 & [email protected] & [email protected] & [email protected] & 49.69 & @xmath14 dor + @xmath138 & [email protected] & [email protected] & [email protected] & 38.00 & @xmath14 dor + @xmath139 & [email protected] & [email protected] & [email protected] & 33.78 & @xmath14 dor + @xmath140 & [email protected] & [email protected] & [email protected] & 28.96 & @xmath14 dor + @xmath141 & [email protected] & [email protected] & [email protected] & 29.45 & @xmath14 dor + @xmath142 & [email protected] & [email protected] & [email protected] & 24.79 & @xmath14 dor + @xmath143 & [email protected] & [email protected] & [email protected] & 13.22 & @xmath13@xmath144 + @xmath145 & [email protected] & [email protected] & [email protected] & 15.46 & + @xmath146 & [email protected] & [email protected] & [email protected] & 17.03 & + @xmath147 & [email protected] & [email protected] & [email protected] & 18.09 & + @xmath148 & [email protected] & [email protected] & [email protected] & 10.64 & + @xmath149 & [email protected] & [email protected] & [email protected] & 9.05 & + @xmath87 & [email protected] & [email protected] & [email protected] & 11.73 & @xmath14 dor + @xmath150 & [email protected] & [email protected] & [email protected] & 11.38 & + @xmath151 & [email protected] & [email protected] & [email protected] & 11.71 & + @xmath152 & [email protected] & [email protected] & [email protected] & 12.13 & + @xmath88 & [email protected] & [email protected] & [email protected] & 11.99 & 4@xmath153 + @xmath91 & [email protected] & [email protected] & [email protected] & 8.39 & + @xmath154 & [email protected] & [email protected] & [email protected] & 7.29 & @xmath155 + @xmath156 & [email protected] & [email protected] & [email protected] & 13.67 & + @xmath157 & [email protected] & [email protected] & [email protected] & 9.40 & 4@xmath158 + @xmath159 & [email protected] & [email protected] & [email protected] & 8.10 & + @xmath160 & [email protected] & [email protected] & [email protected] & 12.31 & + @xmath161 & [email protected] & [email protected] & [email protected] & 8.13 & 4@xmath162 + @xmath163 & [email protected] & [email protected] & [email protected] & 7.49 & 4@xmath164 + @xmath165 & [email protected] & [email protected] & [email protected] & 5.85 & 2@xmath166 + @xmath167 & [email protected] & [email protected] & [email protected] & 5.22 & + @xmath168 & [email protected] & [email protected] & [email protected] & 5.72 & @xmath169 + @xmath170 & [email protected] & [email protected] & [email protected] & 7.31 & 4@xmath171 + @xmath172 & [email protected] & [email protected] & [email protected] & 5.18 & + @xmath173 & [email protected] & [email protected] & [email protected] & 5.80 & + @xmath174 & [email protected] & [email protected] & [email protected] & 6.78 & 4@xmath175 + @xmath176 & [email protected] & [email protected] & [email protected] & 7.56 & + @xmath177 & [email protected] & [email protected] & [email protected] & 4.70 & + @xmath178 & [email protected] & [email protected] & [email protected] & 4.72 & + @xmath179 & [email protected] & [email protected] & [email protected] & 7.39 & @xmath133@xmath144 + @xmath180 & [email protected] & [email protected] & [email protected] & 4.72 & @xmath181 + @xmath182 & [email protected] & [email protected] & [email protected] & 4.67 & + @xmath183 & [email protected] & [email protected] & [email protected] & 7.22 & + @xmath184 & [email protected] & [email protected] & [email protected] & 4.22 & 3@xmath158 + lccccccc @xmath85 & 1.97460 & 0.25 & 0.9351 & ( 26 , 2 ) & 0.9245 & + 0.0106 & 776.75 + @xmath102 & 2.11165 & 0.24 & @xmath15 & ( 24 , 2 ) & @xmath15 & @xmath15 & 767.97 + @xmath105 & 2.08421 & 0.24 & 0.9870 & ( 14 , 1 ) & 0.9755 & + 0.0115 & 777.01 + @xmath104 & 1.89372 & 0.26 & 0.8968 & ( 27 , 2 ) & 0.8909 & + 0.0059 & 773.05 + @xmath103 & 2.03560 & 0.24 & 0.9640 & ( 25 , 2 ) & 0.9608 & + 0.0032 & 770.53 + @xmath86 & 1.92342 & 0.26 & 0.9109 & ( 15 , 1 ) & 0.9126 & @xmath150.0017 & 766.52 + @xmath87 & 1.85446 & 0.27 & 0.8782 & ( 39 , 3 ) & 0.8772 & + 0.0010 & 768.88 + & & & & & average & + [email protected] & [email protected] +
we investigate the current status of the light neutralino dark matter scenario within the minimal supersymmetric standard model ( mssm ) taking into account latest results from the lhc . a discussion of the relevant constraints , in particular from the dark matter relic abundance , leads us to a manageable simplified model defined by a subset of mssm parameters . within this simplified model we reinterpret a recent search for electroweak supersymmetric particle production based on a signature including multi - taus plus missing transverse momentum performed by the atlas collaboration . in this way we derive stringent constraints on the light neutralino parameter space . in combination with further experimental information from the lhc , such as dark matter searches in the monojet channel and constraints on invisible higgs decays , we obtain a lower bound on the lightest neutralino mass of about . this limit is stronger than any current limit set by underground direct dark matter searches or indirect detection experiments . with a mild improvement of the sensitivity of the multi - tau search , light neutralino dark matter can be fully tested up to about . ulb - th/13 - 10 , mpp-2013 - 162 , stupp-2013 - 213 * cornering light neutralino dark matter at the lhc * + lorenzo calibbi , jonas m. lindert , toshihiko ota , yasutaka takanishi + _ service de physique thorique , universit libre de bruxelles , _ _ bld du triomphe , cp225 , b-1050 brussels , belgium _ _ max - planck - institut fr physik ( werner - heisenberg - institut ) , _ _ fhringer ring 6 , d-80805 mnchen , germany _ _ department of physics , saitama university , _ _ shimo - okubo 255 , 338 - 8570 saitama - sakura , japan _ _ max - planck - institut fr kernphysik , _ _ saupfercheckweg 1 , d-69117 heidelberg , germany _
1307.4119
searches for supersymmetry ( susy ) performed by the large hadron collider ( lhc ) experiments on the center - of - mass energy @xmath5 tev run data have started to set stringent constraints to the strongly - interacting susy particles . they translate to limits on the susy masses up to @xmath6 tev for gluinos and squarks of the first family @xcite and @xmath7 gev for third generation squarks @xcite . comparatively weaker constraints were obtained for the states that can be only produced through electroweak ( ew ) interactions , such as sleptons , ew gauginos and higgsinos . the resulting bounds are up to @xmath8 gev for selectrons and smuons @xcite and up to @xmath9 gev for charginos @xcite . they often depend on the assumption of available on - shell decays . nevertheless , it is remarkable that the lhc experiments have started especially with the 2012 run data at @xmath10 tev to go considerably beyond lep in testing the ew sector of the theory . such direct searches for ew production of susy particles are of major importance , as the ew - interacting particles can be in principle much lighter than the strongly - interacting ones . in this paper , we want to discuss how the 8 tev run data constrain the parameter space that is compatible with a light neutralino as candidate for cold dark matter ( dm ) . in other words , we aim at answering the following question : how light can the lightest neutralino still be after the 8 tev run of the lhc ? the framework we adopt for our study can be defined as follows : * only the field content of the mssm is considered ; * dm is a thermal relic and a standard history of the universe is assumed ; * the abundance of the lightest mssm neutralino , whose stability is guaranteed by r - parity conservation , is required to not exceed the observed dm relic abundance . we are not going to make any assumption on the origin and the relations among the susy - breaking parameters : in particular we drop the hypothesis of gaugino mass unification that would imply the lower bound for the lightest neutralino mass reported by the pdg @xcite : @xmath11 . instead , we treat all susy soft - masses as free low - energy parameters . as we discuss in the next section , relic density constraints from cosmic microwave background ( cmb ) observations identify the parameter space compatible with light neutralino dark matter and thus the features of the susy spectrum and the lhc phenomenology . in particular , we are going to argue that , under the assumptions listed above , neutralino dm with @xmath12 is only possible in a specific region of the supersymmetric parameter space , featuring relatively light staus and higgsinos . hence , searches based on multi - tau plus missing energy events are particularly promising in order to test such a scenario , as we are going to discuss in detail . a similar study has been presented last year in ref . @xcite . here , we want to generalize the approach of ref . @xcite , by considering a simplified model defined only by the subset of susy parameters relevant for the determination of the relic abundance . the rest of the spectrum is allowed to be heavy , so that our study is very conservative as it considers only ew production of staus and neutral and charged higgsinos . moreover , the atlas collaboration has recently performed a search for new physics in a final state with at least two taus and large missing transverse momentum @xcite , that we will translate into a stringent test of the light neutralino parameter space . at @xmath10 tev , has been presented by cms @xcite . the exclusion limit they obtain seems to be perfectly compatible with the results of ref . @xcite . ] the rest of this paper is organized as follows . in section [ sec : lightdm ] we qualitatively review the features of the parameter space compatible with light neutralino dm and we define the simplified model we are going to employ in the rest of the study . a quantitative discussion of the relevant constraints and the result of a numerical scan of the parameter space are presented in section [ sec : relicdensity ] , while bounds from direct and indirect dark matter searches are briefly discussed in [ sec : dmsearches ] . section [ sec : lhc_pheno ] is dedicated to a discussion of the spectrum of the light dm scenario and the consequences for susy searches at the lhc . in section [ sec : otherlhc ] , we discuss the indirect limits on the parameter space from invisible higgs decays and we show the possible impact of direct dm searches at the lhc in the monojet plus missing energy channel . in section [ sec : lhc_numerics ] , we present a monte carlo study reproducing the limits from ref . afterwards we translate these limits into bounds on the light neutralino parameter space . we conclude summarizing our results in section [ sec : conclusions ] . in this section , we discuss how the parameter space of the mssm is constrained by the requirement of neutralino dark matter with @xmath13 gev . the first obvious condition is that , due to the lep bound on the chargino mass @xcite that implies for the wino and higgsino mass parameters @xmath14 gev , the lightest neutralino is mainly bino - like with the bino mass @xmath15 approximately : @xmath16 the other susy parameters depend strongly on how the dm relic density constraints are fulfilled and hence on the annihilation processes of the lightest neutralino in the early universe . in fact , a bino - like neutralino is typically overproduced in thermal processes so that an efficient annihilation mechanism is required in order to reproduce the observed relic density . there are mainly the following three categories for the annihilation mechanism that select different regions of the parameter space : ( i ) @xmath17-channel higgs mediation , ( ii ) co - annihilation with a light sfermion , ( iii ) @xmath18-channel sfermion mediation . [ [ higgs - mediated - annihilation . ] ] higgs - mediated annihilation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + case ( i ) , in which the neutralino pair - annihilation is mediated by higgs bosons ( mainly the cp - odd one , as the s - wave @xmath19 initial state is cp - odd ) , is an attractive possibility , because the parameter space selected by this scenario unavoidably corresponds to a large scattering cross - section with nuclei ( up to @xmath20 @xmath21 ) , so that the signals at dama @xcite , cogent @xcite , cresst - ii @xcite and the three events recently reported by cdms @xcite could be nicely explained @xcite . however , recent lhc results exclude light dm in this parameter region of the mssm . the reason is the following : @xmath22 gev requires a light cp - odd higgs boson @xmath23 ( with @xmath24 gev ) and quite large values of @xmath25 ( @xmath26 ) to make the annihilation cross - section efficient enough @xcite . this setup has been recently excluded by extra higgs boson searches at the lhc @xcite . on the other hand , larger values of @xmath27 ( @xmath28 gev ) still correspond to a direct detection cross - section @xmath29 @xmath21 that is excluded by xenon100 @xcite for that mass range @xcite . with br @xmath30 @xcite and the observation of a sm - like higgs with @xmath31 gev @xcite are also incompatible with the parameter space of the higgs mediation scenario , as one can see in ref . @xcite . ] in other words , lhc searches for the cp - odd higgs in combination with direct dark matter searches constrain @xmath32 to values that can not efficiently mediate neutralino annihilation for @xmath13 gev . [ [ co - annihilation - with - sfermions . ] ] co - annihilation with sfermions . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the co - annihilation scenario ( ii ) with a light stau has recently awaken a great interest , because the parameter choice can be also compatible with an enhancement of the @xmath33 decay rate , as discussed in ref . however , in order to have an efficient co - annihilation for a neutralino lighter than 30 gev , one would need @xmath34 gev @xcite , below the limit set by the lep experiments @xcite . on the other hand , if the stau - neutralino mass splitting is small enough to evade direct lep searches , the scenario is seriously challenged by @xmath35 width measurements implying @xmath36 gev @xcite ( unless the stau left - right mixing is tuned such that the coupling to @xmath35 is strongly suppressed ) and would anyway correspond to a too efficient annihilation , i.e. to a dm relic density way below the observed value . as we will discuss in the following , neutralino dm with @xmath13 gev also requires a light stau . however , the selected parameter space requires small values of the higgsino mass parameter @xmath37 , while very large values of the stau left - right mixing @xmath38 are needed in order to have an effective enhancement of @xmath39 @xcite . we have checked that the parameter region we consider does not drive the co - annihilation process and does not give a significant enhancement of @xmath39 . nevertheless , the current lhc measurements of the higgs decay rates , especially @xmath40 still provide important information for our scenario . we will come back to this point later . [ [ sfermion - mediation . ] ] sfermion mediation . + + + + + + + + + + + + + + + + + + + the only option left is @xmath18-channel sfermion mediation ( iii ) . light neutralino dm is indeed possible in presence of a light stau , as recently shown in refs . the reasons why the exchange of other sfermions can not give enough enhancement to the annihilation cross - section are twofold : ( a ) lep and lhc searches put severe limits on the masses of the other sfermions , ( b ) efficient annihilation of very light neutralinos requires the contribution of yukawa interactions . the annihilation cross - section is inverse - proportional to the mass of the mediating sfermion . therefore , it is constrained by the bound on the mass of the mediation field , the so - called lee - weinberg bound @xcite . the lhc places strong limits on the masses of squarks and first and second generation sleptons . for instance , direct slepton searches with @xmath41 and @xmath37 in final states at the lhc imply @xmath42 gev ( @xmath43 ) for @xmath44 gev @xcite . similarly , the bounds on direct production of stop and sbottom subsequently decaying to the lightest susy particle ( lsp ) are @xmath45 gev @xcite . if the sfermion - neutralino mass - splitting is small enough ( @xmath46 gev ) , these collider bounds can be evaded , as the outgoing fermion is too soft to be detected ( see e.g. @xcite ) . however , lep observations of the @xmath35 decays imply the lower bound on any sfermion @xmath47 gev . this means that for @xmath44 gev it is anyway not possible to have efficient annihilation mediated by light sfermions and to evade at the same time the collider constraints . this justifies _ a posteriori _ our choice of concentrating on a neutralino lighter than 30 gev . interaction gets strongly suppressed @xcite . this possibility might be challenged by indirect dm searches through antiproton and gamma - ray production @xcite . see however @xcite . ] relevant neutralino annihilation processes mediated by a light stau . ] the only option left is that the lightest neutralino , that is mainly bino , does pair - annihilate into @xmath48 via the @xmath18-channel stau exchange diagram driven by gauge interactions shown in the left panel of fig . [ fig : annihilation ] . since the amplitude is proportional to the square of the hypercharge of the mediating sfermion , the right - handed stau mediation is much more efficient than the left - handed one , giving a factor 16 larger cross - section . as a consequence , the relic density constraints select the mass of the right - handed stau to be not far above the lep lower bound @xcite : @xmath49 however , the gauge interaction diagram might not be enough to have an efficient neutralino annihilation , especially for even lower neutralino masses , e.g. @xmath50 gev . if the lightest neutralino has a sizeable higgsino component , this can contribute through higgsino - bino and higgsino - pair annihilation diagrams ( shown as the middle and the right diagrams in fig . [ fig : annihilation ] ) . these further contributions assist in realizing the correct relic density with a very light neutralino , in particular for large values of @xmath25 . vanishes if higgsinos decouple . ] a significant higgsino component in the lightest neutralino requires a small @xmath37-term . this implies that , in addition to the lightest @xmath51 , the lightest chargino @xmath52 and two higgsino - like neutralinos ( @xmath53 and @xmath54 , if @xmath55 ) are necessarily light . in short , in order to enhance the annihilation process driven by the higgsino component in the lightest neutralino , the following values of the above parameters are selected : @xmath56 in summary , the parameter space compatible with light neutralino dm we aim at studying is essentially determined by the values of the four parameters given in eqs . ( [ eq : our - parameter - region - nr1]-[eq : our - parameter - region - nr3 ] ) , ( @xmath57 , @xmath37 , @xmath58 , and @xmath25 ) and does not depend on the detail of the other susy parameters . in the following we illustrate numerically the parameter space qualitatively sketched above . we then make use of the simplified model defined by this subset of the parameters in order to discuss the limits set by the lhc on neutralino dm . in order to investigate the parameter space selected by requiring light neutralino dark matter , we perform a numerical scan by means of the suspect @xcite and micromegas codes @xcite . the low - energy values of the four parameters identified above were randomly varied in the following ranges : @xmath59 and the susy breaking scale within suspect is set to the approximate scale of the relevant low - mass states ( @xmath60 gev ) . the other susy parameters have a marginal role in fulfilling the relic density constraints and they were set to the following constant values : @xmath61 where @xmath62 denotes all sfermion soft masses besides @xmath63 , @xmath64 are the gaugino masses , @xmath32 the cp - odd higgs mass . the values chosen for @xmath65 and @xmath62 give the higgs mass consistent with the lhc measurements , @xmath66 gev . all other @xmath23-terms are set to zero . in the following we describe various ( possible ) constraints on the resulting parameter space , where , as motivated above , we limit our discussion on the light neutralino regime defined in eq . ( [ eq : mchi30 ] ) . [ [ dm - relic - density . ] ] dm relic density . + + + + + + + + + + + + + + + + + assuming a standard thermal history of the universe , we compute with micromegas @xcite the neutralino relic density and impose a conservative 3@xmath67 upper bound taken from ref . @xcite , @xmath68 [ [ direct - susy - searches - at - lep . ] ] direct susy searches at lep . + + + + + + + + + + + + + + + + + + + + + + + + + + + + the 95% cl lep bounds on the lightest stau and chargino masses listed in ref . @xcite are , respectively , @xmath69 in addition , we consider bounds from searches of @xmath70 associated production at lep , followed by the decay @xmath71 . the conservative limit on this process is about 100 @xmath72 for the @xmath73 mass range we are interested in @xcite . explicitly this bound reads @xmath74 we calculate the associated production cross - sections at lep using the leading order formulae reported in refs . @xcite , the branching fractions were computed by means of the susy - hit package @xcite . [ [ z - invisible - width . ] ] @xmath35 invisible width . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + width form the process @xmath75 as a function of the higgsino mass parameter @xmath37 . the dashed line corresponds to the lep limit of ref . @xcite.[fig : zchichi],scaledwidth=60.0% ] as discussed above , a sizeable higgsino @xmath76 component in @xmath73 is required . this gives rise to the coupling @xmath77 and also to @xmath78 that originates from the gauge vertex @xmath79 . as a consequence , the decays @xmath80 and @xmath81 can occur at relevant rates and we expect stringent constraints from the lep measurement of the invisible @xmath35 width as well as from higgs observations at the lhc , as recently discussed in ref . the decay width of the @xmath35 boson into a lightest neutralino pair reads @xcite : @xmath82 where @xmath83 is the neutralino mixing matrix defined by @xmath84 . in fig . [ fig : zchichi ] we plot the resulting @xmath85 as a function of @xmath37 , after applying the other constraints described above . the orange points correspond to @xmath73 dm with @xmath86 gev , the blue points to @xmath87 gev . the figure is to be compared to the lep bound on the new physics contribution to @xmath88 , @xmath89 @xcite : @xmath90 as we see , only a small number of points ( corresponding to very low @xmath37 and particularly light @xmath73 masses ) is excluded by this observable . is missing in the tree - level result stated in eq . a similar constraint from @xmath81 will be discussed in section [ sec : invh ] . [ [ flavor - processes . ] ] flavor processes . + + + + + + + + + + + + + + + + + rare decays such as @xmath91 , recently measured at @xmath92 by lhcb @xcite , the partially correlated processes of the kind @xmath93 , charged current processes like @xmath94 and @xmath95 set stringent constraints on supersymmetric models with large @xmath96 @xcite ( for a discussion in the context of light neutralino dm , see @xcite ) . nevertheless , these processes are mediated by the charged or cp - odd higgs boson ( and also squarks in the case of @xmath93 ) , hence the rates depend on the mass - scale of the extended higgs sector ( and squarks ) . these parameters can be set to arbitrarily high values , as the @xmath73 annihilation cross - section is not affected by them . as a consequence , in general flavor processes do not constrain the light neutralino parameter space we consider here . [ [ lhc - searches . ] ] lhc searches . + + + + + + + + + + + + + recent limits from chargino searches in multi - leptons + missing transverse energy @xmath97 events at the lhc @xcite depend on the mass of the first and second generation sleptons and assume wino - like charginos . hence they do not apply to our scenario . bounds from lhc searches based on at least one pair of hadronically decaying taus @xcite are not applied to the scan we are presenting . they will be discussed in detail in the next sections . plane.[fig : stau - chi],scaledwidth=60.0% ] [ [ section ] ] the resulting parameter space compatible with light neutralino dark matter and the bounds listed above is shown in fig . [ fig : stau - chi ] , for the physical masses @xmath98 and @xmath99 . again the orange points correspond to @xmath86 gev , the blue points to @xmath100 gev . the upper boundary of the relevant parameter space is set by the cmb constraint , eq . ( [ eq : planck ] ) , that requires low values for either @xmath98 or @xmath101 . hence , light neutralino dark matter implies upper bounds on the stau and higgsino masses : @xmath102 notice that , below the threshold where @xmath103 decays into a real stau are possible , some points are excluded by the limit from lep neutralino searches stated in eq . ( [ eq : opal ] ) . this is a consequence of the fact that , as far as @xmath103 decays into a real @xmath35 are kinematically allowed while the decays into a real stau are forbidden , one obtains @xmath104 , so that the lep bound is maximized . on the other hand , for very low @xmath105 , where this decay is not open , or where this decay competes with on - shell decays into staus , constraints from eq . ( [ eq : opal ] ) vanish . plane.[fig : tb - chi],scaledwidth=60.0% ] in fig . [ fig : tb - chi ] , the same parameter space is shown in the @xmath106 plane . different ranges for @xmath37 are displayed in different colors . as a consequence of the higgsino contribution to the @xmath73 annihilation , the larger @xmath96 is , the smaller @xmath107 can be . taking @xmath108 gev , required for perturbativity of the bottom yukawa coupling , one then gets the following lower bound on the mass of neutralino dark matter : @xmath109 from the figure we see that neutralino masses close to this lower limit require a particularly light higgsino sector , namely @xmath110 gev . in sections [ sec : lhc_pheno]-[sec : lhc_numerics ] , we are going to discuss the consequences for susy searches at the lhc of the spectrum discussed above , including all constraints listed here . but first , let us briefly discuss the possible impact of direct and indirect dm searches on the light neutralino parameter space . several works have been recently dedicated to light neutralinos ( see e.g. @xcite ) , aiming at a possible explanation of the annual modulation signal at dama / libra @xcite and also reported by cogent @xcite and the excess of nuclear recoil events observed by cogent itself and cresst @xcite . recently , three signal events have been also reported by cdms @xcite . although the situation is not conclusive yet ( the above claims are not consistent with the xenon100 results @xcite and do not even seem to be perfectly compatible with each other @xcite ) , the above results might be broadly accounted for by a light wimp with a mass in the range of @xmath111 gev and a spin - independent ( si ) elastic dm - nucleon cross - section , @xmath112 , of the order of @xmath113 to @xmath114 @xmath21 . as previously mentioned , it is not possible to realize the above configuration within the mssm @xcite : in fact , such a relatively large spin - independent cross - sections require a light extended higgs sector , which is now excluded for moderate to large @xmath96 by ( i ) searches at the lhc for extra higgs bosons decaying into pairs of taus @xcite , and ( ii ) the recent observation of the decay @xmath115 at lhcb @xcite with a branching fraction compatible with the sm prediction @xcite . direct detection experiments might still set relevant constraints on the region we are exploring , as there is an irreducible contribution to @xmath112 mediated by the exchange of the light higgs @xmath116 , whose coupling with the lightest neutralino is set by the relic density bound . other contributions , such as the squark - mediated ones , are model - dependent as they are controlled by parameters that are not constrained by @xmath117 . the calculation of the scattering cross - section is still affected by a residual model dependence , coming from the fact that the decoupling of the contribution mediated by the heavy cp - even higgs @xmath118 is very slow : in fact both @xmath116 and @xmath118 contributions are enhanced by a small value of @xmath37 ( that implies a large higgsino component in @xmath73 ) , though solely the @xmath118 contribution is enhanced by large @xmath96 . therefore , if one keeps , as in our scan , @xmath119 tev ( which might be a reasonable choice in the light of our susy spectrum ) and large values of @xmath96 ( as required by the relic density constraint ) , the size of the two contributions is comparable , and the si cross - section can be a powerful probe of the corner of the parameter space with small values of @xmath37 . indeed , such a region is difficult to directly probe at the lhc , as we will see in section [ sec : lhc_numerics ] . as an illustration , in the left panel of fig.[fig : dmsearches ] we plot @xmath112 ( weighted by @xmath120 ) versus @xmath107 for @xmath121 tev and the default micromegas value of the quark masses and the hadronic matrix elements . by about a factor of two . ] as we can see , xenon100 already represents a relevant constraint on the light neutralino parameter space and the expected sensitivity of xenon1 t has the potential of completely testing our scenario . we checked , however , that , if we allow @xmath32 to be @xmath111 tev , the @xmath118 contribution would finally decouple and the si cross - section would be decreased by almost a factor of three , so that most of the points would escape from the xenon100 bound , although probably still being within the future xenon1 t sensitivity . . the rescaling factor @xmath120 is taken into account . see the text for the details.[fig : dmsearches ] , title="fig:",scaledwidth=45.0% ] . the rescaling factor @xmath120 is taken into account . see the text for the details.[fig : dmsearches ] , title="fig:",scaledwidth=45.0% ] contrary to @xmath112 , the prediction for the spin - dependent ( sd ) cross - section , @xmath122 , is much more robust , as the relevant dimension-6 operator @xmath123 is in our case mediated by a @xmath35 exchange , which at leading order does not depend on the susy parameters except for the neutralino mixing parameters . bounds on the @xmath122 from direct detection experiments such as picasso @xcite , coupp @xcite , simple @xcite , kims @xcite icecube @xcite , xenon100 @xcite ( see also @xcite ) and the baksan neutrino observatory @xcite are not much constraining yet for light neutralino dm . nevertheless , lhc searches for events with a monojet plus missing transverse momentum @xcite can be translated into bounds on neutralino - nucleon scattering cross - sections @xcite . in the case of the operator given above , the bounds on @xmath122 translated from the monojet searches are much stronger than the ones obtained from underground direct dm detection experiments . as we are going to see in section [ sec : monojet ] , searches for monojet events performed by cms and atlas might set a strong constraint on the small-@xmath37 region of the parameter space . let us now briefly discuss indirect detection limits . our scenario predicts as a unique annihilation channel @xmath124 . the most stringent bound on this channel is provided by the observation of gamma - rays from satellite galaxies performed by the fermi - lat collaboration @xcite . the limit is shown in the right panel of fig . [ fig : dmsearches ] and it is compared with our parameter space prediction for the velocity - averaged annihilation cross - section ( weighted by @xmath125 ) as computed with micromegas . as we can see , the fermi - lat limit slightly increases the lower bound of the neutralino mass to about 13 gev . still , a conservative estimate of about @xmath126 of possible uncertainties affecting the theoretical computation and the experimental limit could easily make such a bound milder . nevertheless , the model predictions lie certainly on the border of the current fermi - lat sensitivity so that the light neutralino scenario might be tested independently in the near future by gamma - ray observations . a detailed discussion of experimental and theoretical uncertainties affecting direct and indirect dm searches is beyond the scope of this paper . what we can conclude from this brief discussion is that direct and indirect dm searches are approaching the sensitivity for testing the light neutralino scenario , so that we can expect this to be achieved in the upcoming years , but at the moment we can conservatively consider most of our parameter space unconstrained by such searches . therefore , we are not going to impose the bounds from xenon100 and fermi - lat further in this study and concentrate on the constraints that can be obtained from the lhc data . as illustrated in the previous sections , light neutralino dm in the mssm is only consistent with wmap and planck observations in a region of the parameter space with peculiar features : a small soft mass for the right - handed stau , a small higgsino mass parameter @xmath37 and moderate to large @xmath96 . this leads us naturally to the following spectrum at the electroweak scale : besides the lightest neutralino that is assumed to have mass @xmath127 gev , there are only the lighter stau ( mainly right - handed ) , the higgsino - dominated neutralinos @xmath103 , and the lighter chargino @xmath52 , which take masses that are not much above @xmath128 gev . more specifically , these states are assumed to be in the range between the lep lower bounds given in eq . ( [ eq : leplimits ] ) and the upper limits given in eq . ( [ eq : up - bounds ] ) , i.e. @xmath129 . all the other states ( in particular the strong - interacting superpartners ) play no role in satisfying the relic density constraint and can be in principle too heavy to be detected by lhc experiments . still , direct electroweak production for @xmath130 , @xmath103 and @xmath52 with masses of @xmath131 gev can be sizeable , with cross - sections up to the pb level for proton - proton collisions at @xmath132 tev @xcite . this represents the only unavoidable contribution to the total susy production in our scenario . in order to test light neutralino dm at the lhc , it is therefore sufficient to consider the particle content described above with the following electroweak drell - yan production modes : @xmath133 where @xmath134 . the decays of these particles clearly depend on the detail of the spectrum . a typical mass spectrum of the light neutralino dark matter scenario . the shown particles masses are mainly controlled by three susy parameters : @xmath57 , @xmath135 , and @xmath37 . all the other fields are assumed to be heavy.,scaledwidth=55.0% ] let us first consider the hierarchy depicted in fig . [ fig : spectrum ] , i.e. , @xmath136 that we typically observe in the parameter region consistent with planck ( cf . the points above the dashed - line in fig . [ fig : stau - chi ] ) . in this case the stau decays with almost 100% probability into a tau lepton and the lsp , @xmath137.\end{aligned}\ ] ] charginos and heavier neutralinos can instead decay first to an on - shell stau , @xmath138 , \\ { \widetilde{\chi}^\pm}_1 & \to \nu_\tau { \widetilde{\tau}}_1^\pm \quad [ { \rm br}\approx 75\% ] , \end{aligned}\ ] ] where in parenthesis we show typical values for the branching fractions for the case @xmath139 , as computed with the susy - hit package @xcite . the other possible channels are @xmath140 for the neutralinos and @xmath141 for the chargino . from the above , we see that the pair production of higgsino - like neutralinos can lead with high probability ( @xmath142 80% ) to a striking 4@xmath143 + @xmath97 signal at the lhc from the decay chain : @xmath144 still , event rates for such a signature are suppressed by hadronic tau reconstruction efficiencies , which are typically of the order of @xmath145 for each hadronic tau @xmath146 . similarly , @xmath147 production gives about 70% of times events with 3@xmath143 + @xmath97 . other combinations of production and decay modes result in a smaller number of taus ( e.g. two from @xmath148 production followed by @xmath149 ) , hence they are , in principle , more difficult to disentangle from the sm background . for instance , production and decays of sm gauge bosons and @xmath150 can copiously give signatures like 2@xmath143 @xmath151 @xmath97 . nevertheless , the @xmath152 cut , we are going to discuss in the next section , turns out to be very efficient in discriminating between signal and background also in this case . moreover , we expect a very large number of such events as several combinations of production and decay modes contribute to this category . thus , 2@xmath143 @xmath151 @xmath97 events also have an important role in testing our scenario . let us now consider the case that neutralinos can not decay to a real stau : @xmath153 such hierarchies are consistent with the relic density bound in a corner of the parameter space , as shown with the points below the dashed - line in fig . [ fig : stau - chi ] . the 3-body decays @xmath154 through an off - shell @xmath130 can still give multi - tau events . in this case , the exact branching ratios depend on the masses of the other sfermions that can mediate 3-body decays at comparable rates even if much heavier than @xmath130 . we checked that even for squarks and sleptons above the tev scale , the branching ratio br(@xmath155 ) does typically not exceed the @xmath156 level . furthermore , the decay of @xmath103 into @xmath157 will dominate , if kinematically allowed . this further suppresses a possible multi - tau signal . however , corresponding parameter regions are partly already covered by the limit stated in eq . ( [ eq : opal ] ) . the remaining corner of this parameter region at very small @xmath37 should be much more difficult to directly probe at the lhc . nevertheless , other lhc observables already disfavor this scenario , as we illustrate in the following . ) versus the mass ratio @xmath159 , all the constraints of section [ sec : relicdensity ] are applied . right : different values for br(@xmath81 ) displayed on the same @xmath160 plane of fig . [ fig : stau - chi ] . [ fig : hchichi ] , title="fig:",scaledwidth=45.0% ] ) versus the mass ratio @xmath159 , all the constraints of section [ sec : relicdensity ] are applied . right : different values for br(@xmath81 ) displayed on the same @xmath160 plane of fig . [ fig : stau - chi ] . [ fig : hchichi ] , title="fig:",scaledwidth=45.0% ] as discussed in section [ sec : parspace ] , relic density constraints require sizeable higgsino components in @xmath73 that can induce large branching ratios for the decay @xmath161 @xcite . this is in particular the case for heavier @xmath130 and thus for the hierarchy @xmath162 . here , fulfilling the relic density bound requires a larger higgsino component ( small @xmath37 ) to contribute to @xmath73 annihilation . the light higgs decay width into the lightest neutralinos is given by @xcite : @xmath163 where in the decoupling regime @xmath164 @xmath165 thus , we see that , in contrast to @xmath166 , the contribution from the @xmath167 component is dominant as it is @xmath96 enhanced with respect to the @xmath168 one . in fig . [ fig : hchichi ] ( left ) we plot br(@xmath81 ) versus @xmath159 using the parameter scan illustrated in section [ sec : parspace ] . br(@xmath81 ) has been computed using susy - hit @xcite . we find that amost all the allowed points with @xmath169 ( i.e. the points below the dashed - line in the right plot of fig . [ fig : hchichi ] ) correspond to a branching ratio larger than 30% . this is to be compared to the constraints from fits of @xmath170 to the observed higgs decay rates @xcite : @xmath171 from this we see that the case @xmath172 is strongly disfavored , while @xmath173 is still viable but partly constrained , as shown in fig . [ fig : hchichi ] ( right ) .. ] ( right ) , compared to the current bound from monojet searches at the lhc.[fig : monojet ] , title="fig:",scaledwidth=45.0% ] ( right ) , compared to the current bound from monojet searches at the lhc.[fig : monojet ] , title="fig:",scaledwidth=45.0% ] as anticipated in section [ sec : dmsearches ] , direct dm searches at the lhc , based on monojet + @xmath174 events , can be displayed as limits on the neutralino scattering cross - section with nuclei . as we discussed above , the prediction for the spin - dependent cross - section , @xmath122 , is not too sensitive to the susy spectrum under consideration . indeed , the effective proton - neutralino interaction is mediated by a @xmath35 boson and hence is solely determined by the size of the higgsino components in @xmath73 . we expect that the larger the components are corresponding to small values of @xmath37 , the stronger the bound from lhc searches becomes . we computed @xmath175 by means of micromegas , checking that variations of the hadronic matrix elements affect the results only at the level of @xmath176 % .in the interpretation of the atlas @xcite and cms @xcite analyses the latter giving a slightly stronger limit in terms of @xmath122 , we take into account the fact that neutralinos are majorana particles and thus the published limits are in our case weaker by a phase - space factor of two @xcite . the result is shown in fig . [ fig : monojet ] , from which we see that the parameter space is considerably constrained by the monojet data . in particular , the lower bound of the neutralino mass ( left panel ) is raised to @xmath177 and points with low values of the higgsino mass namely @xmath110 gev are excluded ( right panel ) . as in the case of the invisible higgs decay discussed above , this observable is therefore complementary to the multi - tau searches in testing the light neutralino scenario : in fact , the parameter region with @xmath172 , which is kinematically unfavorable for the multi - tau searches , requires @xmath178 gev ( cf . [ fig : stau - chi ] ) . from the comparison between fig . [ fig : stau - chi ] and the left panel of fig . [ fig : monojet ] , it is reasonable to conclude that the null result of the monojet searches at the lhc would not be compatible with the parameter region of the light neutralino dark matter scenario with @xmath172 . however , for a conclusive statement of the potential of the lhc monojet searches on direct neutralino production and its relation with direct dark matter detection limits , a detailed study beyond the effective field theory approximation carried out here is necessary . recently the atlas collaboration presented a ( preliminary ) analysis of a search for new physics in a final state with multi - taus and large missing transverse energy @xcite employing a data sample of @xmath179 at @xmath10 tev . in ref . @xcite at least two reconstructed hadronic taus are required together with a missing transverse energy of @xmath180 gev ; the final event selection and background suppression are based on a cut on @xmath152 @xcite and different jet vetoes . resulting limits are presented in different simplified models and a parameter region of the `` phenomenological mssm '' ( pmssm ) with light charginos , neutralinos and staus . the presented analysis is inclusive in the number of reconstructed hadronic taus and relevant for the light neutralino scenario investigated in this study . in the following we reproduce the pmssm limits presented in ref . @xcite and reinterpret them in the light neutralino parameter space discussed above . for the signal event simulation we use @xcite and include production of neutralinos , charginos and sleptons . for any scenario we consider , squarks and gluinos are assumed to be heavy and their productions do not contribute . everywhere full spin correlations in the decays , initial - state - radiation ( isr ) , final - state - radiation ( fsr ) , hadronization effects , and underlying - event - simulation are included . obtained event samples are normalized to inclusive nlo cross - sections calculated with @xcite . via the format @xcite signal events are passed to @xcite for ( fast ) detector simulation and event reconstruction . we use the default atlas detector card within and tune all efficiencies to the values given in ref . signal events are reconstructed and selected , as in ref . @xcite , according to the following criteria . light central jets are required to have transverse momentum @xmath181 and pseudorapidity @xmath182 . forward jets must satisfy @xmath183 and @xmath184 . for tagged b - jets we require @xmath185 and @xmath186 . for electrons and muons we require @xmath187 , and @xmath188 and @xmath189 , respectively . hadronic tau candidates are required to have @xmath190 and @xmath182 . now we select events with at least two opposite sign ( os ) taus . additional light leptons are vetoed and we require @xmath191 . the variable @xmath152 is calculated using @xcite and the assumed mass of the neutralino is set to @xmath192 . in events where more than two taus are reconstructed , @xmath152 is computed taking all possible os tau pairs into account and then choosing the largest value . for the final event selection two signal regions are defined as in ref . @xcite : * sr1 : veto of central light jets and forward jets , and a cut on @xmath193 , * sr2 : veto of b - jets , and a cut on @xmath194 . in these signal regions the analysis presented in ref . @xcite sets at @xmath195 cl the following limits on the number of signal events : @xmath196 these limits are then interpreted in an @xmath197 plane of the pmssm with @xmath198 , @xmath199 and @xmath200 ( together with @xmath201 this yields a lighter stau mass of @xmath202 ) . all other parameters are decoupled . parameter plane of the pmssm . the bounds obtained by @xcite and the limits reproduced by our simulation are shown.[fig : atlas_exclusion ] ] with our monte carlo framework we perform a scan in the very same pmssm plane and compare the resulting number of signal events with the above stated limits at @xmath195 cl . the result is shown in fig . [ fig : atlas_exclusion ] . besides the exclusions obtained with our framework we show the expected and observed exclusions of ref . the agreement between the reproduced exclusions and the ones given in ref . @xcite seems to be sufficient in order to allow a reinterpretation of the underlying limits in the light neutralino parameter space . parameter plane . the light neutralino parameter space is also shown : the gray points fulfill the constraints of section [ sec : relicdensity ] , the orange points additionally give @xmath203 , see section [ sec : invh].[fig : our_exclusion ] ] plane : the gray points fulfill the constraints of section [ sec : relicdensity ] , the orange points correspond to @xmath203 , the dark - green points evade the atlas multi - tau limit of fig . [ fig : our_exclusion].[fig : our_exclusion2 ] ] as discussed in detail in section [ sec : lhc_pheno ] , we expect striking multi - tau signals for the considered light neutralino parameter space , and consequently possibly strong constraints on this parameter space from the atlas analysis in ref . @xcite . in fig . [ fig : our_exclusion ] we reinterpret the limits of @xcite in the @xmath204 plane discussed above . in this reinterpretation , we set @xmath205 , @xmath206 , and as before all other mass parameters with the exception of the relevant stau and higgsino masses are assumed to be heavy . we checked that limits for an even lighter neutralino are at least as strong as the obtained ones . moreover , the limits obtained for our simplified model do hardly depend on the choice of @xmath96 . from the figure , we see that parameter regions with @xmath207 , where the heavier neutralinos can decay into on - shell lighter staus are excluded up to @xmath208 . for @xmath209 decays of the heavier neutralinos into taus are only possible via off - shell decays , where various decay modes compete and corresponding exclusions limits are much weaker . however , as discussed in section [ sec : invh ] and [ sec : monojet ] , such parameter regions are strongly disfavored by recent limits on the invisible width of the light higgs , @xmath116 , and potentially by monojet searches at the lhc . in fig . [ fig : our_exclusion ] the orange points give @xmath203 ( cf . the right panel of fig . [ fig : stau - chi ] ) , while the gray points fulfill the relic density constraints ( and the other constrains introduced in section [ sec : relicdensity ] ) . additionally , in the region where @xmath210 limits from dedicated searches in final states with sm gauge bosons and missing transverse energy could become relevant @xcite . still , in the light neutralino scenario this is a very small parameter region . multi - tau searches with higher luminosity and center - of - mass energy @xmath211 , in combination with monojet searches and higgs decay measurements , are the most promising way to test light neutralino dm up to @xmath212 gev . indeed , only a tiny corner of the parameter space is left unprobed by the different experimental information discussed above : this is better depicted in fig . [ fig : our_exclusion2 ] , where we show the impact of the atlas limit on the @xmath213 plane . as in fig . [ fig : our_exclusion ] , the gray points fulfill the constraints discussed in section [ sec : relicdensity ] and the orange points additionally correspond to @xmath203 . the dark - green points are the only ones left after the atlas multi - tau search . this allows us to set the present lower bound on the neutralino mass assuming eqs . ( [ eq : planck ] ) and ( [ eq : inv ] ) at about @xmath214 in addition , we see that the few points left require a very light stau with @xmath215 gev , a value very close to the lep exclusion limit reported in ref . in this paper , we have shown that , assuming a light neutralino as dm candidate and the particle content of the mssm , the bounds from the dm relic abundance can be fulfilled only in a limited and well - defined region of the parameter space . for a mass of the lightest neutralino smaller than about 30 gev a handful of parameters suffices to define this parameter space . in turn , this allows to employ current searches for susy at the lhc to set definite limits on light neutralino dm and give a lower bound on its mass . the allowed region of the parameter space is characterized by a relatively light right - handed stau , a light higgsino - dominated chargino and light higgsino - dominated neutralinos , cf . ( [ eq : up - bounds ] ) . therefore , the recent search for multi - tau @xmath151 @xmath174 events performed by the atlas collaboration @xcite can be interpreted as a test of the light neutralino dm scenario . by means of a monte carlo simulation including fast detector simulation , we have shown how this atlas analysis strongly constrains the relevant parameter space . we have also highlighted the complementarity of such tests with other new physics observables at the lhc like the search for the decay @xmath216 , or searches in the monojet @xmath151 @xmath174 channel . in combination with this experimental information , the multi - tau atlas search excludes most of the parameter space with @xmath217 gev , setting a lower bound on the neutralino dm mass : @xmath218 . the plots in figs . [ fig : our_exclusion ] and [ fig : our_exclusion2 ] present our final results and show how tiny the remaining parameter space is . clearly , a small increase of the sensitivity in this channel at the future @xmath219 tev run of the lhc can completely test the light neutralino dm scenario up to @xmath220 gev . larger values of the lightest neutralino mass can not be probed in such a unique way : in fact , as we argued in section [ sec : lightdm ] , @xmath221 gev would open the possibility of satisfying the relic density constraints with compressed spectra that can , at the same time , ( i ) evade the lep searches for light sfermions , ( ii ) be insensitive to constraints from @xmath35-pole observables , ( iii ) be very hard to be tested at the lhc . furthermore , for even larger @xmath107 , a very efficient neutralino annihilation would be possible through resonant @xmath35 and/or @xmath116 exchange that would just require non - vanishing higgsino components in @xmath73 . a detailed discussion of such possibilities is beyond the scope of this paper and will be given elsewhere . a crucial assumption leading to the above results is that the lightest neutralino constitutes ( mainly ) the observed dm in the universe . dropping such a hypothesis , as for instance in models with ( small ) r - parity violation , would clearly change completely the discussion on how light the neutralino can be . corresponding phenomenology at colliders could be very different , e.g. in the case of a promptly decaying neutralino . for early discussions on ( very ) light neutralinos without cosmological bounds , we refer to refs . @xcite and references therein . concerning direct underground dm searches , we have shown in section [ sec : dmsearches ] that the scattering cross - sections with nuclei predicted in the present scenario are in the range @xmath222 . this implies that ( i ) light mssm neutralinos can not account for the signals reported by several direct detection experiments , ( ii ) the scenario we discussed has cross - sections close to the current xenon100 sensitivity and will be fully tested independently by xenon1 t ( cf . the left panel of fig . [ fig : dmsearches ] ) . similar conclusions can be drawn for indirect searches : the light neutralino scenario lies close to the border of the current fermi - lat sensitivity . thus , it might be complementary tested in the near future also by gamma - ray observations ( cf . [ fig : dmsearches ] , right ) . to conclude , let us remark that the presented study provides an example of the amazing capability of the lhc experiments to test new physics through pure electroweak interactions , as well as of the complementarity of different collider searches and other experimental information ( cmb observations , direct / indirect dm searches , etc . ) in shedding light on nature and properties of dark matter and physics beyond the standard model in general . we thank wolfgang hollik for providing us with the numerical implementation of ref . @xcite . we would also like to thank federica legger , borge gjelsten and xuai zhuang for explanation of various details of ref . furthermore , we acknowledge helpful comments by herbi dreiner on the development of the light neutralino scenario and discussions with michael gustafsson about indirect dm searches . to thanks carlos yaguna for his insightful comments on the invisible higgs decay . lc , to , and yt are grateful to the max - planck - institut fr physik for the support and hospitality during their visit . 99 atlas collaboration , atlas - conf-2012 - 109 ; s. chatrchyan _ et al . _ [ cms collaboration ] , arxiv:1301.2175 [ hep - ex ] . atlas collaboration , atlas - conf-2013 - 024 ; atlas - conf-2012 - 165 ; s. chatrchyan _ et al . _ [ cms collaboration ] , arxiv:1303.2985 [ hep - ex ] . cms collaboration , cms - pas - sus-12 - 022 . atlas collaboration , atlas - conf-2013 - 035 . d. hooper and t. plehn , phys . b * 562 * ( 2003 ) 18 [ hep - ph/0212226 ] . g. blanger , f. boudjema , a. pukhov and s. rosier - lees , hep - ph/0212227 . a. bottino , n. fornengo and s. scopel , phys . d * 67 * ( 2003 ) 063519 [ hep - ph/0212379 ] . atlas collaboration , atlas - conf-2013 - 028 . cms collaboration , cms - pas - sus-13 - 006 . r. bernabei _ et al . _ [ dama collaboration ] , eur . j. * c56 * ( 2008 ) 333 [ arxiv:0804.2741 [ astro - ph ] ] . r. bernabei , p. belli , f. cappella , r. cerulli , c. j. dai , a. dangelo , h. l. he , a. incicchitti _ et al . _ , eur . j. * c67 * ( 2010 ) 39 [ arxiv:1002.1028 [ astro-ph.ga ] ] . c. e. aalseth _ [ cogent collaboration ] , arxiv:1002.4703 [ astro-ph.co ] . g. angloher , m. bauer , i. bavykina , a. bento , c. bucci , c. ciemniak , g. deuter and f. von feilitzsch _ et al . _ , j. c * 72 * ( 2012 ) 1971 [ arxiv:1109.0702 [ astro-ph.co ] ] . r. agnese _ et al . _ [ cdms collaboration ] , arxiv:1304.4279 [ hep - ex ] . d. feldman , z. liu and p. nath , phys . d * 81 * ( 2010 ) 117701 [ arxiv:1003.0437 [ hep - ph ] ] . d. a. vsquez , g. blanger , c. bhm , a. pukhov and j. silk , phys . d * 82 * ( 2010 ) 115027 [ arxiv:1009.4380 [ hep - ph ] ] . l. calibbi , t. ota and y. takanishi , jhep * 1107 * ( 2011 ) 013 [ arxiv:1104.1134 [ hep - ph ] ] . l. calibbi , t. ota and y. takanishi , j. phys . * 375 * ( 2012 ) 012041 [ arxiv:1112.0219 [ hep - ph ] ] . e. aprile _ et al . _ [ xenon100 collaboration ] , phys . * 109 * ( 2012 ) 181301 [ arxiv:1207.5988 [ astro-ph.co ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , phys . * 110 * ( 2013 ) 021801 [ arxiv:1211.2674 [ hep - ex ] ] . m. carena , s. gori , n. r. shah , c. e. m. wagner and l .- t . wang , jhep * 1207 * ( 2012 ) 175 [ arxiv:1205.5842 [ hep - ph ] ] . m. carena , s. gori , n. r. shah and c. e. m. wagner , jhep * 1203 * ( 2012 ) 014 [ arxiv:1112.3336 [ hep - ph ] ] . d. albornoz vsquez , g. blanger and c. bhm , phys . d * 84 * ( 2011 ) 095015 [ arxiv:1108.1338 [ hep - ph ] ] . g. blanger , f. boudjema , a. pukhov , a. semenov , comput . commun . * 176 * ( 2007 ) 367 [ hep - ph/0607059 ] ; g. blanger , f. boudjema , p. brun , a. pukhov , s. rosier - lees , p. salati , a. semenov , comput . commun . * 182 * ( 2011 ) 842 [ arxiv:1004.1092 [ hep - ph ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5076 [ astro-ph.co ] . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. c * 35 * ( 2004 ) 1 [ hep - ex/0401026 ] . j. r. ellis , j. m. frre , j. s. hagelin , g. l. kane and s. t. petcov , phys . b * 132 * ( 1983 ) 436 . s. schael _ et al . _ [ aleph and delphi and l3 and opal and sld and lep electroweak working group and sld electroweak group and sld heavy flavour group collaborations ] , phys . * 427 * ( 2006 ) 257 [ hep - ex/0509008 ] . h. k. dreiner , j. s. kim and o. lebedev , phys . b * 715 * ( 2012 ) 199 [ arxiv:1206.3096 [ hep - ph ] ] . r. barbieri , g. gamberini , g. f. giudice and g. ridolfi , phys . b * 195 * ( 1987 ) 500 . s. heinemeyer , w. hollik , a. m. weber and g. weiglein , jhep * 0804 * , 039 ( 2008 ) [ arxiv:0710.2972 [ hep - ph ] ] . a. j. buras , p. h. chankowski , j. rosiek and . sawianowska , nucl . phys . b * 659 * ( 2003 ) 3 [ hep - ph/0210145 ] . g. isidori and p. paradisi , phys . b * 639 * ( 2006 ) 499 [ hep - ph/0605012 ] . c. bhm , p. s. b. dev , a. mazumdar and e. pukartas , jhep * 1306 * ( 2013 ) 113 [ arxiv:1303.5386 [ hep - ph ] ] . j. kopp , t. schwetz and j. zupan , jcap * 1002 * ( 2010 ) 014 [ arxiv:0912.4264 [ hep - ph ] ] . t. schwetz and j. zupan , jcap * 1108 * ( 2011 ) 008 [ arxiv:1106.6241 [ hep - ph ] ] . m. t. frandsen , f. kahlhoefer , c. mccabe , s. sarkar and k. schmidt - hoberg , arxiv:1304.6066 [ hep - ph ] . a. j. buras , j. girrbach , d. guadagnoli and g. isidori , eur . j. c * 72 * ( 2012 ) 2172 [ arxiv:1208.0934 [ hep - ph ] ] . s. archambault _ et al . _ [ picasso collaboration ] , phys . b * 711 * ( 2012 ) 153 [ arxiv:1202.1240 [ hep - ex ] ] . e. behnke _ et al . _ [ coupp collaboration ] , phys . rev . d * 86 * ( 2012 ) 052001 [ arxiv:1204.3094 [ astro-ph.co ] ] . m. felizardo , t. a. girard , t. morlat , a. c. fernandes , a. r. ramos , j. g. marques , a. kling and j. puibasset _ et al . _ , phys . * 108 * ( 2012 ) 201302 [ arxiv:1106.3014 [ astro-ph.co ] ] . s. c. kim , h. bhang , j. h. choi , w. g. kang , b. h. kim , h. j. kim , k. w. kim and s. k. kim _ et al . _ , * 108 * ( 2012 ) 181301 [ arxiv:1204.2646 [ astro-ph.co ] ] . m. g. aartsen _ et al . _ [ icecube collaboration ] , arxiv:1212.4097 [ astro-ph.he ] . e. aprile _ et al . _ [ xenon100 collaboration ] , arxiv:1301.6620 [ astro-ph.co ] . m. garny , a. ibarra , m. pato and s. vogl , phys . d * 87 * ( 2013 ) 056002 [ arxiv:1211.4573 [ hep - ph ] ] . m. m. boliev , s. v. demidov , s. p. mikheyev and o. v. suvorova , arxiv:1301.1138 [ astro-ph.he ] . g. aad _ et al . _ [ atlas collaboration ] , jhep * 1304 * ( 2013 ) 075 [ arxiv:1210.4491 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , jhep * 1209 * ( 2012 ) 094 [ arxiv:1206.5663 [ hep - ex ] ] . a. birkedal , k. matchev and m. perelstein , phys . rev . d * 70 * ( 2004 ) 077701 [ hep - ph/0403004 ] . m. beltran , d. hooper , e. w. kolb , z. a. c. krusberg and t. m. p. tait , jhep * 1009 * ( 2010 ) 037 [ arxiv:1002.4137 [ hep - ph ] ] . a. rajaraman , w. shepherd , t. m. p. tait and a. m. wijangco , phys . d * 84 * ( 2011 ) 095013 [ arxiv:1108.1196 [ hep - ph ] ] . giardino , k. kannike , i. masina , m. raidal and a. strumia , arxiv:1303.3570 [ hep - ph ] . a. falkowski , f. riva and a. urbano , arxiv:1303.1812 [ hep - ph ] . a. djouadi and g. moreau , arxiv:1303.6591 [ hep - ph ] . j. goodman , m. ibe , a. rajaraman , w. shepherd , t. m. p. tait and h .- b . yu , phys . b * 695 * ( 2011 ) 185 [ arxiv:1005.1286 [ hep - ph ] ] . j. goodman , m. ibe , a. rajaraman , w. shepherd , t. m. p. tait and h .- b . yu , phys . d * 82 * ( 2010 ) 116010 [ arxiv:1008.1783 [ hep - ph ] ] . o. buchmueller , m. j. dolan and c. mccabe , arxiv:1308.6799 [ hep - ph ] . w. beenakker , m. klasen , m. krmer , t. plehn , m. spira and p. m. zerwas , phys . lett . * 83 * ( 1999 ) 3780 [ erratum - ibid . * 100 * ( 2008 ) 029901 ] [ hep - ph/9906298 ] . m. dobbs and j. b. hansen , comput . commun . * 134 * ( 2001 ) 41 . s. ovyn , x. rouby and v. lematre , arxiv:0903.2225 [ hep - ph ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 718 * ( 2013 ) 841 [ arxiv:1208.3144 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , jhep * 1211 * ( 2012 ) 147 [ arxiv:1209.6620 [ hep - ex ] ] . h. k. dreiner , s. heinemeyer , o. kittel , u. langenfeld , a. m. weber and g. weiglein , eur . j. c * 62 * ( 2009 ) 547 [ arxiv:0901.3485 [ hep - ph ] ] .
we present an overview on some basic properties of massive neutrinos and focus on their flavor issues , including the mass spectrum , flavor mixing pattern and cp violation . the lepton flavor structures are explored by taking account of the observed value of the smallest neutrino mixing angle . the impact of on the running behaviors of other flavor mixing parameters is discussed in some detail . the seesaw - induced enhancement of the electromagnetic dipole moments for three majorana neutrinos is also discussed in a tev seesaw scenario .
1211.4331
it is well known that two important experimental results were in the news in 2012 : * on 8 march 2012 , the daya bay collaboration announced a @xmath1 discovery of @xmath2 for this smallest neutrino mixing angle @xcite , @xmath3 which is equivalent to @xmath4 . the convincing daya bay result puts the preliminary results of t2k @xcite , minos @xcite and double chooz @xcite , which all hinted at @xmath2 in 2011 , on solid ground . in particular , the fact that @xmath0 is not strongly suppressed is a good news to the experimental attempts towards a measurement of cp violation in the lepton sector . * on 4 july 2012 , the atlas @xcite and cms @xcite collaborations at the large hadron collider ( lhc ) independently announced the discovery of a higgs - like boson at the mass scale of 125 gev to 127 gev . if this result turns out to be true , it will have an important impact on the development of neutrino physics because most of the neutrino mass models depend on the existence of the higgs particle(s ) and yukawa interactions . therefore , a brief overview of where we are standing and where we are expecting to go makes sense . the remaining parts of this review paper are organized as follows . in section 2 we give a fast overview of some fundamental neutrino properties , such as the speed of neutrinos , the nature of massive neutrinos and the number of neutrino species . section 3 is devoted to a brief description of the flavor issues of charged leptons and neutrinos , including the mass spectrum , flavor mixing pattern and cp violation . we compare the observed pattern of quark flavor mixing with that of lepton flavor mixing . in section 4 we go into details of possible lepton flavor structures by outlining two phenomenological strategies and taking a number of typical examples . the impact of large @xmath0 on the running behaviors of other flavor mixing parameters is discussed in section 5 by using the one - loop renormalization - group equations ( rges ) in the framework of the minimal supersymmetric standard model ( mssm ) . section 6 is devoted to the seesaw - enhanced electromagnetic dipole moments of three majorana neutrinos based on a tev seesaw scenario . a summary and some concluding remarks are given in section 7 . the constancy of the speed of light @xmath5 in vacuum and the independence of physical laws from the choice of inertial systems are two fundamental propositions of the special relativity ( sr ) @xcite . if our world is lorentz invariant , a free particle s energy @xmath6 , momentum @xmath7 and rest mass @xmath8 satisfy the relationship @xmath9 . the velocity of this particle turns out to be @xmath10 , implying that it can not travel faster than light in vacuum . could a particle be superluminal ? the answer would be yes if the particle had an imaginary mass ( called a `` tachyon '' @xcite ) or if the lorentz invariance were broken . the opera collaboration claimed a convincing " measurement of the superluminal neutrinos in september 2011 @xcite . but five months later this story ended up with a mistake of the bad connection of the optical fiber . the opera paper was updated in july 2012 by including the new sources of errors , and the new result was in agreement with the sr . here let us quote steven weinberg s comments on the original result of the opera experiment : `` the report of this experiment is pretty impressive , but it bothers me that there is plenty of evidence that all sorts of other particles never travel faster than light , while observations of neutrinos are exceptionally difficult . it is as if someone said that there are fairies in the bottom of their garden , but they can only be seen on dark , foggy nights . '' an early measurement of the neutrino speed was done by using the pulsed pion beams ( produced by the pulsed proton beams hitting a target ) at the fermilab in the 1970s @xcite . in this experiment the speed of muons was compared with that of neutrinos and antineutrinos . the same measurement was repeated in 2007 by using the minos detector @xcite . in 2011 the speed of neutrinos was also measured in a few other long - baseline neutrino experiments , such as the icarus @xcite , borexino @xcite and lvd @xcite experiments . but the most stringent constraint on the speed of neutrinos was from the observational data of the supernova 1987a @xcite : @xmath11 obtained by comparing the arrival time of light with that of neutrinos . the neutrinos are massless in the standard model ( sm ) as a result of its simple structure and renormalizability . on the one hand , the sm does not contain any right - handed neutrinos , and thus there is no way to write out the dirac neutrino mass term . on the other hand , the sm conserves the @xmath12 gauge symmetry and only contains the higgs doublet , and thus the majorana mass term is forbidden . although the sm accidently possesses the @xmath13 symmetry and naturally allows neutrinos to be massless , the vanishing of neutrino masses in the sm is not guaranteed by any fundamental symmetry or conservation law . today we have achieved a lot of robust evidence for neutrino oscillations from solar , atmospheric , reactor and accelerator neutrino experiments . the phenomenon of neutrino oscillations implies that at least two of the three neutrinos must be massive and the lepton flavors must be mixed . this is the first convincing evidence for new physics beyond the sm . 1 is a schematic plot of the mass spectrum of the sm leptons and quarks at the electroweak scale . one can see that the span between @xmath14 and @xmath15 is at least twelve orders of magnitude . furthermore , there exists an obvious `` desert '' spanning six orders of magnitude between the neutrino masses and the masses of the charged fermions . why do the sm fermions have such hierarchy and desert puzzles ? the answer to this important question remains open . in particular , the tiny neutrino masses must have a peculiar origin ( e.g. , via the seesaw mechanisms @xcite ) . moreover , there might exist one or more kev sterile neutrinos in the desert as a natural candidate for warm dark matter @xcite . a _ pure _ dirac mass term added into the sm is in general disfavored , unless the theory is built by introducing extra dimensions . such a mass term in a renormalizable model of electroweak interactions would worsen the fermion mass hierarchy problem . an effective majorana mass term given by the right - handed neutrinos and their charge - conjugated counterparts is not forbidden by the sm gauge symmetry , unless the contrived assumption of lepton number conservation is imposed on the theory . hence most theorists believe that massive neutrinos are more likely to be the majorana particles and their salient feature is lepton number violation . the unique window to verify the majorana nature of massive neutrinos is to observe the neutrinoless doube - beta ( @xmath16 ) decay . so far we have not obtained very convincing evidence for this lepton - number - violating process . even if the @xmath16 decay were never observed , one would still be unable to conclude that massive neutrinos are the dirac particles @xcite . the effective mass of the @xmath16 decay could vanish if the majorana cp - violating phases lie in some specific regions . on the other hand , there are some other mechanisms which can lead to the @xmath16 decay . such new physics effects could be of the same order as or even larger than the standard light - neutrino - exchange effect @xcite . given the sm interactions , a massive dirac neutrino can have a tiny ( one - loop ) magnetic dipole moment @xmath17 , where @xmath18 is the bohr magneton @xcite . in contrast , a massive majorana neutrino can not have magnetic and electric dipole moments , because its antiparticle is just itself . both dirac and majorana neutrinos can have _ transition _ dipole moments ( of a size comparable with @xmath19 ) @xcite , which may give rise to neutrino decays , scattering effects with electrons , interactions with external magnetic fields ( red - giant stars , the sun , supernovae , and so on ) , and contributions to neutrino masses . current experimental bounds on the neutrino dipole moments are at the level of @xmath20 . it is well known that three " is a mystically popular number in particle physics , such as three @xmath21 quarks , three @xmath22 quarks , three @xmath23 leptons , three @xmath24 neutrinos , three colors and three forces in the sm . in this case , why do people want to go beyond @xmath25 ? the study of light sterile neutrinos has become a popular direction in neutrino physics @xcite . one is motivated to consider such exotic " particles for several reasons . on the theoretical side , the type - i seesaw mechanism @xcite provides a very elegant interpretation of the small masses of @xmath26 ( for @xmath27 ) with the help of two or three heavy sterile neutrinos , and the latter can even help account for the observed matter - antimatter asymmetry of the universe via the leptogenesis mechanism @xcite . on the experimental side , the lsnd @xcite , miniboone @xcite and reactor @xcite antineutrino anomalies can all be explained as the active - sterile antineutrino oscillations in the assumption of one or two species of sterile antineutrinos whose masses are below 1 ev @xcite . furthermore , a careful analysis of the existing data on the big bang nucleosynthesis @xcite or the cosmic microwave background anisotropy , galaxy clustering and supernovae ia @xcite seems to favor at least one species of sterile neutrinos at the sub - ev mass scale . on the other hand , sufficiently long - lived sterile neutrinos in the kev mass range might serve for a good candidate for warm dark matter if they were present in the early universe @xcite . if the three known neutrinos have mixing with a few new degrees of freedom above or far above the fermi scale , an exciting window will be open to new physics at high energy scales . in this case , however , the mixing between light and heavy neutrinos violates the unitarity of the @xmath28 light neutrino mixing matrix and might result in some observable effects in the future precision neutrino experiments @xcite . there are three central concepts in flavor physics : mass , flavor mixing and cp violation @xcite . the phenomenon of lepton flavor mixing at low energies is effectively described by a @xmath28 matrix @xmath29 , the so - called maki - nakagawa - sakata - pontecorvo ( mnsp ) matrix @xcite . given the unitarity of @xmath29 , it can be parametrized in terms of three angles and three phases @xcite : @xmath30 where @xmath31 , @xmath32 ( for @xmath33 ) , and @xmath34 is physically relevant if massive neutrinos are the majorana particles . fogli _ et al _ @xcite have recently done a global analysis of current neutrino oscillation data and obtained the ranges of two neutrino mass - squared differences ( @xmath35 and @xmath36 ) and three neutrino mixing angles , as listed in table 1 , where nh and ih stand for the normal hierarchy ( @xmath37 ) and the inverted hierarchy ( @xmath38 ) , respectively . the two mass - squared differences of three neutrinos have been determined , to a very good degree of accuracy , from current experimental data : @xmath39 and @xmath40 . the absolute neutrino mass scale remains unknown and may hopefully be determined in the following experimental or observational ways : the single @xmath41 decay , the @xmath16 decay and the cosmological constraints . 2 shows the parameter space of @xmath42 , @xmath43 ( the effective electron neutrino mass in the @xmath41 decay ) and @xmath44 ( the effective mass of the @xmath16 decay ) . a precision measurement of @xmath45 and @xmath46 in the sub - ev range could determine the neutrino mass hierarchy . in the two lower panels of fig 2 there remains a large vertical spread in the allowed slanted bands , as a result of the unknown majorana cp - violating phases in the @xmath47 components . this observation indicates that more precise data in either the @xmath48 plane or the @xmath49 plane might provide some useful constraints on the majorana phases . before the absolute mass scale is determined , there remain two open questions : ( 1 ) is @xmath50 bigger or smaller than @xmath51 ? ( 2 ) can one neutrino mass ( @xmath51 or @xmath50 ) be vanishing or vanishingly small ? the first question awaits an experimental answer in the foreseeable future , such as a long - baseline neutrino oscillation experiment with appreciable terrestrial matter effects @xcite or a long - baseline reactor antineutrino oscillation experiment with accurate information on the energy spectrum @xcite . a theoretical answer to the second question is strongly model - dependent . examples of this type include the minimal type - i seesaw mechanism with two heavy majorana neutrinos @xcite or the friedberg - lee ansatz with an effective dirac or majorana neutrino mass operator @xcite . the fact that the smallest neutrino mixing angle @xmath0 is not strongly suppressed leads us to some new questions about the feature of lepton flavor mixing : ( 1 ) can the relatively large @xmath0 be understood by an underlying flavor symmetry or is it generated by a symmetry breaking mechanism or quantum corrections ? ( 2 ) does @xmath52 still hold ? ( 3 ) what is the strength of leptonic cp violation ? the structure of the mnsp lepton flavor mixing matrix @xmath29 is significantly different from that of the cabibbo - kobayashi - maskawa ( ckm ) quark flavor mixing matrix @xmath53 . the ckm matrix is nearly the unit matrix up to some small corrections , while the mnsp matrix has an approximate @xmath54-@xmath55 symmetry . the full @xmath54-@xmath55 symmetry of @xmath29 in modulus is described by the equalities @xmath56 equivalent to two independent sets of conditions in the standard parametrization given in eq . ( 2 ) @xcite : @xmath57 or @xmath58 if @xmath59 is exactly equal to @xmath60 , then one may arrive at a partial @xmath54-@xmath55 permutation symmetry in the mnsp matrix @xmath29 ( i.e. , the equality @xmath61 ) . now that @xmath62 has firmly been established by the daya bay experiment @xcite , it becomes crucial to check the deviation of @xmath59 from @xmath60 and ( or ) a possible departure of @xmath63 from @xmath64 . we speculate that @xmath29 might have an approximate @xmath54-@xmath55 symmetry with @xmath65 , in contrast with the approximate off - diagonal symmetry of the ckm matrix @xmath53 in modulus ( i.e. , @xmath66 , @xmath67 ) , the majorana neutrino mass matrix of the form @xmath68 predicts the @xmath54-@xmath55 permutation symmetry of the mnsp matrix @xmath29 with @xmath69 and @xmath70 ; while the mass matrix of the form @xmath71 leads us to the @xmath54-@xmath55 symmetry of @xmath29 with @xmath72 and @xmath70 . in either of the above textures of @xmath73 , its entries have certain kinds of linear correlations or equalities and thus can be generated from some underlying flavor symmetries . in view of the experimental evidence for @xmath62 @xcite , the pattern of @xmath73 in eq . ( 6 ) has to be modified . for a similar reason , the more reliable and accurate experimental knowledge on @xmath59 and @xmath63 will be useful for us to identify the effect of @xmath54-@xmath55 symmetry breaking and build more realistic models for lepton mass generation , flavor mixing and cp violation . if neutrinos are the majorana particles , the @xmath28 mnsp matrix @xmath29 contains three cp - violating phases @xmath63 , @xmath74 and @xmath75 . among them , @xmath63 determines the strength of cp and t violation in neutrino oscillations , because both @xmath76 and @xmath77 are proportional to the leptonic jarlskog invariant @xmath78 in vacuum @xcite . the phases @xmath74 and @xmath75 , which have nothing to do with neutrino oscillations , are associated with the @xmath16 decay . note that @xmath63 itself is also of the majorana nature , although it is usually referred to as the dirac phase : one reason is that @xmath63 may appear in other lepton - number - violating processes , even if it can sometimes be arranged _ not _ to appear in the @xmath16 decay ; and the other reason is that @xmath63 , @xmath74 and @xmath75 are actually entangled with one another in the rge running from one energy scale to another . the fact that @xmath0 is not strongly suppressed is certainly a good news to the experimental attempts towards a final measurement of cp violation in the lepton sector . the reason is simply that the strength of leptonic cp violation ( i.e. , @xmath79 ) is proportional to @xmath80 . in the quark sector one has determined the corresponding jarlskog invariant @xmath81 @xcite and attributed its smallness to the strongly suppressed values of quark flavor mixing angles ( i.e. , @xmath82 , @xmath83 and @xmath84 ) . in the lepton sector both @xmath85 and @xmath59 are large , and thus it is possible to achieve a relatively large value of @xmath86 if the cp - violating phase @xmath63 is not small either . taking @xmath87 , @xmath88 and @xmath89 as a realistic example of @xmath29 , we arrive at @xmath90 , implying that the magnitude of leptonic cp violation can actually reach the percent level in neutrino oscillations if @xmath63 is not strongly suppressed . whether cp violation is significant or not turns out to be an important question in lepton physics , especially in neutrino phenomenology . the relative sizes of the nine elements of the mnsp matrix @xmath29 can not be completely fixed unless we have known @xmath91 or @xmath92 as well as the range of @xmath63 . with the help of the available experimental data and the unitarity of @xmath29 , we find @xmath93 where @xmath94 " implies that the relative magnitudes of @xmath95 and @xmath96 ( for @xmath27 ) remain undetermined at present . in comparison , the nine elements of the ckm matrix @xmath53 are known to have the following hierarchy @xcite : @xmath97 there is a striking similarity between the quark and lepton flavor mixing matrices : the smallest elements of both @xmath53 and @xmath29 appear in their respective top - right corners . in the history of flavor physics it took quite a long time to measure the four independent parameters of @xmath53 , but the experimental development had a clear roadmap : @xmath98 namely , the observation of the largest mixing angle @xmath99 was the first step , the determination of the smallest mixing angle @xmath100 was an important turning point , and then the quark flavor physics entered an era of precision measurements in which cp violation could be explored and new physics could be searched for . interestingly and hopefully , the lepton flavor physics is repeating the same story : @xmath101 where @xmath59 is the largest and @xmath0 is the smallest . the observation of @xmath0 in the daya bay experiment is paving the way for future experiments to study leptonic cp violation and to look for possible new physics ( e.g. , whether the @xmath28 mnsp matrix @xmath29 is exactly unitary or not @xcite ) , in particular through the measurements of neutrino oscillations for different sources of neutrino beams . the majorana nature of three massive neutrinos and their other two cp - violating phases ( i.e. , @xmath74 and @xmath75 ) can also be probed in the new era of neutrino physics . the mnsp matrix @xmath29 actually describes a fundamental mismatch between the flavor and mass eigenstates of six leptons , or a mismatch between diagonalizations of the charged - lepton mass matrix @xmath102 and the effective neutrino mass matrix @xmath73 in a given model , no matter whether the origin of neutrino masses is attributed to the seesaw mechanisms or not @xcite . assuming massive neutrinos to be the majorana particles , we may simply write out the leptonic mass terms as @xmath103 where @xmath104 " stands for the flavor eigenstates of charged leptons , @xmath5 " denotes the charge - conjugated neutrino fields , and @xmath73 is symmetric . by using the unitary matrices @xmath105 , @xmath106 and @xmath107 , one can diagonalize @xmath102 and @xmath73 through the transformations @xmath108 and @xmath109 , respectively . then one arrives at the lepton mass terms in terms of the mass eigenstates : @xmath110 extending this basis transformation to the standard charged - current interactions , we immediately obtain @xmath111 in which @xmath112 . the above treatment is most general at a given energy scale ( e.g. , the electroweak scale ) . there are two different strategies of phenomenologically understanding the structure of the mnsp matrix @xcite . \(1 ) _ the mixing angles of @xmath29 are associated with the lepton mass ratios_. the structure of lepton flavor mixing is directly determined by the structures of @xmath105 and @xmath107 . since these two unitary matrices are used to diagonalize @xmath102 and @xmath73 , respectively , their structures are governed by those of @xmath102 and @xmath73 , whose eigenvalues are the physical lepton masses . therefore , we anticipate that the dimensionless flavor mixing angles of @xmath29 should be certain kinds of functions whose variables include four independent mass ratios of three charged leptons and three neutrinos . namely , @xmath113 where the greek subscripts denote the charged leptons , the latin subscripts stand for the neutrinos , and @xmath114 " implies other dimensionless parameters originating from the lepton mass matrices . such an expectation has proved valid in the quark sector to explain why the relation @xmath115 works quite well and how the hierarchical structure of the ckm matrix @xmath53 is related to the strong hierarchies of quark masses ( i.e. , @xmath116 and @xmath117 ) @xcite . as for the phenomenon of lepton flavor mixing , it is apparently difficult to link two large mixing angles @xmath85 and @xmath59 to @xmath118 and @xmath119 @xcite . hence one may consider to ascribe the largeness of @xmath85 and @xmath59 to a weak hierarchy of three neutrino masses , such as the conjecture @xmath120 @xcite . to establish a direct relation between @xmath121 and lepton mass ratios , one has to specify the textures of @xmath102 and @xmath73 by allowing some of their elements to vanish or to be vanishingly small . a typical example of this kind is the fritzsch ansatz @xcite , @xmath122 which is able to account for the present neutrino oscillation data to an acceptable degree of accuracy ( e.g. , @xmath123 ) @xcite . another well - known and viable example is the two - zero textures of @xmath73 in the basis where @xmath102 is diagonal @xcite . note that the texture zeros of a fermion mass matrix dynamically mean that the corresponding matrix elements are sufficiently suppressed as compared with their neighboring counterparts , and they can be derived from a certain flavor symmetry in a given theoretical framework ( e.g. , with the help of the froggatt - nielson mechanism @xcite or discrete flavor symmetries @xcite ) . \(2 ) _ the lepton flavor mixing matrix @xmath29 consists of a constant leading term @xmath124 and a small perturbation term @xmath125_. in fact , @xmath29 has been conjectured to have the following structure for a quite long time @xcite : @xmath126 in which the leading term @xmath124 is a constant matrix responsible for two larger mixing angles @xmath85 and @xmath59 , and the next - to - leading term @xmath125 is a perturbation responsible for both the smallest mixing angle @xmath0 and the dirac cp - violating phase @xmath63 . so far a lot of flavor symmetries have been brought into exercise to derive @xmath124 , while @xmath125 might originate from either an explicit flavor symmetry breaking scenario or some finite quantum corrections at a given energy scale or from a superhigh - energy scale to the electroweak scale . in this case the mnsp matrix @xmath29 is approximately a constant matrix whose mixing angles are independent of the lepton mass ratios . this conjecture is actually in conflict with the conjecture made in the first strategy . the reason for this conflict " is rather simple : the assumed structures of lepton flavor mixing in eqs . ( 15 ) and ( 17 ) correspond to two different structures of lepton mass matrices . as we have pointed out above , the direct dependence of @xmath121 on @xmath127 and @xmath128 is usually a direct consequence of the texture zeros of @xmath102 and ( or ) @xmath73 . in contrast , a constant flavor mixing pattern @xmath124 may arise from some special textures of @xmath102 and ( or ) @xmath73 whose entries have certain kinds of linear correlations or equalities . for instance , the texture @xcite @xmath129 assures @xmath107 to be of the tri - bimaximal mixing pattern to be discussed in section 4.2 . this neutrino mass matrix has no zero entries , but its nine elements satisfy the sum rules @xmath130 and @xmath131 ( for @xmath132 ) . such correlative relations are similar to those texture zeros in the sense that both of them may reduce the number of free parameters associated with lepton mass matrices , making some predictions for the lepton flavor mixing angles technically possible . in short , one may try to understand the structure of the mnsp matrix @xmath29 by following two phenomenological strategies : 1 . to explore possible relations between the flavor mixing angles and the lepton mass ratios ; 2 . to investigate possible constant patterns of lepton flavor mixing as the leading - order effects . the first possibility points to some vanishing ( or vanishingly small ) entries of @xmath102 and @xmath73 , while the second possibility indicates some equalities or linear correlations among the entries of @xmath102 or @xmath73 . in both cases the underlying flavor symmetries play a crucial role in deriving the structures of lepton mass matrices which finally determine the structure of lepton flavor mixing . of course , how to pin down the correct flavor symmetries remains an open question . it is well known that the special textures of @xmath102 and @xmath73 like that in eq . ( 18 ) can easily be derived from certain discrete flavor symmetries ( e.g. , @xmath133 or @xmath134 ) @xcite . that is why eq . ( 17 ) formally summarizes a large class of lepton flavor mixing patterns in which the leading terms are constant matrices originating from some underlying flavor symmetries . the fact that @xmath0 is not very small poses a meaningful question to us today : can this mixing angle naturally be generated from the perturbation matrix @xmath125 ? the answer to this question is certainly dependent upon the form of @xmath124 in the flavor symmetry limit . here we reexamine five typical patterns of @xmath124 in order to get a feeling of the respective structures of @xmath125 which can be constrained by current experimental data on neutrino oscillations . for the sake of simplicity , we typically take @xmath87 , @xmath135 and @xmath136 as our inputs to fix the primary structure of the mnsp matrix @xmath29 . then we have @xmath137 it makes sense to compare a constant mixing pattern @xmath124 with the observed pattern of @xmath29 in eq . ( 19 ) , such that one may estimate the structure of the corresponding perturbation matrix @xmath125 . let us consider five well - known patterns of @xmath124 in the following for illustration . \(1 ) the democratic mixing pattern of lepton flavors @xcite : @xmath138 whose three mixing angles are @xmath139 , @xmath140 and @xmath141 in the standard parametrization as given in eq . ( 2 ) . with the help of eq . ( 19 ) , we immediately obtain the form of @xmath142 as follows : @xmath143 one can see that the magnitude of each matrix element of @xmath125 is of @xmath144 , implying that the realistic pattern of @xmath29 might result from a democratic perturbation to @xmath124 ( i.e. , the nine entries of @xmath125 are all proportional to a common small parameter ) . \(2 ) the bimaximal mixing pattern of lepton flavors @xcite : @xmath145 which has @xmath139 , @xmath146 and @xmath147 in the standard parametrization . comparing eq . ( 22 ) with eq . ( 19 ) , we obtain the perturbation matrix @xmath148 we see that the matrix elements @xmath149 and @xmath150 are highly suppressed . in other words , the initially maximal angle @xmath151 receives the minimal correction , which is much smaller than the one received by the initially minimal angle @xmath152 . such a situation is more or less unnatural , at least from a point of view of model building . \(3 ) the tri - bimaximal mixing pattern of lepton flavors @xcite : @xmath153 whose three mixing angles are @xmath154 , @xmath146 and @xmath155 in the standard parametrization . in a similar way we get the corresponding perturbation matrix @xmath156 it is quite obvious that @xmath157 , @xmath158 , @xmath149 and @xmath159 are highly suppressed . so two initially large angles @xmath160 and @xmath151 are only slightly modified by the perturbation effects , but the initially minimal angle @xmath152 receives the maximal correction . \(4 ) the golden - ratio mixing pattern of lepton flavors @xcite : @xmath161 which has @xmath162 \simeq 31.7^\circ$ ] , @xmath140 and @xmath163 in the standard parametrization . in this case the perturbation matrix @xmath125 turns out to be @xmath164 similar to the tri - bimaximal mixing pattern , two initially large angles of the golden - ratio mixing pattern are only slightly corrected , but the initially minimal angle @xmath152 is significantly modified by the same perturbation . \(5 ) the hexagonal mixing pattern of lepton flavors @xcite : @xmath165 whose mixing angles are @xmath166 , @xmath140 and @xmath155 in the standard parametrization . in this case we obtain the perturbation matrix @xmath167 this result is quite analogous to the one obtained in eq . ( 25 ) or eq . ( 27 ) , simply because the patterns of @xmath124 in these three cases are quite similar . now let us summarize some useful lessons that we can directly learn from the above five typical examples of @xmath29 . * to accommodate the observed value of @xmath0 in a generic flavor mixing structure @xmath168 , one has to choose a proper constant mixing pattern @xmath124 and adjust its perturbation matrix @xmath125 . the phenomenological criterion to do so is two - fold : on the one hand , @xmath124 should easily be derived from a certain flavor symmetry ; on the other hand , @xmath125 should have a natural structure which can easily be accounted for by either the flavor symmetry breaking or quantum corrections ( or both of them ) . * the common feature of the above five patterns of @xmath124 is apparently @xmath169 ( or equivalently , @xmath170 ) , implying that a relatively large perturbation is required for generating @xmath171 . in this case , the closer @xmath160 and @xmath151 are to the observed values of @xmath85 and @xmath59 , the more unnatural the structure of @xmath125 seems to be . the tri - bimaximal mixing pattern given in eq . ( 24 ) , which is currently the most popular ansatz for model building based on certain flavor symmetries , suffers from this unnaturalness in particular @xcite . in this sense we argue that the democratic mixing pattern in eq . ( 29 ) might be more natural and deserve some more attention . * one may certainly consider some possible patterns of @xmath124 which can predict a finite value of @xmath152 in the vicinity of the experimental value of @xmath0 . in this case the three mixing angles of @xmath124 may receive comparably small corrections from the perturbation matrix @xmath125 , and thus the naturalness criterion can be satisfied . for example , the following two patterns of @xmath124 belong to this category and have been discussed in the literature @xcite . one of them is the so - called correlative mixing pattern @xcite @xmath172 with @xmath173 and @xmath174 , which predicts @xmath175 , @xmath147 and @xmath176 . the three mixing angles in this constant scenario satisfy two interesting sum rules : @xmath177 which are geometrically illustrated in fig . + the other pattern of @xmath124 is the tetra - maximal mixing pattern @xcite @xmath178 which predicts @xmath179 , @xmath147 and @xmath180 \simeq 8.4^\circ$ ] . of course , whether such constant mixing patterns can easily be derived from some underlying flavor symmetries remains an open question . in short , today s model building has to take the challenge caused by the reasonably large value of @xmath0 . note that the rge running effects or finite quantum corrections are not easy to generate @xmath135 from @xmath181 , unless the seesaw threshold effects or other extreme conditions are taken into account @xcite . one may therefore consider a pattern of @xmath124 with nonzero @xmath152 , such as the tetra - maximal mixing pattern @xcite or the correlative mixing pattern @xcite , as a starting point of view to calculate the radiative corrections before confronting it with current experimental data . we shall elaborate on this point in detail in section 5 . note that the perturbation matrix @xmath125 in eq . ( 17 ) is in general a sum of all possible perturbations to the constant flavor mixing matrix @xmath124 . from the point of view of model building , it is helpful to single out a viable @xmath125 whose form is as simple as possible . to do so , let us reexpress eq . ( 17 ) in the following manner : @xmath182 where @xmath183 holds , and it satisfies the condition @xmath184 as a result of the unitarity of @xmath29 itself . therefore , one may achieve a viable but minimal perturbation to @xmath124 by switching off @xmath185 ( or @xmath186 ) and adjusting @xmath187 ( or @xmath185 ) to its simplest form which is allowed by current experimental data . such a treatment is actually equivalent to multiplying @xmath124 by a unitary perturbation matrix , which may more or less deviate from the unit matrix @xmath188 , from either its left - hand side or its right - hand side . the first example of this kind was given before @xcite for the democratic mixing pattern , and its @xmath125 was mainly responsible for the generation of nonzero @xmath0 and @xmath63 . here we concentrate on the typical patterns of @xmath124 discussed above and outline the main ideas of choosing the minimal perturbations to them . * if @xmath124 predicts @xmath189 and @xmath170 together with @xmath190 ( the best - fit value based on current neutrino oscillation data @xcite ) , then the simplest way to generate a relatively large @xmath0 , keep @xmath191 unchanged and correct @xmath160 to a slightly smaller value is to choose a complex @xmath192 rotation matrix as the perturbation matrix : @xmath193 where @xmath194 is a small angle to trigger the perturbation effect . the most striking example in this category is to take @xmath124 to be the tri - bimaximal mixing pattern given in eq . the result is @xcite : @xmath195 which predicts @xmath196 in the standard parametrization . note that the obtained correlation between @xmath85 and @xmath0 is especially interesting because it leads us to @xmath197 when @xmath198 , consistent with the present experimental data . if @xmath59 is allowed to slightly deviate from @xmath155 , then one may simply make the replacement @xmath199 in eq . * if @xmath124 predicts @xmath189 and @xmath170 together with @xmath200 , then the most economical way to generate a relatively large @xmath0 , keep @xmath201 unchanged and correct @xmath160 to a slightly larger value is to choose a complex @xmath202 rotation matrix as the perturbation matrix : @xmath203 taking @xmath124 to be the golden - ratio mixing pattern in eq . ( 26 ) , we immediately arrive at @xmath204 whose predictions include @xmath70 , @xmath205 , and @xmath206 in the standard parametrization of @xmath29 . in this case the correlation between @xmath85 and @xmath0 leads to @xmath207 when @xmath198 , compatible with the experimental data . again , the replacement @xmath208 in eq . ( 38 ) allows one to obtain a somewhat more flexible value of @xmath59 which may slightly deviate from @xmath155 . * if @xmath124 is quite far away from the realistic mnsp matrix @xmath29 , one has to consider a somewhat complicated perturbation matrix including two rotation angles . in the neglect of cp violation , for instance , we may consider @xmath209 where @xmath210 and @xmath211 ( for @xmath212 ) . however , we hope that the resulting structure of @xmath29 still allows us to obtain one or two predictions , in particular for the mixing angle @xmath0 . a simple example of this kind has been given before @xcite by taking @xmath124 to be the democratic mixing pattern , and it predicts an interesting relationship between @xmath0 and @xmath59 in the standard parametrization : @xmath213 typically taking @xmath52 , we can arrive at @xmath214 @xcite . it is certainly easy to accommodate a cp - violating phase in @xmath215 , although its form might not be really minimal anymore . for those constant flavor mixing patterns with @xmath216 from the very beginning , such as the correlative @xcite and tetra - maximal @xcite mixing scenarios given in eqs . ( 30 ) and ( 32 ) , the similar minimal perturbations can be introduced in order to make the resulting mnsp matrix @xmath29 fit the experimental data to a better degree of accuracy . it should be noted that the above discussions about possible patterns of @xmath125 ( or @xmath215 ) with respect to those of @xmath124 are purely phenomenological . from the point of view of model building , it is more meaningful to consider the textures of lepton mass matrices @xmath217 where @xmath218 and @xmath219 can be obtained in the limit of certain flavor symmetries , and their special structures allow us to achieve a constant flavor mixing pattern @xmath124 . the perturbation matrices @xmath220 and @xmath221 play an important role in transforming @xmath124 into the realistic mnsp matrix @xmath29 , and thus their textures should be determined in a simple way and with a good reason . the connection between @xmath222 and @xmath125 ( or @xmath215 ) depends on the details of a lepton flavor model and may not be very transparent in most cases . in the basis where @xmath102 is real and positive , however , @xmath223 can be formally expressed as @xmath224 where @xmath225 and @xmath226 , @xmath227 and @xmath228 . here @xmath229 ( for @xmath27 ) denote the eigenvalues of @xmath219 in the symmetry limit , while @xmath230 and @xmath231 stand for the majorana phases in the same limit . it is therefore possible , at least in principle , to fix the structure of @xmath221 with the help of a certain flavor symmetry and current experimental data . the rge running effects of the neutrino flavor parameters have been discussed in many papers @xcite . it is known that large radiative corrections to those parameters are possible , especially when the neutrino masses are nearly degenerate or the value of @xmath232 is sufficiently big in the mssm . the fact that @xmath0 is not as small as previously expected motivates one to reconsider how it can be generated at the tree level or by quantum corrections . some studies in this regard have recently been done to look at the impacts of a relatively large @xmath0 on the running behaviors of the other two mixing angles and the cp - violating phases @xcite . the masses of the majorana neutrinos are believed to be attributed to some underlying new physics at a superhigh - energy scale @xmath233 ( e.g. , via the canonical seesaw mechanism @xcite ) . but this kind of new physics can all point to the unique dimension-5 weinberg operator for the neutrino masses in an effective field theory after the corresponding heavy degrees of freedom are integrated out @xcite . in the mssm , such a dimension-5 operator reads @xmath234 where @xmath233 denotes the cutoff scale , @xmath235 stands for the left - handed lepton doublet , @xmath236 is one of the mssm higgs doublets , and @xmath237 represents the effective neutrino coupling matrix . one may obtain the effective majorana neutrino mass matrix @xmath238 after spontaneous gauge symmetry breaking . the cutoff scale @xmath233 implies the scale of new physics , such as the mass scale of the heavy majorana neutrinos in the canonical seesaw mechanism . the evolution of @xmath237 from @xmath233 down to the electroweak scale @xmath239 is formally independent of any details of the relevent model from which @xmath237 is derived . below @xmath233 the scale dependence of @xmath237 is described by @xmath240 \ ; % ( 45)\ ] ] at the one - loop level in the mssm @xcite , where @xmath241 . \(1 ) _ the running behaviors of the three mixing angles_. eq . ( 47 ) shows that the rge running behaviors of the three neutrino mixing angles are strongly dependent on the three cp - violating phases . as for the majorana neutrinos , the radiative corrections to the three mixing angles can be adjusted by choosing different values of the cp - violating phases @xmath63 , @xmath74 and @xmath75 . a numerical analysis has been carried out to look at their numerical evolution to @xmath252 via the rges in the mssm with @xmath253 ( denoted as `` mssm10 '' for short ) or @xmath254 ( denoted as `` mssm50 '' for short ) , in which @xmath255 , @xmath256 , @xmath257 are taken as the typical input and the three cp - violating phases are freely adjusted at @xmath239 @xcite . the main results are summarized in table 2 , where the upper ( lower ) lines show the possible ranges of three mixing angles at @xmath252 for the normal ( inverted ) neutrino mass hierarchy . the rge of @xmath85 is dominated by the term @xmath258 , implying that the magnitude of the radiative correction to @xmath85 depends strongly on the phase difference @xmath259 . hence @xmath85 is most sensitive to the rge effect when @xmath260 holds . running from @xmath261 gev down to @xmath239 , the mixing angles @xmath59 and @xmath0 receive less significant radiative corrections in the mssm10 case , as shown in table 2 . while in the mssm50 case @xmath59 and @xmath0 may also receive significant radiative corrections if three cp - violating phases are well turned . here let us consider a typical example of this special case the correlative neutrino mixing pattern with @xmath267 , @xmath70 and @xmath268 @xcite . compared with the best - fit values of the three mixing angles at @xmath239 , the three mixing angles in this correlative mixing pattern at @xmath252 have to receive comparably small radiative corrections during their rge evolution . as we have mentioned in the last paragraph , this requirement can easily be achieved in the mssm10 case provided the condition @xmath269 is satisfied for a nearly degenerate neutrino mass spectrum . such a condition is unnecessary if the neutrino mass spectrum has a strong hierarchy . the numerical results are presented in table 3 , where @xmath270 , @xmath271 and @xmath272 are input at @xmath273 . then @xmath274 , @xmath275 and @xmath276 are obtained at @xmath277 after the radiative corrections . \(2 ) _ the radiative generation of the cp - violating phases_. it is well known that one cp - violating phase can be generated from another @xcite , simply because they are entangled in the rges . an especially interesting example is the dirac phase @xmath63 , which measures the strength of cp violation in neutrino oscillations at the electroweak scale , can be radiatively generated from the nonzero majorana phases @xmath74 and @xmath75 at a superhigh - energy scale . if @xmath0 is very small , however , the running of @xmath63 can be significantly enhanced by the terms that are inversely proportional to @xmath80 . in the mssm10 case it has been found that even @xmath205 can be radiatively generated if @xmath278 is taken @xcite . given @xmath135 at @xmath279 , it is found that @xmath280 at @xmath277 may result from @xmath281 at @xmath252 in the mssm10 case . in the mssm50 case even @xmath282 can be obtained at @xmath277 @xcite . to summarize , in order to obtain a phenomenologically - favored neutrino mixing pattern at the electroweak scale @xmath288 , the radiative corrections should be carefully examined for those mixing patterns at a superhigh - energy scale @xmath289 which might result from a certain flavor symmetry . in the mssm10 case the values of @xmath59 and @xmath0 predicted at @xmath252 are always close to their running values at @xmath239 , while the value of @xmath85 at @xmath252 can be somewhat smaller or larger than its running value at @xmath239 . in the mssm50 case the allowed ranges of the three mixing angles at @xmath252 can be quite wide , as we have discussed above . however , a crucial point is that a given flavor symmetry model should be able to predict the appropriate cp - violating phases at @xmath252 in order to obtain the appropriate mixing angles at @xmath288 after the rge evolution . in general , it is possible to generate @xmath135 at @xmath239 from @xmath290 at @xmath252 through the radiative corrections , in particular when some new degrees of freedom or nontrivial running effects ( such as the seesaw threshold effects @xcite ) are taken into account or the three cp - violating phases are fine - turned . therefore , we argue that it seems more natural for a specific flavor symmetry model to predict a relatively large @xmath0 at @xmath273 . a measurement of the dirac phase @xmath63 in the forthcoming long - baseline neutrino oscillation experiments and any experimental information about the majorana cp - violating phases @xmath74 and @xmath75 are extremely important , so as to distinguish one flavor symmetry model from another through their different sensitivities to the radiative corrections . this observation makes sense in particular after the experimental errors associated with the neutrino mixing parameters are comparable with or smaller than the magnitudes of their respective rge running effects . the most popular mechanism of generating finite but tiny neutrino masses beyond the sm is the canonical seesaw mechanism @xcite , where the small neutrino masses are attributed to the existence of heavy degrees of freedom such as the right - handed majorana neutrinos . in this elegant picture the @xmath28 mnsp matrix @xmath29 has a striking difference from the @xmath28 ckm matrix @xmath53 in the sm : it is not exactly unitary due to small mixing between light and heavy neutrinos as a result of the yukawa interactions . the heavy neutrinos can be searched for at the lhc if the seesaw mechanism works at the tev scale @xcite , and the unitarity of @xmath29 can be tested in the future neutrino oscillation experiments . another possibility is to look at the seesaw - induced non - unitary effects on the electromagnetic dipole moments ( emdms ) and radiative decays @xmath291 @xcite . if the active neutrinos acquire their respective masses in the seesaw mechanism , they should have the emdms through quantum loops . the fact that the majorana neutrinos are their own antiparticles implies that they can only have the _ transition _ emdms between two different neutrino mass eigenstates in an electric or magnetic field . the relevant radiative decays of the heavier active neutrinos , which may contribute to the cosmic infrared background in the universe @xcite , are of particular interest in cosmology . and @xmath325.,title="fig:",scaledwidth=80.0% ] + -@xmath314 self - energy diagrams ( and their charge - conjugate counterparts ) associated with the emdms of massive majorana neutrinos , where @xmath326 denotes all the sm fermions and @xmath325.,title="fig:",scaledwidth=80.0% ] + let us consider the radiative @xmath291 transition , whose electromagnetic vertex can be written as @xmath327 for a real photon satisfying the on - shell conditions @xmath328 and @xmath329 . in eq . ( 56 ) @xmath330 and @xmath331 are the electric and magnetic _ transition _ dipole moments of majorana neutrinos , and their sizes can be calculated via the proper vertex diagrams in fig . 4 ( weak cc interactions ) . the @xmath332-@xmath314 self - energy diagrams in fig . 5 ( weak nc interactions ) do not have any _ net _ contribution to @xmath330 and @xmath331 , but we find that they play a very crucial role in eliminating the infinities because the divergent terms originating from fig . 4 are unable to automatically cancel out in the presence of the seesaw - induced non - unitary effects ( i.e. , @xmath333 and @xmath334 ) unless those divergent terms originating from fig . 5 are also taken into account . this observation is new . it implies that the non - unitary case under discussion is somewhat different from the unitary case ( i.e. , @xmath335 and @xmath336 ) discussed before in the literature @xcite , where the feynman diagrams in fig . 5 are forbidden and the divergent terms arising from fig . 4 can automatically cancel out . after a careful calculation , we arrive at @xmath337 where @xmath338 \ ; % ( 58)\end{aligned}\ ] ] with @xmath339 ( for @xmath340 ) denotes the one - loop function . although this result is _ formally _ the same as that obtained in the literature @xcite , they are _ intrinsically _ different as the seesaw - induced non - unitary effects on @xmath331 and @xmath330 were not considered in the previous works . to see how important such non - unitary effects may be , let us make two analytical approximations . first , @xmath341 holds to a good degree of accuracy for @xmath342 . second , @xmath343 is also a good approximation for small non - unitary corrections to @xmath313 , where @xcite @xmath344 with @xmath31 and @xmath345 ( here @xmath121 and @xmath346 are the mixing angles and cp - violating phases ) . note that the light - heavy neutrino mixing angles @xmath347 ( for @xmath27 and @xmath348 ) are at most of @xmath144 @xcite , such that the deviation of @xmath29 from @xmath313 is at the percent level or much smaller . then we obtain @xmath349 \nonumber\\ & & - \frac{3}{4 } \sum_\alpha \left [ \xi^{}_\alpha \left(x\right)^{}_{\alpha i } \left(x\right)^*_{\alpha j } \right ] \ ; . % ( 60)\end{aligned}\ ] ] the first and second terms on the right - hand side of this equation correspond to the non - unitary and unitary contributions , respectively . while the former is suppressed by @xmath350 ( for @xmath27 and @xmath348 ) hidden in @xmath351 , the latter is suppressed by @xmath352 ( for @xmath353 ) due to the gim mechanism @xcite . we therefore draw a generic conclusion that the seesaw - induced non - unitary effects on @xmath330 and @xmath331 can be comparable with or even larger than the standard ( unitary ) contributions . in this case the rates of radiative @xmath291 decays are given by @xmath354 with @xmath355 being the bohr magneton . the size of @xmath356 can be experimentally constrained by observing no emission of the photons from solar @xmath357 and reactor @xmath358 fluxes . more stringent constraints on @xmath359 come from the supernova 1987a limit on neutrino decays and from the cosmological limit on distortions of the cosmic microwave background radiation ( in particular , its infrared part ) : @xmath360 @xcite . now that more and more interest is being paid to the cosmic infrared background relevant to the radiative decays of massive neutrinos @xcite , it is desirable to evaluate @xmath359 and @xmath361 on a well - defined theoretical ground , such as the canonical seesaw mechanism under discussion . for three active neutrinos . the standard ( unitary ) results correspond to @xmath362 , and their uncertainties come from the three unknown cp - violating phases of @xmath313.,title="fig:",scaledwidth=90.0% ] + for three active neutrinos . the standard ( unitary ) results correspond to @xmath362 , and their uncertainties come from the three unknown cp - violating phases of @xmath313.,title="fig:",scaledwidth=90.0% ] + figs . 6 and 7 illustrate the numerical results of the non - unitary effects on @xmath359 and @xmath363 respectively @xcite . note that in this numerical analysis those small active - sterile neutrino mixing angles in eq . ( 59 ) are constrained by present experimental data @xcite as follows : @xmath364 all @xmath365 in @xmath351 ( for @xmath27 and @xmath348 ) are positive or vanishing . the cp phases @xmath366 are all allowed to vary from zero to @xmath367 , but they must satisfy the above constraints together with the relations @xmath368 and @xmath369 . to assure that radiative corrections to the masses of three light neutrinos ( via the one - loop self - energy diagrams involving the heavy neutrinos ) are sufficiently small ( e.g. , smaller than @xmath370 mev ) and stable , we simply assume that the masses of three heavy neutrinos are nearly degenerate @xcite and not more than @xmath371 tev . this assumption implies that the results shown here are for a limited and safe parameter space of the tev seesaw mechanism , but it is instructive enough to reveal the salient features of the non - unitary effects on the effective emdms @xmath372 and the radiative decay rates @xmath373 . to present our numerical results in a convenient way , let us define @xmath374^{1/2 } \ ; , % ( 63)\end{aligned}\ ] ] which measures the overall strength of the unitarity violation of @xmath29 , and @xmath375 is reasonably taken in our calculations . namely , we allow each @xmath365 ( for @xmath27 and @xmath348 ) to vary in the range @xmath376 . the numerical dependence of @xmath377 and @xmath356 on @xmath378 is shown in figs . 6 and 7 , respectively . some discussions are in order . \(1 ) switching off the non - unitary effects ( i.e. , @xmath362 ) , we obtain the effective electromagnetic dipole moments @xmath379 for the normal mass hierarchy with @xmath380 mev ; and @xmath381 for the inverted mass hierarchy with @xmath382 mev , where the uncertainties mainly come from the unknown cp phases @xmath383 , @xmath384 and @xmath385 . such standard ( unitary ) results are far below the observational upper bound on @xmath359 ( @xmath386 a few @xmath387 @xcite ) , but they serve as a good reference to the non - unitary effects on @xmath359 being explored . \(2 ) figs . 6 and 7 clearly show that @xmath359 and @xmath356 can be maximally enhanced by a factor of @xmath388 and a factor of @xmath389 , respectively , in particular when @xmath390 approaches its upper limit as set by current experimental data . the magnitude of @xmath391 may be strongly suppressed in the inverted neutrino mass hierarchy . the reason is rather simple : on the one hand , @xmath392 holds in this case , and thus @xmath393 must be very small ; on the other hand , @xmath394 depends on @xmath395 , so it can also be very small when the cp - violating phases are around zero or @xmath396 . this two - fold suppression becomes severer for the decay rate @xmath397 , because it is proportional to @xmath398 . \(3 ) the results of @xmath359 and @xmath399 are sensitive to the absolute neutrino mass scale for both normal and inverted mass hierarchies . for instance , @xmath391 and @xmath400 get enhanced when @xmath51 changes from zero to 5 mev in the normal mass hierarchy ; while @xmath401 and @xmath402 are enhanced when @xmath50 changes from zero to 5 mev in the inverted mass hierarchy . this kind of sensitivity is not so obvious if one only takes a look at the expressions of @xmath331 and @xmath330 in eq . the main reason is that a change of @xmath51 or @xmath50 requires some fine - tuning of the active - sterile neutrino mixing angles and cp - violating phases as dictated by the exact seesaw relation @xmath403 , leading to a possibly significant change of @xmath359 . the dependence of @xmath356 on the absolute neutrino mass scale is somewhat more complicated , as one can see from eq . ( 61 ) . \(4 ) the cp - violating phases play a very important role in fitting both the exact seesaw relation and eq . if the heavy neutrino masses @xmath308 are not suppressed , then an appreciable value of @xmath378 requires some fine cancellations in the matrix product @xmath404 such that sufficiently small @xmath242 can be obtained from @xmath405 . on the other hand , we remark that it is actually unnecessary to require @xmath308 to be around or above the electroweak scale . the seesaw - induced non - unitary effects on @xmath359 and @xmath356 can be significant even if one allows one , two or three heavy neutrinos to be relatively light ( e.g. , at the kev mass scale ) . such sterile neutrinos are interesting in particle physics and cosmology . note that it is easier to satisfy the exact seesaw relation with an appreciable value of @xmath378 by arranging @xmath308 to lie in the kev , mev or gev range . this kind of low - scale seesaw scenarios @xcite might be technically natural , but they have more or less lost the seesaw spirit . of course , sufficiently large @xmath308 and sufficiently small @xmath347 can always coexist to make the seesaw mechanism work in a natural way , but in this traditional case the non - unitary effects are too small to have any measurable consequences at low energies . it is also worth pointing out that the seesaw - induced non - unitary effects on @xmath331 and @xmath330 are rather different from the case of making a naive assumption of the flavor mixing between three active neutrinos and a few light sterile neutrinos @xcite . the latter can directly break the unitarity of the @xmath28 mnsp matrix @xmath29 and then lift the gim suppression @xcite associated with @xmath331 and @xmath330 . this kind of non - unitary effects are not constrained by the seesaw relation , and thus they are more arbitrary and less motivated from the point of view of model building . the effective electromagnetic dipole moments of three neutrinos and the rates of their radiative decays can be maximally enhanced by a factor of @xmath388 and a factor of @xmath406 , respectively , no matter whether the seesaw scale is around or below the tev energy scale . this observation is new and nontrivial , and it reveals an intrinsic and presumably important correlation between the electromagnetic properties of neutrinos and the origin of their masses . such a correlation may even serve as a sensitive touch - stone for the highly - regarded seesaw mechanism . after the daya bay measurement of the smallest neutrino mixing angle @xmath0 , it is natural to ask where we are standing and where we are expecting to go in neutrino physics . we have tried to answer these two questions from a phenomenological point of view in this review paper , although our answers are incomplete and full of conjectures . to be specific , we have given a fast overview of some fundamental neutrino properties and paid particular interest to the flavor issues of charged leptons and neutrinos , including the mass spectrum , flavor mixing pattern and cp violation . we have gone into details of possible lepton flavor structures by describing two useful phenomenological strategies and giving a number of typical examples . the impact of large @xmath0 on the running behaviors of other flavor mixing parameters has been discussed in the framework of the mssm . we have also illustrated the seesaw - enhanced electromagnetic dipole moments of three majorana neutrinos based on a viable tev seesaw scenario . , @xmath320 , @xmath407 and @xmath408 ) , six quark masses ( @xmath409 , @xmath410 , @xmath411 and @xmath412 , @xmath413 , @xmath414 ) , six lepton masses ( @xmath415 , @xmath416 , @xmath417 and @xmath51 , @xmath418 , @xmath50 ) , three quark flavor mixing angles ( @xmath99 , @xmath100 and @xmath419 ) , three lepton flavor mixing angles ( @xmath85 , @xmath0 and @xmath59 ) , two cp - violating phases in the quark sector ( @xmath420 and @xmath421 ) , three cp - violating phases in the lepton sector ( @xmath422 , @xmath74 and @xmath75 ) , the higgs mass @xmath423 and the @xmath424-boson mass @xmath425.,scaledwidth=60.0% ] if only the sm particles are taken into account and the massive neutrinos are assumed to be the majorana particles , we are then left with 29 fundamental parameters in nature , as described by the so - called fritzsch - xing plot in fig . the determination of @xmath0 and @xmath423 in 2012 is a milestone in particle physics . but the effective strong cp - violating phase @xmath426 in the quark sector and the three weak cp - violating phases ( i.e. , @xmath422 , @xmath74 and @xmath75 ) in the lepton sector remain unknown . moreover , the absolute mass scale of three neutrinos and their mass ordering have not been determined . we hope that various precision neutrino experiments can help pin down the relevant parameters in the lepton sector and then shed light on the flavor dynamics in the foreseeable future . of course , the picture in fig . 8 may be too simple and too naive because we are not sure whether some of those fundamental " parameters are really fundamental or not . new degrees of freedom , such as the sterile neutrinos or the supersymmetric particles , might be discovered and make the flavor sector much messier . if the underlying flavor theory is regarded as an animal , one has no idea whether it is a donkey or an elephant or something else . in this sense we are blind today and have to make a lot of experimental and theoretical efforts to identify its nose , eyes , ears , legs and so on in order to make sure what animal it is . the road behind has repeatedly told us that the road ahead is always challenging , but it is always exciting . this review paper is essentially based on the plenary talk given by one of us ( z.z.x . ) at the _ susy 2012 _ conference . the work of s.l . is supported in part by the national basic research program ( 973 program ) of china under grant no . 2009cb824800 , the national natural science foundation of china under grant no . 11105113 and the fujian provincial natural science foundation under grant no . the work of z.z.x . is supported in part by the national natural science foundation of china under grant no .
living systems have time - evolving interactions that , until recently , could not be identified accurately from recorded time series in the presence of noise . stankovski et al . ( _ phys . rev . lett . _ * 109 * 024101 ; 2012 ) introduced a method based on dynamical bayesian inference that facilitates the simultaneous detection of time - varying synchronization , directionality of influence , and coupling functions . it can distinguish unsynchronized dynamics from noise - induced phase slips . the method is based on phase dynamics , with bayesian inference of the time - evolving parameters being achieved by shaping the prior densities to incorporate knowledge of previous samples . we now present the method in detail using numerically - generated data , data from an analog electronic circuit , and cardio - respiratory data . we also generalize the method to encompass networks of interacting oscillators and thus demonstrate its applicability to small - scale networks .
1209.4684
systems of interacting oscillators are ubiquitous in science . in the common case where the natural frequencies or amplitudes or inter - oscillator couplings are time - varying , they pose a continuing challenge to the time - series analyst who endeavours to understand the underlying system from the signal(s ) it creates . oversimplifications of hypotheses are often used to render the problem more tractable , but can all too easily result in a failure to describe phenomena that are in fact of central importance given that the strength , direction and functional relationships that define the nature of the interactions can cause qualitatively new states to appear or disappear . time - variability of this kind is especially important in biological applications , though it is by no means restricted to biology . in the absence of time - variability , there are many different methods available @xcite for detecting and quantifying the couplings and directionality ( dominant direction of influence ) between oscillators based , especially , on the analysis of phase dynamics . approaches to the detection of synchronization have mostly been based on the statistical properties of the phase difference @xcite . the inference of an underlying phase model has been used as the functional basis for a number of techniques to infer the nature of the phase - resetting curves , interactions and structures of networks @xcite . however , these techniques inferred neither the noise dynamics nor the parameters characterizing the noise . an additional challenge to these methods can be the time - varying dynamics mentioned above . in a separate line of development , bayesian inference was applied to analyse the system dynamics @xcite , thereby opening the door to inference of noisy time - evolving phase dynamics . methods based on transfer entropy and granger causality have a generality that has facilitated a number of applications , including inference of the coupling strength and directionality @xcite . these techniques provide measures of the amount of information in a measured signal , or the causal relationships between measured signals and , in doing so , they infer effect rather than mechanism . in this paper we describe in detail and further extend a recently introduced method @xcite based on dynamical bayesian inference . as we demonstrate below , it enjoys many advantages over earlier approaches . one of these is that it does not require the observable to fill the domain of the probability density function at equilibrium : it can thus provide the same information from a small fraction of the data that is required by the transfer entropy or granger causality approaches . additionally , the dynamical approach has the advantage of characterizing the system completely ( not only in terms of information measures ) . thus , from the inferred dynamics one can deduce self - consistently any information that is of interest , be it coupling functions , or synchronization , or causality , or equilibrium densities , including the equations of motion . we discuss in detail the theoretical background , the technical aspects and limitations of the algorithms , and we demonstrate the wide applicability of the method by consideration of several examples . the coupling functions are of particular importance . their form is uniquely able to describe the functional laws of interaction between the oscillators . earlier theoretical studies have included the work of kuramoto @xcite and winfree @xcite , which used a function defined either by the phase difference or by both phases , and of daido @xcite and crawford @xcite who used a more general form in which the coupling function was expanded in fourier series . other methods for inference of the coupling functions have also been suggested @xcite . the technique described below goes beyond all of these because it is able to follow the time - variability of the coupling functions and hence can reveal their dynamical character where it exists . we will also show how the technique can readily be extended to encompass networks of interacting oscillators . these form a large and important group of physical systems , including neural networks @xcite , electrochemical systems @xcite , crowd synchrony on the millennium bridge , and networks of fireflies @xcite . the large scale of the networks can introduce a higher complexity , both in structure and functional behavior . for example in neuronal networks , the existence of spatial and spatial - temporal correlations , collective or partially collective ( clustering ) behavior , synchronization or desynchronization , and time - variability has been reported @xcite . in such cases , and given the kinds of phenomenon to be studied , there is an increasing need for powerful techniques that can infer the time - varying dynamics of the oscillatory networks . in sec . [ sc : dynamics ] we provide details about the phase decompositions , the implementation of the bayesian framework and how the time - varying information is propagated . the synchronization detection through a map representation of the phase dynamics is discussed in sec . [ sc : synchdetection ] , while the method for describing the interactions is demonstrated in sec . [ sc : interactions ] . before the method is applied , we consider in sec . [ sc : technical ] some important technical aspects and limitations . the wide applicability of the method is demonstrated in sec . [ sc : applications ] , through the analysis of time - series from numerical phase and limit - cycle oscillators , analogue simulation and cardio - respiratory interactions . the generalization of the approach to networks of oscillators , as exemplified by two numerical examples , is presented in sec . [ sc : networks ] . finally , we summarise and draw conclusions in sec . [ sc : conclusion ] . the algorithm used for the detection of synchronization is described in appendix [ app : fixedpoint ] . consider an @xmath0-dimensional oscillator @xmath1 whose solution @xmath2 admits a limit cycle . such an oscillator can usually be represented by a constant phase velocity @xmath3 and a vector coordinate that defines the limit cycle as a function of the phase @xmath4 : @xmath5 . when two such oscillators mutually interact sufficiently weakly , their motion is commonly approximated just by their phase dynamics @xcite . we note that , in general , if we describe the phase of a system through a generic monotonic change of variables , than the dynamical process can be written as @xmath6 eq . ( [ eq : phi ] ) explicitly includes a noise term @xmath7 to enable it to represent a process in a real system . the noise can be e.g. a white gaussian noise @xmath8 , where the symmetric matrix @xmath9 encloses the information about correlation between the noises on different oscillators , which we will refer to as _ spatial correlation_. the phase - dynamics decomposition technique is highly modular from the algorithmic point of view , and each module will be explained separately in the sections that follow . the overall procedure comprises the following steps : 1 . assumption that the dynamics can be precisely described by a finite number of fourier terms ( see sec . [ s : fourier ] ) . 2 . inference , given the data , of the fourier terms , the noise amplitude , and their correlation in form of a parameter probability distribution ( see sec . [ sec : bayesian ] for stationary dynamics , and sec . [ s : timevarying ] for time - varying dynamics ) . 3 . integration of the probability that this parameter set lies inside the arnold tongue defining synchronization . this effectively yields the cumulative probability of the synchronization state of the dynamics ( see sec . [ sec : discrimination ] ) . use of the parameter information as obtained in step 2 to create a description of the interactions , leading to detection of the predominant directionality and coupling function estimation among the oscillators ( see sec . [ sc : interactions ] ) . the periodic behaviour of the system suggests that it can appropriately be described by a fourier decomposition . decomposing both @xmath10 and @xmath11 in this way leads to the infinite sums @xmath12 it is reasonable to assume that , in most cases , the dynamics will be well - described by a finite number @xmath13 of fourier terms , so that we can rewrite the phase dynamics of eq.([eq : phi ] ) as a finite sum of base functions @xmath14 where @xmath15 , @xmath16 , @xmath17 , and the rest of @xmath18 and @xmath19 are the @xmath13 most important fourier components . in order to reconstruct the parameters of eq . ( [ eq : phif ] ) we make extensive use of the approach to dynamical inference presented in @xcite . in this section we briefly outline the technique as adapted to the present case . the fundamental problem in dynamical inference can be defined as follows . a @xmath20-dimensional ( in general @xmath21-dimensional ) time - series of observational data @xmath22 ( @xmath23 , @xmath24 ) is provided , and the unknown model parameters @xmath25 are to be inferred . bayesian statistics employs a given _ prior _ density @xmath26 that encloses expert knowledge of the unknown parameters , together with a _ likelihood _ function @xmath27 , the probability density to observe @xmath28 given the choice @xmath29 of the dynamical model . bayes theorem @xmath30 then enables calculation of the so - called _ posterior _ density @xmath31 of the unknown parameters @xmath32 conditioned on the observations . for independent white gaussian noise sources , and in the mid - point approximation where @xmath33 the likelihood is given by a product over @xmath34 of the probability of observing @xmath35 at each time . the likelihood function is constructed by evaluation of the stochastic integral of the noise term over time , as @xmath36 where @xmath37 is the cholesky decomposition of the positive definite matrix @xmath38 , and @xmath39 is a vector of normally - distributed random variables with zero mean and unit variance . the joint probability density of @xmath39 is used to find the joint probability density of the process in respect of @xmath40 by imposing @xmath41 , where @xmath42 is the jacobian term of the transformation of variables that can be calculated from eq . ( [ eq : fourierdecomposition ] ) . if the sampling frequency is high enough , the time step @xmath43 tends to zero , and the determinant of the jacobian @xmath42 can be well - approximated by the product of its diagonal terms @xmath44 this transformation leads to an extra term in a least squares likelihood , and the minus log - likelihood function @xmath45 can thus be written as @xmath46 { ( { e}^{-1})}_{ij } [ \dot{\phi}_{j , n } - c^{(j)}_k { \phi}_{j , k}({\phi}_{\cdot , n}^{\ast } ) ] \big ) \ , , \end{split } \label{eq : likelihood}\ ] ] where summation over the repeated indices @xmath47,@xmath48,@xmath49,@xmath50 is implicit . the log - likelihood ( [ eq : likelihood ] ) is a quadratic form of the fourier coefficients of the phases . hence if a multivariate prior probability is assumed , the posterior probability is a multivariate normal distribution as well . this is highly desirable for two reasons : ( i ) a gaussian posterior is computationally convenient because it guarantees a unique maximum , with the mean vector and covariance matrix completely characterizing the distribution and giving us the most significant information ; ( ii ) all the multivariate normal posteriors can be used again as priors in the presence of a new block of data , and knowledge about the system can easily be updated . this last feature is essential for any real - time application because it ensures that the complexity of the algorithm does not change with the length of the input data - stream . from @xcite , and assuming a multivariate normal distribution as the prior for parameters @xmath51 , with means @xmath52 , and covariances @xmath53 , the stationary point of @xmath54 can be calculated recursively from @xmath55 where the covariance is @xmath56 , summation over @xmath34 from 1 to @xmath0 is assumed and summation over repeated indices @xmath47,@xmath48,@xmath49,@xmath50,@xmath57 is again implicit . we note that a noninformative `` flat '' prior can be used as the limit of an infinitely large normal distribution , by setting @xmath58 and @xmath59 . the multivariate probability @xmath60 given the readout time series @xmath22 explicitly defines the probability density of each parameter set of the dynamical system . because each of them can be discriminated , as belonging or not belonging to the arnold tongue region we can define the binary property @xmath61 , and can obtain the posterior probability of the system being synchronized or not by evaluating the probability of @xmath62 @xmath63 the computation of @xmath64 will be discussed in sec . [ sc : synchdetection ] . the multivariate probability described by @xmath65 for the given time series @xmath66 explicitly defines the probability density of each parameter set of the dynamical system . when the sequential data come from a stream of measurements providing multiple blocks of information , one applies ( [ eq : cd ] ) to each block . within the bayesian theorem , the evaluation of the current distribution relies on the evaluation of the previous block of data , i.e. the current prior depends on the previous posterior . thus the inference defined in this way is not a simple windowing , but each stationary posterior depends on the history of the evaluations from previous blocks of data . in classical bayesian inference , if the system is known to be non - time - varying , then the posterior density of each block is taken as the prior of the next one : @xmath67 . this full propagation of the covariance matrix allows good separation of the noise , and the uncertainties in the parameters steadily decrease with time as more data are included . if time - variability exists , however , this propagation will act as a strong constraint on the inference , which will then fail to follow the variations of the parameters . this situation is illustrated in fig . [ fig1:tv_propg](a ) . shows inference of two coupled noisy poincar oscillators with the frequency of one oscillator time - varying for clarity and compactness of presentation the details are not shown here , but the reader can refer to the model and other details in sec . [ sc : applications ] . ] in such cases , one can consider the processes between each block of data to be independent ( i.e. markovian ) . there can not then be any information propagation between the blocks of data , and each inference starts from the flat distribution @xmath68 . the inference can thus follow more closely the time - variability of the parameters , but the effect of noise and the uncertainty of the inference will of course be much larger , as shown in fig . [ fig1:tv_propg](b ) . where the system s parameters are time - dependent , we may assume that their probability diffuses normally accordingly to the known diffusion matrix @xmath69 . thus , the probability density of the parameters is the convolution of the two normal multivariate distributions , @xmath70 and @xmath69 @xmath71 the covariance matrix @xmath69 expresses our belief about which part of the dynamical fields that define the oscillators has changed , and the extent of that change . its elements are ( @xmath72 = @xmath73 , where @xmath74 is the standard deviation of the diffusion of the parameter @xmath75 after the time window @xmath76 that has elapsed from the first block of information to the following one . @xmath77 is the correlation between the change of the parameters @xmath75 and @xmath78 ( with @xmath79 ) . in relation to the latter , a special example of @xmath69 will be considered : we assume that there is no correlation between parameters , i.e. @xmath80 , and that each standard deviation @xmath74 is a known fraction of the parameter @xmath75 : @xmath81 ( where @xmath82 indicates that @xmath83 is referred to a window of length @xmath76 ) . it is important to note that this particular example is actually rather general because it assumes that all of the parameters ( from the @xmath84 diagonal ) can be of a time - varying nature which corresponds to the inference of real ( experimental ) systems with _ a priori _ unknown time - variability . there are two obvious limits in modeling the knowledge assumed with respect to possible time variation of parameters . the first of these is to assume no time - variability : in this case the full information propagation matrix is used , @xmath85 . if the assumption proves wrong , the inferred parameters may accumulate a bias when the real system varies in time . the other limit is to assume each time window to be completely independent of the previous signal history . in this case no propagation is used , @xmath86 , ( i.e. @xmath87 ) , and there is no bias but , because much information is forgotten , the probability of the inferred parameters has a large covariance matrix . an optimal assumption must lies in between these two limits : @xmath88 ; where the choice of @xmath89 is parameterized with the values of the @xmath82 s . if a diffusion matrix is assumed , we allow the method some freedom for the time - variability to be followed , while restricting it to be unbiased . the amount of variability is part of the model , like the number of free parameters in any standard method . [ fig1:tv_propg ] illustrates the two extreme limits , and a possible trade - off . ) . the gray lines represent the actual parameter in the numerical simulation , whereas the black lines indicate the time - varying parameter inferred from the resultant time series , for : ( a ) full propagation , @xmath67 ; ( b ) no propagation , @xmath68 ; and ( c ) propagation for time - varying processes , @xmath90 . ] the inference in fig . [ fig1:tv_propg](c ) demonstrates that the time - variability is captured correctly and that the uncertainty is reduced because more data have been included . if one knows beforehand that only one parameter is varying ( or , at most , a small number of parameters ) , then @xmath69 can be customized to allow tracking of the time - variability specifically of that parameter . this selective propagation can be achieved if , for example , not all but only the selected correlation @xmath91 from the diagonal has a non - zero value . it is important to note that finite noise can induce phase slips in a system that would be synchronized in the noiseless limit . rather than focusing on the presence and statistics of phase - slips , we propose to detect synchronization from the nature of the phase - slip itself . a novel feature of the present study is that it proposes evaluation of the probability that the equations driving the dynamics are _ intrinsically _ synchronized and thus of whether any phase - slips that may possibly be observed are dynamics - related or noise - induced . after performing the inference , one can use the reconstructed parameters , derived in the form of a multivariate normal distribution @xmath65 , to study the interactions between the oscillators under study . in general , the border of the arnold tongue may not have an analytic solution . in practice , we estimate @xmath64 numerically , sampling from the parameter space many realizations @xmath92 , where @xmath93 labels each parameter vector tested . for every set of @xmath94 we compute @xmath95 numerically . let us assume for now that @xmath95 is given . to find @xmath64 with arbitrary precision , it is enough to generate a number @xmath96 of parameters @xmath97 , with @xmath98 sampled from @xmath99 , since @xmath100 . however , this @xmath101-dimensional integration quickly becomes inefficient with an increasing number of fourier components . moreover , as we will discuss in sec . [ sec : discrimination ] , the computation time of the variable @xmath95 is not insignificant . on the other hand , if the posterior probability @xmath102 is sharply peaked around the mean value @xmath103 , then @xmath64 will be indistinguishable from @xmath104 , and the evaluation of @xmath104 will suffice . we now illustrate a simple numerical technique to recognize whether a coupled phase oscillator system is synchronized , or not . the technique itself amounts to a simple check by numerical integration of the system of ordinary differential equation defined by eq . ( [ eq : phi ] ) through one cycle of the dynamics , and testing whether the 1:1 synchronization condition @xmath105 is always obeyed . [ sec : torus ] let us assume we are observing motion on the torus @xmath106 defined by the toroidal coordinate @xmath107 , and the polar coordinate @xmath108 . for assessment of possible 1:1 synchronization the phase difference @xmath108 will be defined as @xmath109 . fig.[fig2:torus ] provides a schematic representation of the phase dynamics on the torus . and polar coordinate @xmath110 . the white circle denotes the poincar cross section . , width=151 ] let us consider a poincar section defined by @xmath111 and assume that @xmath112 for any @xmath113 . this means that the direction of motion along the toroidal coordinate is the same for every point of the section . ideally we would follow the time - evolution of every point and establish whether or not there is a periodic orbit ; if there is one , and if its winding number is zero , then the system is synchronized . if such a periodic orbit exists , then there is at least one other periodic orbit , with one of them being stable and the other unstable . the solution of the dynamical system over the torus yields a map @xmath114\to [ 0,2\pi]$ ] that defines , for each @xmath115 on the poincar section , the next phase @xmath116 after one circuit of the toroidal coordinate @xmath117 . fig.[fig8:maps](b),(c ) illustrates the map @xmath96 as evaluated computationally in two situations , corresponding to no synchronization , or synchronization , respectively . the map @xmath96 is continuous , periodic , and has two fixed points ( one stable and one unstable ) if and only if there is a pair of periodic orbits for the dynamical system , i.e. synchronization is verified if @xmath118 exists such that @xmath119 and @xmath120 . the existence of the fixed point @xmath118 is established through the simple algorithmic procedure described in appendix [ app : fixedpoint ] . inferring the parameters of the system not only allows for evaluation of the synchronization as an epiphenomenon in its own right , but their probability @xmath65 also describes the interaction properties of the oscillators . because their dynamics is reconstructed separately , as described by eq . ( [ eq : phi ] ) , use can be made only of those inferred parameters from the base functions @xmath121 and @xmath122 that are linked to influences between the oscillators . one can seek to determine the properties that characterize the interaction in terms of a strength of coupling , predominant direction of coupling , or even by inference of a coupling function . the analysis of information propagation allows inference of the time - varying dynamics , and the interactions properties can be traced in time as well . this is especially important for the inference of open interacting oscillatory processes where the time - variability of the interactions can lead to transitions between qualitatively different states , such as synchronization @xcite . the coupling amplitude quantifies the total influence between the oscillators in a particular direction , e.g. how much the dynamics of the first oscillator affects the dynamical behavior of the second oscillator ( @xmath123 ) . depending on whether the coupling is in only one direction , or in both directions , we speak of unidirectional or bidirectional coupling , respectively . in the inferential framework that we propose , the coupling amplitudes are evaluated as normalized measures , based on the interacting parameters inferred from the coupling base functions . the influence of one oscillator on the other can either be direct through @xmath121 , or can arise through the combined interacting base functions @xmath122 . in what follows , the base functions @xmath121 and @xmath122 are described with a common notation @xmath124 . the quantification is calculated as a euclidian norm : @xmath125 where the odd inferred parameters are assigned to the base functions @xmath126 for the coupling that the second oscillator imposes on the first ( @xmath127 ) , and _ vice versa _ ( @xmath128 ) . the directionality of coupling @xcite often provides useful information about the interactions . it is defined as normalization about the predominant coupling amplitude @xmath129 if @xmath130 $ ] the first oscillator drives the second ( @xmath131 ) , or if @xmath132 the second ( @xmath133 ) drives the first . the quantified values of the coupling strengths @xmath134 or the directionality @xmath135 represent measures of the combined relationships between the oscillators . thus , a non - zero value can be inferred even when there is no interaction . such discrepancies can be overcome by careful surrogate testing @xcite by rejection of values below an surrogate acceptance threshold , which can be specified e.g. as the mean plus two standard deviations among many realization of the measure . in addition to the coupling strength and the directionality , one can also infer the coupling function that characterizes the interactions , i.e. the law that describes the functional relationships between the oscillators . its characteristic form reflects the nature of the oscillators and how their dynamics reacts to perturbations . the coupling function should be @xmath136-periodic . in the inferential framework under study , the coupling functions were decomposed into a finite number of fourier components . the function describing the interactions between the two oscillators was decomposed by use of the odd parameters @xmath137 and the corresponding base functions @xmath138\in$ ] @xmath139 @xmath140 up to order @xmath34 of the decomposition . the reverse function @xmath141 @xmath142 @xmath143 was similarly decomposed . the technique is quite generally applicable to a broad class of problems , and so there are a number of technical aspects and choices to bear in mind . we now discuss three of them in particular : the number of base functions to be employed in the inference process ( which is part of the model ) ; the intensity of the noise characterizing the system ( which is an externally imposed constraint ) ; and the time resolution . [ [ number - of - base - functions . ] ] number of base functions . + + + + + + + + + + + + + + + + + + + + + + + + + selection of the optimal set of base functions to describe the problem is far from trivial . in general one wishes to have the minimal set that describes sufficiently well the model to be tested . where the length of the data series is very long or effectively infinite , one can include an excessive number of base functions without immediate penalties . in reality , however , any unneeded base function jeopardizes the precision of the coefficients that really are relevant for the model , and the picture is further complicated when the model to be adopted is expected to be an outcome of the inference machinery . where one deals with a long data series , possibly with a high signal - to - noise ratio , a relative large number of base functions can be used . the speed of computation is also an important aspect to keep under consideration , given that having a large number of base functions vastly increases the parameter space , and that iterative calculations ( especially matrix inversion ) slow the speed of processing by the third power of the number of coefficients . note that , even though the bayesian inference is generally popular in real - time applications , computational speed limitations mean that our inference framework for general phase dynamics can not yet be used in this way . [ [ sec4_sub2_cph3:base_noises ] ] role of noise intensity . + + + + + + + + + + + + + + + + + + + + + + + + in general , the greater the noise intensity , the bigger the covariance of the inferred parameters . for a repeated experiment ( e.g. generation of a synthetic signal , and parameter inference based on that signal ) the variance of a particular parameter would increase monotonically with noise amplitude , as shown in fig . [ fig5:noises ] . and coupling @xmath144 for different noise intensities @xmath38 . the signal to be analysed is generated from eq . ( [ equ : num_model_app ] ) . the dotted line shows the actual values of the parameters . the boxplots refer to the descriptive statistics ( median , quartiles , max . and min . ) of @xmath145 different runs of the generation - and - inference loop . ] there are , however , a few notable exceptions . the inferential capabilities rely on the volume of phase - space spanned by the variables . a state of synchronization would represent a limit cycle for the global system , and parameter inference of neither oscillator would reach satisfactory precision . in such cases a minimal amount of noise is typically needed , sufficient to drive the system out of equilibrium at least once . during the resultant phase - slip , the data would be filling the phase space sufficiently for correct parameter reconstruction . [ [ sec4_sub3_cph3:base_time_res ] ] time resolution . + + + + + + + + + + + + + + + + we now summarize the limits of an idealized data acquisition . the time step @xmath43 is much smaller than any of the sequential time - windows used as data blocks for inference , so that each block contains many data points . also , @xmath43 is much smaller than either of the oscillator periods . each inference block is big enough to contain many cycles of the dynamics ( in particular , more cycles that those typical of a phase slip ) while , at the same time , each block is small enough to provide the desired resolution of parameter change . it can happen that the time resolution of the change in dynamical parameters is incompatible with an acquisition time - window that would guarantee precision for other parameters . the choice of the time - window must therefore be done on a case - by - case basis , depending on the type of information that is required from the system . ) for four different lengths of the inference windows . the sizes of the windows are shown in the box . for clearer presentation , the curves are separated by shifting them vertically from each other through equal offsets in @xmath146 . ] [ fig6:time_res ] illustrates such a compromise . we use the numerical model ( [ equ : num_model_app ] ) that will be introduced in sec . [ sec5_sub2_cph3:app_num ] to investigate the time - resolution for the case where the frequency @xmath147 and coupling amplitude @xmath148 were varying periodically at the same time . the parameters were : @xmath149 , @xmath150 , @xmath151 , @xmath152 , @xmath153 , @xmath154 @xmath155 and noise strengths @xmath156 . the parameters were reconstructed using four different window lengths for the inference . the results presented in fig . [ fig6:time_res ] demonstrate that , for small windows ( 0.5s ) , the parameters are sparse and sporadic , while for very large windows ( 100s ) the time - variability is faster than the size of the window and there is cut - off on the form of the variability . the optimal window length will lie between these two . another interesting feature is that , for the smallest window ( 0.5s ) , the coupling amplitude improves with information propagation as time progresses , while the frequency inferred ( as a constant component without base function ) remains sparse throughout the whole time interval . the technique is first applied to synthetic data to test the performance of the algorithm , and then real data are analyzed . to create the synthetic data , we used both numerical and analogue electronic simulations . in the examples that follow , except where stated otherwise , we used the phase model eqs . ( 1)-(3 ) with fourier expansion to second order @xmath157 , propagation constant @xmath158 and window length of @xmath159s . numerically generated data were obtained from models of phase oscillators and limit - cycle oscillators . the phase oscillator model provides a sufficient basis for the description of synchronization while being , at the same time , analytically traceable . we thus test the detection of synchronization ( as explained in sec . [ sc : synchdetection ] ) through bayesian inference of synthetic data whose synchronization is already known . the model is given by two coupled phase oscillators subject to white noise @xmath160 the parameters are @xmath161 , @xmath162 , @xmath163 ; parameter @xmath164 is chosen so that the system lies close to the border of the arnold tongue ( either just inside or just outside ) . because we aim to demonstrate the precision of synchronization detection , we add no time - variability to the model , the inference is applied to a single block of data , and there is no spatial noise correlation with noise intensities @xmath165 . the dynamics of the phase difference is described as @xmath166 , where @xmath167 is the frequency mismatch and @xmath168 is the resultant coupling . it is evident that the analytic condition for synchronization , i.e. the existence of a stable equilibrium solution @xmath169 , is @xmath170 . for @xmath171 ( outside the arnold tongue ) the reconstructed map @xmath172 ( fig . [ fig8:maps](b ) ) after parameter inference has no root @xmath173 : hence the oscillators are not synchronized . when @xmath174 , even though the system was inside the arnold tongue , noise triggered occasional phase slips ( see fig . [ fig8:maps](d ) ) . we tested synchronization detection on the same signals using the methods already available in the literature , based on the statistics of the phase difference @xcite , but none of them was able to detect the presence of synchronization under these conditions . for example , one of the most widely - used methods for synchronization detection @xcite gives a normalized index of 0.7539 , well below the 0.9183 threshold ( evaluated as the mean plus two sds of surrogate realizations ) for acceptance of synchronization . in spite of the phase slips , our technique correctly detects the root @xmath173 from the inferred parameters , revealing that the oscillators are _ intrinsically _ synchronized as shown in fig . [ fig8:maps](c ) : the phase slips are attributable purely to noise ( whose inferred intensity is given by the matrix @xmath175 ) , and not to deterministic interactions between the oscillators . to demonstrate the capabilities of the technique in tracing time - varying parameters , coupling functions , directionality and synchronization , we analyzed data from a numerical model of two coupled , non - autonomous , poincar oscillators subject to white noise , @xmath176 we tested several possibilities for the parameters : while letting the frequencies @xmath177 and coupling parameters @xmath178 be time - varying , we ran numerical experiments with the coupling function @xmath179 either fixed or time - varying . as a first numerical experiment , we considered bidirectional coupling ( 1@xmath1802 ) , where the natural frequency of the first oscillator , and its coupling strength to the second one , vary periodically at the same time : @xmath181 and @xmath182 . the other parameters were : @xmath183 , @xmath184 , @xmath185 , @xmath186 , @xmath187 , @xmath188 , @xmath189 and noise @xmath190 . the coupling function was proportional to the difference in the state variables : @xmath191 and @xmath192 ( the same coupling function was used for construction of fig . [ fig1:tv_propg ] , fig . [ fig5:noises ] and fig . [ fig6:time_res ] ) . the phases were estimated as the angle variable @xmath193 ( where @xmath194 is defined as the four - quadrant inverse tangent function ) . with @xmath195 , in a state of no synchronization , the time - varying parameters @xmath196 and @xmath197 are accurately traced as it can be seen in fig . [ fig9:num_sync](a ) and ( b ) . ) . the frequency @xmath196 ( a ) and coupling @xmath198 ( b ) are independently varied . the dotted and full lines plot the parameters when the two oscillators are synchronized for part of the time ( @xmath199 ) , and not synchronized at all ( @xmath195 ) , respectively . the regions of synchronization , found by calculation of the synchronization index , are indicated by the gray shaded regions . ( c ) and ( d ) show the coupling functions @xmath126 and @xmath141 for time windows centered at @xmath200 . in both cases , the window length was @xmath201 and the coupling was @xmath202 . ] the coupling amplitude of @xmath199 corresponds to a state of intermittent synchronization , where the two oscillators are synchronized for part of the time . the precision of the reconstructed time - variable parameters is satisfactory during the non - synchronized intervals . during the synchronized intervals , however , the oscillators do not span sufficient phase - space to allow precise inference of the parameters ( fig . [ fig9:num_sync](a ) and ( b ) , dashed lines ) . within these synchronized intervals , the posterior probability distribution of the parameters was not peaked ; however , it was sensibly different from zero only in that parameter region for which the corresponding noiseless dynamics is synchronized . hence , despite the impossibility of accurate parameter tracking , the detection of a synchronized state ( @xmath203 ) is always precise ( fig . [ fig9:num_sync](a ) and ( b ) , grey shaded regions ) . the reconstructed sine - like functions @xmath126 and @xmath141 are shown in figs . [ fig9:num_sync](c ) and ( d ) for the first and second oscillators , respectively . they describe the functional form of the interactions between the two poincar systems in eq . ( [ equ : num_model_app ] ) . the reconstructed form of the coupling functions was evaluated dynamically for each block . for discretely time - varying coupling amplitudes @xmath204 and @xmath205 . different unidirectionally and bidirectionally coupled states are reached for different values of @xmath204 and @xmath205 , as indicated by the square insets . ] next , the method was applied to deduce the predominant direction of coupling as specified from the norm of the inferred coupling base parameters . to illustrate the precision of the directionality detection , the frequencies were now set constant , while both of the coupling strengths remained discretely time - varying . the parameters were @xmath206 , @xmath207 , @xmath208 , and the coupling functions were , as in the previous example , @xmath191 and @xmath192 . synchronization was not reached , however , for these parameters . the couplings undergo changes at particular times , but otherwise remain constant , as shown in fig . [ fig10:num_dirc ] . the detected directionality index @xmath135 was consistent with the actual values . note that , for unidirectional coupling , @xmath135 does not quite reach unity on account of the noise . to further investigate the ability to track subtle changes of time - varying coupling functions , we used the same model as in eq . ( [ equ : num_model_app ] ) to generate a synthetic signal where the coupling functions are absolute values of the state difference to a power of the time - varying parameter : @xmath209 where @xmath210 and @xmath211 . the exponent parameter varied linearly with time @xmath212 , and the other parameters remained constant : @xmath184 , @xmath213 , @xmath214 , @xmath215 and @xmath216 . the reconstructed phase coupling functions @xmath217 were calculated from the inferred parameters for the interacting terms of the base functions . the results for four consecutive windows are presented in fig . [ fig11:num_cfs ] . it can readily be seen that their complex form now is not constant , but varies with time . comparing them in neighboring ( consecutive ) pairs : ( a ) and ( b ) , then ( b ) and ( c ) , then ( c ) and ( d ) , one can follow the time - evolution of the functional form . even though we can follow their time - variability , the two most distant functions fig . [ fig11:num_cfs](a ) and ( d ) are of substantially different shapes . note also that , beside their form , the functions norm i.e. coupling strength also varies ( cf . the height of the maxima in fig . [ fig11:num_cfs](a ) and ( d ) ) . thus we have validated the technique on numerical models whose deterministic dynamics and time - variability were already known , thereby demonstrating the usefulness , precision and comprehensiveness of the method . we found that it can produce a good description of noise - induced phase - slips , synchronization , directionality and coupling functions even when the dynamics is subject to deterministic time - varying influences . we also tested the technique on signals emanating from analog models . these are real , highly controllable , oscillatory systems and the noise on their signals is real rather than contrived , as in the case of numerical models . it is attributable to environmental disturbances , thermal fluctuations , and the inherent nonidealities of the circuit components . during the process of data acquisition and discretization , measurement noise can be introduced as well noise which has no links with the actual dynamics of the interacting oscillators . such signals provide a good test of our analysis capabilities . ) . ( a ) phase portrait from the oscilloscope ; ( b ) frequency @xmath218 from the external signal generator ; ( c ) detected frequency @xmath219 of the second driven oscillator ; ( d ) fast fourier transform ( fft ) of the detected frequency @xmath219.,width=302 ] we analyzed data from an analog experimental simulation of two coupled van der pol oscillators . details of the electronic implementation are given elsewhere @xcite . the noise here arises mainly from the imperfections of the electronic components and there is also measurement noise . [ fig12_analog](a ) shows the phase portrait derived from the first oscillator , with time - varying frequency , which drives the second oscillator @xmath220 ^ 2x_1=0,\nonumber\\ & \frac{1}{c^2}\ddot{x}_{2}-\mu_2(1-x_2 ^ 2)\frac{1}{c}\dot{x}_{2 } + \omega_2 ^ 2x_2+\varepsilon(x_1-x_2)=0,\hspace{+0.8em}\end{aligned}\ ] ] where the periodic time - variability @xmath221 ( fig . [ fig12_analog](b ) ) comes from an external signal generator . the parameters were @xmath222 , @xmath223 , @xmath224 , @xmath225 , @xmath226 and @xmath227 is constant resulting from the analogue integration . the phases were estimated as @xmath228 . for the given parameters the oscillators were synchronized , so that the second driven oscillator changed its frequency from being constant to being time - varying . applying the inferential technique showed , correctly , that the oscillators were indeed synchronized ( @xmath203 ) throughout the whole time period . the frequency of the driven oscillator was inferred as being time - varying fig . [ fig12_analog](c ) . performing a simple fft ( fig . [ fig12_analog](d ) ) showed that @xmath219 is periodic with period 0.2hz ( exactly as set on the signal generator ) . clearly , the technique reveals information about the nature and the dynamics of the time - variability of the parameters and is still able to do so using a more realistic signal than that from a numerical simulation . having tested our technique on two quite different kinds of synthetic data , we now apply it to a real physiological problem , to investigate the cardiorespiratory interaction . the analysis of physiological signals of this kind has already been found useful in relation to several different diseases and physiological states ( see e.g. @xcite and references therein ) . transitions in cardiorespiratory synchronization have been studied in relation to ansthesia @xcite and sleep cycles @xcite . it is also known that modulations and time - varying sources are present , and that these can affect the synchronization between biological oscillators @xcite . for comprehensive and reliable analysis a technique is needed that is able , not only to identify the time - varying information , but which will allow evaluation of the interacting measures ( e.g. synchronization and directionality ) , based solely on the information inferred from the signals . we will show that our technique meets these criteria . we analyse cardiorespiratory measurements from human subject under ansthesia . their breathing rate was held constant , being determined by a respirator . for such systems the analytic model is unknown , in contrast to analogue and numerical examples , but the oscillatory nature of the signal is immediately evident . the instantaneous cardiac phase was estimated by synchrosqueezed wavelet decomposition @xcite of the ecg signal . similarly , the respiratory phase was extracted from the respiration signal . the final phase time - series were reached after protophase - phase transformation @xcite . a more detailed explanation of the phase estimation procedure is given in appendix [ app : phases ] . synchrogram . ( b ) synchronization index for ratios 2:8 and 2:9 as indicated . ( c ) time - evolution of the cardiac @xmath229 and respiratory @xmath230 frequency . , width=302 ] application of the inferential technique reconstructs the phase parameters that govern the interacting dynamics . [ fig13:cvs_sync](c ) shows the time - evolution of the cardiac and respiration frequencies . it is evident that the constant pacing of the breathing is well - inferred , and that the instantaneous cardiac frequency , i.e. `` heart rate variability '' , increases with time . the inferred parameters , and their correlations , is used to detect the occurrence of cardiorespiratory synchronization and the corresponding synchronization ratio . the synchronization evaluation @xmath231 , shown in fig . [ fig13:cvs_sync](b ) reveals the occurrence of transitions between the synchronized and non - synchronized states , and transitions between different synchronization ratios : from 2:8 ( i.e. 1:4 ) at the beginning to 2:9 in the later intervals . because the evaluation of the synchronization state is based on all of the given details about the phase dynamics , the proposed method not only detects the occurrence of transitions , but also describes their inherent nature . the results for @xmath232 were consistent with the corresponding synchrogram shown in fig . [ fig13:cvs_sync](a ) , but provided a clearer and less ambiguous indication of synchronization . and @xmath233 respectively . ( a)-(c ) coupling function @xmath126 from the first oscillator , and ( d)-(f ) @xmath141 from the second oscillator . the window time intervals were calculated at : @xmath234s for ( a ) and ( d ) ; @xmath235s for ( b ) and ( e ) ; and at @xmath236s for ( c ) and ( f ) . ] the functional relationships that describe the cardiorespiratory interactions are shown in fig . [ fig14_cvs_cfs ] . evaluated for three different time windows , the upper figures ( a)-(c ) show the coupling function @xmath126 from the cardiac oscillating activity , and the lower figures ( d)-(f ) show @xmath141 from the respiration oscillator . the form of the functions is complex , and it changes qualitatively over time [ fig14_cvs_cfs](a ) with ( b ) and ( c ) , or ( d ) with ( e ) and ( f ) . the influence from respiration to heart ( @xmath126 ) has a larger norm ( i.e. coupling strength ) than in the opposite direction , indicating that the predominant direction of coupling is from respiration to heart . one can also observe that @xmath126 in ( b ) and ( c ) is of a fairly regular sinusoidal form with a strong influence from respiration . this arises from the contribution of those base functions describing the direct influence of respiration ( for a detailed discussion see @xcite ) . furthermore , fig . [ fig14_cvs_cfs ] shows that the functional relationships for the interactions of an open ( biological ) system can in themselves be time - varying processes . we conclude that the method is effective , not only when applied to digital and analogue synthetic signals , but also in the analysis of signals from human cardiorespiratory system . unlike the synthetic signals , the cardiorespiratory signals are real , unpredictable , and subject to considerable time - variability . in this way , we were able to reconstruct the cardiac and respiratory frequency variabilities , estimate the direction of coupling , and detect the presence of cardiorespiratory synchronization and transitions between its different states . we also found that the form of the coupling functions themselves is a time - varying dynamical process . our parameter inference procedure can be applied with only minimal modification to any number @xmath0 of interacting oscillators within a general coupled - network structure . the notation of eq . ( [ eq : phi ] ) is readily generalized for the @xmath0 oscillators , and the inference procedure , eq . ( [ eq : cd ] ) , is then applied to the corresponding @xmath0-dimensional phase observable . for example , if one wants to include all @xmath47-tuple interactions with @xmath237 , then eq . ( [ eq : phi ] ) would be generalized into @xmath238 every function @xmath239 is periodic on the @xmath47-dimensional torus , and can be decomposed in the sum of fourier @xmath47-dimensional series of trigonometric functions . although , this decomposition is theoretically possible , it becomes less and less feasible in practice as the number of oscillators and the number of @xmath47-tuples are increased . as a general approach , one could limit the number of base functions to the most significant fourier terms per @xmath239 functions ; but the task of finding the most significant component is algorithmically demanding in itself . first , a very fast algorithm for the @xmath47-dimensional space ( such as @xcite ) is required . secondly , since we have the value of each @xmath240 only at sparse values of the @xmath4 s that appear as argument in each @xmath239 , the algorithm should be adapted to deal with sparse , @xmath47-dimensional data , as the one recently developed in @xcite . it is needless to say that , a part of the computational speed aspects , the overall number of base functions should anyway be much less than the number of observed data . in view of these difficulties , automatic selection of the most important fourier terms to be used as base functions is hard to achieve on a network of more than just a few oscillators . known information about the system should be used to reduce the number of base functions such that only those terms relevant to the @xmath0-oscillator dynamics are included . other sub - procedures like the time - varying propagation , and the noise inference , apply exactly as before . we note that the computational power required increases very fast with @xmath0 , as discussed in appendix [ app : comput ] , which makes the method unsuitable for the inference of large - scale networks . however , for relatively small networks , a standard high - performance personal computer will suffice for useful inference . we first demonstrate the inference on three interacting poincar oscillators subject to noise @xmath241 where many of the coefficients @xmath242 and @xmath243 are initially set to zero ; but some are non - zero , such as when the first oscillator is pairwise coupled to the second and third oscillators . the second oscillator is coupled also to the first ( forming a bidirectional interaction ) . the third oscillator is influenced by the join contribution from the first and second oscillators . the latter coupling means physically that part of the network ( cluster ) exhibits a common functional influence on the other oscillators . the inference of this cross - coupling is the direct benefit of network ( rather than pairwise ) coupling detection . [ tab : par ] .parameters used for numerical simulation of systems ( [ equ : num_model ] ) . note that the indexes @xmath50 and @xmath47 are introduced only for easier notation of the generic coupling amplitudes @xmath244 . [ cols="^,^,^,^,^,^,^,^",options="header " , ] the inference of the three - dimensional phase variables from a numerical simulation of the network ( [ equ : num_model ] ) is presented in fig . [ fig : net_cfs ] . the plots present the specific forms of coupling function that govern the interactions within the network . the coupling strengths are evaluated as partial norms from the relevant base functions . note that the cross - couplings ( c ) , ( f ) and ( i ) are shown for visual presentation as functions dependent on two phases , whereas the coupling strengths include also the base function dependent on all the three phases . in order to determine whether the inferred couplings are real or spurious , we conducted surrogate testing . the detected couplings were tested for significance in respect of 100 couplings evaluated from surrogate phases . cyclic surrogates @xcite were generated from each of the phases , randomizing the temporal cross correlations , while preserving the frequencies and statistical characteristics unchanged . recently , kralemann et al . @xcite discussed the notion of effective and structural connectivity in networks . effective couplings are those that are detected , while not present in the original structure e.g.indirectly-induced coupling . in our numerical examples , the structural couplings are the parameters from the numerical simulation , while the effective are those evaluated as partial norms from the inferred parameters . the question posed was : are the effective couplings real , or are they artifacts ? our analysis showed that when one applies appropriate surrogate testing , the technique is able to distinguish the structural couplings as being significant . the resultant coupling strengths and significance @xmath83-values in fig.[fig : net_cfs ] suggest that the connectivity ( black - boxed couplings ) of the network ( [ equ : num_model ] ) was inferred correctly . note that some relations as in ( g ) have relatively large strength , even though they are less significant then some lower couplings as in ( e ) . if the possibility of effective couplings can not be excluded , then our technique ( with use of surrogate testing ) provides a consistent way of inferring the true structure of the network . it is also important to note that the coupling strength is evaluated as a partial norm and its value is not necessarily equal to the structural value , but is only proportional to it . when one infers complex networks , it is not only important what the structural coupling value is , but also how the oscillators are coupled and what are the coupling functions between the oscillators . more importantly , the use of our method allows one to follow the time - variability of the structural and functional connectivity within the network . this is especially important when inferring the interactions of biological oscillators , for which it is known that the dynamics is time - varying @xcite . to illustrate the latter we infer the following network of four phase oscillators subject to white gaussian noise @xmath245 note that , because the coupling strengths are functions of time , we were effectively changing the structural connectivity of the network by varying their values . the parameter values for the simulations were : @xmath246 , @xmath247 , @xmath248 , @xmath249 , @xmath250 , and noise strengths @xmath251 . the couplings were varied discreetly in three time - segments , as follows . ( i ) for 0 - 500s : @xmath252 , @xmath253 , @xmath254 and @xmath255 . ( ii ) for 500 - 1000s : @xmath256 , @xmath257 , @xmath258 and @xmath255 . ( iii ) for 1000 - 1500 : @xmath259 , @xmath257 , @xmath258 and @xmath260 . the coupling @xmath261 was continuously varied between @xmath262 . note also that in eq . ( [ equ : num_model2 ] ) the coupling functions are qualitatively different i.e. the arguments in the sine functions are not the same for each oscillator . for example the coupling functions for @xmath263 , @xmath264 and @xmath265 have one phase argument , while the coupling functions for @xmath261 and @xmath266 have the phase difference as their argument . the last two are additionally different because the coupling function with @xmath261 for the second oscillator contains its own phase @xmath233 in the phase difference . the results are presented in fig . [ fig : net_tv ] . in the first interval ( 0 - 500s ) we inferred three pairwise coupling amplitudes @xmath261 , @xmath263 and @xmath265 , and also one joint coupling @xmath266 which results from the joint influences of the second and fourth oscillators on the third one . the schematic diagram above the 0 - 500s time interval represents the structural connectivity , where the arrows indicate the direction of influence between the oscillators . on the transition to the second interval ( 500 - 1000s ) two of the couplings @xmath263 , @xmath266 disappear and one new one @xmath264 appears . this change occurs discretely at the instant of transition between the two regions . two couplings continue to exist : @xmath265 at a constant level , while @xmath261 decreases linearly and continuously . the second schematic diagram shows the structure of the network in this interval . comparing the diagrams describing the first two intervals one may note that the method infers correctly the couplings and their time - variability , and by doing so it infers the network connectivity even though it is changing with time . similarly the transition to the third interval ( 1000 - 1500s ) detects the alternations of two couplings @xmath265 and @xmath263 . this leads to a new connectivity state of the network , as presented in the third schematic diagram . the results from the whole time span demonstrate that the method follows the time - variability of the couplings effectively and precisely . the dynamical variations are taking the network structure through various different connectivity states , and the different topologies are detected reliably throughout their time - evolution . ( color online ) inference of time - varying coupling structure for the network ( [ equ : num_model2 ] ) . the color / grayscale code for the couplings is presented in the box at the top , where @xmath261 is represented by a dotted line , @xmath263 by a dashed line , @xmath264 by a dash - dotted line , @xmath266 by a bold full line and @xmath265 by a light full line . the four couplings @xmath263 , @xmath264 , @xmath266 and @xmath265 were held constant at different values within three time segments each of length 500s . however , @xmath261 was varied continuously through the whole time interval . for each segment the structure of the network is presented schematically on the diagrams in the dashed grey boxes . the parameters are given in the text . ] starting from the perspective of dynamical systems inference , we have built an algorithm able to detect synchronization , to describe the functional form of the mutual interactions between oscillators , and to perform such tasks successfully in the presence of a time - evolving dynamics . the algorithm differs substantially from earlier approaches with respect both to synchronization detection capabilities , and to the estimation of coupling and directionality . most other techniques are based on information flow ( e.g. transfer entropy , or granger causality ) providing them with great generality . while limiting ourself to the hypothesis of continuous time differential equations driving the dynamics ( correspondingly restricting the domain of applicability ) , we can optimally exploit the benefits of this assumption . unlike all other approaches , our technique does not require the observable to fill the domain of the probability density function at equilibrium . thus , in oscillators with a limit cycle ( van der pol , fitz - hugh nagumo , poicar , etc ... ) even one single extreme path is sufficient to characterize the parameters of the dynamics . hence , we can determine uniquely the limit - time equilibrium distribution , i.e. the fokker - plank equation associated with the sde . thus an immediate advantage is that we can extract the same information from a fraction of the volume of data that is typically required by earlier methods . because a very wide range of natural and artificial systems are describable in terms of continuous time differential equations ( e.g. oscillatory processes in nature , mechanical systems , analogue voltage systems ) , the loss of generality in our approach is actually minimal , compared to the advantage gained in terms of informational efficiency . we have applied the algorithm successfully to a representative classes of oscillators , testing it on synthetically - generated data created from various models , and on data from an analogue circuit device with known dynamics . in each case , we were able to demonstrate the precision of parameter detection , the temporal precision of synchronization detection , and the accuracy of directionality identification . we have also demonstrated the efficacy of the technique in relation to cardiorespiratory time series data . synchronization phenomena were already well - known in such systems , but the details of functional coupling were not . from the inferred parameters we were able to reconstruct the extent of the cardiac and respiratory variability , estimate the direction of coupling , and detect the presence of and type of intermittent cardiorespiratory synchronization . because the whole enterprize is built on an inference algorithm for an @xmath0-dimensional dynamical system , the technique was readily extensible to the study of a network of oscillators whose parameters and coupling functions may be changing in time . an example of such application we considered a network of poincar oscillators , generated by numerical simulation . we were able to demonstrate effective coupling detection , cross - validating the results by surrogate testing . although the implementation itself might see future improvements ( e.g. in terms of speed of calculation , or automatic base function selection ) , it is worth emphasizing that the method allows one to designate which components of the system are expected to be time - variable . such selection is optional , but it provides an effective means by which to incorporate previous knowledge available for any particular system , and enables the algorithm to adapt itself optimally to the externally - imposed constraints . given the advantages that the dynamical approach offers in tackling synchronization detection and coupling identification , we believe that the framework presented above will be found valuable for a wide range of future applications . the procedure of synchronization detection between two oscillators generating phase time - series reduces to the investigation of synchronization of the synthetic phase model using the parameters returned by the bayesian algorithm . to calculate @xmath267 for any of the sampled parameter sets , one can proceed as follows : * from an arbitrary fixed @xmath268 , and for an arbitrary @xmath269 , integrate numerically ( using the standard fourth - order runge - kutta algorithm ) the dynamical system prescribed by the phase base function ( eq . ( [ eq : phif ] ) without the noise ) for one cycle of the toroidal coordinate , obtaining the mapped point @xmath270 . * repeat the same integration for multiple @xmath271 coordinates next to the initial one , obtaining the map @xmath272 * based on finite difference evaluation of @xmath273 , use a modified version of newton s root - finding method to analyse the function @xmath274 . the method is modified by calculating @xmath96 at the next point @xmath116 such that @xmath275 the coefficient @xmath276 is an arbitrary constant that we found to be particularly efficient for solution of the problem . note that in this version , newton s method can only test the function by moving forward ; in actual fact ( a ) the existence of the root is not guaranteed ; and ( b ) we are not interested in the root itself but only in its existence . * if there is a root , @xmath203 is returned . if a root is not found , @xmath277 is returned . in order to infer the phase dynamics , one needs to have good estimates of the phases from the observable time - series . this is even more important when the oscillatory dynamics is time - varying and the analysis requires instantaneous phases . potential difficulties for phase estimation arise when the signals emanate from complex , highly nonlinear and/or mixed - mode oscillatory dynamics . although the phase from the respiration signal is relatively easy to detect , obtaining the instantaneous phase from the ecg signal is considerably more difficult . we used the synchrosqueezed wavelet transform @xcite to estimate phases from the complex and nonlinear ecg and respiration signals . given a signal @xmath278 we first calculate its wavelet transform in the scale - time domain @xmath279 , @xmath280 where the @xmath281 represents the complex conjugate of the mother wavelet @xmath282 @xmath283 we use the morlet mother wavelet @xmath284{\pi}}\text { } e^{-i 2\pi f_0 u}\cdot e^{-u^2/2 } , \nonumber \label{equ : wavelet}\ ] ] where the central frequency was set to be @xmath285 . the synchrosqueezed transform aims to `` squeeze '' the wavelet around the intrinsic frequency in order to provide better frequency localization . for any @xmath279 for which @xmath286 , a candidate instantaneous frequency for the signal @xmath287 can be calculated as @xmath288 the information from the time - scale plane is transferred to the time - frequency plane , according to a map @xmath289 , in an operation called synchrosqueezing . the synchrosqueezed wavelet transform is then expressed as @xmath290 where @xmath291 , and @xmath292 is as defined in ( [ equ : freq ] ) above , for @xmath293 such that @xmath294 . the complex ( as with real and imaginary values ) nature of the synchrosqueezed transform allows one to extract the phase of the signal as the angle of the transform @xmath295.\ ] ] the transform s great advantage lies in its ability to determine instantaneous characteristics from complex signals with non - harmonic waveforms . evaluated through such a procedure the phases @xmath296 may , however , be observable - dependent and non - universal , i.e. they can retain premises resulting from the phase - detection technique ( in this case the synchrosqueezed transform ) but not from the genuine phases . they are therefore treated as protophases , and a special technique is applied to transform the protophases into true phases @xmath297 that are independent of the observable and are universally defined @xcite . the transformation can be written as @xmath298 where @xmath299 are coefficients from a fourier expansion of the averaged phase relationships . for further detail see @xcite . for a sufficiently large number of parameters @xmath96 , the complexity of the algorithm for parameter estimation is substantially dominated by o@xmath300 , which is of the order of the time required for an @xmath96-sized matrix inversion . for a network of @xmath0 oscillators , if one consider all possible pairwise connections then @xmath301 , where @xmath302 is a proportionality coefficient to account for by the truncation order of the fourier decomposition . similarly , if one considers all pairwise connections and every double connection up to a truncation order of @xmath303 , then @xmath304 . with recursive reasoning , if one considers all the @xmath47-tuples with @xmath47 up to @xmath305 , each with a truncation order of @xmath306 , then the number of coefficients would grow as @xmath307 it is clear that even for a modest network , considering just a few @xmath47-tuples of possible connections would be unfeasible in practice . very careful selection of the base functions is therefore always to be recommended .
star formation efficiency controlled by the protostellar outflow in a single cloud core is investigated by three - dimensional resistive mhd simulations . starting from the prestellar cloud core , the star formation process is calculated until the end of the main accretion phase . in the calculations , the mass of the prestellar cloud is parameterized . during the star formation , the protostellar outflow is driven by the circumstellar disk . the outflow extends also in the transverse direction until its width becomes comparable to the initial cloud scale , and thus , the outflow has a wide opening angle of . as a result , the protostellar outflow sweeps up a large fraction of the infalling material and ejects it into the interstellar space . the outflow can eject at most over half of the host cloud mass , significantly decreasing star formation efficiency . the outflow power is stronger in clouds with a greater initial mass . thus , the protostellar outflow effectively suppresses star formation efficiency in a massive cloud . the outflow weakens significantly and disappears in several free - fall timescales of the initial cloud after the cloud begins to collapse . the natal prestellar core influences the lifetime and size of the outflow . at the end of the main accretion phase , a massive circumstellar disk comparable in mass to the protostar remains . calculations show that typically ,% of the initial cloud mass is converted into the protostar and% remains in the circumstellar disk , while% is ejected into the interstellar space by the protostellar outflow . therefore , a single cloud core typically has a star formation efficiency of% .
1108.3564
many protostellar outflows are observed in star - forming regions . these outflows are believed to be universally driven by a protostar and play a critical role in star formation@xcite . the protostellar outflow may determine star formation efficiency , especially in low - mass star formation process . the similarity between the core mass function and the initial mass function implies that only a certain fraction of the prestellar core is converted into the star @xcite . recent observations have shown that , in a single prestellar core , star formation efficiency is limited to @xmath5 @xcite . @xcite argued that star formation efficiency might be limited by the protostellar outflow . with a simple analytical approach , they showed that a wide - opening - angle outflow sweeps up the gas in the infalling envelope and ejects it into the interstellar space . they concluded that the protostellar outflow can limit star formation efficiency to @xmath6 . in addition , feedback from the protostellar outflow may maintain interstellar turbulence and affect subsequent star formation @xcite . thus , the protostellar outflow is crucial in ( low - mass ) star formation . however , it is quite difficult to model the outflow using only an analytical approach because the outflow driving depends on various conditions such as the infalling mass rate onto the circumstellar disk , the size of the circumstellar disk , the configuration and the strength of the magnetic field around the protostar . thus , numerical simulation is necessary to understand the driving and evolution of the outflow during star formation and to determine star formation efficiency , as controlled by the protostellar outflow , in more detail . some authors have used a spherically symmetric calculation to investigate the evolution of a prestellar cloud until protostar formation @xcite . now , we can directly calculate the star formation from the prestellar cloud stage until protostar formation with a multi - dimensional calculation @xcite . before protostar formation , the first ( adiabatic ) core forms @xcite and drives a wide - opening - angle outflow @xcite . after protostar formation , the first core evolves directly into the circumstellar disk after the protostar formation @xcite . since the first core transitions smoothly to the circumstellar disk , the outflow driven by the first core before protostar formation is driven by the circumstellar disk after protostar formation without a transient disappearance . in summary , a wide - opening - angle outflow appears before protostar formation and continues to be driven after protostar formation until the end of the main accretion phase @xcite . this type of outflow is believed to correspond to the molecular outflow frequently observed in star - forming regions @xcite . recently , a bipolar molecular outflow was observed around the candidate first core per - bolo 58 @xcite . in addition , some first core candidates were reported by several authors @xcite . thus , to estimate the mass ejection rate from the host cloud or star formation efficiency , we need to calculate the evolution of the protostellar outflow from the prestellar core stage because a wide - opening - outflow appears before protostar formation . in addition to the wide - opening - angle outflow , well collimated high - velocity jets are observed in the star forming region @xcite . such a jet also appears during star formation process , and is driven in very close proximity to the protostar @xcite . the driving source and its spatial scale of each flow ( wide - opening - angle outflow and well collimated jet ) are considerably different : the driving source for the wide - opening - angle flow is the first core or the circumstellar disk , which has a size of @xmath7au , whereas the driving source for the well collimated jet is the protostar and the disk near the protostar in the region of @xmath8au @xcite . in addition , the mass ejection rate of the wide - opening - angle flow is much higher than that of the well collimated jet . because the jet has a good collimation , it can not accumulate sufficient mass in the infalling envelope to reduce star formation efficiency @xcite . note that the difference in collimation between the wide - opening - angle outflow and well collimated jet is caused by the driving mechanism and the configuration of the magnetic field lines around the driving source @xcite . a statistical study of molecular outflows observed in various star forming regions also indicates that the ( wide - opening - angle ) molecular outflows have a sufficient energy or momentum to control the star formation @xcite . therefore , the wide - opening - angle outflow and not the well collimated jet is expected to be important in determining star formation efficiency . thus , to estimate the mass ejection rate ( or star formation efficiency ) , we must resolve the driving source of the wide - opening - angle outflow , whereas we do not necessarily need to resolve the protostar itself , which drives the well collimated jet . since previous studies spatially resolved the protostar , they could not directly estimate the resulting mass ejection rate by the protostellar outflow and star formation efficiency @xcite . they calculated the evolution of the protostellar outflow only for a short duration of @xmath9yr at most , whereas the main accretion phase lasts for at least @xmath10yr . it is difficult to calculate the evolution of the protostellar outflow for a long duration by resolving the protostar itself because the timescales ( time steps ) in the regions around the protostar and the molecular cloud are quite different . on the other hand , without resolving the driving source of the protostellar outflow , some studies focused on the long - term evolution of the protostellar outflow propagating into the interstellar medium . in these studies , the outflow is input artificially in the computational domain ( e.g. , * ? ? ? in addition , the circumstellar disk , the configuration , and the strength of the magnetic field are also adopted arbitrarily . even with this type of calculation , we can not estimate star formation efficiency because the mass ejection rate is artificially assumed in such studies . at the expense of spatial resolution around the protostar , @xcite calculated the evolution of the collapsing cloud from the prestellar stage until the end of the main accretion phase . in their study , the driving source of the outflow ( i.e. , the first core and the circumstellar disk ) was well resolved spatially , whereas the protostar was not resolved and was replaced by sink cells . they pointed out that the protostellar outflow reduces star formation efficiency to @xmath11 . however , they investigated only the evolution of a low - mass cloud core ( @xmath12 ) and did not investigate the evolution of a cloud with a typical mass scale of @xmath13 ( e.g. , * ? ? ? * ; * ? ? ? * ) . in this study , using conditions similar to those in @xcite , we calculate the evolution of a cloud with various initial masses and investigate the evolution of the outflow and the impact of the protostellar outflow on star formation efficiency . this paper is structured as follows . the framework of our models and the numerical method are described in 2 . the numerical results are presented in 3 . the mass ejection rate and star formation efficiency are discussed in 4 . we summarize our results in 5 . to investigate star formation efficiency and long - term evolution of the protostellar outflow , we calculate the star formation process from the prestellar core stage until the end of the main accretion phase for clouds with various masses using three - dimensional resistive mhd equations , including self - gravity : @xmath14 where @xmath15 , @xmath16 , @xmath17 , @xmath18 , @xmath19 , and @xmath20 denote the density , velocity , pressure , magnetic flux density , resistivity , and gravitational potential , respectively . to mimic the temperature evolution calculated by @xcite , we adopt the piece - wise polytropic equation of state ( see * ? ? ? * ; * ? ? ? * ) as@xmath21 , \label{eq : eos}\ ] ] where @xmath22ms@xmath23 and @xmath24 ( @xmath25 ) . equation ( [ eq : eos ] ) shows that the gas behaves isothermally for @xmath26 and adiabatically for @xmath27 . for the realistic evolution of the magnetic field during star formation , we adopt a resistivity ( @xmath19 ) of the fiducial value in @xcite , in which ohmic dissipation becomes effective for @xmath28 ( for details , see eqs . [ 9 ] and [ 10 ] and fig . 1 of @xcite ) . this study investigates star formation efficiency in a single cloud . thus , we assume an isolated cloud core embedded in the interstellar medium . as the initial state , we take a spherical cloud with a critical bonnor ebert ( be ) density profile @xmath29 , in which a uniform density is adopted outside the sphere ( @xmath30 , where @xmath31 is the critical be radius ) to mimic the interstellar medium . the gravitational force is ignored outside the be sphere ( @xmath32 ) to avoid the inflow of gas from outside the isolated core ( i.e. , from the interstellar medium ) . in addition , to strictly avoid mass inflow from outside the core , we prohibit mass inflow at @xmath33 . note that we do not prohibit mass outflow at the boundary between the be sphere and the interstellar medium ; the gas escapes freely from the be sphere by the protostellar outflow . hereafter , we call the gravitationally bound gas cloud within @xmath34 the host cloud . we confirmed that the total mass of the host cloud is well conserved during the calculation before the protostellar outflow reaches the cloud boundary . note that the total mass inside the host cloud is not conserved after the protostellar outflow propagates into the interstellar space ( @xmath35 ) , because the mass is ejected from the host cloud by the protostellar outflow . since the critical be sphere is in equilibrium , we increase the density by a factor of @xmath36 to promote contraction , where @xmath36 is the density enhancement factor that represents the stability of the initial cloud . the density profile of the initial cloud is described as @xmath37 where @xmath38 is the density distribution of the critical be sphere , and @xmath39 is the axisymmetric density perturbation . an initial cloud with larger @xmath36 is more unstable against gravity . the cloud stability is generally represented by a parameter @xmath40 ( @xmath41 ) , which is the ratio of thermal ( @xmath42 ) to gravitational ( @xmath43 ) energy . as shown in @xcite , when the be density profile is adopted , the density enhancement factor is related to the parameter @xmath40 as @xmath44 we adopt a density enhancement factor as @xmath45 , which corresponds to @xmath46 . the density contrast between the center of the cloud ( @xmath47 ) and the cloud boundary ( @xmath48 ) is @xmath49 . in addition , a uniform density of @xmath50 ( i.e , @xmath51 of the central cloud density ) is adopted outside the sphere ( @xmath30 ) . to break the axial symmetry , we add a small amount of @xmath52-mode non - axisymmetric density perturbation to the initial core . for the @xmath52-mode , in equation ( [ eq : dens ] ) , we choose @xmath53 where @xmath54 ( = 0.01 ) represents the amplitude of the perturbation . the radial dependence is chosen so that the density perturbation remains regular at the origin ( @xmath55 ) at one time - step after the initial stage . this perturbation develops into a non - axisymmetric perturbation in the circumstellar disk that contributes to angular momentum transfer . in addition , this @xmath52 perturbation ensures that the center of gravity is always located at the origin . for a dimensional be density profile , we adopt an isothermal temperature of @xmath56k and a central number density of @xmath57 , where @xmath58 is a parameter in the range @xmath59 . since the temperature of each initial cloud is fixed , the size ( i.e. , radius and mass ) of the be sphere is uniquely determined only by the parameter @xmath58 . thus , each model is characterized only by the initial central density , @xmath60 . with these parameters , the critical be radius ( or radius of the host cloud ) is @xmath61au . the mass inside @xmath62 for each model is @xmath63 . the host cloud radius and mass for each model are listed in table [ table:1 ] . in each model , the cloud rotates rigidly around the @xmath64-axis in the @xmath65 region and a uniform magnetic field parallel to the @xmath64-axis ( or rotation axis ) is adopted in the entire computational domain . the magnetic field strength and the rotation rate are scaled using the central density @xmath66 as @xmath67 @xmath68 in all models , we adopt @xmath69 and @xmath70 , which are the most appropriate parameters for driving strong outflow in the collapsing cloud @xcite . since the magnetic field and rotation are normalized by the central density and each model is characterized by only the central density , each cloud has a different magnetic field strengths and angular velocity . the magnetic field ( @xmath71 ) and angular velocity ( @xmath72 ) for each model are summarized in table [ table:1 ] . however , all models have the same ratios of rotational and magnetic energies to the gravitational energy , @xmath73 ( @xmath74 ) and @xmath75 ( @xmath76 ) , where @xmath77 and @xmath78 are rotational and magnetic energies , respectively . in addition , all models have the same mass - to - flux ratio @xmath79 . there exists a critical value of @xmath79 below which a cloud is supported against gravity by the magnetic field . for a cloud with uniform density , @xcite derived a critical mass - to - flux ratio @xmath80 where the constant @xmath81 @xcite . the mass - to - flux ratio normalized by the critical value @xmath82 is described as @xmath83 all models have @xmath84 , which is slightly larger than the typical value of observation . the observations indicate that molecular cloud cores have the mass - to - flux ratio in the range of @xmath85 with a median value of @xmath86 @xcite . the model names and parameters are also listed in table [ table:1 ] . we believe that these energy ratios are adequate for comparing cloud evolution among models with different masses . we can match the magnetic field strength ( @xmath71 ) and rotation rate ( @xmath72 ) for clouds with different masses . however , since doing so changes the evolution of the cloud and the protostellar outflow , it is difficult to compare the star formation efficiency of the models . for example , when the initial magnetic field strength is fixed among the models , no outflow may appear in a less massive cloud because it has a small ratio of magnetic energy to gravitational energy . in this paper , to compare cloud evolution among models , we usually use the freefall timescale at the center , @xmath47 ( @xmath87 ) , and boundary , @xmath33 ( @xmath88 ) , of the initial cloud as the unit of time . since the density contrast between the center of the cloud and its boundary is 14 , the freefall timescale of the cloud boundary is about 3.7 times longer than that at the center of the cloud ( @xmath89 ) . the freefall timescale at the center of the initial cloud ( @xmath87 ) for each model is listed in table [ table:1 ] . to realize the long - term calculation of star formation , we adopt a sink at the center of the cloud . we start the calculation without a sink and calculate the cloud evolution for the prestellar gas collapse phase without a sink . later , we identify protostar formation in the collapsing cloud core when the number density exceeds @xmath90 at the cloud center , where @xmath91 is the threshold density . after protostar formation , we calculate the cloud evolution with the sink . to model the protostar , we adopt a fixed sink with a radius of @xmath92au composed of sink cells only around the center of the computational domain . since we add only @xmath52 density perturbation , as described in [ sec : setting ] , the protostar ( or center of gravity ) does not move and remains at the center of the computational domain during the calculation . in the region @xmath93au , the gas having a number density of @xmath94 is removed from the computational domain and added to the protostar as a gravitating mass in each time step ( for details , see * ? ? ? thus , for each time step , the accretion mass onto the protostar is calculated as @xmath95\ , dv.\ ] ] in addition , inside the sink , the magnetic flux is removed by ohmic dissipation because this region has a magnetic reynolds number @xmath96 exceeding unity @xmath97 ( for details , see * ? ? ? * ) . for calculation on a large spatial scale , the nested grid method is adopted ( for details , see * ? ? ? * ; * ? ? ? each level of a rectangular grid has the same number of cells ( @xmath98 ) . the calculation is first performed with five grid levels ( @xmath99 ) . in all models , the fifth level of the grid ( @xmath100 ) has a box size of @xmath101 and just covers the entire region of the isolated be sphere . the first level of the grid has a box size of @xmath102 and is filled with low - density interstellar medium outside @xmath32 . thus , we can calculate the propagation of the protostellar outflow in the region of @xmath103 . the protostellar outflow never reaches the computational boundary by the end of the calculation . in addition , we have checked that the alfv@xmath104n waves generated at the center of the cloud ( or the computational boundary ) never reaches the computational boundary ( or the center of the cloud ) during the calculation ( for details , see * ? ? ? * ) . after the calculation starts , a new finer grid is generated before the jeans condition is violated @xcite . although the maximum grid level differs among the models , each model has a spatial resolution of @xmath105au in the finest grid . in this section , we present the evolution of the collapsing cloud and the outflow for typical models in [ sec : typical ] ( model n08 , less massive cloud case ) and [ sec : typical2 ] ( model n06 , massive cloud case ) . then , we describe the mass accretion rate ( [ sec : accretion ] ) and masses of the protostar and circumstellar disk ( [ sec : mass ] ) , and the mass ejected by the protostellar outflow ( [ sec : out ] ) . the properties of the protostellar outflow are described in [ sec : out2 ] . figure [ fig:1 ] plots the cloud evolution for model n08 from the initial state until the end of the main accretion phase ; only the @xmath106 region is shown . note that although the structure only around the initial cloud scale ( @xmath107 ) is plotted in figure [ fig:1 ] , the computational domain has a size of @xmath108 . as seen in figure [ fig:1]_a _ , we adopted a spherical cloud with the be density profile and the radius of @xmath109 as the initial state . as described in [ sec : setting ] , since we ignored the gravity outside the be sphere ( @xmath110 ) , only the gas in the @xmath111 region ( inside the white dotted line in fig . [ fig:1 ] ) can collapse to fall onto the center of the cloud . for this model , the number density exceeds @xmath90 and a protostar forms at @xmath112 , where @xmath87 ( @xmath113yr ) is the freefall timescale of the initial cloud at the center . figure [ fig:1]_b _ shows the density distribution just after protostar formation . the red contours of @xmath114 in figure [ fig:1]_a _ and _ b _ indicate that the cloud gradually contracts toward the center with time . the blue contour in the figure corresponds to the boundary of the protostellar outflow , inside which the gas moves outward against the center of the cloud . thus , this contour shows the shape of the protostellar outflow . the outflow is driven by the circumstellar disk that originates from the first core , which appears before protostar formation @xcite . after protostar formation , the first core becomes the circumstellar disk @xcite . the outflow is then driven by the circumstellar disk after a smooth transition from the first core to the circumstellar disk @xcite . the protostellar outflow remains in the host cloud ( @xmath111 ) for @xmath115yr ( fig . [ fig:1]_c _ ) ; it reaches the boundary of the host cloud at @xmath116yr ( fig . [ fig:1]_d _ ) and penetrates the host cloud boundary and propagates into the interstellar space ( @xmath110 ) for @xmath117yr ( fig . [ fig:1]_e_-_h _ ) . the outflow driving halts inside the host cloud at @xmath118yr ( @xmath119 ) . the infalling envelope is depleted and the mass accretion is almost over by this epoch ( [ sec : mass ] ) . thus , the circumstellar disk can not drive the outflow at this epoch because the outflow is powered by gas accretion onto the circumstellar disk . since the freefall timescale of the cloud boundary is @xmath120yr , a large fraction of the gas inside the host cloud has already fallen onto either the circumstellar disk or the protostar by this epoch . although the outflow at the host cloud scale completely disappears in @xmath121yr ( = @xmath122@xmath87 = @xmath123 ) , the density cavity formed by the protostellar outflow remains around the host cloud as seen in figure [ fig:1]_i_. figure [ fig:1 ] also shows that the infalling envelope in the host cloud depletes with time . part of the gas inside the initial host cloud is ejected by the protostellar outflow , and the remainder falls onto either the circumstellar disk or the protostar . in figures [ fig:1]_h _ and _ i _ , we can see only a disk - like structure with the size of @xmath124200au , because the infalling gas is depleted by this epoch . at these epochs , the density of the infalling envelope ( @xmath125 ) is less than @xmath126 the initial cloud density ( @xmath127 ) . at the end of the calculation ( @xmath128 ) , the mass ratio of the infalling envelope to the total mass of the initial cloud is only @xmath129 . thus , the main accretion phase has already ended by this epoch . the evolution of the outflow configuration is shown in figure [ fig:2 ] . each panel corresponds to the same epoch as in figure [ fig:1]_d _ , _ e _ , _ f _ , and _ h_. figure [ fig:2 ] shows that the protostellar outflow propagates into the interstellar space while maintaining good collimation . however , figure [ fig:2]_a _ and _ b _ indicate that , early in the evolution , the outflow also extends in a transverse direction and increases in width . after its width becomes comparable to the host cloud scale , the outflow extends only in the vertical direction while maintaining its width . the magnetic field lines that guide the outflow are anchored by the host cloud ( or the gravitationally bound sphere ) . thus , the final width of the outflow is comparable to that of the host cloud . therefore , the outflow collimation improves with time after the outflow width becomes comparable to the host cloud scale , whereas the collimation is not good when its width is smaller than the host cloud scale or the outflow remains within the host cloud . figures [ fig:3 ] and [ fig:4 ] show the density and velocity distributions for model n06 at @xmath130yr ( @xmath131=@xmath132 ) . for this model , the protostar forms at @xmath133yr after the cloud begins to collapse . thus , the figures show the structure @xmath134yr after the protostar formation . by this epoch , the outflow penetrated the host cloud that has a radius of 4800au and reaches @xmath135au from the center of the cloud , as seen in figure [ fig:3 ] left panel . the figure also shows the bow shocks caused by the protostellar outflow at @xmath136au . the upper right panel in figure [ fig:3 ] plots the structure around the host cloud and shows that the gas flows out from the host cloud in the vertical direction by the protostellar outflow . in addition , inside the white dotted line that denotes the size of the initial cloud , the gas density is considerably lower than the ambient medium . this is because a part of the gas falls onto the central region to form the protostar and circumstellar disk , while the remainder is ejected from the host cloud by the outflow . the lower right panel is 16 times magnified view of the central region of the upper right panel . this panel shows that the outflow is strongly driven by the disk - like structure around the protostar . on this scale , the gas accretes onto the protostar through the circumstellar disk , and a part of the accreting mater is ejected by the outflow . figure [ fig:4 ] shows the structure around the circumstellar disk on @xmath137 ( left panel ) and @xmath138 ( right panel ) planes . these panels show that the rotating disk exits around the protostar and drives the outflow . thus , it is expected that the outflow is mainly driven by the magnetocentrifugal mechanism @xcite . figures [ fig:3 ] and [ fig:4 ] indicate that large scale outflow with a size of @xmath139au ( fig . [ fig:3 ] left panel ) originates from the disk wind driven by the circumstellar or rapidly rotating disk with a size of @xmath140au . the disk wind propagates into the infalling envelope , and thus it sweeps and collects a larger fraction of the infalling matter . finally , a large fraction of the swept material is expelled from the host cloud . thus , the outflow significantly affects the final protostellar mass or star formation efficiency . figure [ fig:5 ] shows the mass accretion rate onto the protostar and the protostellar mass against the time after protostar formation @xmath141 . here , we describe the time after the protostar formation as @xmath141 , which is defined as @xmath142 where @xmath143 is the elapsed time after the cloud begins to collapse ( or the calculation starts ) , and @xmath144 is the protostar formation epoch and is listed in table [ table:1 ] . in each model , the gas density exceeds @xmath145 , and the protostar forms @xmath146 after the cloud begins to collapse ( table [ table:1 ] ) . note that model n39 shows no continuous collapse and no protostar appears ; this is because the initial cloud density for model n39 is too high to induce continuous collapse . in our models , since the gas becomes adiabatic at @xmath147 ( [ sec : setting ] ) , the initial cloud density ( @xmath148 ) for model n39 is very close to this critical value @xmath149 . as a result , model n39 shows repeated contraction and expansion around the initial configuration , not continuous collapse . below , we describe only the models showing protostar formation ( n35 , n06 , n36 , n07 , n37 , n08 , n38 , n09 ) and do not mention model n39 again . as described in [ sec : sink ] , we removed the gas having a number density of @xmath145 in the region @xmath150au from the computational domain . we regarded the removed gas as the accreted mass onto the protostar and estimated the mass accretion rate in each time step . figure [ fig:5 ] shows that , in each model , the mass accretion rate is as high as @xmath151 just after protostar formation ( @xmath152yr ) . in theoretical analyses , the mass accretion rate is defined as @xmath153 , where @xmath36 is a numerical factor ( e.g. , @xmath154 in @xcite , @xmath155 in @xcite ) . since gas clouds have temperatures of @xmath56k ( @xmath156 ) , the accretion rate in the main accretion phase is @xmath157 . thus , the accretion rate derived in our calculations corresponds well to the theoretical expectation . then , in each model , the mass accretion rate gradually decreases with time , to @xmath158 at @xmath159 and @xmath160 at @xmath161 . thus , gas accretion almost halts and the protostar rarely increases in mass at @xmath162 . as shown in figure [ fig:5 ] , all models show a qualitatively and quantitatively similar mass accretion rate trend when the mass accretion rate onto the protostar is normalized by the freefall timescale of the initial cloud . however , since the real ( or dimensional ) time of the freefall timescale depends on the initial cloud density ( or the initial cloud mass ) , the duration of the main accretion phase is different in models with different cloud masses . a cloud with a larger mass ( or lower density ) has a longer duration of the main accretion phase . for example , model n35 with @xmath163 has the mass accretion of @xmath164 for @xmath165yr , while model n09 with @xmath166 has that only for @xmath167yr . figure [ fig:5 ] also shows that the protostar formed in a massive cloud is more massive than that formed in a less massive cloud . this is because the massive cloud has a longer gas accretion phase , so the protostar has enough time to acquire sufficient mass and evolves into a relatively massive star . in figure [ fig:5 ] , the mass accretion rate oscillates in models having an initially massive cloud . this is caused by a non - axisymmetric structure appearing in the circumstellar disk . since the circumstellar disk is more massive than the protostar just after protostar formation @xcite , a non - axisymmetric or spiral structure develops owing to gravitational instability . such structure effectively transfers angular momentum outward and intermittently promotes mass accretion onto the protostar . as a result , these models show time variability in the mass accretion rate @xcite . in figure [ fig:5 ] , models n35 , n06 , n36 , n07 and n37 show time variability in the mass accretion rate . the mass of the protostar , circumstellar disk , protostellar outflow , and infalling envelope are plotted against time after protostar formation @xmath141 in figure [ fig:6 ] . in addition , these masses at @xmath88 after protostar formation are listed in table [ table:2 ] . the circumstellar disk mass is estimated according to the formula in @xcite . to estimate the outflowing gas @xmath168 , we integrated the gas with velocity @xmath169 for the entire computational domain and subtracted the gas swept by the protostellar outflow outside the host cloud @xmath35 from the integrated mass . for later convenience , we divide the outflowing mass into two parts , @xmath170 where @xmath171 is the mass ejected from the host cloud , and @xmath172 is the outflowing mass having @xmath173 inside the host cloud ( @xmath111 ) . to calculate the mass of the infalling envelope , we calculated the total mass @xmath174 inside the host cloud ( @xmath34 ) . then , we subtracted the disk mass @xmath175 and the outflowing mass @xmath172 in the @xmath34 region from the total mass , @xmath176 in our models , the sum of the protostellar mass , circumstellar mass , and infalling envelope mass is not conserved inside the host cloud ( i.e. , @xmath34 ) because part of the gas is ejected from the host cloud by the protostellar outflow . however , we confirmed that the sum of the total mass and the protostellar mass is well conserved before the outflow reaches the boundary of the host cloud , as described in [ sec : setting ] . figure [ fig:6 ] shows that , in each model , the mass of the infalling envelope decreases to @xmath177 at @xmath161 ( table [ table:2 ] ) , where @xmath178 is the initial cloud mass . thus , the mass of the infalling envelope is depleted and the main accretion phase is almost over by this epoch . note that , in figure [ fig:6 ] , we subtracted the time ( @xmath144 ) until the protostar forms from the time ( @xmath143 ) after the cloud begins to collapse . thus , at the epoch @xmath88 , indicated by the arrow in figure [ fig:6 ] , a longer time than @xmath88 has passed since the cloud begins to collapse ( table [ table:1 ] ) . therefore , it is reasonable that almost all the gas has already fallen onto the center of the cloud at this epoch @xmath179 , because the gas falls onto the cloud center in several times freefall timescale @xcite and the freefall timescale at the cloud boundary ( @xmath88 ) has already passed by this epoch . figure [ fig:6 ] also indicates that the circumstellar disk mass dominates the protostellar mass just after protostar formation in each model . the circumstellar has been reported to originate from the first core @xcite , which is about 10 - 100 times more massive than the protostar at the protostar formation epoch @xcite . @xcite showed that the circumstellar disk is inevitably more massive than the protostar in the early main accretion phase because the first core evolves directly into the circumstellar disk after protostar formation . in the later main accretion phase , however , in some models , the protostellar mass dominates the circumstellar disk mass . in the initially massive clouds ( models n35 , n06 , and n36 ) , the protostellar mass continues to increase , whereas the circumstellar disk mass gradually decreases until the end of the main accretion phase . in addition , in models n35 and n06 , the protostar becomes more massive than the circumstellar disk for @xmath180 . the massive circumstellar disk becomes gravitationally unstable and tends to exhibit a non - axisymmetric structure that can transfer angular momentum outward ; the gas in the circumstellar disk effectively falls onto the protostar . therefore , the accretion rate onto the protostar from the circumstellar disk dominates the accretion rate onto the circumstellar disk from the infalling envelope , and the mass of the disk begins to decrease . figure [ fig:7 ] shows the density distribution on the equatorial plane around the center of the cloud at @xmath181 for models n35 , n06 , n08 and n09 . the figure shows that a rotating disk with a size of @xmath182au forms around the center of the cloud by this epoch . as seen in figure [ fig:7]_a _ , two clumps appear in model n35 , in which fragmentation occurs in the circumstellar disk about @xmath165yr after protostar formation . in models n35 and n06 , the circumstellar disk is rather massive in the early main accretion phase . as shown in figure [ fig:6 ] , the circumstellar disk has a mass of @xmath183 in model n35 and @xmath184 in model n06 in the early main accretion phase for @xmath185 . such a massive disk becomes gravitationally unstable and tends to develop a non - axisymmetric structure and show subsequent fragmentation , as seen in figure [ fig:7]_a_. although model n06 shows no fragmentation until the end of the calculation , non - axisymmetric ( or spiral ) structure develops in the circumstellar disk , as seen in figure [ fig:7]_b_. the circumstellar disk mass begins to decrease after the non - axisymmetric structure develops because such structure transfers angular momentum outward , and the gas in the circumstellar disk effectively falls onto the central protostar . finally , the protostellar mass dominates the circumstellar disk mass in the later main accretion phase for models n35 and n06 ( fig . [ fig:6 ] ) . for model n36 , the circumstellar disk mass gradually decreases in the main accretion phase ( fig . [ fig:6 ] ) , whereas it is greater than the protostellar mass at @xmath186 . the efficiency of angular momentum transfer for model n36 is considered to be lower than those for models n35 and n06 because only a weak spiral structure appears in this model . on the other hand , for models n07 , n37 , n08 and n38 , the circumstellar disk mass does not decrease greatly in the main accretion phase and is slightly greater than or comparable to the protostellar mass at the end of the main accretion phase . although the circumstellar disk is more massive than the protostar during the main accretion phase , no non - axisymmetric structure develops in these models . this is because the size of the circumstellar disk is not sufficiently larger than the jeans length , and the disk is stable against gravity ( @xcite ) . as shown in figure [ fig:7 ] , model n08 has an almost axisymmetric structure . thus , angular momentum transfer due to non - axisymmetric structure is not so effective , and a massive disk remains until the end of the main accretion phase . as seen in figure [ fig:6 ] , for model n09 , the protostar becomes more massive than the circumstellar disk just after the protostar formation , and the protostellar mass dominates the circumstellar disk mass by the end of the calculation . since model n09 has no sufficient cloud mass at the initial state , the circumstellar disk can not increase its mass significantly by gas accretion . for this model , the protostar and the circumstellar disk just after the protostar formation has a mass of @xmath187 and @xmath188 , respectively . thus , the sum of protostellar and circumstellar disk masses ( @xmath189 ) is comparable to the initial cloud mass ( @xmath190 ) . therefore , mass accretion onto the circumstellar disk from the infalling envelope rapidly declines immediately after the protostar formation . on the other hand , the protostar gradually increases its mass by the mass accretion from the circumstellar disk . figure [ fig:6 ] indicates that , in each model , the protostellar mass rapidly increases for @xmath191 , and slightly increases for @xmath192 . for @xmath162 , the protostellar mass rarely increases because the infalling envelope is almost depleted by this epoch . to compare the protostellar mass evolution among models , the time averaged mass accretion rate for @xmath193 is estimated as @xmath194 , where @xmath195 is the protostellar mass at @xmath186 . for all models , the averaged mass accretion rates are in the range of @xmath196 . the difference of the mass accretion rate among models is only a factor of about 2 . thus , a protostar has a similar accretion history when the protostellar evolution is normalized by the freefall timescale of the initial cloud . in reality , however , initially massive cloud has a longer freefall timescale and longer duration of the main accretion phase . thus , the protostar formed in massive cloud has a greater mass at the end of the main accretion phase . in the main accretion phase , the mass of the protostellar outflow dominates , or is comparable to , both the protostellar and circumstellar disk masses for models n35 , n06 , n36 , n07 , and n37 , as shown in figure [ fig:6 ] . this indicates that the protostellar outflow greatly affects star formation efficiency because it ejects a large fraction of the mass of the host cloud . in other words , the protostellar outflow controls the protostellar mass or star formation efficiency in the star formation process as pointed by @xcite . figure [ fig:8 ] shows the density and velocity distribution on the @xmath137 cutting plane for models n35 , n06 , n08 , and n09 at @xmath197 @xmath198 . note that the models and epochs in figure [ fig:8 ] are the same as those in figure [ fig:7 ] , but the spatial scales differ . in each panel , the outflowing region is denoted by a white contour , inside of which radial velocity of the gas is supersonic ( @xmath169 ) . figure [ fig:8 ] shows that the gas is strongly ejected from the host cloud by the protostellar outflow in any model . in addition , the outflow width ( i.e. , the horizontal length in the direction of travel ) is comparable to the radius of the host cloud ( @xmath199 ) , and thus , the outflow has a wide opening angle . at this epoch , on the cloud scale of @xmath33 , the outflow has opening angles of @xmath200 ( n35 ) , @xmath201 ( n06 ) , @xmath202 ( n08 ) , and @xmath203 ( n09 ) . note that the opening angle continues to increase until it becomes comparable to the cloud radius . as a result , the protostellar outflow with a wide opening angle sweeps up a large fraction of the infalling material and ejects it into the interstellar space . in addition to this swept material , a part of the mass in the circumstellar disk is ejected directly by the outflow . figure [ fig:9 ] shows the ratio of the outflowing to infalling masses for models n35 , n36 , n37 , and n38 . to investigate outflow efficiency and mass ejection rate , we calculated the outflowing / infalling mass rate on the @xmath204 and @xmath205 grid surfaces . in each model , the @xmath204 grid covers the entire circumstellar disk , and @xmath206 grid has a size of @xmath207 times the disk radius . since the protostellar outflow is originally driven by the circumstellar disk , the mass ejected from the @xmath204 grid almost corresponds to the outflowing mass directly driven by the circumstellar disk . on the other hand , the mass ejected from the @xmath206 is the sum of the outflowing mass directly driven by the circumstellar disk and the swept mass by the outflow that propagates into the dense ( or massive ) infalling envelope . according to @xcite , we estimated the outflow @xmath208 and inflow @xmath209 masses as @xmath210\ , ds,\ ] ] and @xmath211\ , ds,\ ] ] respectively , where @xmath212 is the unit vector outwardly normal to the surface of the @xmath204 or @xmath205 grid . figure [ fig:9 ] shows that the ratio of the outflowing to inflowing mass around the circumstellar disk ( @xmath204 , thin line ) has a peak of @xmath213 at @xmath214 . thus , the mass ejection rate is comparable to the mass infalling rate at this epoch . note that since the circumstellar disk and the outflow driving region are embedded in the grid of @xmath204 , the mass swept up by the outflow may be slightly included in the estimation of the outflowing rate . then , in all the models , the ratio in the @xmath204 grid gradually decreases with time . the ratio decreases to @xmath215 at @xmath216 and reaches @xmath217 at @xmath218 . figure [ fig:9 ] also shows that the ratio of @xmath219 in the @xmath220 grid exceeds that in the @xmath221 grid for a short period after the protostar appears . thus , the outflowing mass ( rate ) increases with distance from the driving source . this indicates that a wide - opening - angle outflow sweeps up the infalling gas and ejects it toward the center of the cloud . especially , for @xmath180 , the rate in the @xmath206 grid is about 10 times larger than that in the @xmath206 grid . this indicates that the protostellar outflow collects the matter in the infalling envelope 10 times more massive than that directly blown away from the circumstellar disk . note that since the boundary of the host cloud is located more far away from @xmath206 grid , the outflow sweeps more matter in the infalling envelope to be ejected from the host cloud . the thick line in figure [ fig:9 ] indicates that the outflow on a larger scale gradually decreases for @xmath222 and disappears in @xmath223 . thus , the lifetime of the outflow is about 10 times the freefall timescale of the center of the host cloud , or 3 - 5 times of the freefall timescale of the outer edge of the host cloud . the protostar forms @xmath146 after the cloud begins to collapse ( table [ table:1 ] ) . thus , roughly speaking , the protostellar outflow continues to be driven by the circumstellar disk until the freefall timescale of the cloud boundary passes after the protostar formation . this is natural that the almost all the gas inside the cloud fall onto the circumstellar disk in @xmath88 ( figs . [ fig:5 ] and [ fig:6 ] ) , and the protostellar outflow is powered by the mass accretion onto the circumstellar disk . figure [ fig:10 ] shows the outflow length ( upper panel ) and width ( lower panel ) against the time normalized by the freefall timescale at the center of the cloud for all models . note that the length and width in figure [ fig:10 ] right panels are normalized by the initial each cloud radius @xmath109 . we defined the outflow as the gas having the ( positive ) radial velocity larger than the sound speed ( @xmath169 ) . the outflow expands with time and reaches @xmath224au at @xmath225 ( fig . [ fig:10 ] upper left panel ) . the final size of the outflow depends on the size ( or mass ) of the initial cloud . the outflow extends up to about 10 times the initial cloud radius except for model n09 ( fig . [ fig:10 ] upper right panel ) . the outflow in a massive cloud has a longer lifetime to reach a greater distance from the protostar because a massive cloud has a longer freefall timescale and the outflow continues to be driven on @xmath226 times the freefall timescale . the lower panels of figure [ fig:10 ] indicates that the outflow also expands in the horizontal direction and becomes comparable in size to the host cloud radius @xmath199 . the outflow propagates into the infalling envelope along the magnetic field lines . although the magnetic field lines that drive the outflow are strongly bundled around the circumstellar disk , they spread with the distance from the center of the cloud up to @xmath227 because they are connected to the cloud scale lines . therefore , the outflow width also spreads in the horizontal direction and has an opening angle of @xmath228 on the host cloud scale . figure [ fig:11 ] shows the evolution of the collimation factor of the outflow , which is defined as @xmath229 where @xmath230 and @xmath231 are the length and width of the outflow , respectively . the figure indicates that the collimation factors remain @xmath232 for @xmath233 . then , they begin to increase for @xmath234 . at the end of the calculation , the collimation factors reach @xmath235 . as seen in figure [ fig:2 ] , although the outflow has a wide - opening - angle in the early main accretion phase , it is a very well collimated in the later main accretion phase . in addition , the width of the protostellar outflow reflects the size of the host cloud . thus , we can acquire information on the prestellar cloud core from the size and width of the protostellar outflow . to investigate the evolution of the outflow strength , we estimated the outflow momentum , which is defined as @xmath236 where @xmath237 is the outflow velocity . figure [ fig:12 ] upper panel shows that the evolution of the outflow momentum against the elapsed time after the outflow appears @xmath238 , which is defined as @xmath239 where @xmath240 is the time at the moment of the outflow appearing . this panel indicates that an initially massive cloud has a larger outflow momentum . the outflow momentum increases for @xmath241 after the outflow appears , whereas it gradually weakens for @xmath242 . since the main accretion phase is almost over at @xmath243 , the outflow momentum has a peak at @xmath243 . at its peak , the outflow momentum is @xmath244kms@xmath23 , depending on the initial cloud mass . the outflow originating from an initially massive cloud has a larger momentum and reaches further , while that from an initially less massive cloud has a weak momentum and disappears in a short duration . @xcite observed many protostellar outflows in the perseus molecular cloud complex and statistically investigated them in detail . they showed that a large fraction of protostellar outflows have momenta in the range of @xmath245 . @xcite also investigated outflows in perseus molecular cloud and showed that outflow momenta are distributed around @xmath246 . figure [ fig:12 ] upper panel shows that models with typical cloud mass of @xmath247 ( models n35 , n06 , n36 , n07 and n37 ) have the outflow momentum of @xmath248 for @xmath249yr . thus , the observation of outflow momentum well agrees with simulation results . note that the observation show a snapshot of many protostellar outflows , while figure [ fig:12 ] upper panel shows the momentum evolution of individual outflow in each cloud . figure [ fig:12 ] lower panel shows the momentum flux of the outflow , which is defined as @xmath250 the figure shows that the protostellar outflow has a momentum flux in the range of @xmath251 . the momentum flux gradually decreases with time . roughly speaking , the outflow has a momentum flux of @xmath252 in the early main accretion phase and @xmath253 in the late stage of the star formation . @xcite showed that , with 45 observed outflow samples , the outflow momentum flux is typically @xmath254 at the early class 0 stage and @xmath255 at the late class i stage . @xcite also shows the similar trend of the momentum flux . thus , the momentum flux derived from our simulation well agrees with observations . in the calculations , we adopted the minimum spatial scale of @xmath256au as described in [ sec : setting ] . thus , we could not resolve protostar and its neighborhood . therefore , no high velocity jet that is driven near the protostar appears . however , outflow momenta derived in our simulation well agree with observations . this indicates that observed outflow momenta can be explained only by the outflow driven by the circumstellar disk without both high velocity jet and mass entrained by the jet . in addition , since the outflow driven by the circumstellar disk has a wide - opening - angle , it greatly contributes to determine the star formation efficiency . the masses of the protostar , circumstellar disk , and outflow at @xmath257 are listed in table [ table:2 ] . the table also lists the mass fraction of the protostar ( @xmath258 ) , protostar plus circumstellar disk ( @xmath259 ) , outflow ( @xmath260 ) and infalling envelope ( @xmath261 ) to that of the initial cloud at @xmath257 . as shown in figure [ fig:5 ] , since the mass accretion rate at this epoch ( @xmath257 ) decreases to @xmath262 , the protostar can not acquire additional mass from the infalling envelope after this epoch . figure [ fig:6 ] and table [ table:2 ] indicate that more than 90% of the initial cloud mass is already depleted at this epoch . part of the cloud mass is ejected from the host cloud by the protostellar outflow ; the reminder falls onto either the circumstellar disk or the protostar . in summary , the main accretion phase is almost over by this epoch . the upper panel of figure [ fig:13 ] shows the masses of the protostar , outflow , and circumstellar disk at the end of the main accretion phase ( @xmath257 ) against the initial cloud mass . the protostar and outflow increase in mass as the initial cloud mass increases . on the other hand , the mass of the circumstellar disk increases with the initial cloud mass when @xmath263 , whereas it saturates when @xmath264 for @xmath265 . as seen in figure [ fig:7 ] , a massive protostar formed in a massive host cloud has a massive , gravitationally unstable circumstellar disk . such a disk shows spiral structure and subsequent fragmentation . the spiral structure or fragments effectively transfer angular momentum outward and promote mass accretion from the circumstellar disk onto the protostar . therefore , the massive circumstellar disk is self - regulated : when it is sufficiently massive , mass accretion onto the protostar is amplified , and the disk mass begins to decrease . the upper panel of figure [ fig:13 ] also shows that the mass of the protostellar outflow is larger than or comparable to the protostellar mass for @xmath266 . this indicates that , in a massive host cloud , the protostellar outflow sweeps up a large fraction of the infalling mass and ejects it into the interstellar space . thus , the protostellar outflow significantly affects star formation efficiency in such a cloud . on the other hand , in a less massive cloud , the mass of the protostellar outflow is smaller than the protostellar mass . as described in [ sec : intro ] , the first core is formed before the protostar formation . at its formation , the first core has a mass of @xmath267 which is comparable to the jeans mass at this epoch . the first core or the circumstellar disk drives a wide - opening - angle outflow . however , when the initial cloud mass is comparable to the mass of the first core , the infalling envelope does not contain enough mass to be swept up by the protostellar outflow . as a result , the outflow can not accumulate enough mass and only a small fraction of the host cloud mass is ejected . thus , higher star formation efficiency is realized in a less massive cloud without significant mass ejection from the host cloud . figure [ fig:13 ] lower panel shows the mass ratio of protostar , circumstellar disk , and protostellar outflow at @xmath268 . as shown by the blue line , the circumstellar disk mass is comparable to the protostellar mass for @xmath269 ; in this case , the circumstellar disk has an almost axisymmetric structure . on the other hand , for @xmath270 , the mass ratio of the circumstellar disk to the protostar decreases rapidly . the mass ratio of disk - to - protostar for model n09 is @xmath271 , and that for model 35 is @xmath272 . the difference is the result of the non - axisymmetric structure and fragmentation appearing in the circumstellar disk . such structure greatly promotes mass accretion from the circumstellar disk onto the protostar and decreases the circumstellar disk mass . the cloud with @xmath273 has a marginally gravitationally stable disk and has a maximum mass ratio of the circumstellar disk to the protostar . the black line in figure [ fig:13 ] lower panel shows the mass ratio of the protostellar outflow to the protostar . the ratio increases with the initial cloud mass . as seen in figure [ fig:9 ] , there are no significant difference of the mass ejection rate from the circumstellar disk by the outflow during main accretion phase among models with different initial masses ; in each model , @xmath274% of the accreting matter is blown away from the circumstellar disk for @xmath241 . however , the remaining mass of the infalling envelope and outflow driving period are different among models . in an initially massive cloud , a large fraction of the cloud mass remains in the infalling envelope even after the protostar or circumstellar disk formation . the outflow with a wide - opening - angle sweeps and collects the infalling material . thus , a large fraction of the cloud mass swept by the outflow is ejected from the host cloud . the red line of figure [ fig:13 ] also indicates that a massive cloud has a large fraction of the outflowing mass . therefore , the protostellar outflow effectively suppresses the star formation efficiency in a massive cloud . as a result , the star formation efficiency in a massive cloud is lower than that in a less massive cloud . figure [ fig:14 ] shows the star formation efficiency for each model . in the figure , the diamond and square symbols indicate the ratio of the protostellar mass to the initial cloud mass ( @xmath275 ; @xmath276 ) and the mass of the protostar plus circumstellar disk to the initial cloud mass ( @xmath277/ m_{\rm cl}$ ] , @xmath278 ) at @xmath279 , respectively . at this epoch , the infalling envelope is already depleted , and gas accretion from the infalling envelope onto the circumstellar disk or protostar has already stopped . thus , in a subsequent evolutionary phase of star formation ( class ii and class iii phases ) , the protostar acquires its mass only from the circumstellar disk , not the infalling envelope . thus , solid lines correspond to the lower and upper limits of star formation efficiency . the upper limit ( square symbol ) is realized when the entire circumstellar disk finally falls onto the protostar , and the lower limit ( diamond symbol ) is realized when the entire circumstellar disk is blown away or disappears without falling onto the protostar by any mechanism such as photo evaporation , jets around the protostar , or magnetorotational instability . in fact , the star formation efficiency is expected to fall between the lower and upper limits . figure [ fig:14 ] shows that the upper limit of star formation efficiency decreases with the initial cloud mass . this is because the mass ejection rate owing to the protostellar outflow increases with the host cloud mass . as described in [ sec : intro ] , the protostellar outflow is originally driven by the first core . thus , no outflow appears before first core formation in the collapsing cloud . the first core has a mass of @xmath280 @xcite at its formation . thus , in a less massive host cloud , a very slight mass remains as the infalling envelope after first core formation . for example , when the initial cloud mass is @xmath281 and the first core has a mass of @xmath282 , only 20% ( @xmath283 ) of the initial cloud mass remains as the infalling envelope . after first core formation , part of the gas accreted onto the first core ( or the circumstellar disk ) is blown away by the outflow . since the outflow is powered by the release of the gravitational energy of the accreting matter , no powerful outflow appears unless sufficient accreting matter exists around the driving source . in addition , a protostellar outflow with a wide - opening - angle sweeps up the gas of the infalling envelope as it propagates into the host cloud . however , when the infalling envelope is already depleted , the outflow sweeps up only a slight mass of the infalling envelope and ejects it into the interstellar space . the outflow power and the amount of mass swept up by the outflow increase with the mass of the infalling envelope . as a result , in a massive cloud , a large fraction of the initial cloud mass is ejected from the host cloud and star formation efficiency is effectively suppressed by the protostellar outflow . as seen in table [ table:2 ] , for the less massive cloud model n09 , 92% of the initial cloud mass accretes onto the circumstellar disk and the protostar and only 8% of the initial cloud mass is ejected by the protostellar outflow . on the other hand , about half of the initial cloud mass is ejected from the host cloud by the protostellar outflow for the massive cloud model n35 . as seen in figure [ fig:14 ] , the lower limit of star formation efficiency also decreases with the initial cloud mass for @xmath269 , whereas it increases slightly for @xmath270 . the lower limit is determined by the efficiency of mass accretion from the circumstellar disk onto the protostar . a less massive disk appears in a less massive cloud and is stable against gravity . in such a disk , angular momentum is transferred by magnetic effects such as magnetic braking and protostellar outflow and the mass accretes steadily onto the protostar @xcite . in contrast , a massive circumstellar disk appears in a massive cloud and is unstable against gravity . in such a massive circumstellar disk , a non - axisymmetric structure appears because of gravitational instability , as shown in figure [ fig:7 ] . this structure effectively transfers angular momentum outward , promoting mass accretion from the circumstellar disk onto the protostar . in addition , the gas accretes unsteadily onto the protostar , as described in @xcite and @xcite ; these authors pointed out the possibility of episodic accretion in such massive disks . thus , in addition to the magnetic effects , the dynamical structure of the circumstellar disk contributes to angular momentum transfer in the massive circumstellar disk that forms in an initially massive cloud . as a result , the accretion rate from the circumstellar disk onto the protostar increases with the initial cloud mass , and thus the protostellar mass and star formation efficiency also increases with initial cloud mass . star formation efficiency in the least less massive cloud is @xmath284 and that in the most massive cloud is @xmath285 . figure [ fig:14 ] shows that both the lower and upper limit of the star formation efficiency tend to decrease as the initial cloud mass increases . this is because , in a massive cloud , a protostellar outflow with a wide - opening - angle can sweep up a large amount of mass and eject it into the interstellar space . the red line in the lower panel of figure [ fig:13 ] shows that the mass ejection rate owing to the protostellar outflow increases with the initial cloud mass . the mass ejection rate owing to the protostellar outflow exceeds @xmath286 in a relatively massive cloud . thus , at most half of the initial cloud mass can be ejected by protostellar outflow . since we can not estimate the mass ratio finally falling onto the protostar from the circumstellar disk , the realistic value of star formation efficiency is unclear . however , our result indicates that the protostellar outflow contributes greatly to the protostellar mass and star formation efficiency in a single cloud core . in this study , we fixed the initial ratio of magnetic and rotational energies to the gravitational energy in each cloud . as described in [ sec : sfe ] , star formation efficiency depends strongly on outflow properties , which in turn depend on magnetic and rotational energies of the initial cloud . thus , the star formation efficiency described in [ sec : sfe ] may be just a lower limit because we selected magnetic and rotational energies most suitable for driving a powerful outflow according to the previous studies @xcite . for example , when a very weak ( negligible ) magnetic field exists in the initial cloud , only a weak ( or negligible ) outflow appears , and higher star formation efficiency is realized without significant mass ejection from the host cloud . even with a slower rotation rate , a weak outflow realizes higher star formation efficiency . in addition , the initial strength of the magnetic field affects the collimation of the outflow : outflow appearing in a weakly magnetized cloud has a relatively narrow opening angle @xcite . thus , to investigate the relationship between star formation efficiency and protostellar outflow in more detail , we may need to investigate cloud evolution in terms of magnetic and rotational energies . however , such calculation incurs a very high cpu cost . in this study , we showed that the protostellar outflow can suppress star formation efficiency to @xmath287% ( @xmath288 including the circumstellar disk ) at most . this indicates that the protostellar outflow greatly affects star formation , and we need to consider the protostellar outflow in order to investigate the protostellar mass . we will investigate cloud evolution with a large parameter space in the future . in this study , to realize long - term evolution of the cloud until the end of the main accretion phase , we adopted sink treatment instead of resolving the protostar and the structure around it ( @xmath289au ) . however , as described in @xcite , @xcite , and @xcite , another flow component called the high - velocity jet may appear around the protostar . since the high - velocity jet is well collimated @xcite , it can not sweep up a large amount of mass in the infalling envelope when it propagates into the host cloud . however , about 10% of the accreting matter is expected to be ejected by the collimated jet . thus , the high - velocity jet may further lower star formation efficiency during the star formation . however , the mass ejected by the jet may rarely affect the star formation efficiency , because a large fraction of the cloud mass is already ejected by the outflow in the early stage of the star formation . we need a considerably higher spatial resolution to include the effect of high - velocity jets to estimate the star formation efficiency in more detail . in this study , we investigated the impact of protostellar outflow on the star formation process . we constructed nine models with different initial cloud masses in the range of @xmath63 , and calculated the cloud evolution until the cloud mass is depleted . in the calculation , without artificially inputting outflow ( momentum ) to the computational domain as seen in any other studies , outflow is naturally or automatically driven by the circumstellar disk . as a result of the calculation , we found that a large fraction of the initial cloud mass is ejected from the host cloud into the interstellar space by the protostellar outflow . thus , the protostellar outflow significantly affects the star formation process and determination of the star formation efficiency . the following results are obtained . the protostellar outflow continues to be driven by the circumstellar disk for about 10 times the freefall timescale of the initial cloud after the cloud begins to collapse . the final size of the outflow is different among models , because different models have different initial central densities and different freefall timescales . in each model , the protostellar outflow reaches @xmath224au far from the protostar in @xmath290 . when the outflow remains inside the host cloud , the outflow extends also to the vertical direction toward the direction of travel , because the outflow is anchored by large - scale ( host cloud scale ) magnetic field lines . after the outflow penetrates the host cloud and propagates into the interstellar space , it extends only in the direction of the travel keeping its width . thus , in this period , the outflow collimation improves to reach @xmath274 . before disappearing , the outflow has a length of 10 times the natal cloud radius and a maximum width comparable to the natal cloud radius . thus , we can expect natal cloud size ( or mass ) from the length and width of observed outflow . in this study , since we did not resolve protostar itself , no high - velocity jet appears . however , outflow momentum and momentum flux derived in our simulation well agree with observations . this indicates that the outflow driven by the circumstellar disk is responsible for total outflow momentum , and high velocity jet rarely affects the star formation efficiency . in addition , it is expected that the entrained mass by the high velocity jet can be ignored to estimate the mass ejection from the host cloud , because observed outflow momentum can be explained only by the disk driven outflow . although the high velocity jet may promote the mass ejection further , a larger fraction of the mass can be ejected only by the outflow . the protostellar outflow can eject @xmath291% of the host cloud mass into the interstellar space . the mass ejection rate increases as the initial cloud mass increases ; a large fraction of the initial cloud mass is ejected in a massive cloud . this is because a massive cloud retains a large amount of the infalling matter even after the protostar formation , and outflow can sweep and collect a large fraction of the infalling matter when it propagates into the infalling envelope . thus , the protostellar outflow can suppress the star formation efficiency to @xmath292 . in addition , a massive circumstellar disk comparable to the protostellar mass remains even after the infalling envelope is depleted . although we need to calculate further evolution of the protostellar system to determine the star formation efficiency , the star formation efficiency of @xmath293% may be realized when ( a part of ) the circumstellar disk is blown away in further evolution stage . plane are plotted with the initial cloud scale . the white dashed circle represents initial cloud radius ( i.e. , the host cloud ) . the blue line is the boundary of the outflow inside which the gas moves outwardly toward the center of the cloud ( or the protostar ) with a supersonic velocity . the elapsed time @xmath143 in unit of the freefall timescale ( @xmath87 ) and year is plotted on the upper side of each panel . the white squares in each panel denote the outer boundary of the subgrid . , width=566 ] inside which the gas is outflowing from the center of the cloud with the supersonic velocity @xmath294 . the blue sphere corresponds to the initial host cloud . the elapsed time @xmath143 in unit of the freefall timescale ( @xmath87 ) and year is plotted on the upper side of each panel . , width=566 ] plane at @xmath295 for model n06 with different spatial scales . the white dashed circle represents initial cloud radius ( i.e. , be radius ) . the red contours denote the outflow inside which the gas moves outwardly toward the center of the cloud with a supersonic velocity ( thick line ) and half of the supersonic velocity ( thin line ) . , width=566 ] ( left ) and @xmath138 ( right ) planes at the same epoch as in fig . [ fig:3 ] . the red contour denotes the outflow inside which the gas moves outwardly toward the center of the cloud with a supersonic velocity . , width=566 ] ( = @xmath296 , where @xmath144 is time at the protostar formation ) after the protostar formation for each model . the vertical lines in each panel corresponds to the freefall timescale of the host cloud at the center ( @xmath87 , solid line ) and cloud boundary ( @xmath88 , broken line ) . the initial host cloud mass is plotted by the cross ( @xmath297 ) on the right axis . , width=566 ] (= @xmath296 ) after the protostar formation for each model . the vertical lines in each panel corresponds to the freefall timescale of the host cloud at the center ( @xmath87 , solid line ) and cloud boundary ( @xmath88 , broken line ) . , width=566 ] plane for models n35 , n06 , n08 and n09 . the elapsed time @xmath143 in unit of @xmath87 and year is plotted on the upper side of each panel . the white dashed circle represents initial cloud radius ( i.e. , be radius ) . the white solid line is the boundary of the outflow inside which the gas moves outwardly toward the center of the cloud ( or the protostar ) with a supersonic velocity . the white squares in each panel denote the outer boundary of the subgrid . , width=566 ] and @xmath205 grid against the time for model n35 , n36 , n37 and n38 . the vertical lines are the freefall timescale of the host cloud at the center ( @xmath87 , solid line ) , cloud boundary ( @xmath88 , broken line ) , and 10 times the freefall timescale of the host cloud at the center ( @xmath298 , dotted line ) . , width=566 ]
an automated , real - time , multiple sensor data source relying and globally applicable earthquake loss model and visualiser is desirable for post - event earthquake analysis . to achieve this there is a need to support rapid data ingestion , loss estimation and integration of data from multiple data sources and rapid visualisation at multiple geographic levels . in this paper , the design and development of the automated post - event earthquake loss estimation and visualisation ( ape - elev ) system for real - time estimation and visualisation of insured losses incurred due to earthquakes is presented . a model for estimating ground up and net of facultative losses due to earthquakes in near real - time is implemented . since post - event data is often available immediately from multiple disparate sources , a geo - browser is employed to facilitate the visualisation and integration of earthquake hazard , exposure and loss data . the feasibility of ape - elev is demonstrated using a test case earthquake that occurred in tohoku , japan ( 2011 ) . the ape - elev model is further validated for ten global earthquakes using industry loss data . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore
1308.1846
research in estimating losses for catastrophes have led to the development of a wide variety of earthquake loss models . earthquake loss models can generate loss values before an event occurs or while an event is evolving or after an event occurs . earthquake loss models can be classified as probabilistic , deterministic and real - time models . probabilistic models produce a maximum probable loss value using a stochastic event catalog which represents a sample of possible future earthquakes . models such as capra - central american probabilistic risk assessment @xcite , eqrm - earthquake risk model @xcite and riskscape @xcite are probabilistic models . in deterministic models the losses caused by a specific event that occurred are estimated . lnecloss @xcite , redars - risks from earthquake damage to roadway systems @xcite and nhematis @xcite are deterministic models . real - time models estimate losses soon after ( near real - time ) an earthquake has occurred . examples include eler - earthquake loss estimation routine @xcite , emergeo @xcite and pager - prompt assessment of global earthquakes for response @xcite . a hybrid of the former models are seen in hazus ( combines deterministic , probabilistic and real - time models ) @xcite , koeriloss @xcite and maeviz @xcite . in this paper , a loss estimator which produces loss values in near real - time and can model past earthquake events is presented . models that focus on generating a probable loss value use a catalog of possible future earthquakes . in such models , there is no focus on a specific event and any analysis is done before an earthquake may occur and is called pre - event analysis . examples include air @xcite , dbela - displacement - based earthquake loss assessment @xcite and mdla @xcite . for quick and imminent decision making it is desirable that loss estimates be accurately generated as an event evolves . post - event analysis presents a timely evaluation of losses due to an earthquake in the minutes , hours , days and weeks immediately following an earthquake . examples of post - event models are inlet - internet - based loss estimation tool @xcite , pager @xcite and extremum @xcite . models combining both pre - event and post - event analysis are available in epedat - early post - earthquake damage assessment tool @xcite , hazus - @xcite and selena - seismic loss estimation using a logic tree approach @xcite . the model proposed in this paper focuses on analysing the effects of an earthquake soon after it occurs and modelling the effects of a past earthquake . pre - event models are of limited interest in the context of estimating losses in real - time . in this paper the focus is on post - event analysis since it is different from pre - event analysis in a number of important ways : * the focus is on a single earthquake event which has just occurred rather than a catalog of possible future events , or on a past earthquake event which can be modelled from archived sensory data . * there is an evolving view of the event as it unfolds , and therefore the sensor data related to the event changes hours , days and weeks after the event , * there is a need for rapid estimation of losses to guide early responses @xcite , and * since post - event data is available from multiple sources , there is a need to visualise and integrate hazard , exposure and loss data from these multiple sources . the 2011 tohoku earthquake that struck off the pacific coast of japan at 05:46 utc on friday , 11 march 2011 is a recent example that illustrates the importance of post - event analysis . figure 1 presents the timeline of the earthquake . fifteen alerts @xmath0 were issued by pager / shakemap in time periods ranging from within an hour to six months after the earthquake . the first alert was issued twenty three minutes after the event and reported a magnitude 7.9 earthquake . additional information such as initial peak ground velocity and peak ground acceleration maps of the ground shake was also available with the alert . further , over the course of the first day alone four additional alerts were issued each updating the data available . not only did the earthquake event unfold over time but the data describing the event and our knowledge of the event evolved . the earthquake data alone was not sufficient to produce reliable loss estimates because between 06:15 utc and 07:52 utc a tsunami struck the coastal towns . additional data sources are required for complete loss estimation . estimating loss values of a future earthquake is based on using a static catalog containing data related to historic events and is employed in pre - event analysis . for example , models such as air @xcite , dbela @xcite and eqrm @xcite employ static catalogs . a static catalog therefore is not sufficient to estimate accurate losses as an earthquake evolves over hours and days of its occurrence . there is a need for up - to - date information of an earthquake as it evolves . one possibility is to make use of seismic sensor networks which can provide earthquake information as soon as minutes after it has occurred . shakemaps @xcite , for example , are a representation of earthquake sensory information . models that employ real - time models include emergeo @xcite , inlet @xcite and pager @xcite . a few models incorporate both historic and sensor data such as in hazus @xcite , mdla @xcite and selena @xcite . in this paper , we investigate how sensor data from multiple sources can be used for timely estimation of losses . the use of regional seismic sensor networks can provide a model with only region specific data and thereby restricts loss estimation to regions . this may be due to the nature of the research where the project was undertaken and therefore only a country or a region was considered . models such as openrisk @xcite , tefer - turkish emergency flood and earthquake recovery programme earthquake model @xcite and teles - taiwan earthquake loss estimation system @xcite are examples that analyse earthquakes in a region . to ensure global applicability of the model it needs to rely on global sensor networks . epedat @xcite , radius @xcite and qlarm - earthquake loss assessment for response and mitigation @xcite are a few examples . further , full - fledged global applicability also implies being able to use the model to estimate losses at different geographic levels ( for example , loss estimation at cities , counties , states and countries ) . the model presented in this paper explores how global applicability can be achieved . among the earthquake loss estimation models that have been referenced , eler , emergeo , epedat , extremum , hazus , inlet , pager , qlarm , quake - loss , selena and teles support post - event analysis . among these , models such as , eler , epedat , hazus , inlet and teles are region restricted . while these models may provide close to accurate loss estimates , yet they do not support global earthquakes . this may be due to the reliance of the models on regional seismic networks . the emergeo earthquake model produces maps of mmi and peak ground acceleration ( pga ) and can predict damages . loss estimates are not a focus in the model . both the extremum and quakeloss models rely on multiple data sources but are focused on structural and human losses . financial loss estimates are not considered in both models . pager ( prompt assessment of global earthquakes for response ) provides fatality and economic loss impact estimates . however , pager does not determine region specific loss data . global financial and economic organisations need to know the losses ( estimates ) incurred at different geographical levels . the qlarm model calculates human losses and damage in a given human settlement . however , qlarm does not focus on estimating financial losses . the selena model and the complementing rise ( risk illustrator for selena ) @xcite visualisation software computes real - time loss estimates and presents the losses visually . however , there seems to be less automation along the pipeline from obtaining real - time data to visualising the losses . the real - time data needs to be provided by the user to the selena model . research that is pursued for automated post - event estimation of financial losses globally is sparse at best , though many loss models are available in the public domain @xcite . the research reported in this paper is motivated towards the development of ( a ) a real - time , ( b ) a post - event , ( c ) a multiple sensor data relying and ( d ) a globally applicable loss model . to achieve this there is a need to support rapid data ingestion , rapid loss estimation , rapid visualisation and integration of data from multiple data sources and rapid visualisation at multiple geographic levels . the * * a**utomated * * p**ost-**e**vent * * e**arthquake * * l**oss * * e**stimation and * * v**isualisation ( ape - elev ) system is proposed , which comprises three primary modules , namely the earthquake loss estimator ( ele ) , the earthquake visualiser ( ev ) and the elev database ( elev - db ) . the ele module is built on pager and shakemap for accessing real - time earthquake data and estimating losses at different geographic levels . the ele module computes financial losses . visualisation of the losses is facilitated by the ev module . the elev - db module aids the functioning of the ele and ev modules . the remainder of this paper is organised as follows . section [ centralisedarchitecture ] proposes a centralised architecture for the automated post - event earthquake loss estimator and visualiser ( ape - elev ) . the loss estimation module is presented in section [ estimator ] and the loss visualiser module is presented in section [ visualiser ] . section [ distributedarchitecture ] presents a distributed architecture for the ape - elev and how estimation and visualisation are distributed across the server and the client respectively . section [ experiments ] presents one test case using ape - elev and a validation study of the model using ten global earthquakes . section [ conclusion ] concludes the paper . the automated post - event earthquake loss estimator and visualiser ( ape - elev ) is a system that determines expected losses due to the occurrence of an earthquake ( on building that are exposed to the earthquake , otherwise called exposure ) and graphically display these losses . decision makers in financial organisations , governmental agencies working toward disaster management and emergency response teams can benefit from interpreting the output produced by ape - elev for aiding imminent decision making . the output can also be adjusted for the benefit of the decision maker by changing the exposure data . the ape - elev system determines two types of losses . firstly , the ground up loss , referred to as gul which is the entire amount of an insurance loss , including deductibles , before applying any retention or reinsurance . secondly , the net of facultative loss , referred to as nfl which is the entire amount of an insurance loss , including deductibles , primary retention and any reinsurance . the determined losses can be visualised at four geographic levels , namely country , state , county and city , on a geo - browser . the country , state and county levels are sometimes referred to as regions , while the city level is referred to as both point and population centre . indicators are defined to facilitate visualisation at the region level ; indicators are either event - specific ( for example , losses at regions ) or geography - specific ( for example , population at cities or regions ) . ape - elev is composed of three primary modules , namely the earthquake loss estimator and visualiser database ( elev - db ) , the earthquake loss estimator ( ele ) and the earthquake visualiser ( ev ) . figure 2 shows the architecture of ape - elev . the elev - db module is a collection of tables related to an event and geographic data . the ele model ( see figure 2 ( top ) ) as the name suggests estimates the losses incurred when an earthquake occurs . the ev model ( see figure 2 ( bottom ) ) again as the name suggests facilitates the visualisation of the loss estimates generated by the ele model . the elev - db module comprises seven tables which contribute to the working of the ele and the ev modules . the tables are : 1 . @xmath1 , which consists of industrial data for ground up exposure , 2 . @xmath2 , which consists of industrial data for net of facultative exposure , 3 . @xmath3 , which consists of event data , 4 . @xmath4 , which consists of a set of indicators , 5 . @xmath5 , which consists of geographic information that is used to map lower geographic levels onto higher geographic levels ( for example , mapping of cities onto counties or counties onto state ) , 6 . @xmath6 , which consists of data that is generated from the jaiswal and wald mean damage ratio ( mdr ) model @xcite , and 7 . @xmath7 , which comprises loss data populated by the ele module . the ele module , as shown in figure 2 ( top ) , comprises three sub - modules , namely the hazard , vulnerability and loss modules . the hazard module receives two inputs , firstly , the data on cities ( i.e. population centres with more than one thousand people ) affected by the earthquake , and secondly , geographic information required for mapping lower geographic levels onto higher geographic levels . the hazard module produces the measure of severity of an earthquake , otherwise referred to as the modified mercalli intensity ( mmi ) , in a city and region . the mmi values along with data from @xmath6 are used by the vulnerability module to produce mdr values . this data is employed by the loss module along with two types of exposure data , namely ground up exposure and net of facultative exposure to generate both the gul and nfl losses . the event data extractor receives the notification of the event and initiates the ele . the ev module , as shown in figure 2 ( bottom ) , comprises five sub - modules , namely the exposure data visualiser , loss data visualiser , hazard data visualiser , static data visualiser and the portfolio visualiser . the visualiser modules employ a geo - browser for graphical display . the exposure data visualiser presents the exposure for different geographic levels . the loss data visualiser presents the gul and nfl for different geographic levels . the hazard data visualiser presents the mmi and mdr for different geographic levels . static data visualiser is employed for presenting geography - specific indicators , and as the name implies these indicator values do not change from one event to another . the portfolio visualiser presents a comparison of losses and exposures . the earthquake visualiser mapping engine ( ev - me ) module facilitates visualisation of data on a geo - browser . having presented the architecture of ape - elev , it is also necessary to consider how the ele , ev and elev - db modules and their sub - modules glue together for coherent functioning . the data required to kick - start ape - elev is obtained before the occurrence of an earthquake or in a pre - event phase . an accumulation model is used to generate the ground up and net of facultative exposures at the region level . casualties are proportional to the number of people present in the affected area and the quantity and value of buildings , infrastructure and other property in this area . the accumulation model quantifies regional exposure based on the whether economic losses need to be determined for the assets insured by the insurance / reinsurance company . in the research reported in this paper , the accumulation model is a black box used by the industrial partner supporting this research and the model generated gul and nfl exposures for a given region . the region level exposure is then disaggregated into cities ( i.e. , population centres that fall within the region ) based on the percentage of population . the city level exposure is further used by the ele module in the post - event phase . for an earthquake event , @xmath8 , that has just occurred or is unfolding we firstly need to be notified of the event . an automated system for notifying earthquakes is shakecast lite @xcite . the ele module employs shakecast lite for notification alerts which are received by the event data extractor . when the notification alert is received the ele module is instantiated . further , we require real - time data of the earthquake . the prompt assessment of global earthquakes for response ( pager ) is an automated system that can provide such real - time data . the ele module employs the real - time data from pager / shakemap that is acquired as an .xml file . the .xml file is then parsed to extract event related information that is stored in @xmath3 of elev - db . information such as an affected city , represented as @xmath9 ( @xmath9 represents city , @xmath10 represents counties , and @xmath11 represents states and @xmath12 represents countries ) , population of the city , represented as @xmath13 and mmi of the city , represented as @xmath14 is provided to the hazard module . the hazard module computes the mmi at higher geographic levels using the mmi of affected cities . if the geographic level is represented as @xmath15 , where @xmath16 and @xmath17 , the population at the geographic level @xmath15 is represented as @xmath18 and the mmi at the geographic level @xmath15 is represented as @xmath19 , then @xmath20 where @xmath21 ( @xmath22 is the total no of affected regions ) , and @xmath23 ( @xmath24 is the number of affected cities in a region @xmath25 ) . the geographic data to evaluate whether an affected city lies within a given region is provided through @xmath5 . the double subscript notation is used to capture the idea that there are population centres which are affected due to the earthquake within a large affected region . for example , consider an earthquake that affects two counties , @xmath26 and @xmath27 . in the equation counties are represented by @xmath28 and since there are two affected counties , @xmath29 , and @xmath25 iterates two times . assume there are three cities in @xmath26 , namely @xmath30 , @xmath31 and @xmath32 , their populations denoted as @xmath33 , @xmath34 and @xmath35 and their mmis denoted as @xmath36 , @xmath37 and @xmath38 respectively . for this county @xmath39 ( three cities are in the affected region , and @xmath40 iterates three times for this county ) . the mmi at the county levels @xmath41 for @xmath26 is equal to @xmath42}{p(city_{1 } ) + p(city_{2 } ) + p(city_{3})}\end{aligned}\ ] ] assume four cities in @xmath27 , namely @xmath43 , @xmath44 , @xmath45 and @xmath46 , their populations denoted as @xmath47 , @xmath48 , @xmath49 and @xmath50 and their mmis denoted as @xmath51 , @xmath52 , @xmath53 and @xmath54 respectively . for this county @xmath55 ( four cities are in the affected region , and @xmath40 iterates four times for this county ) . the mmi at the county levels @xmath41 for @xmath27 is @xmath56}{\big(p(city_{4 } ) + p(city_{5 } ) + } \\ & & p(city_{6 } ) + p(city_{7})\big)\end{aligned}\ ] ] consider that both counties , @xmath26 and @xmath57 , are in the same state , @xmath58 , the population of the counties denoted as @xmath59 and @xmath60 , and the mmis of the counties obtained from the above equations . the mmi at the state level @xmath61 for @xmath58 is @xmath62}{p(county_{1 } ) + p(county_{2})}\end{aligned}\ ] ] the @xmath19 , where @xmath63 and @xmath17 is then utilised by the vulnerability module to compute @xmath64 . unlike the hazard module , the city level is considered in the vulnerability module , and therefore @xmath65 ranges from 1 to 4 . it is worthwhile to note that mmi values range from i to xii . @xmath6 which was originally generated by the jaiswal and wald mdr model provides the mdr value corresponding to an integer mmi value . should a floating point mmi value be obtained during computations from the hazard module , then the mdr values are computed by linear interpolation in the vulnerability module . for example , if mmi is obtained as 7.5 from the hazard module , then the mdr values corresponding to mmi vii and mmi viii are interpolated in the vulnerability module to obtain the mdr value for mmi-7.5 . such a technique is employed in hazus @xcite . the mdr value of a city is provided to the loss module , along with the ground up and the net of facultative exposure data from @xmath1 and @xmath2 . the gul and nfl of a city are computed by multiplying the mdr values for a city with the exposure of the city . the city losses are then aggregated onto higher geographic levels using @xmath5 to compute the losses on the county , state and country levels . the total loss corresponding to an event is provided to @xmath3 , while the regional losses corresponding to an event is provided to @xmath4 and losses related to a specific line of business in @xmath7 . line of business refers to a statutory set of insurance / reinsurance policies to define coverage . the coverage may or may not affect a strategic business unit . the hierarchies structures of lines of business are property - fire insurance , business interruption and natural catastrophes ; casualty - liability , motor , non - life accident and health ; special lines - aviation , engineering , marine ; credit and surety . these lines of business are either industrial , personal or commercial coverages . the elev - db module plays an important role in providing data to and receiving data from the ele module . during the period from the notification of an event until completion of computing losses , tables @xmath3 , @xmath4 and @xmath7 are modified . tables @xmath1 , @xmath2 , @xmath5 and @xmath6 provide input to the ele module . the five sub - modules of ev , namely the exposure data visualiser , the loss data visualiser , the hazard data visualiser , the static data visualiser and the portfolio visualiser operate in parallel . this is unlike the ele sub - modules that operate in sequence . the functioning of the sub - modules of ev are nevertheless presented sequentially in this section for the sake of convenience . the exposure data visualiser utilises @xmath1 and @xmath2 for displaying two types of exposures , the ground up exposure and the net of facultative exposure . the latitude , longitude and geography related indicators of all regions are extracted from @xmath5 and provided to the earthquake visualiser mapping engine ( ev - me ) . the ev - me module generates a .kml ( keyhole markup language ) file that contains place marks which highlight the exposure of the regions . the .kml format is compatible for visualisation on geo - browsers @xcite , and in this research google earth is employed . the thematic mapping engine ( tme ) is the underlying building block of ev - me @xcite . a number of visualisation techniques such as bar , prism , choropleth , collada and push pins are made available for facilitating analysis of the data . the loss data visualiser utilises @xmath4 from which regional loss data is extracted for displaying the ground up and net of facultative losses . similar to the exposure data visualiser , the ev - me module generates a .kml file that is viewable on google earth . the hazard data visualiser utilises @xmath4 and @xmath5 from which regional and point hazard data are extracted respectively for displaying mmi and mdr at all geographic levels . similar to the above modules a .kml file is generated by the ev - me module . the static data visualiser again utilises @xmath4 and @xmath5 from which cities affected by the event and static - data related to the affected cities are extracted respectively . a .kml file is generated by the ev - me module and the extracted data is visualised . the portfolio visualiser that is incorporated within the ev module compares losses and exposure ( of areas affected by the event ) by line of business . data related to the distribution of total losses by line of business such as industrial , personal and commercial is extracted from @xmath7 . since visualisations are provided on pie - charts , the ev - me module is not employed . the distributed ape - elev comprises the server system and the client system , as shown in figure 3 , and are considered in the following sub - sections . the ape - elev server system consists of the elev - db database , the ele module and an ev module . the elev - db and the ele module are similar to those employed in the centralised architecture . the ev module is different from the centralised architecture as the geo - browser , the web browser and the portfolio visualiser are located on the client system . to facilitate the handling of client requests , an additional sub - module is required on the server visualiser system , and therefore the data handler is employed which acts as an interface between client requests and the data available for visualisation that is stored in the database . four handlers are available , namely the exposure data handler , the hazard data handler , the loss data handler and the static data handler . the exposure data handler retrieves the exposure for different geographic levels . the loss data handler retrieves gul and nfl for different geographic levels . the hazard data handler retrieves mmi and mdr for different geographic levels . the static data handler retrieves geography - specific indicators . the mapping engine receives data from the handlers and facilitates the visualisation of data on the client system . it is built on the thematic mapping engine ( tme ) @xcite and generates .kml files . the kml file repository stores the .kml files generated by the mapping engine . the portfolio generator is built on the google chart api and presents a comparison of losses and exposures as pie - charts . the client system in the distributed ape - elev is a client visualiser that consists a geo - browser , an event viewer and a portfolio viewer . figure 4 is the illustration of interactions between the client and server modules . the loss estimation module executes step 1 to step 5 after it receives an earthquake notification , thereby storing loss values in the database . the client system can raise two type of visualisation requests , those to the data handler and to the portfolio generator . a visualisation request to the data handler is made by the event viewer . based on the type of data that needs to be visualised , the exposure , loss , hazard or static data handlers are invoked . the handler retrieves data from elev - db and a .kml file is generated in the kml file repository . the event viewer after receiving a .kml file link requests to read the file and is accessed by the geo - browser on the client system . a visualisation request to the portfolio generator again retrieves loss and exposure data from elev - db . the google chart api is used to generate pie - charts in a repository . the portfolio viewer can then access the pie - charts on the client system . there are seven benefits of distributing the modules of ape - elev on a server and a client : 1 . the server system can facilitate archiving for multiple users . this presents the opportunity for a user to manage his workspace and archive earthquakes of his interest . the server system is accessible to the client but is concealed from the client . therefore the installation of third party softwares such as shakecast lite and the thematic mapping engine which are used in the development of ape - elev is not required on the client system as they are made available from the server . it needs to be however noted that the installation of a geo - browser is mandatory to view .kml files on the client system . 3 . there is no data management on the client system . since multiple external data sources including real - time earthquake data , exposure data , geography data and geometry data are ingested by ape - elev , user management of these data sources would be cumbersome . in distributed ape - elev , data management is carried out at the server . there are no repositories on the client system . should a user require to analyse a large number of earthquakes , then the kml file and pie - chart repositories can be large . the client system is granted access to the repositories that are situated on the server . the database consisting of voluminous data created by ape - elev is resident on the server system . the data is voluminous due to the integration of geometry , geographic , exposure and event data which further produces loss and hazard data at multiple geographic levels . ape - elev can be made globally accessible by hosting the server system on the world wide web . 7 . the client system can be made available on multiple platforms such as tablets , smartphones and personal digital assistants ( pdas ) . the availability of ape - elev essentially requires internet access . kml data will require a geo - browser enabled platform . administrative privileges to the server will be required for decision makers to be able to use the distributed ape - elev to their benefit of not merely interpreting the output of ape - elev using the default exposure set but using a custom exposure . as of where the current development of distributed ape - elev stands the data management facilitated by the server limits the user ability to adjust input data and customise the output data ; the centralised system lends itself more to such custom user requirements . consequently , multiplier indices considered in section [ validationstudy ] can not be set by the user and this flexibility needs to be incorporated in future research . this section in the first instance considers the experimental platform and the user interface of ape - elev , followed by feasibility and validation studies of the ape - elev model . the feasibility of ape - elev is confirmed using a test case earthquake of magnitude 9.0 that occurred on 11th march 2011 , commonly known as the tohoku earthquake or referred to as near the east coast of honshu , japan with an event i d usc0001xgp in pager . the validation study considers 10 global earthquakes and the expected losses computed by ape - elev is compared against normalised historic loss data . the validation study is also pursued to determine the probability of the expected losses falling within a pre - defined loss threshold . the data related to the earthquake was available on the pager archive @xcite and shakemap archive @xcite . the event data extractor in the ape - elev architecture fetches data related to the event from the pager archive in .xml format and instantiates the ele module . after the ele module is instantiated , the losses are estimated as considered in section [ estimator ] . the ev module is then employed to visualise the estimated losses . geometry data for the geographic levels was obtained from the global administrative areas database @xcite , as shapefiles . the shapefiles obtained were large in size containing accurate boundary specification . since the experiment reported here was a preliminary test , approximate boundary specifications were sufficient , and therefore the shapefile was simplified using the mapshaper tool @xcite . figure 5 is a screenshot of the visualiser . the inline map shown on the screenshot represents the shakemap representation of the earthquake . the earthquake related data is shown on the right - hand side of the map . the four visualisers of the ev module are listed under google earth visualisation as static data , exposure data , hazard data and loss data . the visualisation techniques ( choropleth in the screenshot ) are available in a drop - down box . the shakemap link presents the shakemap on the google earth application . the ground up and net of facultative losses computed by the ele module are displayed under global earthquake loss model . the portfolio loss link presents four pie charts that compares the losses and exposures by line of business such as industrial , personal and commercial . the test case employed in the feasibility study is magnitude 9.0 , which occurred in tohoku , japan on 11 march 2011 that struck off the pacific coast of japan at 05:46 utc on friday , 11 march 2011 . this recent earthquake was a major catastrophe and affected 28 prefectures . it is worthwhile to note that the catastrophe was due to both a tsunami and an earthquake . the ape - elev model does not incorporate any mechanism to differentiate between the tsunami and the earthquake related losses . this differentiation , however , is achieved in the model since the input data from usgs pager and shakecast differentiates the catastrophe by producing earthquake related data . therefore , the model inherently produces loss estimates for the catastrophe data provided and its accuracy is dependent on the input . figures 6 - 10 are a set of screenshots obtained from the visualiser . figure 6 shows the mmi of the affected prefectures using the prism visualisation technique . the gradient scale on the left hand side shows the mmi at the prefectures . the right most pop - up shows gul and nfl for the earthquake . the pop - up in the centre shows the exposure , population and hazard data of shizuoka prefecture . figure 7 shows the mdr of the affected prefectures . the choropleth visualisation technique is employed for representing the mdr . the gradient scale on the left hand side shows the mmi at the prefectures . the pop - up shown on the right side shows information relevant to the earthquake for japan and the pop - up in the centre shows regional information for the fukushima prefecture . figure 8 shows the superimposition of mdr and population of the affected prefectures . choropleth is employed for visualising mdr of the prefectures , prisms are employed for visualising nfl and push - pins are used for visualising populations . the two gradient scales on the left side show the scale of mdr and populations . the pop - up shown on the right side shows information relevant to the earthquake and the pop - up in the centre shows regional information relevant to miyagi prefecture . figure 9 shows the mmi of the affected prefectures using choropleth , the population in the prefectures using human push - pins and the estimated losses using prisms . the two gradient scales on the left side show the scale of mmi and population . the pop - up on the right side shows the estimated loss information for the entire event in the gul and nfl categories . the pie charts indicate the losses for industrial , personal , commercial and other lines of business for the exposure data used . figure 10 shows a different view of information visualised in figure 9 . the mmi of the affected prefectures using choropleth , the population in the prefectures using human push - pins and the estimated losses using prisms . mmi and population are shown on the gradient scale . while the right - most pop up showing the pie charts indicates the loss for the entire event , the pop up in the centre shows the losses specific to the saitama prefecture . the gul and nfl aggregated for the prefecture along with information relevant to the prefecture and the event are presented . figures 11 - 18 are screenshots of different alert versions , @xmath0 of the test - case earthquake which shows the evolving view of the earthquake and how losses can be rapidly estimated . the mmi of the affected prefectures are shown using choropleth visualisation technique and the height of the prisms are indicative of the ground up losses . @xmath66 were received within the first day after the event , @xmath67 within the same week after the event , @xmath68 within the same month after the event and the remaining alerts within six months after the event . figure 11 is based on the first alert , @xmath69 which presented data for an overall magnitude of 7.9 twenty two minutes and fifty eight seconds after the event occurred . in this alert , as shown in the figure fourteen prefectures are affected - six prefectures with mmi vii ( dark yellow ) , six prefectures with mmi vi ( light yellow ) and two prefectures with mmi v ( green ) . the ground up loss for the prefectures are estimated and presented above the prisms indicative of the magnitude of the loss . the estimated losses are highest for the chiba and kanagawa prefectures . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] figure 12 is based on the third alert , @xmath70 which presented data data for an overall magnitude of 8.8 one hour and fifteen minutes after the event occurred . in this alert , more data was available and was used to update the first alert . while there is a difference in the data showing the magnitude of the earthquake , the mmi data and the estimates for the ground up loss remained the same . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] figure 13 is based on the fifth alert , @xmath71 which presented data for an overall magnitude of 8.9 two hours and forty four minutes after the event . the mmi information of the prefectures were updated - six prefectures with mmi vii ( dark yellow ) , eight prefectures with mmi vi ( light yellow ) , five prefectures with mmi v ( light green ) and three prefectures with mmi iv ( light blue ) . the loss estimates for the prefectures have rapidly changed after this alert . for example , for the chiba and kanagawa prefectures the ground up loss estimates have increased by approximately 8 times after the first and third alert . the sensor data in this alert has gathered more information about the prefectures which are land - locked . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] figure 14 is based on the seventh alert , @xmath72 which presented data for an overall magnitude of 9.0 four days and nine hours after the event . again the mmi information of the prefectures are updated with more accurate information gathered by the sensors . one prefecture has an mmi viii and the ground up loss estimates of the prefectures around chiba and kanagawa prefectures have increased . more prefectures to the south of the island have an mmi iv though the losses estimated here are zero . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] figure 15 is based on the ninth alert , @xmath73 which presented data for magnitude similar to the previous alert and was received one week and one day after the event . the data for the next alerts will remain almost similar with minor details updated . while in the previous alerts an evolving view of the hazard , vulnerability and loss were visualised from this alert a constant view is obtained . again loss estimates in the prefectures to the vicinity of the coastal prefectures are updated . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] figure 16 , figure 17 and figure 18 are based on alerts , @xmath74 , @xmath75 and @xmath76 respectively . the overall data visualised in these alerts are more or less the same with minimal updates to the mmi and losses estimated for the prefectures . of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] of magnitude 9.0 , tohoku , japan , 11 march 2011 earthquake , scaledwidth=50.0% ] a study that compares the predicted losses of ten global earthquakes against historic loss data was pursued in order to validate the ape - elev model . table 1 shows the list of earthquakes selected for this study , their date of occurrence ( dd - mm - yyyy ) , magnitude , latitude and longitude , historic losses in millions of usd in the year of occurrence of the earthquake , adjustment multipliers to normalise the historic losses to 2012 usd , predicted losses in millions of usd and percent error between the normalised historic and predicted losses . the earthquakes were selected such that ( a ) they were distributed geographically across different continents , ( b ) their magnitude was over 5.5 , ( c ) and had occurred in the last 30 years . m1.4 cm | m1.4 cm | m1 cm | * region affected * & * country * & * date * & * mag * & * lat * & * long * & * historic losses in millions of usd for year y , @xmath77 * & & * normalised historic losses in millions of 2012 usd , @xmath78 * & * predicted losses in millions of 2012 usd * & * percent error % * + & & & & & & & * * & * * & * * & * * & & & + libertador ohiggins & chile & 11/03/2010 & 6.9 & -34.2592 & -71.9288 & 16.0500 & 1.0558 & 0.9957 & 0.9651 & 1.0318 & 16.8732 & 238.8136 & 1315.33 + wnw of ferndale & usa & 09/01/2010 & 6.5 & 40.6520 & -124.6920 & 25.0000 & 1.0352 & 0.9850 & 0.9683 & 1.0172 & 25.4904 & 16.8655 & -33.84 + california & usa & 28/06/1992 & 7.3 & 34.2012 & -116.4360 & 37.8403 & 1.4926 & 1.1539 & 0.9368 & 1.2316 & 65.1718 & 601.4143 & 822.81 + ne of san simeon & usa & 22/12/2003 & 6.5 & 35.7058 & -121.1010 & 120.7670 & 1.1981 & 1.0003 & 0.9268 & 1.0793 & 144.7416 & 46.4220 & -67.93 + & usa and mexico & & & & & 400.0000 & - & - & - & - & 421.4479 & 488.3228 & 15.87 + & usa & & & & & 250.0000 & 1.0302 & 0.9897 & 0.9729 & 1.0172 & 254.9038 & 370.2505 & - + & mexico & & & & & 150.0000 & 1.0853 & 1.0230 & 0.9893 & 1.0341 & 166.5441 & 118.0722 & - + california & usa & 18/10/1989 & 6.9 & 37.0400 & -121.8800 & 2,510.0000 & 1.6103 & 1.2119 & 0.9525 & 1.2724 & 4,898.6913 & 7,316.6145 & 49.36 + south island of new zealand & new zealand & 13/06/2011 & 6.0 & -43.5800 & 172.7400 & 2,816.4549 & 0.9909 & 1.0165 & 1.0099 & 1.0066 & 2,836.7903 & 3,132.1219 & 10.41 + south island of new zealand & new zealand & 21/02/2011 & 6.1 & -43.6000 & 172.7100 & 13,000.0000 & 1.0025 & 1.0047 & 0.9976 & 1.0070 & 13,093.8628 & 17,660.6445 & 34.88 + california & usa & 17/01/1994 & 6.7 & 34.2130 & -118.5360 & 22,920.0000 & 1.4381 & 1.1106 & 0.9204 & 1.2066 & 36,606.3931 & 4,787.6419 & -86.92 + tohuku & japan & 11/03/2011 & 9.0 & 38.2970 & 142.3730 & 37,200.0000 & 0.9935 & 0.9978 & 0.9873 & 1.0106 & 36,877.4566 & 4,611.4482 & -87.49 + the historic data related to all the earthquakes were collected from multiple sources , namely the national geophysical data centre ( nsdc ) @xcite , united states geological survey ( usgs ) @xcite , pager @xcite , shakemap @xcite , em - dat @xcite and cat - dat @xcite . the information collected includes , event data , exposure data , hazard data and loss data . the collected loss data is denoted as @xmath77 which are in usd of year @xmath79 in which the earthquake occurred . normalisation of loss data is reported by @xcite , @xcite , @xcite and @xcite . in this paper , the historic loss data is normalised to 2012 usd , denoted as @xmath78 using the normalisation method described by @xcite and @xcite . three adjustment multipliers are used for the normalisation . firstly , the inflation multiplier , denoted as @xmath80 , which uses the implicit price deflator ( ipd ) for gross domestic product metric sometimes also referred to as gdfdef . using this metric any output obtained at the current price is converted into constant - dollar gdp by taking inflation into account . how much change in a base year s gdp is dependent on the changes in the price level is captured by the metric . this metric is available from economic research of the federal reserve bank of st . louis @xcite and the us bureau of economic analysis @xcite are employed . secondly , the population multiplier , denoted as @xmath81 , which is the ratio of the population in 2012 and the year of occurrence of the earthquake . the population data is available from the census data published by governmental agencies . thirdly , the wealth multiplier , denoted as @xmath82 is computed as @xmath83 . @xmath84 for year , @xmath79 normalised to 2012 is the inflation - corrected wealth adjustment obtained as @xmath85 . the fixed asset and consumer durable goods ( facdg ) metric in a year is used indicative of the wealth in that given year . the computation of fixed assets capture private and governmental assets and the computation of consumer durable goods take into account non - business goods consumed by households . this metric is obtained from the us bureau of economic analysis ( bea ) . the sole use of the measure of wealth is not indicative of inflation adjustments and therefore the consumer price index ( cpi ) is taken into account . further the wealth multiplier are adjusted for population to a per capita basis . the per capita adjustment is taken into account since increase in wealth is dependent on population and the rate of change of wealth and population are different . the normalisation equation is @xmath86 or can be restated as @xmath87 if the implicit price deflator ( ipd ) index of the gdp is taken into account for computing the inflation - corrected wealth adjustment instead of the consumer price index ( cpi ) , then the normalisation equation is @xmath88 in the research reported in this paper , however , @xmath78 is computed using equation which uses both ipd and cpi . the equation takes into account the effect of population based on the consumption ( definition of cpi ) in normalisation . however , there is no direct dependence on population as seen in equation and equation . there are challenges in considering the population for earthquake losses . for example , consider an area that was affected by a major earthquake 20 years ago and was sparsely populated then which resulted in minimal ground up loss . for normalising the loss of that earthquake in 2012 factors such as how densely populated that area was in 2012 and the ground up loss if the earthquake occurred in 2012 needs to be considered . for such a consideration regional population statistics will need to be incorporated into the equation . consider for example the earthquake that affected wnw of ferndale , usa on 9 january 2010 with a magnitude of 6.5 . the historic loss for this earthquake in 2010 us dollars is 25 million , represented as @xmath89 . the @xmath89 value needs to be normalised for 2012 usd denoted as @xmath78 . the implicit price deflator index in 2010 normalised for 2012 , represented as @xmath90 can be obtained as the ratio of the implicit price deflator in 2012 ( @xmath91 ) to the implicit price deflator in 2010 . in 2012 , @xmath93 and @xmath94 ^ 1^. therefore , @xmath95 . computing the wealth multiplier index for 2010 normalised to 2012 denoted as @xmath96 requires the computation of two indices , namely the inflation corrected wealth multiplier index ( @xmath97 ) and the population multiplier index ( @xmath98 ) . the wealth of usa in 2012 is 51,117.4 billion usd and the wealth in 2010 is 48,758.9 billion usd computed from the fixed assets and consumer durable goods account^2^. therefore , the ratio of wealth of 2012 to 2010 is @xmath99 . the consumer price index ( cpi ) for 2012 is 231.227 and for 2010 is 217.230 ^ 3^. the ratio of the consumer price index of 2012 to 2010 is computed as @xmath100 . @xmath97 is obtained by dividing the ratio of wealth and the ratio of cpis of 2012 to 2010 , which is @xmath101 . the population of us in 2012 was 314,055,800 and the population in 2010 was 308,745,538 . therefore , the population multiplier index , @xmath102 . the wealth multiplier index , @xmath96 can then obtained as @xmath103 . therefore , for the us earthquake in 2010 , normalisation in 2012 us dollars is obtained as @xmath104 pager data ( mmi at city level , affected cities due to an earthquake ) for global earthquakes are only available after 2007 . therefore , for earthquakes prior to 2008 a in - house computer script was developed to extract data from two sources . the first source was a list of cities whose population is greater than one thousand people . this list is provided by geonames @xcite and contains all the cities in the world whose population is more than one thousand . the model assumes population as point values for cities in all its computations . however , in reality population is a gradient , and the loss estimation technique presented can not take into account its continuous nature and underestimates the computation of loss taking into account centres with less than a thousand people . the second source was the shakemap file which is a representation of the affected grid on a map due to an earthquake and comprises a large set of point data ( latitude , longitude and the mmi at that point ) . the script extracts the list of cities that are affected within the grid and their mmis . the cities are mapped onto their respective regions using the latitude and longitude information . the exposure data for the geographic levels are collected from publicly available sources . the above inputs were used to calculate losses using the method in the ape - elev model . as shown in equation , the mmi at the city level is used to compute the mdr at the same level using the jaiswal and wald mdr model , either by direct comparison or by interpolation . the exposure data , which is available for higher geographic levels , is disaggregated onto the city level based on population . the losses for a region are then computed by calculating the sum of the losses for individual cities ( loss for individual cities can be computed by the product of the exposure and mdr at the city ) within that region . a number of obstacles were encountered during the validation study , which are as follows : * exposure data had to be collected from a number of disparate sources and was not easy to obtain . * hazard data is not readily available for events preceding 2008 . to collect data for events prior to 2008 , as presented above , an in - house script had to be developed . * as data obtained from multiple sources which do not follow a standard convention were integrated in the validation study , significant efforts had to be made towards ordering and organising data and eliminating irrelevant information from the sources . despite the above obstacles , ( a ) event data was easily collected , ( b ) population data was publicly available and ( c ) the mmi to mdr was straight forward to calculate based on the vulnerability curves used in pager . two column charts were generated based on increasing historic losses . in figure 19 , the predicted and historic losses are shown in millions of usd for events with historic losses less than 1 billion usd , and in figure 20 , for events with historic losses greater than 1 billion usd . there are multiple sources of error in the validation study and are as follows : 1 . input errors , which refer to the flaws and inaccuracy in the input data to the model . cities with a population of over 1000 were only considered . this data is constructed on the assumption that population is a discrete distribution , while in reality it is continuous ( population outside a city with less than 1000 human inhabitants is not considered ) . the population data obtained from geonames was inaccurate since a large number of cities presented zero population . this was partially overcome by doing manual look - ups with other reliable sources . however , conflicts with the dates of census of the geonames and the source of the manual look - ups persisted . application errors , which refer to the inaccuracies and assumptions that exist within the model . the mmi of a city was converted to a mdr value using country - based mmi - mdr curves . the assumption here is that every city follows the same curve ( values ) as of its country . the losses for a few events are calculated in the currency of its country of origin . the value of the currency is then converted to us dollars based on an average conversion rate for the year in which the event occurred . benchmark errors , which refer to the assumptions that exist in setting a benchmark . a range of values are available for historic insured losses . it is difficult to determine which value needs to be selected as the benchmark for comparison against the predicted loss . for certain events , historic insured losses were not available , and therefore , the total economic losses were used to estimate the insured loss . this was based on a countrywide take - up rate which may not be accurate for certain regions in a country . it is observed that there are two events from the sample which have over 100% error . the first event affected california in 06/28/1992 with a magnitude of 7.3 , have significant error . this is likely because the most recent exposure for california was only available for the validation study , thereby leading to a significant over - prediction . the second event occurred on 03/11/2010 in chile with a magnitude of 6.9 . the over - prediction is in part likely due to the fact that exposure was disaggregated based on population . in this case , the assumption that exposure is proportional to population is less accurate since only one city with a population of over 1000 was affected . the seven events that have less than 100% error indicate the model is feasible . further accuracy can be achieved by calibrating the model . the loss predicted by the ape - elev model is a mean value for an earthquake . to study the probability of a loss threshold @xmath105 the @xmath106 distribution which is the standard normal cumulative distribution function is employed as follows @xcite : @xmath107 - \phi \left [ \frac{ln(a ) - \mu_{ln(l)}}{\zeta}\right]\ ] ] where @xmath108 is the predicted value of the logarithm of loss obtained from the model and is assumed to be a lognormal random variable , and @xmath109 is the normalised standard deviation of the logarithm of loss obtained from @xcite . figure 21 shows the estimate of probability of different loss thresholds ( @xmath110 , @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 , @xmath116 ) represented in millions of usd for the earthquakes of table 1 . these loss thresholds best represent magnitude losses and are therefore chosen for validating the results in this paper . different thresholds can be used by appropriately setting @xmath117 and @xmath118 values in equation . in this section we have evaluated the performance on ape - elev both in terms of how well its data acquisition and visualisation facilities are able to capture the evolving history of earthquake alerts and the performance of its simplistic loss model . the tohoku earthquake used in evaluating the feasibility demonstrates how data can be rapidly ingested from multiple sources to visualise earthquake alerts as the data related to the event evolves over hours , days and months after its occurrence . evaluation of loss models is tricky at best due to the inherent difficulty in collect consistent exposure and loss data for historic events . in the case of ape - elev is important to remember that the goal is to produce on a global basis a crude loss estimate rapidly , as an event evolves , based on very limited information . in this context , the distribution of expected losses is much more important than the point estimates . our validation demonstrated that the methodology pioneered in pager @xcite for economic loss can be usefully applied in the context of portfolio losses . in 50% of our evaluation events the observed historical losses and the predicted losses fall into the same loss threshold . in 90% of our test events the observed historical losses and the predicted losses fall into the two highest loss thresholds . given the limited data , the loss model gives reasonable order of magnitude estimates , but it is important that users be aware of the inherent limitations of the underlying approach . in the time line of an earthquake the sensory data provided by sources such as pager / shakemap evolves over time . for example , sensory data was updated fifteen times for the tohoku earthquake ranging from within an hour to six months after the earthquake . the data was first issued twenty three minutes after the earthquake and updated four times during the first day alone . not only did the earthquake event unfold over time but the data describing the event and our knowledge of the event evolved . the data available initially alone is not sufficient to produce reliable loss estimates . therefore , analysis of an event soon after it has occurred is challenging and important to generate reliable loss estimates . for an earthquake model to be useful in days and weeks after the event , it needs to support ( a ) rapid data ingestion , ( b ) rapid loss estimation , ( c ) rapid visualisation and integration of hazard , exposure , and loss data from multiple sources , and ( d ) rapid visualisation of hazard , exposure and vulnerability loss data at multiple geographic levels . this paper has presented the design and development of such a model , ape - elev ( automated post - event earthquake loss estimation and visualisation ) . the model comprises three modules , firstly , the earthquake loss estimator ( ele ) , the earthquake visualiser ( ev ) and the elev database ( elev - db ) . the ele module is built on relying multiple data sources for accessing real - time earthquake data . financial losses relevant to the insurance and reinsurance industry are particularly taken into account in the model and are estimated at different geographic levels . the visualisation of the losses on a geo - brower is facilitated by the ev module . the elev - db module aids the cohesive functioning of the ele and ev modules . the recent tohoku earthquake is used as a test case to demonstrate the feasibility of the ape - elev model and how an evolving view of the event is generated using the model . two types of losses , namely ground up and net of facultative losses are computed for the earthquake . further , a set of ten global earthquakes are chosen to validate the model by ( a ) computing the percentage error between the predicted loss and historic loss values and ( b ) estimating the probability of loss thresholds for the earthquakes . in the study , all historic loss values are normalised to 2012 us dollars . the key observation is that the model produces reasonable order of magnitude estimates . a video demonstrating a prototype of the distributed ape - elev is available at http://www.blessonv.com/software/ape-elev . future work will aim to refine the model by calibrating the pager vulnerability curves ( for economic losses ) for a more accurate use in portfolio insured loss models . a comparison study of estimated losses against normalised historic losses for a larger number of recent earthquake events will be pursued . extending ape - elev for secondary hazards such as tsunamis and floods will be pursued . efforts will also be made towards augmenting the loss model results with any available historical data points . the distributed ape - elev system will be extended for taking custom user input for exposure and catastrophe data and for adjusting the output presentation as required . a study to quantify the input , benchmark and application errors and consider their impact on the estimated loss will be pursued . amini , j. , karami , j. , sarab , a. a. , and safarrad , t. , an evaluation of the radius model in assessing the damages caused by earthquake via gis ( case study region1 tehran ) , urban - regional studies and research journal , number 11 ( 2012 ) boomer , j. , spence , r. , erdik , m. , tabuchi , s. , aydinoglu , n. , booth , e. , re , d. del , and peterken , o. , development of an earthquake loss model for turkish catastrophe insurance , journal of seismology , volume 6 , number 3 , pp . 431 - 446 ( 2002 ) eguchi , r. t. , goltz , j. d. , seligson , h. a. , flores , p. j. , blais , n. c. , heaton , t. h. , and bortugno , e. , real - time loss estimation as an emergency response decision support system : the early pot - earthquake response tool ( epedat ) , earthquake spectra , volume 13 , pp . 815 - 832 ( 1997 ) erdik , m. , aydinoglu , n. , fahjan , y. , sesetyan , k. , demircioglu , m. , siyahi , b. , durukal , e. , ozbey , c. , biro , y. , akman , h. , yuzugullu , o. , earthquake risk assessment for istanbul metropolitan area , earthquake engineering and engineering vibration , volume 2 , issue 1 , pp.1 - 23 ( 2003 ) frolova , n. , larionov , v. , and bonnin , j. , earthquake casualties estimation in emergency mode , human casualties in earthquakes , advances in natural and technological hazards research 29 , spence , r. , so , e and scawthorn , c. ( editors ) , springer , pp . 107 - 124 ( 2011 ) huyck , c. k. , chung , h. c . , cho , s. , mio , m. z. , ghosh , s. , eguchi , r. t. and mehrotra , s. , centralized web - based loss estimation tool : inlet for disaster response , proceeding of spie 6178 , 61780b ( 2006 ) kaestli , p. , wyss , m. , bonjour , c. , wiemer , s. , and wyss , b. m. , a new tool for estimating losses due to earthquakes : quakeloss2 , american geophysical union , fall meeting 2007 , abstract # s51a-0222 ( 2007 ) kamer , y. , demircioglu , m. b. , erdik , m. , hancilar , u. , harmandar , e. , sesetyan , k. , tuzun , c. , yenidogan , c. , and zulfikar , a. c. , earthquake loss estimation routine eler v 3.0 user manual , department of earthquake engineering , bogazici university , turkey ( 2010 ) miller , s. , muir - wood , r. , and boissonnade , a. , an exploration of trends in normalized weather - related catastrophe losses , climate extremes and society , diaz , h. f. , and murnane , r. j. , ( editors ) , pp . 225 - 247 ( 2008 ) molina , s. , lang , d.h . , and lindholm , c.d . , selena - an open - source tool for seismic risk and loss assessment using a logic tree computation procedure , computers & geosciences , volume 36 , issue 3 , pp . 257 - 269 ( 2010 ) muto , m. krishnan , s. , beck , j. l. , and mitrani - reiser , j. , seismic loss estimation based on end - to - end simulation , proceedings of the 1st international symposium on life - cycle civil engineering , lake como , italy ( 2008 ) pielke , r. a. , jr . , rubiera , j. , landsea , c. , fernandez , m. l. , and klein , r. , hurricane vulnerability in latin america and the carribean : normalized damage and loss potentials , natural hazards review , vol . 4 , issue 3 . pp . 101 - 114 ( 2003 ) sousa , m. l. , campos costa , a. , carvalho , a. , and coelho , e. , an automatic seismic scenario loss methodology integrated on a geographic information system , proceedings of the 13th world conference on earthquake engineering , vancouver , canada , paper no . 2526 ( 2004 ) spencer , b.f . , myers , j. d. , and yang , g. , maeviz / neesgrid and applications overview , proceedings of the 1st international workshop on an earthquake loss estimation program for turkey , istanbul , turkey ( 2005 ) trendafiloski , g. , wyss , m. , and rosset , ph . , loss estimation module in the second generation software qlarm , human casualties in earthquakes , advances in natural and technological hazards research 29 , spence , r. , so , e and scawthorn , c. ( editors ) , springer , pp . 95 - 106 ( 2011 ) wald , d. , lin , k. -w . , porter , k. and turner , l. , shakecast : automating and improving the use of shakemap for post - earthquake decision - making and response , earthquake spectra , volume 24 , no . 533 - 553 ( 2008 ) wald , d. j. , earle , p. s. , allen , t. i. , jaiswal , k. , porter , k. , and hearne , m. , development of the u.s . geological survey s pager system ( prompt assessment of global earthquakes for response ) , proceedings of the 14th world conference on earthquake engineering , beijing , china ( 2008 )
we study the impact of quantum and thermal fluctuations on properties of quantum phase transitions occurring in systems of itinerant fermions with main focus on the order of these transitions . our approach is based on a set of flow equations derived within the functional renormalization group framework , in which the order parameter is retained as the only degree of freedom , and where the effective potential is parametrized with a form allowing for both first and second order scenarios . we find a tendency to turn the first order transitions within the bare model into second order transitions upon accounting for the order parameter fluctuations . we compute the first and second order phase boundary lines as a function of a non - thermal control parameter in the vicinity of a quantum phase transition . we analyze crossovers of the shift exponent governing the shape of the line when the system is tuned close to a quantum tricritical scenario , where a second order phase transition line terminates at a quantum tricritical point .
0811.0388
quantum phase transitions in systems of itinerant fermions continue to attract considerable interest.@xcite on one hand this attention is due to the relevance of quantum critical points for understanding collective phenomena like high-@xmath1 superconductivity and other non - conventional properties of experimentally investigated compounds.@xcite on the other , it is related to peculiarities that so far prohibited the emergence of a fully convincing theoretical description of most systems exhibiting quantum criticality.@xcite the standard approach in describing quantum critical phenomena in systems of itinerant fermions - the hertz - millis theory @xcite - relies on an order parameter field , which is introduced by a stratonovich - hubbard transformation . in a subsequent step the fermionic degrees of freedom are integrated out . the resulting representation of the partition function as a path integral over a bosonic field is usually useful only provided the action may be expanded in the order parameter , and the relevant vertex functions can be evaluated for the values of momentum and frequency that correspond to the anticipated instability . the validity of such an expansion was questioned for both zero and finite momentum instabilities in the magnetic channel . in case of ferromagnetic transitions it was argued that the correct effective action involves additional terms @xcite leading to a first order transition , @xcite while for the case of antiferromagnetic transition in @xmath4 integrating out gapless fermionic modes can yield singular vertex functions all of which were argued to be marginal . @xcite despite these theoretical insufficiencies , hertz - millis theory is successful in explaining a number of nonconventional properties of many systems exhibiting quantum criticality . @xcite in recent years this theory was further extended to account for a number of systems not described by the original approach . these include metamagnetic quantum critical points,@xcite field - tuned quantum critical points where a term describing precession of the order parameter has to be retained in the action,@xcite and phase transitions induced by a nonequilibrium drive . @xcite an interesting issue concerns the actual order of specific quantum phase transitions . hertz - millis theory by construction allows for continuous transitions only . however , microscopic models , which usually rely on fermionic degrees of freedom and meanfield - like treatment , often predict a first order scenario in case of both magnetic and charge instabilities occurring at @xmath5 , and for specific cases corresponding to @xmath6 . @xcite theoretical studies of the interplay between first and second order scenarios are further motivated by experimental results , where the order of the quantum phase transitions in specific compounds is altered by varying external magnetic field @xmath7.@xcite upon increasing @xmath7 , the tricritical point separating the first order transition at @xmath8 from the second order at @xmath9 , is shifted towards lower values of @xmath10 and finally vanishes at @xmath11 . therefore , the order of the transition at small @xmath10 can be controlled by varying @xmath7 . of particular interest is the case @xmath12 , where the transition is second order for all @xmath13 , but scaling behavior of physical quantities at low temperatures is in variance with that predicted by hertz - millis theory . @xcite an interesting scenario of quantum criticality occurs also in systems exhibiting a metamagnetic first order transition terminating with a critical end point.@xcite by tuning pressure the critical end point is suppressed to @xmath11 resulting in a distinct type of a quantum critical point . in this work we investigate the possibility of altering the order of quantum phase transitions by order parameter fluctuations . subsequently , we study shapes of phase boundaries , focusing mainly on the cases where such change occurs . we rely on the conventional bosonic hertz action adapted to phases with broken symmetry and retaining also a @xmath0 term in the effective potential . we analyze two- and three- dimensional systems where discrete symmetry - breaking occurs at @xmath5 ( the dynamical exponent @xmath14 ) . our results should apply whenever an action of this type can be constructed . possible examples are magnetic transitions with ising - like symmetry and the so - called pomeranchuk instabilities @xcite where the discrete point - group lattice symmetry of the fermi surface is broken . for these transitions mean - field studies typically predict a first order scenario at @xmath11 but second order for slightly higher temperatures ( for @xmath9).@xcite we argue that the tricritical temperature is lowered by including the order - parameter fluctuations . under appropriate conditions the tricritical point may be suppressed to @xmath11 , in which case a quantum critical or tricritical point is realized . our study is based on the one - particle irreducible version of the functional renormalization group ( rg).@xcite this framework was applied extensively in the context of classical critical phenomena , where it yields a unified description of @xmath15-symmetric scalar models including systems in @xmath4.@xcite it also provides a suitable framework to treat quantum criticality.@xcite we apply the derivative expansion to quadratic order , and the effective potential is parametrized in a form allowing for the occurrence of both - first and second order transition . an advantage of the functional rg is that it allows control over effects occurring at different energy scales , and treatment of quantities that need not be universal . the outline of this work is as follows : in sec . ii we introduce the bare action to be applied in the rg procedure and provide a phase diagram of the unrenormalized theory . in sec . iii we discuss the applied truncation of the functional rg flow equation and derive equations governing the evolution of quantities parametrizing the effective potential and the inverse propagator upon reducing the cutoff scale . iv is devoted to results obtained for the renormalized phase diagrams in the case @xmath11 . in sec . v we consider @xmath13 . we discuss the computed phase diagrams for cases where the phase transition is second order for @xmath13 and terminates at a quantum critical or tricritical point at @xmath11 . we also analyze the case where a tricritical point situated at @xmath13 occurs . for @xmath16 and a choice of system parameters corresponding to proximity to the quantum tricritical scenario we compute the crossovers of the shift exponent @xmath3 , characterizing the shape of the critical line , from the hertz - millis value @xmath17 to @xmath18 . this latter value of the shift exponent is specific to quantum tricritical behavior in @xmath19 . the obtained value of @xmath20 is justified by invoking a scaling argument in sec . vi , where we also classify and discuss scaling regimes emergent from our analysis . finally , sec . vii contains a summary of the work . in this study we rely on the conventional description of quantum critical points in systems of itinerant fermions in terms of the bosonic hertz action.@xcite this is derived by applying a stratonovich - hubbard transformation to the path - integral representation of the partition function of a fermionic system.@xcite this way the order - parameter field is introduced and the original fermionic degrees of freedom are decoupled allowing one to integrate them out . subsequently , one expands the resulting action in powers of the order - parameter field keeping only the most relevant dependence of the propagator on frequency and momentum , and neglecting this dependence in case of higher - order vertices . in the case of instabilities occurring at wavevector @xmath5 , which is considered here , this leads to the following action @xmath21 = \frac{t}{2 } \sum_{\omega_n } \int\frac{d^dp}{(2\pi)^d } \ , \phi_p \left ( \frac{|\omega_{n}|}{|\mathbf{p}| } + \mathbf{p}^{2 } \right ) \phi_{-p } + u[\phi ] \ ; . \label{eq : lagrangian}\end{aligned}\ ] ] here @xmath22 is the scalar order parameter field and @xmath23 with @xmath24 its momentum representation ; @xmath25 with integer @xmath26 denotes the ( bosonic ) matsubara frequencies . momentum and energy units are chosen such that the prefactors in front of @xmath27 and @xmath28 are equal to unity . the action is regularized in the ultraviolet by restricting momenta to @xmath29 . the term @xmath27 effectively accounts for overdamping of the order - parameter fluctuations by fermionic excitations across the fermi surface . the expression eq . ( [ eq : lagrangian ] ) is valid for @xmath30 , and @xmath27 sufficiently small , which is the limit relevant for the physical situation considered here.@xcite the potential @xmath31 $ ] is usually parametrized by a quartic form with a single minimum at @xmath32 or with two minima at @xmath33 . as we wish to allow for both , first and second order transitions , we expand @xmath31 $ ] to sixth order in @xmath22 , imposing @xmath34=\int_0^{1/t } d\tau \int \ ! d^d x \left[a_6\phi^6+a_4\phi^4+a_2\phi^2\right ] . \label{eq : ef_potential}\ ] ] we consider @xmath35 to stabilize the system at large @xmath36 . in eq . ( [ eq : ef_potential ] ) we assumed the existence of an expansion of @xmath31 $ ] , which correctly captures the properties of @xmath31 $ ] also around the non - zero minima . one should be aware , that in many physical situations such expansion necessarily has to involve also terms of higher order in @xmath22 , or simply can not be constructed . the latter , more severe case occurs whenever the radius of convergence around @xmath32 is smaller than the distance between @xmath32 and the minimum at @xmath37 . our belief is that except for the vicinity of the tricritical point , such complication is rather generic . it arises for example in case of models describing symmetry - breaking fermi surface deformations in @xmath4 ( see e.g. ref . ) . while the model defined by eq . ( [ eq : ef_potential ] ) is one of the simplest correctly accounting for universal critical and tricritical properties , its applicability in the context of specific microscopic models has to be verified for each individual case in the regime of parameters where a first - order transition occurs . also note , that when @xmath38 is expanded around the non - zero minimum @xmath39 , terms involving odd powers of field are generated . the coefficients of these terms can be expressed as functions of the coupling @xmath2 , @xmath40 , @xmath41 . there arise just three independent couplings because of the assumed inversion symmetry @xmath42=u[\phi]$ ] . this model does not apply to physical situations where the symmetry of @xmath38 is explicitly broken by including a field coupling linearly to @xmath22 . for example the present work does not apply to the metamagnetic transitions . @xcite the bare effective potential is now analyzed as a function of the parameters @xmath2 , @xmath40 , @xmath41 . it exhibits a single minimum for @xmath43 , @xmath44 , or @xmath43 , @xmath45 . for @xmath46 there are two minima located at @xmath47 for @xmath48 , @xmath49 , one encounters three minima at @xmath32 and @xmath50 . if in addition @xmath51 holds , @xmath52<u[0]$ ] and a configuration corresponding to @xmath53 is stable , otherwise @xmath52>u[0]$ ] and @xmath32 is the stable state . along the line @xmath54 , @xmath45 a second - order phase transition occurs , while a first - order transition line is located along the parabola @xmath55 for @xmath48 . the tricritical point occurs at @xmath56 . this simple analysis of the bare phase diagram is summarized in fig.1 . model . the line @xmath54 , @xmath45 marks the locus of second - order phase transitions , while the curve @xmath55 , @xmath48 corresponds to a first order transition . the tricritical point is located at @xmath56 . the curves @xmath54 and @xmath57 for @xmath48 mark boundaries of the region where metastable states are present.,width=307 ] in course of the subsequent analysis we investigate the stability of this phase diagram upon including fluctuations , depending on the system dimensionality ( for @xmath58 ) and temperature ( for @xmath59 ) . our analysis is based on approximate flow equations derived by truncating the exact functional rg evolution equation . we use the one - particle irreducible scheme . the flow equation @xcite @xmath60= \frac{1}{2}\text{tr}\frac{\partial_\lambda r^{\lambda}}{\gamma^{(2)}\left[\phi\right ] + r^{\lambda } } \label{eq : flow_eqn}\end{aligned}\ ] ] describes the evolution of the effective action @xmath61 $ ] , the generating functional for one - particle irreducible vertex functions in the presence of an infrared cutoff at scale @xmath62 , under reducing the cutoff scale . in eq . ( [ eq : flow_eqn ] ) @xmath63 denotes the cutoff function added to the inverse propagator to cut off modes with momentum below the scale @xmath62 , and @xmath64 = \delta^{2}\gamma^{\lambda}[\phi]/ \delta \phi^{2}$ ] . in momentum representation the trace sums over momenta and matsubara frequencies , that is @xmath65 . in what follows , we use a regulator term with the optimized litim cutoff function @xcite @xmath66 where @xmath67 is the wave function renormalization . the effective action interpolates smoothly between the bare action eq . ( [ eq : lagrangian ] ) for @xmath68 , and the full effective action for @xmath69 . the quantity @xmath70 $ ] can be interpreted as the gibbs free energy of the rapid modes , i.e. modes with momentum above the cutoff scale @xmath62 . by decreasing @xmath62 fluctuations of lower momentum are included and in the limit @xmath71 the full gibbs free energy functional is recovered . the general strategy employed in this paper for computing phase diagrams amounts to evaluating @xmath72 $ ] as function of the ( bare ) parameters describing the system . the equilibrium order parameter is extracted by finding the global minimum of the resulting free energy . ( [ eq : flow_eqn ] ) describes the flow of an infinite number of couplings and to make progress we shall impose a suitable parametrization in terms of a finite number of variables . namely , we apply a truncation in which the effective potential @xmath31 $ ] , that is @xmath70 $ ] evaluated for momentum - independent field , is taken as a sixth - order polynomial as in eq . ( [ eq : ef_potential ] ) , with flowing coefficients , while the inverse propagator is assumed to maintain the form due to hertz , as in eq . ( [ eq : lagrangian ] ) , where the coefficients in front of @xmath27 and @xmath28 may be renormalized when necessary . thus , we put @xmath73 = z_{\omega } \frac{|\omega_{n}|}{|\mathbf{p}| } + z\mathbf{p}^2 + 2a_2 + r^{\lambda}(\textbf{p } ) \ ; . \label{green}\ ] ] the quantities @xmath67 , @xmath74 depend on the scale @xmath62 only . also note , that the allowed forms of the effective potential , as illustrated in fig . [ fig : eta_vs_lambda ] are typical to @xmath0 like truncations . in particular , if no truncation of the functional rg flow equation was performed , the effective potential should obey bounds on the curvature at @xmath75,@xcite and become convex for @xmath71 . an analogous truncation ( retaining terms up to quartic order in the effective potential ) was applied in ref . , where it yields a simple extension of hertz - millis theory to phases with broken discrete symmetry , capturing the non - gaussian fixed point behavior at finite temperatures . upon evaluating eq . ( [ eq : flow_eqn ] ) for a momentum - independent field @xmath22 we obtain an exact equation governing the flow of the effective potential @xmath31 $ ] ( see ref . ) @xmath76=2v_d\lambda^d l_0^d(w)\ ; , \label{ef_pot_flow}\ ] ] where @xmath77 , and @xmath78 here @xmath79 , @xmath80 + 2\rho u''[\rho])$ ] , and @xmath81 . for our parametrization , the partial differential equation eq . ( [ ef_pot_flow ] ) can be projected onto a set of three ordinary differential equations . following ref . we neglect @xmath82 in @xmath83 . the neglected terms would yield corrections to the flow equations that are linear in @xmath84 and therefore irrelevent except close vicintiy of second order transition in @xmath13 . in the latter case these corrections are small as compared to other included terms involving @xmath84 due to the presence of additional factors involving the interaction couplings and mass . we use the condition @xmath85 = 0 $ ] , to write @xmath86 = \partial_t u'[\rho_0 ] + u''[\rho_0 ] \ , \partial_t\rho_0 $ ] . inserting @xmath87 $ ] as obtained by differentiating eq . ( [ ef_pot_flow ] ) with respect to @xmath88 at @xmath89 , one obtains the flow equation for @xmath90 @xmath91 where the threshold functions @xmath92 are defined by @xmath93 @xmath94 from eq . ( [ flow_rho_0 ] ) we conclude that @xmath90 decreases under the flow . this follows from @xmath95 , and @xmath96 . analogously , by evaluating the second and third derivatives of eq . ( [ ef_pot_flow ] ) at @xmath90 and using @xmath97=48a_6\rho+8a_4 $ ] , @xmath98=48a_6 $ ] , which follows from the ansatz , eq . ( [ eq : ef_potential ] ) , we project out the equations governing the flow of @xmath40 and @xmath41 : @xmath99 - 6\rho_0\partial_t a_6\ ; , \label{flow_a4}\ ] ] @xmath100 , \label{flow_a6}\ ] ] where the threshold functions are evaluated at @xmath101 corresponding to @xmath90 , that is @xmath102 . the flowing coupling @xmath2 is calculated using eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] ) , and eq . ( [ phi_0 ] ) , which yields @xmath103 the contributions to the flow of the couplings @xmath2 , @xmath40 , @xmath41 are illustrated in terms of feynman diagrams in fig . [ feynm_diag ] . 20070802_loops_1 @xmath104 % a_4 & : & \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(20,12 ) \fmfpen{thin } \fmfleftn{l}{2}\fmfrightn{r}{2 } \fmfrpolyn{full , tension=0.9}{g}{4 } \fmfpolyn{full , tension=0.9}{k}{4 } \fmf{dbl_wiggly}{l1,g1}\fmf{dbl_wiggly}{l2,g2 } \fmf{dbl_wiggly}{k1,r1}\fmf{dbl_wiggly}{k2,r2 } \fmf{dbl_wiggly , left=.5,tension=.2}{g3,k3 } \fmf{dbl_wiggly , right=.5,tension=.2}{g4,k4 } \end{fmfgraph * } } } + \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(20,12 ) \fmfpen{thin } \fmfleftn{l}{2}\fmfrightn{r}{2 } \fmfrpolyn{full , tension=1.1}{v}{4 } \fmfpolyn{full , tension=0.5}{w}{3 } \fmfpolyn{full , tension=0.5}{y}{3 } \fmf{dbl_wiggly}{l1,v1}\fmf{dbl_wiggly}{l2,v2 } \fmf{dbl_wiggly}{w1,r1}\fmf{dbl_wiggly}{y1,r2 } \fmf{dbl_wiggly , left=.5,tension=.4}{v3,y2 } \fmf{dbl_wiggly , right=.5,tension=.4}{v4,w3 } \fmf{dbl_wiggly , right=.5,tension=.4}{w2,y3 } \end{fmfgraph * } } } \nonumber\\[+2 mm ] & + & \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(20,12 ) \fmfpen{thin } \fmfleftn{l}{1}\fmfrightn{r}{3 } \fmfrpolyn{full , tension=0.4}{v}{3 } \fmfpolyn{full , tension=1.3}{w}{5 } \fmf{dbl_wiggly}{l1,v1 } \fmf{dbl_wiggly}{w1,r1}\fmf{dbl_wiggly}{w2,r2}\fmf{dbl_wiggly}{w3,r3 } \fmf{dbl_wiggly , left=.5,tension=.4}{v2,w4 } \fmf{dbl_wiggly , right=.5,tension=.4}{v3,w5 } \end{fmfgraph * } } } + \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(15,10)\fmfpen{thin } % pp - with rpa with ladder , inset \fmfsurroundn{v}{6 } \fmfpolyn{full , tension=2.0}{g}{6 } \fmf{dbl_wiggly , left=0.5}{v1,g1 } \fmf{dbl_wiggly , right=0.5}{v4,g4 } \fmf{dbl_wiggly , straight , tension=0.2}{v5,g5 } \fmf{dbl_wiggly , straight , tension=0.2}{v6,g6 } % \fmffreeze \fmf{dbl_wiggly , tension=0.1,right=0.7}{g2,v2 } \fmf{dbl_wiggly , tension=0.1,right=0.7}{v2,v3 } \fmf{dbl_wiggly , tension=0.1,right=0.7}{v3,g3 } \end{fmfgraph * } } } \nonumber\\[+3 mm ] % a_6 & : & \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(20,12 ) \fmfpen{thin } \fmfleftn{l}{2}\fmfrightn{r}{4 } \fmfrpolyn{full , tension=1.4}{v}{4 } \fmfpolyn{full , tension=2.0}{w}{6 } \fmf{dbl_wiggly}{l1,v1}\fmf{dbl_wiggly}{l2,v2 } \fmf{dbl_wiggly}{w1,r1}\fmf{dbl_wiggly}{w2,r2}\fmf{dbl_wiggly}{w3,r3}\fmf{dbl_wiggly}{w4,r4 } \fmf{dbl_wiggly , left=.5,tension=.4}{v3,w5 } \fmf{dbl_wiggly , right=.5,tension=.4}{v4,w6 } \end{fmfgraph * } } } + \parbox{25mm}{\unitlength=1mm\fmfframe(2,2)(1,1 ) { \begin{fmfgraph*}(20,15 ) \fmfpen{thin } \fmfleftn{l}{4}\fmfrightn{r}{4}\fmfbottomn{b}{4 } \fmfrpolyn{full , tension=1.1}{v}{4 } \fmfpolyn{full , tension=1.1}{w}{4 } \fmfpolyn{full , tension=0.8}{y}{4 } \fmf{dbl_wiggly}{l2,v1}\fmf{dbl_wiggly}{l3,v2 } \fmf{dbl_wiggly}{w1,r2}\fmf{dbl_wiggly}{w2,r3 } \fmf{dbl_wiggly}{y1,b2}\fmf{dbl_wiggly}{y2,b3 } \fmf{dbl_wiggly , left=0.7,tension=.4}{v3,w3 } \fmf{dbl_wiggly , tension=.4}{v4,y4 } \fmf{dbl_wiggly , tension=.4}{w4,y3 } \end{fmfgraph * } } } \nonumber\\[-5mm]\nonumber\end{aligned}\ ] ] the threshold functions can be split into classical contributions , that is contributions from @xmath105 , and the quantum parts . the matsubara sums are then evaluated analytically , yielding the following expression for the function @xmath106 : @xmath107 where @xmath108 and @xmath109 here we substituted @xmath110 and @xmath111 . the polygamma function @xmath112 originated from evaluating matsubara sum of the type @xmath113 by taking derivatives with respect to @xmath101 and using @xmath114 , we evaluate the higher order threshold functions . for any @xmath115 one finds a classical contribution proportional to @xmath10 , and therefore vanishing in the zero temperature limit ; and a quantum contribution that survives the limit @xmath116 , however , at finite @xmath10 , vanishes quickly as the cutoff scale is reduced . one technical remark is in place here . in the derivation sketched above it is assumed that @xmath117\neq 0 $ ] . this condition is not fulfilled whenever the minimum of @xmath118 $ ] at @xmath119 disappears in course of the flow , that is the metastable states are swept away by fluctuations . indeed , by inspecting eq . ( [ flow_rho_0 ] ) we realize that in such case a singularity in @xmath120 develops for some finite @xmath121 . while solving the flow equations ( see sec . iv , v ) we will simply terminate the flow whenever such singularity is encountered and conclude that the system is in the symmetric phase with no metastable states present , that is we assume that once the metastable states at @xmath90 are eliminated by fluctuations , they would not appear again . at this point it is also worthwhile making a comment concerning some qualitative properties of eq . ( [ flow_a4 ] ) . its right - hand side contains terms proportional to powers of @xmath122 and a single negative term involving @xmath41 . in the language of feynman diagrams , this term originates from contracting the external legs of the 6-point vertex with a single - scale propagator ( the last diagram contributing to the flow of @xmath40 in fig . [ feynm_diag ] ) . once we choose the mean field ( bare ) tricritical point coordinates @xmath123 , @xmath124 as the initial condition of the evolution equations eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] ) , only this term survives , and the system is driven into a region of the phase diagram with @xmath45 , where a second - order transition occurs for some value of @xmath2 . the same happens if one starts with a negative , though sufficiently close to zero value of the coupling @xmath40 . this tendency occurs only in the initial stage of the flow . however , it is unlikely to be inverted in later stages of the evolution , because @xmath125 vanishes quickly as @xmath71 , thus , significant renormalizations occur mainly at initial stages of the flow - see sec . iv and v. within the truncation proposed here , the inverse propagator is parametrized with the flowing @xmath67-factors . in what follows , we will in addition neglect the renormalization of @xmath74 . as was argued in ref . , this is not expected to influence the results for the phase diagram . renormalization of the factor @xmath67 will be considered only at @xmath13 , where non - gaussian critical behavior occurs . the evolution equation for @xmath67 is obtained by considering the laplacian of the inverse propagator @xmath126 \right|_{\textbf{p } = 0 } \ ; , \label{zp}\ ] ] and plugging in the flow equation for @xmath127 : @xmath128 . \label{gamma2flow}\ ] ] the latter is obtained by evaluating the second functional derivative of eq . ( [ eq : flow_eqn ] ) . in our calculations we use the order parameter anomalous dimension , related to @xmath67 by @xmath129 by carrying out the matsubara sums one arrives at the following expression for @xmath84 : @xmath130 - \frac{1}{2d } { \tilde t}^{-5 } \int_0 ^ 1 dy y^{(d+3z)/2 - 7 } \nonumber\\ & & \bigg[-6 y^2 ( d-1 ) \tilde{t}^2 \psi_2(x ) - 2y^{5/2 } \big(2(8-d ) \nonumber\\ & & ( 1 + w - y ) + 3[(d-14)(1 + w ) + 8y ] + 27(1+w ) \big ) \nonumber\\ & & \tilde{t } \psi_3(x ) - y^{3 } [ 2y + w+1]^2 \psi_4(x ) \bigg]\bigg\ } \ ; , \label{eq : etas}\end{aligned}\ ] ] where @xmath131 . as before , one identifies a classical contribution , originating from @xmath105 , and the remaining quantum contribution ( involving @xmath132 ) . in this section we present results for the renormalized effective potential at @xmath11 . as the critical behavior is governed by a gaussian fixed point in this case,@xcite we put @xmath133 . the classical contributions to the threshold functions vanish , while the integrals in their quantum parts can be carried out analytically . for @xmath134 one obtains : @xmath135 as @xmath116 . the remaining threshold functions are evaluated by taking consecutive derivatives of eq . ( [ threshold_11 ] ) with respect to @xmath101 . the flow equations are then solved numerically in @xmath4 , and @xmath16 for @xmath136 , @xmath137 , and for different choices of initial couplings corresponding to a first order transition at mean - field level . namely , we consider the initial value of @xmath138 , @xmath40 negative , and @xmath2 ranging between small negative and small positive values . for the numerical computations we set the upper cutoff @xmath139 . the renormalized phase diagrams are constructed by inspecting the form of the renormalized effective potential ( at @xmath71 ) for the different choices of the initial couplings . the locus of phase transition is identified with the line in the @xmath140 plane , where the global minimum of @xmath38 becomes nonzero , which happens continuosly or discontinuously as function of @xmath2 or @xmath40 . the metastable states ( corresponding to local minima of @xmath38 ) persist within a region in the @xmath140 plane in the vicinity of the first order transition line . an example solution of the flow equations in case where the effective potential is renormalized to a form corresponding to a state in the vicinity of a second - order transition , is provided in fig . [ flow_d2z3t0 ] . , [ flow_a4 ] , [ flow_a6 ] ) for @xmath4 , @xmath11 and the initial couplings chosen as follows : @xmath141 , @xmath142 , @xmath143 , and where @xmath144 . the parameter @xmath90 is renormalized towards lower values , which leads to expanding the region of the phase diagram where no symmetry - breaking occurs . the coupling @xmath40 increases and reaches a positive value ( specific to systems exhibiting a second - order transition ) in the final part of the flow . , width=307 ] in fig . [ flow_d2z3t0_dod ] we present a solution for a situation where the absolute minimum of @xmath38 becomes zero in course of the flow , while the metastable states persist for @xmath71 . therefore , the average order parameter exhibits a discontinuity as function of @xmath62 . , [ flow_a4 ] , [ flow_a6 ] ) for @xmath4 , @xmath11 and the initial couplings chosen as follows : @xmath145 , @xmath146 , @xmath143 , and where @xmath144 . the parameter @xmath147 corresponding to the global minimum of @xmath38 exhibits a jump as @xmath148 $ ] is shifted above @xmath149 $ ] . , width=307 ] the results for the renormalized phase diagrams in the variables @xmath140 for @xmath4 and @xmath16 are presented in figs . [ fig : plot_d2z3 ] , [ fig : plot_d3z3 ] . model in @xmath4 . the renormalized tricritical point is shifted towards negative @xmath40 . the dashed lines and crosses mark boundaries of the regions where metastable configurations are present in bare and renormalized theory , respectively . , width=307 ] model in @xmath16 . the renormalized tricritical point is shifted towards negative @xmath40 . the dashed lines and crosses mark boundaries of the regions where metastable configurations are present in bare and renormalized theory , respectively . , width=307 ] as anticipated , the region corresponding to the disordered phase in the phase diagram is extended in course of renormalization . one also finds that upon including order parameter fluctuations , the tricritical point is shifted towards negative values of @xmath40 . the magnitude of this shift is nearly one order of magnitude larger in the case @xmath4 , where fluctuations are more pronounced . we also note that the renormalization of the system parameters occurs at the beginning of the flow only . for the present case @xmath11 it is caused exclusively by quantum fluctuations . a consequence of the results in figs . [ fig : plot_d2z3 ] , [ fig : plot_d3z3 ] is that there exists a regime of parameters , for which the quantum phase transition is first order at mean - field level ( within the bare theory ) , and turns continuous when fluctuations are accounted for . this change of the quantum phase transition s character does not occur in the other direction within the framework applied here . in the subsequent section , we analyze the system at finite temperatures and argue that the tendency towards second order scenario is even stronger when thermal fluctuations are present . therefore two interesting possibilities are conceivable - either the transition is second order for @xmath59 , or it becomes of second order for temperatures exceeding a tricritical temperature @xmath150 . of special interest is the case @xmath12 , where the finite @xmath10 second order phase transition line terminates at @xmath11 at a quantum tricritical point . in this section we discuss renormalization of the effective potential @xmath31 $ ] in case of finite temperatures , where both quantum and thermal fluctuations are present . conceptually , the analysis proceeds along the same lines as for @xmath11 , however the threshold functions now contain both thermal and quantum parts and the integrals they involve can not be carried out analytically . moreover , the anomalous dimension @xmath84 has to be accounted for , as we encounter non - gaussian critical behavior in the vicinity of the transition line . we also note , that even slightly off criticality , the presence of anomalous dimension influences the results for the phase diagram , in particular in @xmath4 . this happens because @xmath84 attains nonzero values at the intermediate stages of the flow , and vanishes only in the infrared limit . as a result , the factor @xmath67 acquires finite , though not negligibly small renormalization . as we checked by explicit calculations , a truncation with @xmath136 , which breaks down only in the immediate vicinity of a second - order transition , is not sufficient to determine the order of the transition . namely , at least in @xmath4 , it gives a very weakly first order transition in the regime where the truncation retaining the flow of @xmath84 predicts a second order scenario . as we also checked , the present truncation taking into account the flow of @xmath67 reproduces the results of sec . iv in the limit @xmath116 . the relative difference in the numerical results for the points exhibited in figs 5,6 using the two different truncations is below 1% . therefore , we analyze the set of flow equations eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) with the aim of determining the phase diagrams in the variables @xmath140 for different temperatures . in a subsequent step we reinterpret our results by considering fixed ( initial ) @xmath48 and plotting phase diagrams in the variables @xmath151 . as before , we set the upper cutoff @xmath139 , for the initial condition we choose @xmath138 , @xmath136 , and consider negative initial values of @xmath40 , and @xmath2 ranging between small negative and small positive values . the factor @xmath74 is set equal to unity . an example solution to the flow equations is given in fig . [ flow_d3z3t0.08 ] . , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) for @xmath16 , @xmath152 and the initial couplings chosen as follows : @xmath153 , @xmath154 , @xmath143 , and where @xmath144 . this choice of parameters corresponds to the phase with broken symmetry , but close to the transition line ( within the renormalized theory ) . the coupling @xmath40 increases and reaches a positive value in the final part of the flow . the anomalous dimension @xmath84 attains nonzero values in the intermediate stages of the flow , although the system is slightly off criticality , and vanishes in the infrared limit . the system exhibits a second order phase transition upon varying @xmath2 , although for the same choice of @xmath155 , @xmath156 , the transition is of first order at @xmath11 - see fig . [ fig : plot_d3z3].,width=307 ] in the solution illustrated in fig . [ flow_d3z3t0.08 ] , the choice of @xmath40 corresponds to a situation where , despite the fact that @xmath40 is negative , one encounters a second order transition upon varying @xmath2 . moreover , for this value of @xmath40 , the transition is of first order at @xmath11 , which can be read off from fig . [ fig : plot_d3z3 ] . the initial value of @xmath2 was chosen so that the system is slightly separated from the transition , but @xmath84 attains nonzero values at an intermediate stage of the flow . by performing such an analysis for different values of @xmath10 , @xmath40 , and @xmath2 , we obtain ( for different fixed @xmath10 ) , phase diagrams analogous to those plotted in figs [ fig : plot_d2z3 ] , [ fig : plot_d3z3 ] for @xmath11 . qualitatively these phase diagrams are similar , however the position of the renormalized tricritical point @xmath157 depends on temperature . we find that both for @xmath4 and @xmath16 @xmath158 is a decreasing function , and therefore the tendency towards a second order transition is enhanced upon increasing @xmath10 . in fig . [ a_4_tri ] , we plot the evaluated @xmath158 for @xmath4 and @xmath16 . the quantity @xmath158 may in general be a complicated function of @xmath41 as well as the upper cutoff @xmath159 , and for more sophisticated truncations also other parameters characterizing the effective action . therefore , its precise form depends on all these parameters . we believe however , that the function would remain monotonously decreasing also with better truncations , provided the qualitative structure of the phase diagram for fixed @xmath10 is as in fig . 1 . this could change for example after including a @xmath160 term and allowing for negative @xmath41 , which however corresponds to a physical situation different to the one studied in this paper . computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) for ( a ) @xmath4 and ( b ) @xmath16 . for @xmath161 the system exhibits a second order transition upon varying @xmath2 , while for @xmath162 the transition is of first order.,title="fig:",width=288 ] computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) for ( a ) @xmath4 and ( b ) @xmath16 . for @xmath161 the system exhibits a second order transition upon varying @xmath2 , while for @xmath162 the transition is of first order.,title="fig:",width=288 ] we proceed by evaluating the phase diagrams in the variables @xmath163 for a fixed ( initial ) coupling @xmath40 . for this aim one solves the flow equations for different @xmath10 and @xmath2 , inspects the resulting renormalized @xmath38 and identifies the relevant global minimum . we consider two possibilities : @xmath164 , in which case the transition is of second order for @xmath59 ; and @xmath165 , where the quantum phase transition is first order and one encounters a tricritical point at some @xmath166 . the results are summarized in figs [ phase_diag_d2_temp ] , [ phase_diag_d3_temp ] . as computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) in @xmath4 . the ordered phase is situated below the transition line . the plots correspond to ( a ) @xmath167 , and ( b ) @xmath168 . the transition line ( a ) terminates at a quantum critical point at @xmath11 , while in case ( b ) the quantum phase transition is of first order , and a tricritical point is present at @xmath169 . , title="fig:",width=288 ] as computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) in @xmath4 . the ordered phase is situated below the transition line . the plots correspond to ( a ) @xmath167 , and ( b ) @xmath168 . the transition line ( a ) terminates at a quantum critical point at @xmath11 , while in case ( b ) the quantum phase transition is of first order , and a tricritical point is present at @xmath169 . , title="fig:",width=288 ] as computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) in @xmath16 . the ordered phase is situated below the transition line . the plots correspond to ( a ) @xmath170 , and ( b ) @xmath171 . the transition line ( a ) terminates at a quantum critical point at @xmath11 , while in case ( b ) the quantum phase transition is of first order , and a tricritical point is present at @xmath172 . , title="fig:",width=288 ] as computed from eqs . ( [ flow_rho_0 ] , [ flow_a4 ] , [ flow_a6 ] , [ eq : etas ] ) in @xmath16 . the ordered phase is situated below the transition line . the plots correspond to ( a ) @xmath170 , and ( b ) @xmath171 . the transition line ( a ) terminates at a quantum critical point at @xmath11 , while in case ( b ) the quantum phase transition is of first order , and a tricritical point is present at @xmath172 . , title="fig:",width=288 ] in cases where the transition is first order for low temperatures we find approximately linear behavior of the phase boundary in the limit @xmath116 both in @xmath4 and @xmath16 . however , in @xmath4 the slope of the transition line is infinite ( or very large ) , unlike in @xmath16 . an important conclusion is that a first order transition occurs for @xmath8 in the considered situations . we do not expect the shapes of the first order transition lines to be universal and therefore , in our opinion , it is not quite justified to compare the results of the simple @xmath0 model to other calculations without first assessing the validity of this approach in the considered context . one may note however , that mean field calculations of the first order phase boundaries in two dimensional systems exhibiting discrete symmetry breaking fermi surface deformations yield predictions qualitatively similar to our results @xcite ( very steep , linear behavior of the phase boundary for @xmath116 ) . our expectation is that the results of fig . [ phase_diag_d2_temp ] ( b ) may apply to the system considered in ref . for nonzero uniform fermionic repulsion term @xmath173 , which is required to assure that the factor @xmath41 in the corresponding landau expansion is positive for temperatures reaching down to zero . the other requirement is that the tricritical point is located close to @xmath11 ( as in fig . 4 ( b ) in the abovementioned work ) . when the results for @xmath16 are compared to data from experiments on compounds exhibiting ferromagnetic properties ( see e.g. refs . ) , one finds qualitative agreement in that the first order transition is located at lower @xmath10 , reaching down to @xmath11 , while a second order transition occurs at slightly higher @xmath10 . however , another mechanism driving the transition first order is present for the case of ferromagnets , and the corresponding free energy functional involves terms logarithmic in @xmath22.@xcite this fact is not accounted for within the present approach . as regards to the situations where a quantum critical point is realized , we recover approximately linear behavior of the phase boundary in @xmath4 , as dictated by the hertz - millis theory . the shapes of the phase boundaries in such case for @xmath16 are discussed in more detail in subsection v b. in this subsection we analyze the behavior of the order parameter upon approaching the phase transition line along isotherms for which different situations occur . we also compute the magnitude of the order parameter jump when approaching the tricritical point along the coexistence curve and extract the corresponding critical exponents . for the purpose of this subsection we put @xmath171 and @xmath168 for @xmath16 and @xmath4 , respectively . this means we focus on the situations depicted in fig . [ phase_diag_d2_temp ] ( b ) and fig . [ phase_diag_d3_temp ] ( b ) . an investigation of the critical exponents characterizing the second - order transition in @xmath58 in the present context is contained in ref . , where the results are also compared to calculations involving more sophisticated functional rg truncations @xcite and exact values from the onsager solution to the ising model . for @xmath16 we computed the order parameter , which is given by @xmath174 ( @xmath39 being the global minimum of @xmath31 $ ] ) as a function of @xmath2 , approaching the phase transition line along the isotherms @xmath175 , @xmath176 , @xmath177 and @xmath152 . the results are plotted in fig . [ order_parameters_d3 ] . plotted vs @xmath2 as the transition line is approached along the isotherms @xmath175 , @xmath176 , @xmath177 and @xmath152 . for @xmath175 , @xmath176 , where the transition is first order , a discontinuity of @xmath39 occurs at the phase boundary . for @xmath177 , corresponding to the tricritical isotherm , the order parameter follows the power law @xmath178 , where @xmath179 . along the isotherm , @xmath152 , one observes a power law @xmath180 , where @xmath181 . mean - field behavior is observed because the plotted points are outside the tiny non - gaussinan regime.,width=288 ] for @xmath175 and @xmath176 the order parameter exhibits a discontinuity , as the minimum at @xmath32 becomes the global one when the phase boundary is reached . the order parameter jump vanishes at the tricritical isotherm @xmath182 , along which the order parameter follows the power law @xmath178 , where @xmath179 , as anticipated within mean - field theory.@xcite to recover the expected logarithmic corrections to the leading power law behavior requires going very close to the tricritical point , which is not achieved here . along the isotherm , @xmath152 , in the scale exhibited in fig . [ order_parameters_d3 ] , one observes a power law @xmath180 , where @xmath181 is the mean - field value . in fact , a different behavior , with @xmath183 is expected to occur sufficiently close to the critical line . this is not visible in fig . [ order_parameters_d3 ] because the plotted points are still outside the tiny region where the ginzburg criterion is violated . the crossover to the non - gaussian behavior as well as sizes of the truly critical region as a function of temperature @xmath10 were studied in detail in ref . both in @xmath4 and @xmath16 . we proceed by analysing the order parameter jump upon varying temperature towards @xmath150 along the coexistence curve . the results for @xmath58 are plotted in fig . [ ord_par_jumps ] . as the tricritical point is approached along the coexistence line for ( a ) @xmath16 , @xmath171 , ( b ) @xmath4 , @xmath168 . in the vicinity of the tricritical point the computed order parameter jumps follow a power law @xmath184 . for @xmath16 one finds @xmath185 , being close to the mean - field value . for @xmath4 @xmath186 , deviating from mean field behavior.,title="fig:",width=288 ] as the tricritical point is approached along the coexistence line for ( a ) @xmath16 , @xmath171 , ( b ) @xmath4 , @xmath168 . in the vicinity of the tricritical point the computed order parameter jumps follow a power law @xmath184 . for @xmath16 one finds @xmath185 , being close to the mean - field value . for @xmath4 @xmath186 , deviating from mean field behavior.,title="fig:",width=288 ] in @xmath16 the jump of the order parameter follows a power law @xmath184 with an exponent @xmath185 close to the mean field value @xmath187 . this behavior breaks down at lower temperatures , where the system s behavior is no longer expected to be universal . for @xmath4 a power law with a non mean field exponent @xmath186 is observed . according to our knowledge , the classical tricritical behavior in @xmath4 was not addressed within the functional rg framework . the obtained result compares relatively well to the accurate value @xmath188 , @xcite bearing in mind the simplicity of the applied approximation . a better estimate of @xmath189 would require a more sophisticated truncation of functional rg . this , and also other aspects of classical tricritical behavior in @xmath4 remain to be addressed in a separate study . here we present results regarding the phase boundaries in @xmath16 in the case , where the transition at @xmath13 is second order . we consider two choices of the initial coupling @xmath40 : @xmath190 , which is separated from @xmath191 , and @xmath192 , a value in the proximity of @xmath191 . we perform a careful numerical analysis of the shape of the transition lines over a relatively wide range of temperatures and extract the effective shift exponent @xmath193 as a function of @xmath194 . the quantity @xmath193 is calculated by fitting a power law to three neighboring points of the transition line , which is computed numerically . scaling behavior of the phase boundary shape occurs if @xmath193 remains constant over a wide range of @xmath194 . our findings are summarized in fig . [ psi_eff ] . plotted as a function of @xmath194 for ( a ) @xmath190 , and ( b ) @xmath195 . case ( b ) corresponds to close vicinity of @xmath191 . the transition line exhibits scaling behavior as dictated by hertz - millis theory for the case ( a ) . for case ( b ) a scaling regime with the shift exponent @xmath196 , specific to quantum tricriticality is observed ( see the main text ) , and the hertz - millis scaling persists only in the narrow vicinity of @xmath197 , which is not visible in the picture scale . , title="fig:",width=288 ] plotted as a function of @xmath194 for ( a ) @xmath190 , and ( b ) @xmath195 . case ( b ) corresponds to close vicinity of @xmath191 . the transition line exhibits scaling behavior as dictated by hertz - millis theory for the case ( a ) . for case ( b ) a scaling regime with the shift exponent @xmath196 , specific to quantum tricriticality is observed ( see the main text ) , and the hertz - millis scaling persists only in the narrow vicinity of @xmath197 , which is not visible in the picture scale . , title="fig:",width=288 ] from fig . [ psi_eff ] ( a ) we read off the standard value of the shift exponent @xmath198 , as dictated by hertz - millis theory . however , for case ( b ) scaling behavior with @xmath196 is observed . this is related to the proximity of the chosen coupling @xmath40 to the tricritical value @xmath191 . we also performed analogous computations for other values of @xmath40 between @xmath199 and @xmath200 , yielding the observation , that the hertz - millis scaling region shrinks as @xmath40 approaches @xmath191 , and the other scaling regime forms sufficiently close to @xmath191 . as @xmath201 , the value of @xmath3 corresponding to the latter scaling platoe approaches @xmath202 . in the subsequent section we provide an analytic argument to further clarify the emergent picture and justify the obtained value @xmath18 in this section we show how the obtained values of the shift exponents @xmath203 and @xmath20 can be recovered by invoking a scaling relation for the free energy in the vicinity of the quantum critical point ( see e.g. ref . ) . the analysis largely follows ref . , where @xmath203 was derived , and is restricted to @xmath204 . at @xmath4 additional logarithmic corrections occur , which are not accounted for here . the homogeneity relation for the free energy density @xmath205 } ) \label{sc_rel}\ ] ] is written including the magnetic field @xmath7 and the most relevant interaction coupling @xmath206 . although the scaling dimension @xmath207<0 $ ] , it influences the critical behavior because @xmath206 acts as a so - called dangerously irrelevant variable . @xcite now we differentiate eq . ( [ sc_rel ] ) with respect to @xmath7 and put @xmath208 , thus obtaining a homogeneity relation for the average order parameter @xmath39 at @xmath208 . in the next step we assume , that the dependence of @xmath39 on @xmath2 and @xmath206 occurs only via the ratio @xmath209 . this holds within mean field theory , and should not change the value of the shift exponent provided this exponent is the same for the transition and ginzburg lines in the vicinity of the quantum critical point . this issue was investigated in ref . , where the size of the truly critical region was analyzed yielding the conclusion that the ginzburg line provides a very accurate estimate of @xmath1 in @xmath16 . therefore we write @xmath210 } , tb^z)\;. \label{sc_phi}\ ] ] by choosing the scaling factor @xmath211 such that the first argument of @xmath39 on the rhs of eq . ( [ sc_phi ] ) becomes a constant , we obtain @xmath212)})\;,\ ] ] with @xmath213)$ ] . now we demand that the order parameter @xmath39 vanishes along the transition line @xmath214 , that is @xmath215 for any @xmath10 . this yields @xmath216)}=const$ ] , and therefore the @xmath3 exponent is identified as @xmath217}\;.\ ] ] once again we now take advantage of the mean field assumption identifying the transition and ginzburg lines and replace the exponent @xmath218 with its mean field value @xmath219 . again , this step is justified by invoking the results of ref . and noting that it suffices to consider a region of the phase diagram off true criticality , in which mean field scaling holds . this yields @xmath220}\;. \label{gen_psi}\ ] ] by substituting @xmath221 and @xmath222=4-(d+z)$ ] , we recover the formula @xmath223 , derived by millis . @xcite for @xmath19 we obtain @xmath198 . however , if the most relevant coupling corresponds to @xmath224 , the result is @xmath225 which for @xmath19 yields @xmath226 , as obtained numerically in the previous section . the corresponding values of the shift exponents can also be extracted from eq . ( [ gen_psi ] ) for @xmath227 , which is not covered by our study in sec . iii - v . for this case one obtains @xmath228 and @xmath229 . let us also observe , that eq . ( [ gen_psi ] ) , viewed as a function of @xmath207 $ ] , generates a sequence of conceivable shift exponents . for example for @xmath19 the sequence is @xmath230 for @xmath231 . monotonous decreasing of this sequence is a manifestation of the general fact , that fluctuations tend to suppress the critical temperature . indeed , as @xmath26 grows , fluctuation effects get less relevant , @xmath3 decreases , and the @xmath1 line becomes steeper . also note , that for @xmath232 , @xmath3 equals 1 for all @xmath26 and arbitrary @xmath233 . this follows directly from eq . ( [ gen_psi ] ) . therefore ( up to the neglected logarithmic terms ) , the behavior of the @xmath1 line in two dimensions is always expected to be linear . the analysis performed here and supported by the numerical results of sec . v provides a clear physical picture of the system under study . consider that in addition to the uv cutoff @xmath159 there is another scale @xmath234 present . this scale is set by @xmath235 . once @xmath236 , which always holds for @xmath237 , one recovers hertz - millis scaling of @xmath3 for @xmath238 . if @xmath234 is reduced below @xmath159 , one still observes hertz - millis behavior as long as the two scales do nt become well separated , the only difference being that scaling now occurs for @xmath239 . once @xmath240 , two scaling regimes are present : for @xmath239 the hertz - millis scenario persists , while for @xmath241 scaling behavior with @xmath18 occurs . if @xmath242 , the quantum critical point is replaced with a quantum tricritical point and @xmath20 determines the shape of the whole transition line in the vicinity of @xmath11 ( for @xmath238 ) . in this work we applied the @xmath0 model to analyze the effect of thermal and quantum fluctuations on the phase diagram of a system of itinerant fermions exhibiting a quantum phase transition at wavevector @xmath5 . the analysis is restricted to discrete symmetry breaking and relies on an effective bosonic action analogous to that proposed by hertz . @xcite renormalization of the action is computed from a system of flow equations derived by truncating the exact functional rg flow equation in the one particle irreducible scheme . the flow equations capture quantum and thermal fluctuations on equal footing and are applicable for @xmath59 in @xmath58 , also in the vicinity of the first or second order transition line . by analyzing the renormalized phase diagram at @xmath11 we find that a quantum critical point may be realized even if the bare action corresponds to a first order transition . such scenario occurs for slightly negative quartic coupling @xmath40 . we analyzed the dependence of the tricritical quartic coupling @xmath157 as a function of temperature . the obtained decreasing form of @xmath158 implies enhancing the tendency towards a second order transition upon increasing the temperature . therefore , if the transition is first order at @xmath11 , it becomes second order at higher temperatures , namely for @xmath9 . we computed phase diagrams in the variables @xmath151 , where @xmath2 acts as a non - thermal control parameter . in @xmath4 and @xmath16 we considered cases where the quantum phase transition is first ( @xmath162 ) and second ( @xmath161 ) order . in the former situation , the phase boundaries in @xmath58 show linear behavior in the vicinity of @xmath11 , where the slope in @xmath4 is infinite ( or very large ) . approximately linear behavior of the @xmath1-line is also recovered for @xmath161 in @xmath4 . subsequently , we analyzed the case @xmath161 in @xmath16 . numerical computations of the @xmath1 line for different choices of @xmath40 reveal that the phase boundary follows a power law with the shift exponent @xmath198 , as predicted by millis , @xcite as long as @xmath40 is separated from @xmath191 . upon decreasing @xmath40 towards @xmath191 , the region in the phase diagram , where this scaling is observed shrinks . this is accompanied by formation of a different scaling region ( at slightly higher @xmath10 ) , where @xmath226 . when @xmath243 the quantum critical point is replaced by a quantum tricritical point and the shape of the whole transition line in the vicinity of @xmath11 is described by @xmath20 . these observations are understood by invoking scaling analysis relying on mean - field like assumption , by which the ginzburg and transition lines in the vicinity of the quantum critical point are identified with each other . by deriving a formula for @xmath3 as function of the dynamical exponent @xmath233 and scaling dimension @xmath207 $ ] of the most relevant coupling , we recover @xmath198 for quantum criticality in @xmath19 , and @xmath244 for quantum tricriticality . an interesting avenue for future research would be to reconsider these results in the context of specific microscopic fermionic models , from which coefficients of the landau expansion can possibly be extracted as functions of physical parameters like @xmath10 and the chemical potential @xmath189 for instance . one could then check , in what conditions quantum criticality ( and tricriticality ) induced by order parameter fluctuations is conceivable , i.e. one encounters @xmath245 in the vicinity of the transition . an interesting candidate for such a study is the so called @xmath246-model @xcite of discrete symmetry - breaking fermi surface deformations , for which @xmath14 , and which for small @xmath10 and within mean field approximation , exhibits a phase diagram qualitatively similar to the one obtained here in fig . [ phase_diag_d2_temp ] ( b ) . @xcite the author would like to thank h.w . diehl , a.a . katanin , a. maciolek , w. metzner , m. napirkowski , p. strack , and h. yamase for very useful discussions , and j. bauer and s. takei for providing valuable comments on the manuscript . the support from the german science foundation through the research group for 723 is gratefully acknowledged .
sunada s work on topological crystallography emphasizes the role of the ` maximal abelian cover ' of a graph . this is a covering space of for which the group of deck transformations is the first homology group . an embedding of the maximal abelian cover in a vector space can serve as the pattern for a crystal : atoms are located at the vertices , while bonds lie on the edges . we prove that for any connected graph without bridges , there is a canonical way to embed the maximal abelian cover of into the vector space . we call this a ` topological crystal ' . crystals of graphene and diamond are examples of this construction . we prove that any symmetry of a graph lifts to a symmetry of its topological crystal . we also compute the density of atoms in this topological crystal . we give special attention to the topological crystals coming from platonic solids . the key technical tools are a way of decomposing the 1-chain coming from a path in into manageable pieces , and the work of bacher , de la harpe and nagnibeda on integral cycles and integral cuts .
1607.07748
the ` maximal abelian cover ' of a graph plays a key role in sunada s work on topological crystallography @xcite . just as the universal cover of a connected graph @xmath0 has the fundamental group @xmath3 as its group of deck transformations , the maximal abelian cover , denoted @xmath4 , has the abelianization of @xmath3 as its group of deck transformations . it thus covers every other connected cover of @xmath0 whose group of deck transformations is abelian . since the abelianization of @xmath3 is the first homology group @xmath1 , there is a close connection between the maximal abelian cover and homology theory . in this paper we prove that for a large class of graphs , the maximal abelian cover can naturally be embedded in the vector space @xmath2 . we call this embedded copy of @xmath4 a ` topological crystal ' . the symmetries of the original graph can be lifted to symmetries of its topological crystal , but the topological crystal also has an @xmath5-dimensional lattice of translational symmetries . in the 3-dimensional case , the topological crystal can serve as the blueprint for an actual crystal , with atoms at the vertices and bonds along the edges . the most famous example arises from the tetrahedron . here @xmath0 is the complete graph on 4 vertices : 0.5em since @xmath2 is 3-dimensional , the corresponding topological crystal is a graph embedded in 3-dimensional euclidean space . it looks like this : up to translation this crystal has 4 kinds of atoms , corresponding to the 4 vertices of the tetrahedron . each atom has 3 equally distant neighbors lying in a plane at @xmath6 angles from each other . these planes lie in 4 families , each parallel to one face of a regular tetrahedron . this structure was discovered by the crystallographer laves @xcite and dubbed the ` laves graph ' by coxeter @xcite . later sunada called it the ` @xmath7 lattice ' , studied its energy minimization properties , and raised the question of whether it could serve as the pattern for a crystal form of carbon @xcite . this form of carbon has not yet been seen , but the laves graph plays a role in the structure of certain butterfly wings @xcite . the laves graph is exciting because it was studied mathematically before being found in nature . a more familiar example arises from this graph : the corresponding topological crystal provides the pattern for a diamond : up to translation this crystal has two kinds of atoms , corresponding to the two vertices of the original graph . the general construction of topological crystals was developed by kotani and sunada @xcite , who used the term ` standard realization ' , and later by eon @xcite , who used the term ` archetypical representation ' . sunada uses ` topological crystal ' for an even more general concept @xcite , but we only need the following special case . we start with a graph @xmath0 . this has a space @xmath8 of 0-chains , which are formal linear combinations of vertices , and a space @xmath9 of 1-chains , which are formal linear combinations of edges . there is a boundary operator @xmath10 the linear operator sending any edge to the formal difference of its two endpoints . the kernel of this operator is the space of 1-cycles , @xmath11 . there is an inner product on the space of 1-chains such that edges form an orthonormal basis . this determines an orthogonal projection @xmath12 for a graph , @xmath11 is isomorphic to the first homology group @xmath2 . so , to obtain the topological crystal of @xmath0 , we need only embed its maximal abelian cover @xmath4 in @xmath11 . we do this by embedding @xmath4 in @xmath9 and then projecting it down via @xmath13 . to accomplish this , we need to fix a basepoint for @xmath0 . each path @xmath14 in @xmath0 starting at this basepoint determines a 1-chain @xmath15 . it is easy to show that these 1-chains correspond to the vertices of @xmath4 . furthermore , the graph @xmath4 has an edge from @xmath15 to @xmath16 whenever the path @xmath17 is obtained by adding an extra edge to @xmath14 . we can think of this edge as a straight line segment from @xmath15 to @xmath16 . the hard part is checking that the projection @xmath13 maps this copy of @xmath4 into @xmath11 in a one - to - one manner . in theorem [ embedding_theorem ] we prove that this happens precisely when the graph @xmath0 has no ` bridges ' : that is , edges whose removal would disconnect @xmath0 . kotani and sunada @xcite noted that this condition is necessary ; with considerably more work we show it is also sufficient . the main technical tool is lemma [ decomposition_lemma_2 ] , which for any path @xmath14 decomposes the 1-chain @xmath15 into manageable pieces . we call the resulting copy of @xmath4 embedded in @xmath11 a ` topological crystal ' . for example , let @xmath0 be this graph : since @xmath0 has 3 edges , the space of 1-chains is 3-dimensional . since @xmath2 is 2-dimensional , the space of 1-cycles is a plane in this 3-dimensional space . if we consider paths @xmath14 in @xmath0 starting at the red vertex , form the 1-chains @xmath15 , and project them down to this plane , we obtain the following picture : here the 1-chains @xmath15 are the white and red dots . these are the vertices of @xmath4 , while the line segments between them are the edges of @xmath4 . projecting these vertices and edges onto the plane of 1-cycles , we obtain the topological crystal for @xmath0 . the blue dots come from projecting the white dots onto the plane of 1-cycles , while the red dots already lie on this plane . the resulting topological crystal provides the pattern for graphene , a 2-dimensional form of carbon : 2em for any connected graph @xmath0 , there is a covering map @xmath18 the vertices of @xmath4 come in different kinds , or ` colors ' , depending on which vertex of @xmath0 they map to . we study the group of ` covering symmetries ' , @xmath19 , consisting of all symmetries of @xmath4 that map vertices of the same color to vertices of the same color . in lemma [ covering_lemma_1 ] we prove that this group fits into a short exact sequence @xmath20 where @xmath21 is the group of all symmetries of @xmath0 . thus , every symmetry of @xmath0 is covered by some symmetry of @xmath4 , while @xmath1 acts on @xmath4 in a way that preserves the color of every vertex . in theorem [ symmetry_theorem_2 ] we prove that @xmath19 also acts as symmetries of the topological crystal associated to @xmath0 . more precisely , there is an action of @xmath19 as affine isometries of @xmath2 for which the embedding of @xmath4 in this space is equivariant . in this action , the subgroup @xmath1 acts as translations . these results are most interesting in cases where @xmath0 is highly symmetrical . for example , suppose that @xmath21 acts transitively on ` arcs ' , meaning pairs consisting of a vertex and an edge incident to that vertex . in this case @xmath19 acts transitively on the arcs of the corresponding topological crystal . one example of this phenomena is the laves graph , coming from the tetrahedron . the symmetry group of the tetrahedron is the coxeter group @xmath22 thus , the group of covering symmetries of the laves graph is an extension of @xmath23 by @xmath24 . indeed , the vertices and edges of any platonic solid form a graph whose symmetry group acts transitively on arcs . for example , the symmetry group of the cube and octahedron is the coxeter group @xmath25 since the cube has 6 faces , @xmath2 is 5-dimensional for the cube graph . the corresponding topological crystal is thus 5-dimensional , and its group of covering symmetries , an extension of @xmath26 by @xmath27 , acts transitively on arcs . similarly , the octahedron gives a 7-dimensional topological crystal whose group of covering symmetries , an extension of @xmath28 by @xmath29 , acts transitively on arcs . the cuboctahedron can be seen either as a truncated cube or a truncated octahedron , so it too has @xmath26 as its symmetry group , and in fact this group acts transitively on arcs . since the cuboctahedron has 14 faces , we obtain a 13-dimensional crystal whose covering symmetries act transitively on arcs . the symmetry group of the dodecahedron and icosahedron is @xmath30 and these solids give crystals of dimensions 11 and 19 . the icosidodecahedron can be seen either as a truncated dodecahedron or a truncated icosahedron ; it has @xmath31 as its symmetry group , and this group acts transitively on arcs . it has 32 faces , so it gives a 31-dimensional crystal whose covering symmetries act transitively on arcs . there is also an infinite family of degenerate platonic solids called ` hosohedra ' with two vertices , @xmath5 edges and @xmath5 faces . these faces can not be made flat , since each face has just 2 edges , but that is not relevant to our construction : the vertices and edges still give a graph . for example , when @xmath32 , we have the ` hexagonal hosohedron ' : the symmetry group of the @xmath5-gonal hosohedron is @xmath33 , where the coxeter group @xmath34 is a dihedral group , and @xmath35 acts to interchange the two vertices while fixing the edges . the automorphism group of its underlying graph is @xmath36 . the corresponding crystal has dimension @xmath37 , and its group of covering symmetries is an extension of @xmath36 by @xmath38 . the case @xmath39 gives the graphene crystal , while @xmath40 gives the diamond . there are many other examples of graphs whose symmetry group acts transitively on arcs . one is the petersen graph : the symmetry group of the petersen graph is @xmath41 . this graph has 10 vertices and 15 edges , so its euler characteristic is @xmath42 , which implies that its space of 1-cycles is 6-dimensional . it thus gives a 6-dimensional crystal whose group of covering symmetries , an extension of @xmath41 by @xmath43 , acts transitively on arcs . two more nice examples come from klein s quartic curve , a riemann surface of genus three on which the 336-element group @xmath44 acts as isometries . these isometries preserve a tiling of klein s quartic curve by 56 triangles , with 7 meeting at each vertex . this picture is topologically correct , though not geometrically : from this tiling we obtain a graph @xmath0 embedded in klein s quartic curve . this graph has @xmath45 edges and @xmath46 vertices , so it has euler characteristic @xmath47 . it thus gives a 61-dimensional topological crystal whose group of covering symmetries , an extension of @xmath44 by @xmath48 , acts transitively on arcs . there is also a dual tiling of klein s curve by 24 heptagons , 3 meeting at each vertex . this gives a graph with 84 edges and 56 vertices , hence euler characteristic @xmath49 . from this we obtain a 29-dimensional topological crystal whose group of covering symmetries is an extension of @xmath44 by @xmath50 . again , this group acts transitively on arcs . another interesting property of a topological crystal is its ` packing fraction ' . the set @xmath51 of atoms is contained in the lattice @xmath52 obtained by projecting the integral 1-chains down to the space of 1-cycles : @xmath53 we can ask what fraction of the points in this lattice are actually atoms . we call this the ` packing fraction ' , and since @xmath54 acts as translations on both @xmath51 and @xmath52 , we define it to be @xmath55 for example , suppose @xmath0 is this graph : 1em then the packing fraction is @xmath56 , as can be seen here : 1em 1em for any bridgeless connected graph @xmath0 , we prove in theorem [ packing_fraction_theorem ] that @xmath57 where @xmath58 is the set of vertices and @xmath59 is the set of spanning trees . the main tool here is bacher , de la harpe and nagnibeda s work on integral cycles and integral cuts @xcite . here is some data about the examples discussed above . we used mathematica to count the spanning trees . in this table , @xmath60 denotes an extension of the group @xmath61 by @xmath62 . we start by reviewing some concepts from sunada s _ topological crystallography _ it is convenient to work with graphs having two copies of each edge , one pointing in each direction . thus , he defines a * * @xmath63 to consist of a set @xmath58 of * * , a set @xmath64 of * * , maps @xmath65 assigning to each edge its * * and * * , and a map @xmath66 sending each edge to its * * , obeying @xmath67 and @xmath68 for all @xmath69 . if @xmath70 and @xmath71 we write @xmath72 , and draw @xmath73 as an interval with an arrow on it pointing from @xmath74 to @xmath75 . we write @xmath76 as @xmath77 , and draw @xmath77 as the same interval as @xmath73 , but with its arrow reversed . the equations obeyed by @xmath78 say that taking the inverse of @xmath72 gives an edge @xmath79 and that @xmath80 . no edge can be its own inverse . a * * , say @xmath81 , is a pair of functions , one sending vertices to vertices and one sending edges to edges , that preserve the source , target and inverse maps . by abuse of notation we call both of these functions @xmath82 . from a graph @xmath0 we can build a topological space @xmath83 called its * * . we do this by taking one point for each vertex and gluing on one copy of @xmath84 $ ] for each edge @xmath72 , gluing the point @xmath85 to @xmath74 and the point @xmath86 to @xmath75 , and then identifying the interval for each edge @xmath73 with the interval for its inverse by means of the map @xmath87 . any map of graphs gives rise to a continuous map between their geometric realizations , and we say a map of graphs is a * * if this continuous map is a covering map . for simplicity we denote the fundamental group of @xmath83 by @xmath3 , and similarly for other topological invariants of @xmath83 . however , there are contexts in which we need to distinguish between a graph @xmath0 and its geometric realization @xmath83 . any connected graph @xmath0 has a * * , meaning a cover @xmath88 such that every cover factors through this one . the geometric realization of @xmath89 is connected and simply connected . the fundamental group @xmath3 acts as * * of @xmath89 , meaning invertible maps @xmath90 such that @xmath91 . we can take the quotient of @xmath89 by the action of any subgroup @xmath92 and get a cover @xmath93 . in particular , if we take @xmath61 to be the commutator subgroup of @xmath3 , we call the graph @xmath94 the * * of the graph @xmath0 , and denote it by @xmath4 . we obtain a cover @xmath95 whose group of deck transformations is the abelianization of @xmath3 . this is just the first homology group @xmath1 , which is a free abelian group . we need a concrete description of the maximal abelian cover . we start with the universal cover , but first we need some preliminaries on paths in graphs . given vertices @xmath96 in @xmath0 , define a * * from @xmath97 to @xmath98 to be a word of edges @xmath99 with @xmath100 for some vertices @xmath101 with @xmath102 and @xmath103 . we allow the word to be empty if and only if @xmath104 ; this gives the * * from @xmath97 to itself . given a path @xmath14 from @xmath97 to @xmath98 we write @xmath105 , and we write the trivial path from @xmath97 to itself as @xmath106 . we define the composite of paths @xmath107 and @xmath108 via concatenation of words , obtaining a path we call @xmath109 . we call a path from a vertex @xmath97 to itself a * * based at @xmath97 . we say two paths from @xmath97 to @xmath98 are * * if one can be obtained from the other by repeatedly introducing or deleting subwords of the form @xmath110 where @xmath111 . if @xmath112 $ ] is a homotopy class of paths from @xmath97 to @xmath98 , we write @xmath113 { \colon}x \to y$ ] . we can compose homotopy classes @xmath113 { \colon}x \to y$ ] and @xmath114 { \colon}y \to z$ ] by setting @xmath113 [ \delta ] = [ \gamma \delta]$ ] . if @xmath0 is a connected graph , we can describe the universal cover @xmath89 as follows . fix a vertex @xmath115 of @xmath0 , which we call the * * . the vertices of @xmath89 are defined to be the homotopy classes of paths @xmath113 { \colon}x_0 \to x$ ] where @xmath97 is arbitrary . the edges in @xmath89 from the vertex @xmath112 { \colon}x_0 \to x$ ] to the vertex @xmath114 { \colon}x_0 \to y$ ] are defined to be the edges @xmath69 with @xmath116 = [ \delta]$ ] . in fact , there is always at most one such edge . there is an obvious map of graphs @xmath88 sending each vertex @xmath113 { \colon}x_0 \to x$ ] of @xmath89 to the vertex @xmath97 of @xmath0 . this map is a cover . now we are ready to construct the maximal abelian cover @xmath4 . for this , we impose a further equivalence relation on paths , which is designed to make composition commutative whenever possible . however , we need to be careful . if @xmath107 and @xmath117 , the composites @xmath118 and @xmath119 are both well - defined if and only if @xmath120 and @xmath121 . in this case , @xmath122 and @xmath119 share the same starting point and share the same ending point if and only if @xmath123 and @xmath124 . if all four of these equations hold , both @xmath14 and @xmath125 are loops based at @xmath97 . so , we shall impose the relation @xmath126 only in this case . we say two paths are * * if one can be obtained from another by : * repeatedly introducing or deleting subwords @xmath110 where @xmath111 , and/or * repeatedly replacing subwords of the form @xmath127 by those of the form @xmath128 , where @xmath129 and @xmath130 are loops based at the same vertex . our use of the term ` homologous ' is a bit nonstandard , but we shall see some nice relations to homology theory , which we hope justifies this terminology . we denote the homology class of a path @xmath14 by @xmath131 . note that if two paths @xmath105 , @xmath132 are homologous then @xmath123 and @xmath124 . thus , the starting and ending points of a homology class of paths are well - defined , and given any path @xmath107 we write @xmath133 . the composite of homology classes is also well - defined if we set @xmath134 . we construct the maximal abelian cover of a connected graph @xmath0 just as we constructed its universal cover , but using homology classes rather than homotopy classes of paths . fix a basepoint @xmath115 for @xmath0 . the vertices of @xmath4 , or * * , are defined to be the homology classes of paths @xmath135 where @xmath97 is arbitrary . any edge of @xmath4 , or * * , goes from some atom @xmath136 to the some atom @xmath137 . the bonds from @xmath138 to @xmath139 are defined to be the edges @xmath69 with @xmath140 . there is at most one bond between any two atoms . again we have a covering map @xmath18 the homotopy classes of loops based at @xmath115 form a group , with composition as the group operation . this is the * * @xmath3 of the graph @xmath0 . this is isomorphic as the fundamental group of the space associated to @xmath0 . by our construction of the universal cover , @xmath3 is also the set of vertices of @xmath89 that are mapped to @xmath115 by @xmath141 . furthermore , any element @xmath113 \in \pi_1(x ) $ ] defines a deck transformation of @xmath89 that sends each vertex @xmath142 { \colon}x_0 \to x $ ] to the vertex @xmath113 [ \delta ] { \colon}x_0 \to x$ ] . similarly , the homology classes of loops based at @xmath115 form a group with composition as the group operation . since the additional relation used to define homology classes is precisely that needed to make composition of homology classes of loops commutative , this group is the abelianization of @xmath3 . it is therefore isomorphic to the first homology group @xmath143 of the geometric realization of @xmath0 . by our construction of the maximal abelian cover , @xmath1 is also the set of vertices of @xmath4 that are mapped to @xmath115 by @xmath144 . furthermore , any element @xmath145 defines a deck transformation of @xmath4 that sends each vertex @xmath146 to the vertex @xmath147 . given a connected graph @xmath0 , we call the vertices of its maximal abelian cover ` atoms ' because they play the role of atoms in a topological crystal . now we describe a systematic procedure for mapping these atoms into a vector space with inner product , namely the space of 1-cycles @xmath11 . we show that this map is an embedding if and only if the graph has no bridges . we begin with some standard material . let @xmath0 be a graph . the group of * * on @xmath0 , @xmath148 , is the free abelian group on the set of vertices of @xmath0 . the group of * * on @xmath0 , @xmath149 , is the quotient of the free abelian group on the set of edges of @xmath0 by relations @xmath150 for every edge @xmath73 . the * * is the homomorphism @xmath151 such that @xmath152 for each edge @xmath73 , and @xmath153 is the group of * * on @xmath0 . any path @xmath154 in @xmath0 determines an integral 1-chain @xmath155 for any path @xmath14 we have @xmath156 and if @xmath14 and @xmath125 are composable then @xmath157 we define vector spaces of * * and * * by @xmath158 respectively , and extend the boundary map to a linear map @xmath159 we let @xmath11 be the kernel of this linear map , or equivalently , @xmath160 and we call elements of this vector space * * . the space of 1-cycles is isomorphic to the first cohomology with real coefficients , @xmath2 . since @xmath54 is a free abelian group , it forms a lattice in the space of 1-cycles . any edge of @xmath0 can be seen as a 1-chain , and there is a unique inner product on @xmath9 such that edges form an orthonormal basis ( with each edge @xmath77 counting as the negative of @xmath73 . ) there is thus an orthogonal projection @xmath161 we now come to the main construction , first introduced by kotani and sunada @xcite : [ atom_definition ] let @xmath0 be a connected graph with @xmath58 as its set of vertices and a chosen basepoint @xmath162 . let its set of atoms be @xmath163 define the map @xmath164 by @xmath165 that @xmath78 is well - defined follows from lemma [ chain_lemma ] below : homologous paths give the same 1-chain . for a bridgeless graph , @xmath78 embeds the set of atoms in the space of 1-cycles : [ atom_thm ] let @xmath0 be a connected graph . then the following are equivalent : 1 . @xmath0 has no bridges . 2 . the map @xmath166 is one - to - one . the map @xmath78 is one - to - one if and only if for any atoms @xmath167 and @xmath168 , @xmath169 implies @xmath170 . note that @xmath171 is a path in @xmath0 with @xmath172 , so @xmath173 since @xmath174 vanishes if and only if @xmath15 is orthogonal to every 1-cycle , we have @xmath175 on the other hand , lemma [ chain_lemma ] implies that @xmath176 thus , to prove ( 1)@xmath177(2 ) , it suffices to that show that @xmath0 has no bridges if and only if every 1-chain @xmath178 orthogonal to every 1-cycle has @xmath179 . we show this in lemma [ bridge_lemma ] . [ chain_lemma ] let @xmath0 be a connected graph . two paths @xmath180 in @xmath0 are homologous if and only if @xmath181 . the ` only if ' direction is clear , since @xmath182 is homologous to @xmath183 precisely when it can be obtained from @xmath182 by : * repeatedly introducing or deleting a subword @xmath110 where @xmath111 , and/or * repeatedly replacing a subword of the form @xmath127 by a subword of the form @xmath128 , where @xmath129 and @xmath130 are loops based at the same vertex . and these moves do not change the corresponding 1-chain . conversely , suppose @xmath181 . then @xmath184 is a loop based at @xmath97 with @xmath185 . in section [ sec : maximal_abelian_covers ] we saw that the group of homology classes of loops based at any vertex is the abelianization of @xmath3 , namely @xmath1 . for a graph we have @xmath186 , and the cycle associated to any loop @xmath14 is @xmath15 . thus , @xmath187 implies that @xmath14 is homologous to the trivial loop . this implies that @xmath188 is homologous to @xmath182 . it is also easy to see that @xmath189 is homologous to @xmath183 . thus @xmath183 is homologous to @xmath182 . the following lemmas are the technical heart of theorem [ atom_thm ] . we need to show that any nonzero 1-chain coming from a path in a bridgeless graph has nonzero inner product with some 1-cycle . the following lemmas , inspired by an idea of ilya bogdanov @xcite , yield an algorithm for actually constructing such a 1-cycle . this 1-cycle also has other desirable properties , which will come in handy later . to state these , let a * * be one in which each vertex appears at most once . let a * * be a loop @xmath190 in which each vertex except @xmath97 appears at most once , while @xmath97 appears exactly twice , as the starting point and ending point . let the * * of a 1-chain @xmath191 , denoted @xmath192 , be the set of edges @xmath73 such that @xmath193 . this excludes edges with @xmath194 , but also those with @xmath195 , which are inverses of edges in the support . note that @xmath196 thus , @xmath192 is the smallest set of edges such that @xmath191 can be written as a positive linear combination of edges in this set . given 1-chains @xmath191 and @xmath197 , we have @xmath198 if and only if @xmath199 implies @xmath200 and @xmath201 implies @xmath202 . [ decomposition_lemma_1 ] let @xmath0 be any graph and let @xmath191 be an integral 1-cycle on @xmath0 . then for some @xmath5 we can write @xmath203 where @xmath204 are simple loops with @xmath205 . if @xmath206 we have @xmath207 for any trivial loop @xmath208 otherwise @xmath192 is nonempty , and we can construct a simple loop @xmath209 in @xmath0 as follows . we start by picking any edge @xmath210 in @xmath192 and moving to the vertex @xmath211 . in general , when we arrive at the vertex @xmath212 , if it is not one we have visited before we pick an edge @xmath213 in @xmath192 . to see that such an edge must exist , note that since @xmath191 is a cycle , the sum of the coefficients @xmath214 over edges @xmath215 with @xmath212 as source must equal the sum of these coefficients over edges in @xmath192 with @xmath212 as target . since we are assuming @xmath212 is the target of an edge in @xmath192 , namely @xmath216 , it must also be the source of an edge in @xmath192 . since the support of @xmath191 is finite , eventually we must reach a vertex we have visited before . so , eventually @xmath217 for some @xmath218 . then @xmath219 is a simple loop . by construction we have @xmath220 as desired . now let @xmath221 . we claim that also @xmath222 . to see this , note that both @xmath214 and @xmath223 vanish except for edges in @xmath192 and their inverses . if @xmath215 then @xmath224 , while @xmath225 . thus @xmath226 implies @xmath227 , while @xmath228 implies @xmath229 . it follows that @xmath222 . since @xmath230 is again an integral 1-cycle , either @xmath231 or we can repeat the argument just given with @xmath230 replacing @xmath191 . continuing on , we can construct 1-cycles @xmath232 each with support contained in the support of the previous one , and simple loops @xmath233 with @xmath234 and @xmath235 . with each new 1-cycle the coefficient of at least one edge is less than before , so eventually @xmath236 reaches zero and the procedure terminates . we thus obtain simple loops @xmath237 with @xmath238 and @xmath205 . [ decomposition_lemma_2 ] let @xmath239 be a path in a graph @xmath0 . then for some @xmath240 we can write @xmath241 where @xmath242 is a simple path and @xmath204 are simple loops with @xmath243 . suppose that @xmath105 is a path in @xmath0 . if @xmath104 then @xmath15 is a cycle and the lemma follows from lemma [ decomposition_lemma_1 ] , taking @xmath125 to be any trivial path . thus we assume @xmath244 . we claim there is a simple path @xmath245 with @xmath246 . we can construct such a path , say @xmath247 , as follows . start by setting @xmath248 and let @xmath249 in this case . as we proceed we will repeatedly increment @xmath250 and subtract a 1-cycle from @xmath236 to define a new 1-chain @xmath251 with @xmath252 . we start at the vertex @xmath102 . in general , if @xmath212 is neither @xmath98 nor a vertex we have visited before , we pick an edge @xmath213 in @xmath253 and move to the vertex @xmath254 . such an edge must exist , because @xmath255 , so the sum of the coefficients @xmath256 over edges @xmath257 with @xmath212 as source is greater than or equal to the sum of these coefficients over edges in @xmath253 with @xmath212 as target . since the support of @xmath236 is finite , eventually @xmath212 is either @xmath98 or a vertex we have visited before . if @xmath258 we stop : by construction @xmath259 is a simple path from @xmath97 to @xmath98 . moreover , @xmath260 as desired . if @xmath212 is a vertex we have visited before , say @xmath217 for @xmath218 , then @xmath261 is a simple loop . by construction we have @xmath235 . the argument given in lemma [ decomposition_lemma_1 ] shows that @xmath234 has @xmath252 . since @xmath262 , the new 1-chain @xmath251 is still nonzero . we may thus increment @xmath250 by 1 and restart the process of building a simple path from @xmath97 to @xmath98 . this algorithm must eventually succeed in building a simple path @xmath245 , since the sum of the coefficients of @xmath251 is strictly less than that for @xmath236 , yet @xmath251 can never vanish , given that @xmath263 . if we let @xmath264 then @xmath191 is a 1-cycle , and our construction ensures that @xmath265 . thus , we can use lemma [ decomposition_lemma_1 ] to write @xmath266 where @xmath204 are simple loops with @xmath267 . this implies that @xmath241 where @xmath125 and @xmath204 have the desired properties . [ bridge_lemma ] let @xmath0 be a graph . then the following are equivalent : 1 . @xmath0 has no bridges . 2 . for any path @xmath14 in @xmath0 , if @xmath15 is orthogonal to every 1-cycle then @xmath268 . to prove ( 2)@xmath269(1 ) , we assume @xmath0 has a bridge @xmath73 and show @xmath270 for every 1-cycle @xmath191 . this will give a path @xmath271 for which @xmath272 but @xmath15 is orthogonal to every 1-cycle . since every 1-cycle is a linear combination of 1-cycles @xmath273 coming from loops @xmath274 , it suffices to show @xmath275 for every loop @xmath274 . since @xmath73 is a bridge , removing @xmath73 breaks @xmath0 into two connected components . any loop @xmath276 must start and end in the same component , so the number of edges @xmath216 that equal @xmath73 must equal the number of @xmath216 that equal @xmath77 . thus @xmath277 . to prove ( 1)@xmath269(2 ) , suppose @xmath0 has no bridges . it suffices to show that if @xmath105 is a path for which @xmath15 is orthogonal to every 1-cycle , then @xmath268 . we use lemma [ decomposition_lemma_2 ] to write @xmath241 where @xmath242 is a simple path and @xmath204 are simple loops with if any of the 1-chains @xmath278 are nonzero we are done , since then @xmath279 so @xmath15 is not orthogonal to every 1-cycle . thus , we assume @xmath280 . we assume that @xmath281 , and need to construct a 1-cycle that is not orthogonal to @xmath282 . we also assume @xmath244 , since otherwise we can use @xmath282 itself as the desired 1-cycle . write @xmath283 with @xmath100 . these edges are distinct since @xmath125 is a simple path , and there is at least one of them since @xmath284 is not equal to @xmath285 . since the last edge @xmath286 is not a bridge , there must be a path in @xmath0 from @xmath287 to @xmath288 that does not include @xmath286 . if we follow that path only as far as the first vertex @xmath289 that it reaches , we obtain a path @xmath290 that includes no edges @xmath216 , nor their inverses . thus , we have @xmath291 if @xmath292 is the portion of @xmath125 that goes from @xmath212 to @xmath287 , then @xmath293 the path @xmath294 is a loop , so @xmath295 is a 1-cycle , and this 1-cycle is not orthogonal to @xmath282 , since @xmath296 in the previous section we took a connected bridgeless graph @xmath0 and embedded its atoms into the space of 1-cycles via a map @xmath297 these atoms are the vertices of the maximal abelian cover @xmath4 , or equivalently , the vertices of its geometric realization @xmath298 . now we extend @xmath78 to an embedding @xmath299 we call the image of this embedding the * * associated to @xmath0 . the idea is that just as @xmath78 maps each atom to a point in the vector space @xmath11 , @xmath300 maps each edge of @xmath298 to a straight line segment between such points . these line segments serve as the ` bonds ' of a topological crystal . the only challenge is to show that these bonds do not cross each other . let @xmath0 be a connected bridgeless graph , and fix a vertex @xmath115 . technically , an atom is a homology class of paths @xmath167 starting at @xmath115 and ending at any vertex @xmath97 . given an edge @xmath301 in @xmath0 , there is a unique edge in @xmath4 from the atom @xmath302 to the atom @xmath303 . every edge of @xmath4 arises in this manner in a unique way . so , an edge in @xmath4 amounts to a pair @xmath304 as explained in section [ sec : maximal_abelian_covers ] , each edge in @xmath4 gives a closed interval in the geometric realization @xmath298 , with a preferred parametrization @xmath305 \to |\overline{x}|$ ] . the inverse of an edge gives the same closed interval with the reverse parametrization , @xmath306 . we call these closed intervals * * to distinguish from the edges of an abstract graph . given a finite - dimensional vector space @xmath58 , we say a map @xmath307 is * * if for any given bond the composite map @xmath308 \to v$ ] is affine , meaning @xmath309 for some @xmath310 . it follows that @xmath82 maps each bond to a straight line segment in @xmath58 . it also follows that @xmath82 is continuous . note also that that given @xmath82 defined on the atoms , there exists a unique map extending @xmath82 that is affine on bonds . the reason is that an affine map from @xmath84 $ ] to @xmath58 is determined by its values at the endpoints . [ embedding_theorem ] if @xmath0 is a connected graph , the map @xmath311 extends uniquely to a map @xmath312 that is affine on bonds . if @xmath0 is also bridgeless , then @xmath300 is one - to - one . as mentioned , the existence and uniqueness of an extension @xmath300 that is affine on bonds is automatic . the task is to show that @xmath300 is an embedding when @xmath0 has no bridges . each bond in @xmath4 arises from a pair @xmath313 and the points on this bond are parametrized by numbers @xmath314 $ ] . thus , we can describe any point @xmath315 as a triple @xmath316.\ ] ] however , there is some redundancy in this description . if @xmath317 then @xmath141 is an atom , and in this case @xmath73 becomes irrelevant : any choice of @xmath301 gives the same point @xmath141 . in general , any point described by the triple @xmath316\ ] ] is also described by the triple @xmath318.\ ] ] if @xmath319 , the point @xmath141 is not an atom , and the above triples are the only two that describe it . suppose we have a point @xmath315 corresponding to some triple @xmath320 , @xmath301 , @xmath314 $ ] . we define a 1-chain @xmath321 by @xmath322 thanks to lemma [ chain_lemma ] , @xmath323 is independent of the choice of path @xmath183 representing the atom @xmath324 . furthermore , different triples describing the same point @xmath141 give the same result for @xmath323 . thus , @xmath323 is well - defined . define @xmath312 by @xmath325 this equals @xmath326 when @xmath141 is an atom , since @xmath327 when @xmath141 is the atom @xmath328 . the map @xmath300 is affine on each bond since @xmath323 depends affinely on the parameter @xmath329 and @xmath13 is linear . so , @xmath300 is the unique map equalling @xmath78 on atoms that is affine on each bond . to prove the theorem , it suffices to show that if @xmath141 and @xmath144 are points in @xmath4 such that @xmath330 is orthogonal to every 1-cycle , then @xmath331 . to do this , first we show that @xmath332 implies @xmath331 . then we show that if @xmath330 is orthogonal to every 1-cycle then @xmath332 . suppose @xmath332 . these 1-chains are integral if and only if @xmath141 and @xmath144 are atoms , in which case @xmath331 by lemma [ chain_lemma ] . so , we may assume that neither 1-chain is integral and neither @xmath141 nor @xmath144 is an atom . thus , we may write @xmath141 as a triple @xmath333 and @xmath144 as a triple @xmath334 implying @xmath335 working modulo integral 1-chains we have @xmath336 so either @xmath337 and @xmath338 or @xmath339 and @xmath340 . in the first case we have @xmath341 so @xmath342 by lemma [ chain_lemma ] , and thus @xmath331 . in the second case we have @xmath343 so @xmath344 by lemma [ chain_lemma ] , and again @xmath331 . next suppose that @xmath330 is orthogonal to every 1-cycle . write @xmath345 and note that @xmath346 we need to prove that if @xmath191 is orthogonal to every 1-cycle then @xmath331 . form a path @xmath14 by either composing @xmath347 with the edges @xmath348 and @xmath73 , or not , in such a way that the coefficient of every edge in @xmath349 agrees in sign with its coefficient in @xmath191 . more precisely , let @xmath350 thus @xmath351 . next , using lemma [ decomposition_lemma_2 ] , write @xmath241 where @xmath242 is a simple path and @xmath204 are simple loops with @xmath243 . if any 1-cycle @xmath278 is nonzero , then since @xmath352 we have @xmath353 contradicting our assumption that @xmath191 is orthogonal to every 1-cycle . we thus assume there are no nonzero 1-cycles @xmath278 . this implies @xmath280 . if @xmath125 is a path with no edges , then @xmath268 . since @xmath351 , this implies @xmath206 as desired . we are left with the case where @xmath125 has at least one edge . however , we shall see that this leads to a contradiction . write @xmath283 with @xmath100 . there is at least one of these edges , and they are distinct since @xmath125 is a simple path . since the last edge @xmath286 is not a bridge , there must be a path in @xmath0 from @xmath287 to @xmath288 that does not include @xmath286 . if we follow that path only as far as the first vertex @xmath289 that it reaches , we obtain a path @xmath290 that includes no edges @xmath216 , nor their inverses . since @xmath354 , this implies @xmath355 if @xmath292 is the portion of @xmath125 that goes from @xmath212 to @xmath287 , then @xmath356 , and since @xmath354 , we have @xmath357 since @xmath294 is a loop , @xmath295 is a 1-cycle , and @xmath358 this contradicts our assumption that @xmath191 is orthogonal to every 1-cycle . every bridgeless graph gives a topological crystal . how are the graph s symmetries related to those of its crystal ? to tackle this , we start by asking how the symmetries of a graph @xmath0 are related to those of its maximal abelian cover @xmath359 we shall not study all the symmetries of @xmath4 , only those that cover symmetries of @xmath0 . these form a group @xmath19 . as we shall see , this group contains @xmath360 , which acts as deck transformations of @xmath4 . the group @xmath19 also maps onto the group of symmetries of @xmath0 : in other words , every symmetry of @xmath0 can be covered by some symmetry of @xmath4 . in fact , we shall prove that @xmath19 is an extension of the symmetry group of @xmath0 by @xmath1 . we also prove that @xmath19 acts on the vector space @xmath11 as affine isometries in a way that preserves the topological crystal embedded in this space . define an * * of a graph @xmath0 to be a map of graphs @xmath361 that has an inverse . the automorphisms of @xmath0 form a group @xmath21 . we say an automorphism @xmath362 * * an automorphism @xmath361 if @xmath363 there is a group @xmath19 where an element is an automorphism of @xmath4 that covers some automorphism of @xmath0 . we call elements of @xmath19 * * . since @xmath144 is onto , any covering symmetry covers at most one automorphism of @xmath0 . there is thus a map @xmath364 sending any @xmath365 to the automorphism @xmath366 that it covers . it is easy to check that @xmath367 is a group homomorphism . [ covering_lemma_1 ] let @xmath0 be a connected graph with basepoint . the homomorphism @xmath367 is onto and its kernel is @xmath1 , so we have a short exact sequence @xmath368 in other words , @xmath19 is an extension of @xmath21 by @xmath1 . an element @xmath365 is in the kernel of @xmath367 if and only if it covers the identity map , which means that it is a deck transformation of the maximal abelian cover of @xmath0 . in section [ sec : maximal_abelian_covers ] we saw that the group of these deck transformations is @xmath1 . to show that @xmath367 is onto , we choose any automorphism @xmath361 and find an automorphism @xmath369 that covers @xmath82 . we need to describe how @xmath370 acts on vertices and edges . a vertex of @xmath4 is an atom , that is , a homology class of paths in @xmath0 starting from the basepoint , say @xmath320 . the automorphism @xmath82 sends any path @xmath371 to a path we call @xmath372 . unfortunately this path no longer defines an atom unless @xmath373 . to deal with this , arbitrarily choose a path @xmath374 and let @xmath375 an edge @xmath4 is a pair @xmath304 we let @xmath370 map this edge to the edge @xmath376 one can check that @xmath370 is a map of graphs . since it is one - to - one and onto on both vertices and edges , it is an automorphism . one can also check that @xmath370 covers @xmath82 , since the definition of the projection @xmath377 implies that @xmath378 and @xmath379 the proof of the above lemma gives an explicit description of the group @xmath19 and its action on @xmath4 . in the proof , two choices of @xmath374 define the same automorphism of @xmath4 if and only if they are homologous . a homology class of paths @xmath380 is just an atom mapping to @xmath381 via the projection @xmath377 . thus , we have : [ covering_lemma_2 ] elements of @xmath19 are in one - to - one correspondence with pairs @xmath382 where @xmath366 and @xmath383 . such a pair acts on @xmath4 by mapping each atom @xmath384 to the atom @xmath385 and each edge @xmath386 to the edge @xmath387 there may in principle be other automorphisms of @xmath4 not lying in the group @xmath19 ; we have not found examples . automorphisms in @xmath19 have the advantage that they also act naturally as affine transformations of the vector space @xmath11 . furthermore , this space has an inner product coming from the inner product on 1-chains , which in turn defines a metric . the group @xmath19 acts as * * of @xmath11 , that is , affine transformations that preserve this metric : [ symmetry_theorem_1 ] there exists a unique action @xmath388 of @xmath19 as affine transformations of @xmath11 for which the embedding @xmath311 is equivariant , meaning that @xmath389 for all @xmath365 and @xmath390 moreover , the transformations @xmath391 are affine isometries . since @xmath392 , every point in @xmath11 is an affine combination of points @xmath393 where @xmath167 is an atom , so an affine transformation of @xmath11 is uniquely determined by its action on such points . this proves the uniqueness of @xmath388 . for existence , take any @xmath365 and use lemma [ covering_lemma_2 ] to write it as a pair @xmath382 where @xmath366 and @xmath394 . for any @xmath395 , define @xmath396 where @xmath397 is the linear transformation with @xmath398 for any edge @xmath73 of @xmath0 . since @xmath399 preserves the inner product on 1-chains and the subspace of 1-cycles we have @xmath400 where @xmath13 is the projection from 1-chains to 1-cycles . recall from definition [ atom_definition ] that @xmath401 . thus we have @xmath402 where in the fourth step we used the fact that @xmath403 for any path @xmath183 . to check that @xmath388 is an action we need to show @xmath404 for any @xmath405 . this can be done by brute force , but it suffices to check it on points of the form @xmath393 , since every point of @xmath11 is an affine combination of these . for this , note that : @xmath406 finally , is clear that @xmath391 is an affine isometry , since @xmath396 implies that @xmath391 is a linear isometry followed by a translation . we conclude by showing that @xmath19 acts on the space of 1-cycles in a way that preserves the topological crystal . in general , any automorphism @xmath82 of any graph functorially determines an automorphism of its geometric realization , which we call @xmath407 . thus , @xmath19 acts as automorphisms of @xmath298 . in theorem [ embedding_theorem ] we saw that the map @xmath311 extends uniquely to an embedding @xmath408 that is affine on bonds . to show that @xmath19 preserves the topological crystal , that is , the image of this embedding , we prove that @xmath300 is equivariant : [ symmetry_theorem_2 ] the embedding @xmath409 is equivariant with respect to the action of @xmath19 , meaning that @xmath410 for all @xmath365 and @xmath411 as noted in the proof of theorem [ embedding_theorem ] , every point @xmath412 corresponds to a triple @xmath413,\ ] ] and then we have @xmath414 if we let @xmath415 we thus have @xmath416 or in other words @xmath417 by lemma [ covering_lemma_2 ] any element @xmath365 corresponds to some pair @xmath418 . the point @xmath419 then corresponds to the triple @xmath420.\ ] ] since @xmath391 is affine , it follows that @xmath421 in section [ sec : atoms ] we took a connected bridgeless graph @xmath0 with a basepoint @xmath115 and embedded its set of atoms , @xmath51 , into its space of 1-cycles , @xmath11 . now we shall use this embedding to reinterpret @xmath51 as a subset of @xmath11 , as follows : @xmath422 it is interesting to ask how densely packed these atoms are . computing the density of the corresponding sphere packing seems hard , but we can also ask what fraction of ` potential locations for atoms ' are actually filled by atoms . to make sense of this , note that @xmath51 is contained in the lattice @xmath52 obtained by projecting the integral 1-chains down to the space of 1-cycles : @xmath423 where @xmath13 is the projection of the space of 1-chains onto the space of 1-cycles . we can think of points in @xmath52 as potential locations for atoms . we can not define the fraction of potential atoms that are actual atoms to be @xmath424 , because both the numerator and denominator in this fraction are infinite . to deal with this , note that @xmath425 and @xmath54 acts as translations on both @xmath51 and @xmath52 . for @xmath52 this is obvious , because @xmath54 is a sublattice of @xmath52 . for @xmath51 it follows from @xmath426 and the fact that composing paths has the effect of adding their 1-chains : we can compose any loop @xmath427 with any path @xmath371 and get an atom @xmath428 which is the result of translating the atom @xmath429 by @xmath430 . when @xmath0 is a finite graph , @xmath431 and @xmath432 are finite abelian groups . in this case we define the * * to be the ratio @xmath55 this is a regularized version of the meaningless ratio @xmath424 . to compute the packing fraction , we need another description of the lattice @xmath52 . given any lattice @xmath433 in a finite - dimensional real inner product space @xmath58 , the * * @xmath434 is @xmath435 we say @xmath433 is * * if @xmath436 , or in other words , if the inner product of any two vectors in @xmath433 is an integer . we say @xmath433 is * * if @xmath437 . it is easy to see that @xmath149 is a self - dual lattice in @xmath9 , since edges form an orthonormal basis . it follows that @xmath54 is an integral lattice in @xmath11 , where the latter space inherits its inner product from @xmath9 . however , @xmath54 is not in general self - dual , and this gives another description of the lattice @xmath52 : [ dual_lemma ] if @xmath0 is a finite graph then @xmath438 . this was shown by bacher , de la harpe and nagnibeda ( * ? ? ? * lemma 1 ) . using this we can prove : [ packing_fraction_theorem ] if @xmath0 is a finite graph without bridges , then its packing fraction is @xmath57 where @xmath58 is the set of vertices of @xmath0 and @xmath59 is the set of spanning trees in @xmath0 . in fact @xmath439 and @xmath440 to show that @xmath441 , we use several facts . first , @xmath51 is the set of vertices of the maximal abelian cover of @xmath0 . second , @xmath442 . third , via this isomorphism , the action of @xmath1 as deck transformations on vertices of the maximal abelian cover is equivalent to the action of @xmath54 by translations on @xmath51 . thus , the quotient @xmath431 is isomorphic to the set of vertices of @xmath0 . to show that @xmath443 we use the work of bacher , de la harpe and nagnibeda on integral cuts @xcite . besides the already mentioned inner product on 1-chains , there is an inner product on 0-chains for which the vertices of @xmath0 form an orthonormal basis . this lets us define the adjoint @xmath444 concretely , for any vertex @xmath97 we have @xmath445 where we sum over all edges with target @xmath97 . it is well known @xcite that there is an orthogonal direct sum decomposition @xmath446 where @xmath447 is called the space of * * . inside the space of cuts there is a lattice @xmath448 called the lattice of * * . however , the decomposition of 1-chains into 1-cycles and cuts fails at the integral level : @xmath449 ths is the key to understanding the quotient @xmath432 . since @xmath149 is an integral lattice , so are @xmath450 and @xmath54 . if we had @xmath451 then @xmath450 and @xmath54 would be self - dual , but in fact they are usually not . with the help of kirchhoff s spanning tree theorem , bacher , de la harpe and nagnibeda ( * ? 2 ) prove that @xmath452 then they use a very general fact about lattices ( * ? ? ? * lemma 2 ) to construct isomorphisms @xmath453 this implies @xmath454 but @xmath455 by lemma [ dual_lemma ] , so @xmath456 as desired , and thus @xmath457 the concept of ` cut ' has other applications to topological crystals . in the proof above we saw that for any vertex @xmath97 , @xmath445 is a cut , so it projects to zero in @xmath11 : @xmath458 multiplying by @xmath459 , we obtain @xmath460 where we sum over all edges with @xmath97 as source . for any atom @xmath461 arising from a path @xmath371 , the bonds coming out of this atom correspond to edges @xmath301 , and they connect it to atoms @xmath462 . the above equation thus says that the bonds coming out of any atom give vectors in @xmath11 summing to zero . this also follows from kotani and sunada s characterization of topological crystals in terms of energy minimization @xcite . if we think of the bonds as springs , all obeying hooke s law with the same spring constant , the total force on each atom must vanish in equilibrium , so the bonds must give vectors that sum to zero . we leave the reader with two questions . first , when are all symmetries of the maximal abelian cover of @xmath0 covering symetries ? that is , when is @xmath463 ) ? second , when is @xmath19 actually a semidirect product of @xmath21 and @xmath1 ? the short exact sequence @xmath464 splits , making @xmath19 into a semidirect product , whenever every automorphism of @xmath0 fixes the basepoint @xmath115 . it does not split when @xmath0 is this graph : so , the short exact sequence does not describe @xmath19 as a semidirect product in this case . the reason is that while @xmath21 is the symmetry group of a regular hexagon , this group does not act as symmetries fixing some atom in the corresponding crystal : in this case @xmath21 is the symmetry group of a cube , and there is no way to get this group to act as symmetries of the diamond , so there is no way to express @xmath19 as a semidirect product of @xmath21 and @xmath1 . what general phenomenon is at work here ? i thank greg egan , without whose help this project could never have been done : many of the key ideas are due to him . i thank david speyer and ilya bogdanov for their help in proving lemma [ bridge_lemma ] . in particular , speyer found a proof of lemma [ bridge_lemma ] based on the lattice of integral cuts @xcite , while bogdanov noticed the importance of the ` no bridges ' condition and laid the groundwork for the proof here @xcite . i thank pbroks13 and tom ruen for the picture of the hexagonal hosohedron , which is available https://commons.wikimedia.org/wiki/file:hexagonal_hosohedron.svg[on wikicommons ] under a creative commons attribution - share alike 3.0 unported license . i thank leshabirukov for the picture of the petersen graph , which is available https://commons.wikimedia.org/wiki/file:petersen1_tiny.svg[on wikicommon ] under the same license . all the other pictures were created by egan . r. bacher , p. de la harpe and t. nagnibeda , the lattice of integral flows and the lattice of integral cuts on a finite graph , _ bull . france _ * 125 * ( 1997 ) , 167198 . also available at http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.50.6020 . eon , euclidian embeddings of periodic nets : definition of a topologically induced complete set of geometric descriptors for crystal structures , _ acta crystallographica section a _ * 67 * ( 2011 ) , 6886 . m. kotani and t. sunada , standard realizations of crystal lattices via harmonic maps , _ trans . _ , * 353 * ( 2000 ) , 120 . also available at http://www.ams.org / journals / tran/2001 - 353 - 01/s0002 - 9947 - 00 - 02632 - 5/. f. laves , zur klassifikation der silikate geometrische untersuchungen mglicher silicium - sauerstoff - verbnde als verknpfungsmglichkeiten regulrer tetraeder , _ zeitschrift fr kristallographie _ * 82 * ( 1932 ) , 114 .
we present single - dish h spectra obtained with the green bank telescope , along with optical photometric properties from the sloan digital sky survey , of 20 nearby ( d 70 mpc ) luminous compact blue galaxies ( lcbgs ) . thesel , blue , high surface brightness , starbursting galaxies were selected with the same criteria used to define lcbgs at higher redshifts . we find these galaxies are gas - rich , with m ranging from 5 to 8 m , and m l ranging from 0.2 to 2 m l , consistent with a variety of morphological types of galaxies . we find the dynamical masses ( measured within r ) span a wide range , from 3 to 1 m . however , at least half have dynamical mass - to - light ratios smaller than nearby galaxies of all hubble types , as found for lcbgs at intermediate redshifts . by comparing line widths and effective radii with local galaxy populations , we find that lcbgs are consistent with the dynamical mass properties of magellanic ( low luminosity ) spirals , and the more massive irregulars and dwarf ellipticals , such as ngc 205 .
astro-ph0407438
the hubble space telescope and advances in ground based observing have greatly increased our knowledge of the galaxy population in the distant universe . however , the nature of these galaxies and their evolutionary connections to local galaxies remain poorly understood . luminous , compact , star forming galaxies appear to represent a prominent phase in the early history of galaxy formation @xcite . in particular : * the number density of luminous , compact star forming galaxies rises significantly out to z @xmath1 1 @xcite . * the lyman break galaxies at z @xmath12 2 seen in the hubble deep field are characterized by very compact cores and a high surface brightness @xcite . * sub - millimeter imaging has revealed distant galaxies ( z @xmath1 2@xmath134 ) , half of them compact objects , which may be responsible for as much as half of the total star formation rate in the early universe @xcite . however , little is definitively known of their physical properties , or how they are related to subsets of the local galaxy population . a classification for known examples of intermediate redshift ( 0.4 @xmath0 z @xmath0 0.7 ) luminous , blue , compact galaxies , such as blue nucleated galaxies , compact narrow emission line galaxies , and small blue galaxies , has been developed by @xcite in order to be able to choose samples over a wide redshift range . they have found that the bulk of these galaxies , collectively termed luminous compact blue galaxies ( lcbgs ) , can be distinguished quantitatively from local normal galaxies by their blue color , small size , high luminosity , and high surface brightness . ( see 2.1 for more detail . ) from studies at intermediate redshifts , it has been found that lcbgs are a heterogeneous class of vigorously starbursting , high metallicity galaxies with an underlying older stellar population @xcite . while common at intermediate redshifts , they are rare locally @xcite and little is known about the class as a whole , nor their evolutionary connections to other galaxies . lcbgs undergo dramatic evolution : at z @xmath1 1 , they are numerous and have a total star formation rate density equal to that of grand - design spirals at that time . however , by z @xmath1 0 , the number density and star formation rate density of lcbgs has decreased by at least a factor of ten @xcite . since the lcbg population is morphologically and spectroscopically diverse , these galaxies are unlikely to evolve into one homogeneous galaxy class . @xcite and @xcite suggest that a subset of lcbgs at intermediate redshifts may be the progenitors of local low - mass dwarf elliptical galaxies such as ngc 205 . alternatively , @xcite and @xcite suggest that others may be disk galaxies in the process of building a bulge to become local l@xmath2 spiral galaxies . clearly , to determine the most likely evolutionary scenarios for intermediate redshift lcbgs , it is necessary to know their masses and the timescale of their starburst activity . are they comparable to today s massive or low - mass galaxies ? are they small starbursting galaxies which will soon exhaust their gas and eventually fade ? or are they larger galaxies with only moderate amounts of star formation ? only kinematic line widths that truly reflect the masses of these galaxies , as well as measures of their gas content and star formation rates , can answer these questions . using ionized gas emission line widths , @xcite , @xcite , and @xcite , have found that lcbgs have mass - to - light ratios approximately ten times smaller than typical local l@xmath2 galaxies . however , since ionized gas emission lines may originate primarily from the central regions of galaxies , their line widths may underestimate the gravitational potential @xcite . h emission lines provide a better estimate of the total galaxy mass as they measure the gravitational potential out to larger galactic radii . observations of both h and co ( the best tracer of cold h@xmath14 ) , combined with star formation rates , are necessary to estimate the starburst timescales . with current radio instrumentation , h and co can only easily be measured in very nearby lcbgs , at distances @xmath0 150 mpc for h , and @xmath0 70 mpc for co. therefore , to understand the nature and evolutionary possibilities of higher redshift lcbgs , we have undertaken a survey in h 21 cm emission and multiple rotational transitions of co of a sample of 20 local lcbgs , drawn from the sloan digital sky survey @xcite . this work , paper i , reports the optical photometric properties of our sample and the results of the h 21 cm portion of the survey , including dynamical masses and comparisons with local galaxy types . paper ii @xcite will report the results of a survey of the molecular gas conditions . knowledge of the dynamical masses , combined with gas masses and star formation rates , constrains the evolutionary possibilities of these galaxies . nearby blue compact galaxies ( bcgs ) have been studied extensively at radio and optical wavelengths since @xcite originated the term `` compact galaxy '' and @xcite distinguished between `` red '' and `` blue '' compact galaxies . the term bcg typically refers to galaxies with a compact nature , a high mean surface brightness , and emission lines superposed on a blue continuum . however , many different selection criteria have been used , leading to various definitions of bcgs and samples with a range of properties . for example , the term `` dwarf '' has been used to mean bcgs fainter than @xmath1317 ( e.g. thuan & martin 1981 ; kong & cheng 2002 ) or @xmath1318 blue magnitudes ( e.g. taylor et al . 1994 ) , or an optical diameter less than 10 kpc @xcite . the term `` blue '' has been used to mean blue on the palomar sky survey plate ( e.g. gordon & gottesman 1981 ) , or to have emission lines superposed on a blue background ( e.g. thuan & martin 1981 ) . the term `` compact '' has been used to mean smaller than 1 kpc in optical diameter ( e.g. thuan & martin 1981 ) , or qualitatively compact ( e.g. doublier et al . some , for example @xcite , began using the term blue compact dwarf to refer to the common low luminosity , low metallicity bcgs . as high luminosity ( @xmath1l@xmath2 ) , nearby bcgs are rare , many began to use the term blue compact dwarf for all nearby bcgs , regardless of luminosity . there are only a few of the rare local lcbgs ( by the @xcite definition ) in previous bcg surveys . note that @xcite have recently been studying local luminous bcgs . however , their selection criteria are not as stringent as that of @xcite , and they have been focusing on very low metallicity ( less than 15% solar ) galaxies . intermediate redshift lcbgs with rest - frame properties matching the class definition of @xcite have metallicities at least 40% solar @xcite . four of our local lcbgs have metallicities available in the literature , which range from 40 @xmath13 70 % solar @xcite . by using the selection criteria of @xcite , we ensure that our study is of those local lcbgs defined to be analogs to the widely studied higher redshift lcbgs . the observations , including sample selection and data reduction , are described in @xmath152 . the optical photometric properties , h 21 cm spectra , measurements , and derived properties are presented in @xmath153 , and analyzed in @xmath154 . we compare the derived physical properties to local normal galaxies and higher redshift lcbgs in @xmath155 , and conclude in @xmath156 . we assume h@xmath16 = 70 km s@xmath17 mpc@xmath17 throughout . when we compare our results to those of other authors , we scale their results to this value . when @xcite compared intermediate redshift lcbgs with local normal galaxies , they found that lcbgs can be isolated quantitatively on the basis of color , surface brightness , image concentration , and asymmetry . color and surface brightness were found to give the best leverage for separating lcbgs from normal galaxies . specifically , lcbgs can be defined by a region limited to b@xmath13v @xmath18 0.6 , and a b - band surface brightness within the half - light radius , sbe , brighter than 21 b - mag arc sec@xmath19 . this simple definition differs only slightly from the formal definition in @xcite which uses a color - dependent sbe . @xcite also applied a luminosity cut - off of 25% l@xmath20m@xmath21 @xmath18 @xmath1318.5 ) , to distinguish lcbgs from blue compact galaxies . lcbgs are not so extreme that they are completely separated from the continuum of normal galaxies . the sharp borders used to classify them are artificial , but serve the purpose of defining similar objects over a range of redshifts . lcbgs at intermediate redshifts ( z @xmath1 0.6 ) can be studied in deep spectroscopic surveys ( i @xmath1 24 ) , while the brightest lcbgs at high redshifts ( z @xmath1 3 ) require very deep spectroscopic surveys ( i @xmath1 26 ) . therefore , lcbgs selected in this manner are observable over a wide range of redshifts . using these color , surface brightness , and luminosity criteria , we selected our sample of local lcbgs from the sloan digital sky survey ( sdss ) . begun in 2000 , this survey will ultimately image one quarter of the sky using a large format ccd camera on a 2.5 m telescope at apache point observatory in new mexico . images are taken in five broad bands ( u , g , r , i , z ) which range from 3540 ( u ) to 9130 ( z ) . the survey has a limiting magnitude of 22.2 in g ( 4770 ) and r ( 6230 ) , our two bands of interest . after searching through approximately one million galaxies ( @xmath11500 degree@xmath22 on the sky ) , we identified only 16 nearby ( d @xmath0 70 mpc ) lcbgs . this distance cut - off was chosen to ensure our galaxies could be detected quickly in both h and co. we added four markarian galaxies from the literature which fulfilled our selection criteria and were not yet surveyed by sdss , for a total of 20 local lcbgs . the color and surface brightness characteristics of our local sample , as well as higher redshift ( 0.4 @xmath0 z @xmath0 1 ) samples of lcbgs , are compared to other nearby galaxies in figure 1 . we note that while our sample is not complete , it is representative of local lcbgs ( see castander et al . 2004 ) . as at higher redshifts @xcite , our local lcbg sample is a morphologically heterogeneous mixture of galaxies , including sb to sc spirals , s0 galaxies , polar ring galaxies , peculiar galaxies , and h liners and starbursts , as classified by hyperleda and the nasa@xmath23ipac extragalactic database ( ned ) . approximately one quarter of these galaxies are not classified in either database . roughly one third are members of multiple , sometimes interacting or merging , systems . sdss and digitized sky survey images of our sample of 20 galaxies are shown in figures 2 and 3 . the properties of our sample of local lcbgs are described below , and listed in table 1 . _ full sdss galaxy designation of the form sdss [email protected] , in the j2000 system . in the remainder of the paper , individual galaxies are referred to by an abbreviated sdss name of the form sdssjhhmm@xmath24ddmm . the four non - sdss galaxies are referred to by their markarian ( mrk ) names . _ alternate name . _ mrk and/or ngc designation , if any . _ d@xmath25 . _ hubble distance , in mpc , calculated from the optical redshift ( from sdss , except for the non - sdss galaxies for which we use ned redshifts . ) _ m@xmath21 . _ absolute blue magnitude calculated from the apparent blue magnitude , m@xmath21 , using d@xmath25 . for the sdss galaxies , m@xmath21 = g + 0.30(g@xmath13r ) + 0.18 , b@xmath13v , and r@xmath26(b ) are based on synthetic spectra which fit the observed spectral energy distribution of lcbgs . ] . the r and g magnitudes were calculated from the sdss `` petrosian '' and `` model '' magnitudes : r = r(petro ) and g = r(petro ) @xmath13 [ r(model ) @xmath13 g(model ) ] @xcite . the sdss galactic reddening corrections were applied , but no k corrections were applied as our galaxies are nearby , and many are of unknown spectral type . the k correction is at most 0.015 magnitudes in r and 0.038 magnitudes in g @xcite . no correction was applied for extinction due to inclination , also because of the uncertainty in spectral types . for the non - sdss galaxies , the `` total apparent blue magnitudes '' were used from hyperleda , and corrected for galactic reddening using extinction values from ned . _ b@xmath13v . _ the color transformation is b@xmath13v @xmath27 0.900(g@xmath13r ) + 0.145@xmath28 , where the magnitudes were calculated as above , for the sdss galaxies . the `` total b@xmath13v colors '' from hyperleda were used for the non - sdss galaxies . they were corrected for reddening using extinction values from ned . _ average surface brightness in the b band within the half light or effective radius in b - mag arc sec@xmath19 . for the sdss galaxies , sbe(b ) = m@xmath21 + 2.5 @xmath29 [ 2@xmath30r@xmath31(b ) ] , where r@xmath26(b ) , the effective or half - light radius in the b band , was calculated from the `` petrosian '' effective radii in the g and r bands [ r@xmath26(b ) = 1.30 @xmath32(g ) @xmath13 0.300 @xmath32(r)]@xmath28 . we corrected sbe for cosmological dimming by subtracting 7.5 @xmath29(1 + z ) , where the redshift , z , was calculated using the h velocity ( @xmath153.1 ) . for the non - sdss galaxies , we used sbe from hyperleda ( termed the `` mean effective surface brightness '' ) , and then corrected it for cosmological dimming in the same way . _ ned type . _ morphological type as given by ned , which does not use a homogeneous system . _ hyperleda type . _ morphological type as given by hyperleda , which uses the @xcite system . _ other sources in beam ? _ indicates whether or not each galaxy has other sources at similar velocities within the [email protected] beam of the green bank telescope ( used for the h observations ) , as given by ned . we compared hyperleda and sdss magnitudes for our sdss selected galaxies to ensure that there are no systematic offsets between the two . fourteen galaxies had magnitudes measured in both systems . the median difference between sdss and hyperleda magnitudes ( as calculated in @xmath152.1.2 ) is 0.1 magnitudes . since the standard deviation of these differences is @xmath34 0.8 magnitudes , we estimate the uncertainty in the median magnitude difference to be 0.2 magnitudes . therefore , there is no significant difference between the hyperleda and sdss magnitudes . unfortunately , while fourteen galaxies have magnitudes measured in both systems , only two sdss selected galaxies have b@xmath13v and sbe entries in hyperleda ; for these two galaxies there are no significant differences between the hyperleda and sdss values . h 21 cm observations of 19 of the 20 nearby lcbgs in our sample were made with the 100 meter green bank telescope ( gbt ) at the national radio astronomy observatory in green bank , west virginia between 2002 november 30 and december 6 . the h spectrum of sdssj0934@xmath240014 was acquired by @xcite during gbt commissioning on 2001 november 29 . the main beam half power width is [email protected] at 21 cm @xcite . both linear and circular polarizations were observed using the l - band receiver ( 1.15 @xmath13 1.73 ghz ) . position - switching was used with an offset of @xmath1318@xmath33 in right ascension . the spectral processor was used with a bandwidth of 20 mhz ( @xmath14000 km s@xmath17 ) for most galaxies , although a few were observed with a bandwidth of 5 mhz ( @xmath11000 km s@xmath17 ) . the sample time was 60 s , with each galaxy observed for between 6 and 30 minutes , for a peak signal - to - noise of at least five . the only exceptions were sdssj0218@xmath130757 and sdssj0222@xmath130830 , marginal detections , which were observed for 52 minutes each . the initial data calibration and reduction was performed using the aips++ single dish analysis environment `` dish . '' the position - switched data were calibrated in the standard way taking the difference of the on and off scans divided by the off scans . the dish package automatically calibrates the data into temperature units using tabulated values for the position of the telescope . the individual scans were then combined and a first order baseline was fit to the line - free regions and removed . our wide bandwidths ensured a sufficient line - free region . the individual polarizations were then combined , except when one polarization was significantly noisier than the other , or was affected by radio interference . finally , each spectrum was smoothed to @xmath112 km s@xmath17 channels using boxcar smoothing . the rest of the reduction and analysis was done using our own procedures written in interactive data language ( idl ) . to convert our data from temperature to flux density units , we observed the radio galaxy 3c295 . it makes an ideal calibration source as it has not varied by more than @xmath11@xmath35 since 1976 for 2.8 cm @xmath36 @xmath37 @xmath36 21 cm @xcite . comparing our observations of 3c295 with those of @xcite , we found the gain of the gbt to be 1.9 k jy@xmath17 and applied this calibration to our data . finally , obvious noise spikes were clipped out of the data . data of sdssj0934@xmath240014 , the galaxy observed during telescope commissioning , were reduced and calibrated by @xcite . the spectrum was then smoothed to @xmath112 km s@xmath17 channels . the central 800 km s@xmath17 of the final h 21 cm spectra of the 20 local lcbgs are shown in figure 4 . each galaxy s velocity calculated from optical redshifts is indicated with a triangle . dashed lines indicate the 20% crossings used to measure the line widths ( @xmath153.1 ) . all 20 galaxies were detected in the 21 cm line of h , although two , sdssj0218@xmath130757 and sdssj0222@xmath130830 , were only detected at the 3 and 4 @xmath38 ( respectively ) level . h measurements and derived optical and h quantities are listed in tables 2 , 3 and 4 and described in the following sections . since the gbt beam is [email protected] at 21 cm , the emission from the target galaxies and any other galaxies within the beam and at similar velocities are blended into one spectrum . ( see table 1 for those galaxies with other sources at similar velocities within the gbt beam . ) we find fairly broad h profiles ( 126 km s@xmath17 @xmath18 w@xmath39 @xmath18 362 km s@xmath17 ) and a variety of profile shapes , including double - horned , flat - topped and gaussian . over a third of our profiles appear asymmetric and only half of these are from galaxies known to have other sources within the gbt beam . these asymmetries may indicate asymmetries in the gas density distribution and/or disk kinematics , or unidentified galaxies within the beam @xcite . estimates of the dynamical masses of our local sample of lcbgs constrain their evolutionary possibilities . we estimate the dynamical mass , m@xmath40 , within a radius , r , as @xmath41 the constant c@xmath14 is a geometry - dependent factor which depends on the galactic light profile . for example , the king models of @xcite give 0.9 @xmath0 c@xmath14 @xmath0 1.5 , depending on the ratio of tidal to core radius . as we do not have information on the light profiles of our galaxy sample , we simply choose c@xmath14 = 1 . the line width at 20@xmath35 , corrected for the effects of inclination , w@xmath42 , is used as a measure of twice the maximum rotational velocity , v@xmath43 ( e.g. * ? ? ? w@xmath39 was measured at the points equal to 20% of the peak flux . the first crossing at 20@xmath35 on each side of the emission line was used , once the spectrum became distinguishable from the noise . these crossings are indicated by dashed lines in figure 4 . we find w@xmath39 ranging from 126 @xmath13 362 km s@xmath17 . when @xcite studied lcbgs at intermediate redshifts , they found a similar range of line widths , using optical emission lines . the recessional velocities in the barycentric system , v@xmath44 , were calculated as the midpoint between the 20@xmath35 crossings . the uncertainties listed in table 2 for w@xmath39 , v@xmath44 , and the galaxy distance ( d ) are from uncertainties in measuring the exact w@xmath39 crossings . two galaxies have multiple crossings at the 20% flux level , once the spectrum is distinguishable from the noise . one of these , sdssj0834@xmath240139 , has a wing on the h spectrum which extends over 200 km s@xmath17 . neither sdssj0834@xmath240139 nor its nearby companion have optical velocities coincident with the peak of the h spectrum ; instead , both have velocities in the wing of the spectrum ( see figure 4 ) . for this reason , we used w@xmath39 measured from the last crossing , which is at the edge of the wing . the other galaxy with multiple crossings at 20% is sdssj0904@xmath245136 , which has a low level extension to one side of the spectrum . we used the first crossing at 20% . the uncertainties in these two w@xmath39 measurements are reflected in the large errors associated with w@xmath39 and derived quantities . we have initiated an observing program at the very large array ( vla ) to map these local lcbgs in h to disentangle target galaxy emission from any other galaxies in the field , improving the mass estimates . to correct the line width for the effects of inclination ( _ i _ ) , we divided w@xmath39 by @xmath45(_i _ ) , to give w@xmath42 . each sdss galaxy s inclination was approximated from the sdss data by : @xmath46 the sdss isophotal major and minor axes in the g - band ( 4770 ) were used , except for sdssj0943@xmath130215 where no g band data were available . in that case , r - band ( 6230 ) data were used . the sdss isophotal axes are derived from the ellipticity of the 25 magnitude arc sec@xmath19 isophote in each band . inclinations derived from sdss isophotal axes in r , i ( 7630 ) , and z ( 9130 ) bands agree with inclinations derived from g - band data to within 8@xmath47 . the inclinations from hyperleda were used for the markarian galaxies . thirteen of our sdss galaxies also have inclinations available in hyperleda , which are calculated from the apparent flattening and morphological type of the galaxies . we estimate the dispersion of the differences between sdss and hyperleda inclinations is 12@xmath47 . a further correction can be made to w@xmath42 to account for random motions , giving w@xmath48 . as outlined in @xcite , @xmath49 - 2w_t^2\exp-(w_{20}/w_c)^2\ ] ] and @xmath50 w@xmath51 is the random motion component of the line width ; w@xmath51 = 38 km s@xmath17 . w@xmath52 characterizes the transition region between linear and quadrature summation of rotational and dispersive terms ; w@xmath52 = 120 km s@xmath17 . the formula degenerates to linear summation for giant galaxies , and quadrature summation for dwarf galaxies @xcite . this correction for random motions decreases the line width of the local lcbg sample by 26 to 38 km s@xmath17 , depending on the rotational velocity . it is crucial , when comparing dynamical masses from different studies , that the rotational velocities be calculated in the same way . in general , we calculate the rotational velocities from w@xmath53 . however , when we wish to compare our sample with others ( e.g. * ? ? ? * ) who have not applied this correction for random motions to their line widths , we also do not apply this correction . we indicate dynamical masses calculated without correcting w@xmath42 for random motions as m@xmath54 . unless indicated as such , all dynamical masses are calculated using the line width corrected for random motions , w@xmath53 . note that in all cases , the line widths have been corrected for inclination . it is also crucial , when comparing dynamical masses from different studies , that they be measured within the same radius . typically , the h radius ( r@xmath3 ) is measured at 1 m@xmath7 pc@xmath19 and is used to estimate the total enclosed mass of a galaxy . however , it is only possible to measure r@xmath3 with interferometers in nearby galaxies . by practical necessity then , some other radius must be adopted for our sample and the galaxy s mass is assumed to be spherically distributed in that radius . two different radii are commonly used : r@xmath26 , the effective or half - light radius , and r@xmath10 , the isophotal radius at the limiting surface brightness of 25 b - magnitudes arc sec@xmath19 . we have measurements of both r@xmath26 ( in r and b band from sdss ) and r@xmath10 ( in b - band from hyperleda ) for most galaxies . @xcite found that r@xmath10 = 2.5 r@xmath26 for nearby emission line galaxies measured in r - band . when we compare our values of r@xmath26 to r@xmath10 , we find medians of @xmath55 and @xmath56 for the galaxies with no r@xmath10 ( sdssj0218@xmath130757 , sdssj0222@xmath130830 and sdssj1118@xmath246316 ) or r@xmath26 ( non - sdss galaxies ) measurements available , we estimate these quantities from the above relations . we find our local sample of lcbgs have dynamical masses , measured within r@xmath10 , ranging from 3@xmath410@xmath6 to 1@xmath410@xmath11 m@xmath7 , with a median of 3@xmath410@xmath57 m@xmath7 . the dynamical masses measured within the effective radius ( measured in both the r and b - band ) range from 8@xmath410@xmath5 to 3@xmath410@xmath57 m@xmath7 , with a median of 8@xmath410@xmath58 m@xmath7 . figure 5 compares the dynamical masses measured for local lcbgs to nearby galaxies of hubble type s0a through i m , where `` m '' indicates magellanic , or low luminosity @xcite . note that @xcite did not correct line widths for random motions as we did . therefore , in figure 5 we compare m@xmath54(r @xmath59r@xmath10 ) which have not been corrected for random motions . these dynamical masses are higher than those we list in table 3 , which have been corrected for random motions . as seen in figure 5 , some local lcbgs have dynamical masses as large as l@xmath2 galaxies ( @xmath110@xmath11 m@xmath7 , roberts & haynes 1994 ) . however , at least 75@xmath35 have dynamical masses approximately an order of magnitude smaller than typical local l@xmath2 galaxies , consistent with observations of lcbgs at higher redshifts ( as discussed in @xmath15 1.1 ) . however , our sample includes galaxies down to 0.25 l@xmath2 . galaxies of this luminosity typically have dynamical masses ( within r@xmath10 ) @xmath12@xmath410@xmath57 m@xmath7 @xcite . therefore , it is more appropriate to compare mass - to - light ratios ; see @xmath154.1 . we estimate the random errors associated with the dynamical masses to be approximately 50% . this estimate is made from uncertainties in the inclination ( @xmath34 12@xmath47 , from the earlier comparison of hyperleda and sdss inclinations ) ; uncertainties in radii measurements ( as given by hyperleda and sdss ) ; and uncertainties in measuring w@xmath39 ( as discussed above ) . however , the random errors are overwhelmed by the systematic uncertainties when estimating dynamical masses . we have chosen a structural constant of c@xmath14 = 1 , but this could be as high as 1.5 , as discussed earlier . even more significant is that 45% of our galaxy sample have other galaxies at similar velocities in the beam of the gbt . as the h emission from all sources in the beam and within the bandwidth is blended into one spectrum , we may be overestimating the masses of these galaxies . finally , a fifth of our galaxies have low inclinations ( @xmath18 40@xmath47 ) which lead to large corrections and therefore increasing uncertainties in the rotational velocities and overestimations of the dynamical masses . the dynamical masses we report may be viewed as upper limits we are certainly overestimating the masses in many cases , but not underestimating them . we are pursuing a follow - up program with both the arecibo radio telescope and the vla to address these issues . at 21 cm the arecibo radio telescope has a beam size of [email protected] @xmath4 [email protected] , a third the size of the gbt beam . this decreases the number of target galaxies observed with other galaxies in the beam , decreasing the number of galaxies with overestimates of dynamical masses . we have already observed over 40 lcbgs at arecibo , but results are not yet available . the vla in `` b '' configuration provides the ability to map our galaxies in 21 cm emission to a resolution of 5@xmath33@xmath33 . we have begun a survey of those lcbgs with companions . vla maps allow us to disentangle the emission from each galaxy and more accurately estimate the dynamical mass of the target galaxy . along with dynamical mass , a knowledge of the amount of fuel , i.e. atomic and molecular gas , available for the starburst activity , is critical for narrowing down the evolutionary possibilities for lcbgs . when combined with star formation rates , this gives an estimate of the maximum length of the starburst at the current rate of star formation . we defer a full discussion of this until paper ii @xcite where we present our measurements of the co content of these galaxies , but present the h results here . the h masses are given by : @xmath60 @xcite , where the distance ( measured from h ) is d = v@xmath44 h@xmath61 . the total h flux , @xmath62 , was calculated by numerically integrating under the spectrum where it is distinguishable from the noise . for the two marginal detections , sdssj0218@xmath130757 and sdssj0222@xmath130830 , h masses were calculated using the optical ( sdss ) recessional velocities . we find local lcbgs have h masses ranging from 5@xmath410@xmath5 to 8@xmath410@xmath6 m@xmath7 , with a median of 5@xmath410@xmath6 m@xmath7 . these span the range of h masses in nearby galaxies across the hubble sequence ; the median is that of nearby late - type spiral galaxies @xcite . the uncertainties listed for m@xmath3 ( table 3 ) include uncertainties in the distance and integrated flux density , as listed in table 2 . however , as in the discussion of dynamical mass uncertainties in @xmath153.1 , the h masses for nearly half of our galaxies are most likely overestimates due to the presence of other galaxies within the gbt beam . dynamical mass - to - light ratios indicate whether our local sample of lcbgs are under - massive for their luminosities . we calculated the total blue luminosities , l@xmath21 , of our galaxies from m@xmath21 , assuming m@xmath63 = 5.48 @xcite . we find the median mass - to - light ratio , m@xmath40(r @xmath18 r@xmath10 ) l@xmath8 = 3 m@xmath7 l@xmath64 , with a minimum of 0.6 and a maximum of 9 m@xmath7 l@xmath64 . @xcite find that m@xmath40(r @xmath18 r@xmath10 ) l@xmath8 ranges from 3.6 @xmath13 10 m@xmath7 l@xmath64 in local galaxies across the hubble sequence . the median m@xmath40(r @xmath18 r@xmath10 ) l@xmath8 is fairly constant ranging from 5 to 7 m@xmath7 l@xmath64 . however , as discussed in @xmath153.1 , to compare our mass - to - light ratios to those of local , normal hubble - types in @xcite , we must use dynamical masses which have not been corrected for random motions . figure 6 compares such mass - to - light ratios for our local sample of lcbgs and nearby hubble types . while some local lcbgs have mass - to - light ratios equal to or greater than the median for local hubble types , approximately half the local lcbgs have smaller mass - to - light ratios . many of the galaxies with mass - to - light ratios typical of local hubble types have other galaxies within the beam , possibly leading to an overestimation of the dynamical mass - to - light ratio of the target galaxy . therefore , in general , lcbgs tend to have mass - to - light ratios smaller than local normal galaxies . that is , most are small galaxies undergoing large amounts of star formation and not simply large galaxies with moderate amounts of star formation . we calculated the gas mass fraction , m@xmath3 m@xmath65 ( r @xmath18 r@xmath3 ) for the local sample of lcbgs as a measure of their atomic gas richness . we do not have interferometric observations of these galaxies , so we estimated the hydrogen radius , r@xmath3 , from optical radii . @xcite found for nearby spiral galaxies that r@xmath3 = 2 r@xmath10 , where r@xmath3 is measured at the 1 m@xmath7 pc@xmath19 level . however , @xcite found r@xmath3 ranging from 3 @xmath13 5 r@xmath10 for a small sample of local h galaxies which are similar to lcbgs , but much less luminous ( m@xmath21 @xmath66 @xmath1316 ) . @xcite studied a large sample of nearby galaxies and found that galaxies with smaller optical radii have larger h extensions . @xcite have measured r@xmath3 in one of our lcbgs , mrk 314 , using the vla . they find r@xmath3 = 4 r@xmath10 . therefore , although we estimated r@xmath3 as 2 @xmath4 r@xmath10 following @xcite , we note it may be an underestimate , which would cause us to overestimate the gas mass fraction . we are undertaking interferometric observations of some of our local sample of lcbgs to directly measure r@xmath3 . this will be addressed in a future paper . in local lcbgs , we find the gas mass fraction , m@xmath3 m@xmath65 , ranges from 0.03 to 0.2 , with a median of 0.08 . that is , it ranges from normal hubble type spirals ( @xmath18 0.1 @xcite ) to very gas rich galaxies such as nearby h and irregular galaxies studied by @xcite with gas fractions ranging from 0.2 @xmath13 0.4 . this is similar to the range of 0.01 @xmath13 0.5 found by @xcite for local blue compact galaxies , most of them less luminous than lcbgs . the fraction of hydrogen mass to blue luminosity provides a distance independent alternative measure of the gas richness of galaxies . we find our local lcbgs have m@xmath3 l@xmath8 ranging from 0.09 to 2 m@xmath7 l@xmath67 , with a median of 0.4 m@xmath7 l@xmath67 . this spans the range of nearby galaxies from s0a through i m , the median corresponding to late - type spirals @xcite . @xcite studied a range of blue compact galaxies , ranging from faint blue compact dwarfs to lcbgs , and found the same median value , 0.4 m@xmath7 l@xmath67 . the tully - fisher relationship , a relation between line width and luminosity , is a fundamental scaling relation for non - interacting spiral galaxies . an absolute blue magnitude versus h line width version of the tully - fisher relationship is shown in figure 7 . we have compared our local lcbgs with the tully - fisher relation found for @xmath14500 normal galaxies within @xmath140 mpc @xcite . as seen in figure 7 , the location of local lcbgs is consistent with the @xcite relation , although with a large scatter . @xcite have a 1 @xmath38 scatter of 0.3 magnitudes in m@xmath21 , while the local lcbgs have a 1 @xmath38 scatter of 0.9 magnitudes in m@xmath21 . approximately half the galaxies ( without other sources in the beam ) lie to the left of the tully - fisher relationship , indicating they have lower masses than expected from their luminosities , although this result is not statistically significant . this is consistent with the distribution of lcbg and hubble type galaxy mass - to - light ratios in figure 6 . seven of the galaxies lying to the right of the tully - fisher relationship have other galaxies within the gbt beam , possibly leading to overestimations of the line widths . note that in our comparison of magnitudes , masses , and mass - to - light ratios , as in all other comparisons in this paper , we have adjusted all values to a hubble constant of h@xmath16 = 70 km s@xmath17 mpc@xmath17 we have observed 20 local lcbgs chosen with the same selection criteria as those lcbgs common at higher redshifts . lcbgs can not remain lcbgs for a long period of time : the number density of lcbgs decreases by at least a factor of ten from z @xmath12 0.5 to today . that is , while lcbgs were common in the past , there are very few today . out of approximately a million nearby galaxies observed by the sdss , only about a hundred are lcbgs . ( see castander et al . 2004 for a discussion of the local space density of lcbgs . ) therefore , lcbgs must evolve into some other galaxy type . from studies of intermediate redshift lcbgs , @xcite and @xcite proposed that some may be progenitors of local dwarf elliptical galaxies . alternatively , @xcite and @xcite proposed that some may be disk galaxies in the process of forming a bulge to become present - day l@xmath2 spiral galaxies . two pieces of information are crucial to determining if these possibilities are likely . it is necessary to know the dynamical masses and the duration of the starbursts . we have measured the dynamical masses of our local sample of lcbgs using h . the length of the starburst will not be estimated until we discuss our molecular gas survey in paper ii @xcite . however , we can use our measurements of line width and radius to identify local galaxies comparable to lcbgs , independent of the evolutionary stage of their stellar populations . in figure 8 we plot the effective or half - light radius , r@xmath26 , versus the velocity dispersion , @xmath38 , for our sample of local lcbgs , intermediate redshift ( 0.4 @xmath0 z @xmath0 1 ) lcbgs @xcite , and local samples of elliptical , spiral , magellanic spiral , irregular , and dwarf elliptical galaxies ( guzmn et al . 1996 ; hyperleda ) . we measured the @xmath38 of our sample of galaxies from the h spectra by measuring the moments of the spectra . the measurements were made in the same way as we described in @xmath153.2 for measuring @xmath62 , the zeroth moment . we find similar results if we simply scale w@xmath39 to @xmath38 by assuming the spectra are gaussian , that is , dividing w@xmath39 by 3.6 . the values of @xmath38 for the other galaxy samples are from optical line width measurements , except the measurements for magellanic spirals are from 21 cm h line widths . optical line widths may be smaller than h line widths . for example , @xcite studied a sample of nearby blue compact galaxies and found the ionized gas emission line widths to be systematically smaller than the neutral hydrogen emission line widths . on average , the ratio of w@xmath39 measured from h to w@xmath39 measured from h was 0.66 for their sample of 11 blue compact galaxies . however , the galaxies in @xcite s sample tend to be smaller both in effective radius and h line width than our sample of local lcbgs , suggesting that the difference between the optical and h line widths may be smaller as well @xcite . the r@xmath26 versus @xmath38 plot allows us to compare the dynamical mass properties of our sample of lcbgs with the other galaxy types . these properties are expected to remain constant despite luminosity evolution . we find that while some local lcbgs have @xmath38 consistent with local spiral galaxies , they tend to be too small in r@xmath26 . however , local lcbgs are consistent with the smaller spiral galaxies , magellanic spirals . they are inconsistent with elliptical galaxies , but are consistent with the most massive dwarf elliptical and irregular galaxies . there is much confusion by what various authors mean when discussing dwarf elliptical galaxies ( sometimes called spheroidal galaxies ) . we are referring to galaxies such as ngc 205 and _ not _ the less massive and less luminous galaxies like draco and carina . finally , when we compare our local sample of lcbgs with those lcbgs observed at intermediate redshifts , we find that they occupy the same region of r@xmath26 @xmath13 @xmath38 space , suggesting we are indeed examining a similar mass range in both samples of galaxies . we also compared local lcbgs to other galaxy types in w@xmath39 sin@xmath17(_i _ ) versus r@xmath10 space , where w@xmath39 was measured from h emission for all samples . this allowed us to investigate if the wavelength used to measure the line width , or correcting the line width for inclination , was having an effect on our interpretation of which galaxy types most resemble lcbgs . we could only compare local lcbgs , spirals , magellanic spirals , and irregular galaxies , as h observations of intermediate redshift lcbgs and dwarf ellipticals are rare or non - existent . our findings were entirely consistent with the interpretation from figure 8 . local lcbgs are therefore consistent with the dynamical mass properties of the most massive dwarf ellipticals and irregulars , and lower mass or magellanic spirals . these classes vary in color and magnitude . knowledge of the amount of molecular gas , time scale of starburst and fading , along with the ability of the galaxy to retain its interstellar medium , will allow us to discriminate between these remaining possibilities . we will begin this work in paper ii @xcite . given the diverse nature of lcbgs , it is likely that multiple scenarios apply , each to a different subset of lcbgs . we have performed a single dish 21 cm h survey of 20 local lcbgs chosen to be local analogs to the numerous lcbgs studied at intermediate redshifts ( 0.4 @xmath0 z @xmath0 0.7 ) . our findings have verified results from intermediate redshift lcbg studies . we have found that local lcbgs are a morphologically heterogeneous mixture of galaxies . they are typically gas - rich , with median values of m@xmath3 = 5@xmath410@xmath6 m@xmath7 and m@xmath3 l@xmath8 = 0.4 m@xmath7 l@xmath9 . approximately half have mass - to - light ratios approximately ten times smaller than local galaxies of all hubble types at similar luminosities , confirming that these are indeed small galaxies undergoing vigorous bursts of star formation . this proportion is likely an underestimate , as nearly half our sample of galaxies may have dynamical mass overestimates . by comparing line widths and radii with local galaxy populations , we find that local lcbgs are consistent with magellanic spirals , and the more massive irregulars and dwarf ellipticals . measurements of the length of starburst , amount of fading , and ability of these galaxies to retain their interstellar media will help to constrain the evolutionary possibilities of this galaxy class . we begin to address these issues in paper ii @xcite , where we present the results of a molecular gas survey of these same local lcbgs . _ acknowledgments _ we thank the referee for helpful comments which improved the quality of this paper . we thank rick fisher for providing the h spectrum of sdssj0934 + 0014 . we also thank the operators and staff at the gbt for their help with the observing and reduction , and their hospitality . support for this work was provided by the nsf through award gssp02 - 0001 from the nrao . d. j. p. acknowledges generous support from an nsf mps distinguished international postdoctoral research fellowship , nsf grant ast0104439 . r. g. acknowledges funding from nasa grant ltsa nag5 - 11635 . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. we have made extensive use of hyperleda ( http://www-obs.univ-lyon1.fr/hypercat/ ) and the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration ( http://nedwww.ipac.caltech.edu/ ) . the digitized sky surveys were produced at the space telescope science institute under u.s . government grant nag w-2166 . the images of these surveys are based on photographic data obtained using the oschin schmidt telescope on palomar mountain and the uk schmidt telescope . llcccclcc mrk 297 & ngc 6052 , ngc 6064 & 67 & @xmath1321.0 & 0.4 & 20.6 & & sc & y + mrk 314 & ngc 7468 & 30 & @xmath1318.5 & 0.4 & 20.2 & e3 , pec ( polar ring ? ) & e & n + mrk 325 & ngc 7673 , mrk 325 & 49 & @xmath1320.0 & 0.4 & 20.0 & sac ? , pec , h ii starburst & sc & y + mrk 538 & ngc 7714 & 40 & @xmath1320.1 & 0.4 & 20.2 & sb(s)b , pec , h ii liner & sbb & y + sdss j011932.95 + 145219.0 & ngc 469 & 59 & @xmath1318.9 & 0.4 & 20.3 & & & y + sdss [email protected] & & 69 & @xmath1318.8 & 0.5 & 20.2 & & & n + sdss [email protected] & & 67 & @xmath1318.6 & 0.4 & 20.1 & & & n + sdss j072849.75 + 353255.2 & & 56 & @xmath1318.9 & 0.4 & 20.3 & s ? & sbc & n + sdss j083431.70 + 013957.9 & & 59 & @xmath1319.1 & 0.6 & 20.6 & sb(s)b & sbb & y + sdss j090433.53 + 513651.1 & mrk 101 & 68 & @xmath1319.7 & 0.6 & 20.3 & s & sc & n + sdss j091139.74 + 463823.0 & mrk 102 & 61 & @xmath1319.3 & 0.5 & 19.1 & s ? & & n + sdss j093410.52 + 001430.2 & mrk 1233 & 70 & @xmath1319.8 & 0.3 & 19.9 & sb & sbc & y + sdss j093635.36 + 010659.8 & & 71 & @xmath1319.1 & 0.6 & 21.0 & & & y + sdss [email protected] & & 68 & @xmath1319.3 & 0.5 & 20.4 & s0 & s0 & n + sdss j111836.35 + 631650.4 & mrk 165 & 46 & @xmath1318.6 & 0.4 & 19.4 & compact starburst & & n + sdss j123440.89 + 031925.1 & ngc 4538 & 67 & @xmath1319.2 & 0.6 & 21.0 & s pec & sbc & n + sdss j131949.93 + 520341.1 & & 67 & @xmath1318.7 & 0.2 & 20.1 & & & y + sdss j140203.52 + 095545.6 & ngc 5414 , mrk 800 & 61 & @xmath1319.7 & 0.5 & 19.7 & pec & & y + sdss j150748.33 + 551108.6 & & 48 & @xmath1318.9 & 0.4 & 20.8 & s & sbc & n + sdss j231736.39 + 140004.3 & ngc 7580 , mrk 318 & 63 & @xmath1319.3 & 0.6 & 20.4 & s ? & sbc & n + lllll mrk 297 & 4739 @xmath34 17 & 68 @xmath34 0.2 & 362 @xmath34 17 & 6.5 @xmath34 0.08 + mrk 314 & 2081 @xmath34 17 & 30 @xmath34 0.2 & 188 @xmath34 17 & 12 @xmath34 0.1 + mrk 325 & 3427 @xmath34 17 & 49 @xmath34 0.2 & 202 @xmath34 17 & 11 @xmath34 0.2 + mrk 538 & 2798 @xmath34 17 & 40 @xmath34 0.2 & 240 @xmath34 17 & 20 @xmath34 0.2 + sdssj0119 + 1452 & 4098 @xmath34 17 & 59 @xmath34 0.2 & 266 @xmath34 17 & 2.4 @xmath34 0.1 + sdssj0218@xmath130757 & & & & 0.41 @xmath34 0.07 + sdssj0222@xmath130830 & & & & 0.61 @xmath34 0.09 + sdssj0728 + 3532 & 3953 @xmath34 17 & 56 @xmath34 0.2 & 216 @xmath34 17 & 7.9 @xmath34 0.1 + sdssj0834 + 0139 & 4215 @xmath34 160 & 60 @xmath34 2 & 329 @xmath34 160 & 6.9 @xmath34 0.2 + sdssj0904 + 5136 & 4782 @xmath34 103 & 68 @xmath34 2 & 204 @xmath34 103 & 4.5 @xmath34 0.2 + sdssj0911 + 4636 & 4281 @xmath34 17 & 61 @xmath34 0.2 & 151 @xmath34 17 & 2.0 @xmath34 0.1 + sdssj0934 + 0014 & 4860 @xmath34 17 & 69 @xmath34 0.2 & 332 @xmath34 17 & 4.8 @xmath34 0.2 + sdssj0936 + 0106 & 4920 @xmath34 17 & 70 @xmath34 0.2 & 255 @xmath34 17 & 3.3 @xmath34 0.09 + sdssj0943@xmath130215 & 4823 @xmath34 17 & 69 @xmath34 0.2 & 230 @xmath34 17 & 2.6 @xmath34 0.06 + sdssj1118 + 6316 & 3218 @xmath34 17 & 46 @xmath34 0.2 & 152 @xmath34 17 & 2.2 @xmath34 0.07 + sdssj1234 + 0319 & 4685 @xmath34 17 & 67 @xmath34 0.2 & 269 @xmath34 17 & 4.4 @xmath34 0.1 + sdssj1319 + 5203 & 4619 @xmath34 17 & 66 @xmath34 0.2 & 202 @xmath34 17 & 7.7 @xmath34 0.09 + sdssj1402 + 0955 & 4267 @xmath34 17 & 61 @xmath34 0.2 & 343 @xmath34 17 & 6.8 @xmath34 0.2 + sdssj1507 + 5511 & 3373 @xmath34 17 & 48 @xmath34 0.2 & 126 @xmath34 17 & 3.7 @xmath34 0.1 + sdssj2317 + 1400 & 4413 @xmath34 17 & 63 @xmath34 0.2 & 254 @xmath34 17 & 6.2 @xmath34 0.1 + llccccccc mrk 297 & 7.0 @xmath34 0.1 & 42 & 486 & 243 & 8.0 & 11 & 2.0 & 2.7 + mrk 314 & 2.5 @xmath34 0.05 & 65 & 170 & 85 & 3.8 & 0.63 & 0.94 & 0.16 + mrk 325 & 6.3 @xmath34 0.1 & 43 & 242 & 121 & 9.6 & 3.3 & 2.4 & 0.83 + mrk 538 & 7.6 @xmath34 0.1 & 50 & 264 & 132 & 11 & 4.5 & 2.8 & 1.1 + sdssj0119 + 1452 & 2.0 @xmath34 0.1 & 84 & 230 & 115 & 5.6 & 1.7 & 1.6 & 0.48 + sdssj0218@xmath130757 & 0.47 @xmath34 0.08 & 52 & & & 4.9 & & 1.3 & + sdssj0222@xmath130830 & 0.65 @xmath34 0.1 & 48 & & & 4.2 & & 1.2 & + sdssj0728 + 3532 & 6.0 @xmath34 0.1 & 34 & 321 & 160 & 5.4 & 3.3 & 1.2 & 0.74 + sdssj0834 + 0139 & 5.9 @xmath34 0.5 & 83 & 293 & 147 & 6.5 & 3.3 & 1.6 & 0.82 + sdssj0904 + 5136 & 4.9 @xmath34 0.3 & 34 & 301 & 150 & 7.4 & 3.9 & 2.0 & 1.0 + sdssj0911 + 4636 & 1.7 @xmath34 0.1 & 33 & 221 & 111 & 5.6 & 1.6 & 1.2 & 0.35 + sdssj0934 + 0014 & 5.4 @xmath34 0.2 & 51 & 378 & 189 & 6.8 & 5.7 & 1.9 & 1.5 + sdssj0936 + 0106 & 3.8 @xmath34 0.1 & 48 & 293 & 146 & 7.1 & 3.5 & 2.0 & 1.0 + sdssj0943@xmath130215 & 2.9 @xmath34 0.07 & 84 & 194 & 97 & 4.9 & 1.1 & 1.8 & 0.39 + sdssj1118 + 6316 & 1.1 @xmath34 0.04 & 82 & 123 & 61 & 3.1 & 0.27 & 0.91 & 0.08 + sdssj1234 + 0319 & 4.7 @xmath34 0.1 & 52 & 293 & 147 & 6.7 & 3.4 & 2.1 & 1.1 + sdssj1319 + 5203 & 7.9 @xmath34 0.1 & 44 & 239 & 120 & 5.3 & 1.8 & 1.2 & 0.41 + sdssj1402 + 0955 & 6.0 @xmath34 0.2 & 55 & 372 & 186 & 8.7 & 7.0 & 1.6 & 1.3 + sdssj1507 + 5511 & 2.0 @xmath34 0.07 & 44 & 144 & 72 & 7.9 & 0.95 & 1.8 & 0.21 + sdssj2317 + 1400 & 5.8 @xmath34 0.1 & 31 & 420 & 210 & 7.3 & 7.5 & 1.7 & 1.8 + lccccc mrk 297 & 0.03 & 39 & 0.2 & 3 & 0.7 + mrk 314 & 0.2 & 3.9 & 0.7 & 2 & 0.4 + mrk 325 & 0.1 & 16 & 0.4 & 2 & 0.5 + mrk 538 & 0.08 & 17 & 0.5 & 3 & 0.7 + sdssj0119 + 1452 & 0.06 & 5.6 & 0.4 & 3 & 0.9 + sdssj0218 - 0757 & & 5.2 & 0.09 & & + sdssj0222 - 0830 & & 4.3 & 0.2 & & + sdssj0728 + 3532 & 0.09 & 5.6 & 1 & 6 & 1 + sdssj0834 + 0139 & 0.1 & 6.8 & 0.9 & 5 & 1 + sdssj0904 + 5136 & 0.07 & 12 & 0.4 & 3 & 0.9 + sdssj0911 + 4636 & 0.05 & 8.2 & 0.2 & 2 & 0.4 + sdssj0934 + 0014 & 0.05 & 13 & 0.4 & 4 & 1 + sdssj0936 + 0106 & 0.06 & 6.8 & 0.6 & 5 & 2 + sdssj0943 - 0215 & 0.1 & 8.2 & 0.4 & 1 & 0.5 + sdssj1118 + 6316 & 0.2 & 4.3 & 0.3 & 0.6 & 0.2 + sdssj1234 + 0319 & 0.07 & 7.4 & 0.6 & 5 & 1 + sdssj1319 + 5203 & 0.2 & 4.7 & 2 & 4 & 0.9 + sdssj1402 + 0955 & 0.04 & 12 & 0.5 & 6 & 1 + sdssj1507 + 5511 & 0.1 & 5.6 & 0.4 & 2 & 0.4 + sdssj2317 + 1400 & 0.04 & 8.2 & 0.7 & 9 & 2 + . the spectra have been smoothed to a resolution of @xmath112 km s@xmath17 and only the central 800 km s@xmath17 are shown . the dashed lines indicate the 20% crossings used to measure the line width at 20% . the triangles indicate the recessional velocities calculated from sdss redshifts for the sdss galaxies ; velocities from ned redshifts are shown for the non - sdss galaxies . those galaxies with other sources at similar velocities within the gbt beam are indicated by a star in the upper right corner . [ fig4 ] ] ( r @xmath59r@xmath10 ) , for local lcbgs , as measured from h observations , is shown as a gray histogram . for comparison , the range of dynamical masses for local hubble type galaxies are indicated with black boxes . the `` m '' in `` sm '' and `` i m '' indicates magellanic or low - luminosity spirals and irregulars . note that in order to compare lcbg dynamical masses with @xcite s results for local hubble types , the dynamical masses plotted here do not include a line width correction for random motions . [ fig5 ] ] ) to l@xmath21 ratios for local lcbgs is shown as a gray histogram . for comparison , the range of mass - to - light ratios for local hubble type galaxies ( s0a to i m ) @xcite are shown . as in figure 5 , for accurate comparisons , these mass - to - light ratios do not include a line width correction for random motions . many lcbgs tend to have mass - to - light ratios smaller than local normal galaxies , consistent with findings at intermediate redshifts . note that we may have overestimated the dynamical masses for most of the lcbgs with higher mass - to - light ratios . [ fig6 ] ] ) versus absolute blue magnitude ( m@xmath21 ) . local lcbgs are indicated by filled circles . those galaxies with other sources at similar velocities within the gbt beam are circled ; their line widths may be overestimated . the solid line indicates the tully - fisher relationship from @xcite ; the dotted lines indicate their 1 @xmath38 scatter of 0.3 m@xmath21 . the location of our local lcbgs is consistent with the tully - fisher relationship , but with a higher 1 @xmath38 scatter of 0.9 m@xmath21 . [ fig7 ] ] ) versus the the velocity dispersion ( @xmath38 ) is plotted for our local lcbgs ( filled circles ) and intermediate redshift ( 0.4 @xmath0 z @xmath0 1 ) lcbgs ( open circles ) @xcite . the regions occupied by the bulk of other local galaxy types@xmath13ellipticals ( e ) , spirals ( s ) , magellanic spirals ( sm ) , dwarf ellipticals ( de ) , and irregulars ( irr)@xmath13are indicated ( guzmn et al . 1996 , hyperleda ) . we have also indicated the approximate locations for some representative galaxies : draco , ngc 205 and m 31 ( hyperleda ) . local lcbgs are consistent with higher mass irregulars and dwarf ellipticals , and lower mass or magellanic spirals , as well as intermediate redshift lcbgs .
bogoliubov - de gennes theory is used to investigate the effect of the size of a superconducting square on the vortex states in the quantum confinement regime . when the superconducting coherence length is comparable to the fermi wavelength , the shape resonances of the superconducting order parameter have strong influence on the vortex configuration . several unconventional vortex states , including asymmetric ones , giant multi - vortex combinations , and states comprising giant antivortex , were found as ground states and their stability was found to be very sensitive on the value of , the size of the sample , and the magnetic flux . by increasing the temperature and/or enlarging the size of the sample , quantum confinement is suppressed and the conventional mesoscopic vortex states as predicted by the ginzburg - laudau ( gl ) theory are recovered . however , contrary to the gl results we found that the states containing symmetry - induced vortex - antivortex pairs are stable over the whole temperature range . it turns out that the inhomogeneous order parameter induced by quantum confinement favors vortex - antivortex molecules , as well as giant vortices with a rich structure in the vortex core - unattainable in the gl domain .
1303.7153
vortex states in mesoscopic superconductors have been extensively studied in the past two decades , both theoretically and experimentally.@xcite two main interactions have been found to govern vortex behavior in a mesoscopic system . the first one is the vortex - vortex interaction , which causes vortices to form abrikosov triangular lattices in bulk type - ii superconductors . the second one is the interaction between vortices and sample boundaries , which makes vortex configurations strongly dependent on the size and geometry of mesoscopic samples - whose dimensions are of the order of the penetration depth @xmath3 or the coherence length @xmath4 . for example , in square mesoscopic samples , vortex configurations try to best match the @xmath5 symmetry . when there is only one vortex in the sample ( @xmath6 state where @xmath7 is the winding number or vorticity ) , the vortex always sits in the center of the sample in order to balance the boundary effect from all sides . for the @xmath8 state , two vortices sit on the diagonal such that the vortex - vortex separation is maximal in order to minimize the intervortex interaction . a giant vortex with @xmath8 can be induced when the boundary confinement pushes two single vortices together , as predicted theoretically@xcite and observed experimentally.@xcite for @xmath9 state , because of its incompatability with the four - fold symmetry , the theory predicts that the ground state corresponds to an anti - vortex sitting at the center surrounded by four vortices.@xcite in short , the symmetry of the sample largely determines the vortex configurations when the size of the superconductor is reduced . however , the properties of nanoscale superconductors , whose sizes are of the order of the fermi wavelength @xmath10 , are very different from those of mesoscopic superconductors . this is because the distance between electronic levels becomes comparable to the superconducting energy gap due to quantum confinement @xcite . as a consequence , the number of cooper pairs is suppressed which leads to the quantum - size effect ( qse),@xcite quantum - size cascades,@xcite the shell effect@xcite and inhomogeneous spatial distribution of the order parameter.@xcite the latter is the most important for the present work because it is expected to strongly influence the vortex states in nanoscale superconductors . a similar behavior was shown for an isolated vortex core , where oscillations of there order parameter on the order of the fermi wavelength were predicted . @xcite inhomogeneous superconductivity has been studied in various systems in the last decades and shows more complex behavior than homogeneous ones . it is known that vortices tend to migrate and get pinned in areas where superconductivity is suppressed.@xcite the reason is that it is more favorable energetically for a vortex to suppress the superconducting order parameter in a region where it has already been suppressed , although sometimes vortices can be pinned where the gap is large.@xcite some three - dimensional ( 3d ) samples can also be treated as inhomogeneous systems.@xcite for example for a 3d tip geometry , an asymmetric @xmath6 vortex state can be the ground state because the thick region prevents the vortex from penetrating it.@xcite in multi - layered superconductors , vortices enter first and reside favorably in the weak layers . then , vortices will penetrate into the strong layers only after weak layers become saturated and various vortex clusters and asymmetric vortex states are induced.@xcite also , the fabrication of anti - dots in superconductors results in a spatially varying superconducting energy gap with a barrier at the interfaces . in these systems , the combination of the giant vortex , multi - vortex and anti - vortex states can be found as ground state , which depends strongly on the detailed geometry of the antidots . @xcite for conventional superconductors , @xmath11 ( @xmath12 is the fermi wave vector and @xmath13 is the bcs coherence length ) , systems of size comparable to @xmath10 will not be large enough to host a vortex ( being much smaller than the coherence length ) . however , materials with small coherence lengths , e.g. high - t@xmath14 cuprate superconductors , will have @xmath15 and therefore in such systems it is possible to obtain vortex states in the quantum confinement regime . another such system is a graphene flake deposited on top of a superconductor . because of the proximity effect , cooper pairs will diffuse in graphene@xcite . in graphene the scattering length is large , therefore such a system is in the clean limit . more importantly , near the dirac point , the fermi wavelength is very large and can be easily manipulated by doping . in other words , @xmath16 can be tuned , which will allow for different vortex patterns to be realized in the graphene flake in the quantum confinement regime , but for more accessible sample sizes ( even above 100 nm ) . a similar configuration was also recently proposed by knopnin et . al in ref . yet another system where effects of quantum confinement on vortex matter can be probed systematically are the optically trapped cold gases @xcite , which are nowadays extremely controllable . for studying such nanoscale systems , microscopic bogoliubov - de gennes ( bdg ) theory is required . previous works used the bdg method to focus on isolated single vortex lines@xcite , giant vortices@xcite and to describe the local density of states modifications due to vortex - vortex and vortex - boundary interactions@xcite but are in the mesoscopic limit as opposed to the nanoscale limit considered here . although refs . [ ] and [ ] studied the groud state vortex states in a mesoscopic - nanoscopic crossover region by solving the bdg equations self - consistently , quantum confinement effects do not play any role . recently , we investigated@xcite the vortex states in nano - scale superconducting squares . we found unconventional vortex states in the quantum limit due to shape - induced resonances in the inhomogeneous cooper - pair condensate . vortex - antivortex structures , asymmetric vortex states and vortex clusters were found as ground states over a wide range of parameters . they are distinct from previous results obtained in mesoscopic superconductors using the gl theory . however , there are still several aspects that remained unclear . for example , how does the size of the sample affect the vortex states in nanoscale superconductors ? under which conditions , does one recover the conventional gl results ? why are the antivortex states more stable in the nanoscale limit while giant vortex states are unfavorable ? how do the vortex states change if temperature is increased ? in order to answer these questions , in this paper , we study vortex states in nanoscale superconducting squares systematically . vortex states for different sample sizes , @xmath0 parameters and temperatures @xmath17 are investigated and the stability of the symmetry - induced vortex / anti - vortex molecules is discussed . more unconventional states , very different from the ones obtained within gl theory , are found . this study is therefore fully complementary to what is known for vortex matter in superconductors . the paper is organized as follows . in sec . ii , we introduce the theoretical approach , i.e. the bdg approach for a square geometry . in sec . iii , we present the inhomogeneous superconducting state in the absence of the magnetic field in order to better understand the qse in nano - scale superconductors . in sec . iv , the ground states and metastable states are studied at zero temperature and on the sample size dependence of the vortex states is discussed . in sec . v , the finite temperature ground states are studied . in sec . vi , we discuss the generation of vortex / anti - vortex molecules and study the structure of the vortex core . finally , we summarize our findings in sec . in the presence of a magnetic field , the bogoliubov - de gennes ( bdg ) equations read @xmath18 where @xmath19(@xmath20 ) are electron-(hole-)like quasi - particle eigen - wave functions and @xmath21 are the quasi - particle eigen - energies . the single - electron hamiltonian reads @xmath22 with @xmath23 being the fermi energy and @xmath24 the vector potential ( we consider a gauge such that @xmath25 ) . furthermore , we take the electron band - mass to be isotropic ( i.e. , @xmath26 ) and a circular fermi surface . the pair potential is determined self - consistently from the eigen - wave functions and eigen - energies : @xmath27,\ ] ] where @xmath28 is the coupling constant , @xmath29 is the cutoff energy , and @xmath30^{-1}$ ] is the fermi distribution function at temperature @xmath17 . we consider now a two - dimensional superconducting square with the size @xmath1 . the confinement imposes dirichlet boundary conditions ( i.e. @xmath31 ) such that the order parameter vanishes at the surface . in an extreme type - ii superconductor ( and/or very thin sample ) , it is reasonable to neglect the contribution of the supercurrent to the total magnetic field . in this case , the free energy @xcite of the system is given by @xmath32\right ] \nonumber \\ & + & \int dr \left[-2\sum\limits_{n}e_n|v_n|^2 + 2\delta(r)\sum\limits_{n}u^{*}_n v_n[1 - 2f_n ] \nonumber \right . \\ & -&\left . g\sum\limits_{n}u^*_nv_n(1 - 2f_n ) \sum\limits_{n'}u_{n'}v^\ast_{n^\prime}(1 - 2f_{n^\prime } ) \right],\end{aligned}\ ] ] where the spatial dependence of @xmath19 and @xmath20 is implicit . in order to solve eqs . ( [ bdg ] ) and ( [ op ] ) numerically , we expanded the wave functions @xmath19 and @xmath20 as @xmath33 where the basis set @xmath34 is the basis eigen - states of the hamiltonian @xmath35 in the absence of the magnetic field . the corresponding eigen - energies of such states are @xmath36 . through comparison with results obtained by using the finite difference method , we found that the results are converged when we include the states with energies as large as @xmath37 times the cutoff energy @xmath29 , i.e. @xmath38 $ ] where @xmath39 is the flux quantum and @xmath2 the flux through our sample . .the parameters of the considered sample . coupling constant @xmath40 and @xmath23 is measured from the bottom of the quadratic band . [ cols="^,^,^,^",options="header " , ] + in our numerical investigations , we restrict ourselves to the three materials ( or samples ) with parameters given in table [ tab1 ] . for convenience , we measure the distance in units of the bulk coherence length at zero temperature @xmath13 and the energy in units of @xmath41 . here , @xmath42 where @xmath43 is the fermi velocity and @xmath44 is the bulk value of the order parameter at zero temperature . note that @xmath23 and @xmath44 are not independent when @xmath45 up to a constant . to find the different vortex configurations , which include all stable states , we search for the self - consistent solutions in the following two steps . ( 1 ) global scanning : starting from any reasonable vortex state ( usually , we start from the meissner state at @xmath46 ) , we slowly sweep up / down the applied flux with regular intervals @xmath47 and recalculate the superconducting states each time , until a new state is found . then , we repeat the sweeping process from the new state until no new vortex configurations appears . ( 2 ) special initial states : starting from usual states obtained in gl theory@xcite , we sweep up / down the applied flux . if a new state appears , we repeat the step ( 1 ) . in such a way , we are able to trace back and forth all found vortex states in the whole region of their stability and make sure that the usual gl states are always considered . at @xmath48 and @xmath49 . three insets show the initial , intermediate and final vortex configurations . the solid dots and open diamonds in the insets indicate @xmath6 vortex and @xmath50 giant antivortex , respectively . ] the minimum tolerance in the change of the order parameter between two steps in the self - consistent iteration is @xmath51 where @xmath52 and @xmath53 is the order parameter at the @xmath54 and @xmath55 iteration . we use the absolute tolerance since the relative tolerance can be abnormally high in the vortex core where @xmath56 . this is quite strict when we compared to @xmath57 but is necessary in order to ensure the precision in finding the true ground states in the bdg calculation . we show an example the evaluation of the vortex configuration for sample ii and @xmath58 at flux @xmath48 . based on the gl theory , the ground state of @xmath8 for such a system should be giant vortex state or multi vortex state ( for larger squares ) where two vortices sit on the diagonal of the sample . we start the calculation with such the multi vortex state as initial state . as seen from the fig . [ tolerance ] , the vortices merge into a vortex anti - vortex molecule and the result converges quickly . the error between each steps reach as low as @xmath59 . however , the error increases gradually when we continue the self - consistent procedure . after the second - order phase transition , the new state with two vortices sitting parallel to one of the sides has lower energy . finally , the symmetry of the state does not change and the error is always around @xmath60 which comes from non - physical factors , i.e. numerical accuracy . in the calculation , we found some situations where results do not converge and this usually comes from the change in the number of the quasi - states contained in the debye window . to avoid this , we set the smearing energy @xmath61 for the quasi - states . then , the self - consistent condition reads @xmath62 \times f_n(e_n - e_c),\ ] ] where @xmath63^{-1}$ ] is the fermi distribution function . the choice of @xmath61 is empirical . it should be enough small in order not to affect the results . meanwhile , it should be enough large to make the result converge through the iteration . our experience show that @xmath64 is suitable for our current work . with the corresponding diagonal profiles in the absence of applied magnetic field . panels ( a)-(c ) are for sample i with sizes @xmath65 , respectively . panels ( d ) and ( e ) are for sample ii with sizes @xmath66 and @xmath67 , respectively . panel ( f ) corresponds to sample iii with size @xmath66 . ] and the spatially averaged @xmath68 for sample i. ] first , we study the spatial distribution of the order parameter as a function of sample size in the absence of the magnetic field since the vortex configurations will be strongly affected by that distribution once the magnetic field is applied . in fig . [ s201 ] contour plots of the order parameter @xmath69 and the corresponding diagonal profiles are shown for sample i - iii with different sizes . as expected , all the order parameters are fourfold symmetric and show friedel - like oscillations in space which result in four well defined peaks at each corner . hence , the superconductivity is inhomogeneous . for example , figs . [ s201 ] ( a)-(c ) show the results for sample i with size @xmath70 , respectively . for @xmath58 , there are three oscillations in the order parameter along the diagonal and the resonant central peak prevents vortex from sitting here . however , the profile of the order parameter can be changed dramatically when the size of the sample changes . for @xmath71 [ shown in figs . [ s201](b ) ] , we found that the central peak disappears and a relatively flatter area generates in the center . when @xmath72 is increased to @xmath73 , the flat area enlarges and the friedel - like oscillations can be neglected at center when we compared it with oscillations near boundary . it indicates that the oscillations of the order parameter result from the quantum confinement effect ( or boundary effect ) . in order to study the quantum size effect on the order parameter , we show the amplitude of the order parameter in the center of the sample @xmath74 and the spatially averaged value over the whole sample @xmath68 for the sample i with sizes @xmath75 in fig . as seen , @xmath74 changes dramatically with @xmath72 increasing and converge to @xmath76 when @xmath77 . at the same time , the @xmath68 increases gradually with @xmath72 increasing . in principle , both of the parameters will converge as @xmath78 where the boundary effect can be totally neglected . since @xmath74 and @xmath68 show strong quantum size effect between @xmath58 and @xmath73 , we limit ourselves in following sections to study the samples for these particular size . the profile of @xmath79 is also strongly affected by @xmath0 . [ s201](d ) and ( e ) show results for sample ii with size @xmath58 and @xmath73 and ( f ) is for sample iii with size @xmath58 . comparing to the sample i with the same size , we found that the wave number and the amplitude of the oscillations along diagonal are larger . it indicates that the superconducting order parameter shows more inhomogeneous behavior with larger @xmath0 . in this section , we consider the zero temperature case for which the system is always in the quantum limit ( @xmath80 ) . first , we study the sample i with size @xmath81 and show in fig . [ fhall ] the free energy of the stable vortex states for flux @xmath82 $ ] . different curves ( colors ) represent states with different winding number @xmath7 and the states among them which reached the ground state are marked by dots . vertical lines and shadows show the flux range for each @xmath7 state as the ground state . the top dashed line stands for the free energy of the system in the normal state when the coupling constant @xmath28 is set to zero . when compared with the gl theory@xcite , one of the differences is that the free energy of the normal state depends on the magnetic field while it is a constant in gl theory . the reason is that the energy levels of the confined electrons are different for different magnetic fields . in our case , the change of the energy is relatively small when compared with the energy gap ( energy difference between the normal state and the superconducting state ) especially in weak fields . although the shown energy curves look conventional , there are significant differences with the gl theory.@xcite by sweeping the magnetic field up and down , we can get the full energy spectrum and the corresponding vortex states . for a certain magnetic field , it is common to have more than one converged solution . the lowest energy state is the ground state while the states with higher energy are referred to as metastable states . in fig . [ vcg ] , we show the contour plot of the absolute value of the order parameter of the corresponding ground states for various winding numbers . as can be seen from fig . [ fhall ] , the system favors states with winding numbers @xmath83 and @xmath84 because they have relatively large ground state flux range ( excluding the meissner state ) . from fig . [ vcg ] , we observe that these states have fourfold symmetry which is compatible with the sample geometry . one interesting feature of this system is the richness of metastable states . these states appear for all winding numbers @xmath7 except @xmath85 and @xmath86 . the number of metastable states reaches a peak for @xmath87 and equals @xmath88 . from the free energy curves , we notice that the energy difference between the ground state and the metastable states can sometimes be very small this makes the ground state difficult to find in simulations unless we sweep the field up and down many times . we also note that most metastable states are focused at lower magnetic fields from the corresponding ground state flux range . the reason is that vortices get easily stuck at the boundary due to the pronounced oscillations of the order parameter . for the same reason , their stability range is narrower due to asymmetry . the number of metastable states decreases for higher @xmath7 . in this case , the shorter distances between vortices cause strong interaction between them . this limits the choice of stable positions for vortices and therefore the number of metastable states are lower . due to this reason , metastable states are less favorable for smaller samples because of easy saturation with vortices . for example , no metastable states were found in sample i with size @xmath58 . . @xmath89 and @xmath90 are for flux @xmath91 and @xmath92 , respectively . darkest areas indicate the positions of vortices . note that at the center an anti - vortex ( for @xmath93 ) or a giant vortex [ for @xmath94 can spontaneously form . note also that for some vorticities two different ground states are found ( at different magnetic fields ) . ] next we discuss the ground states configurations for different vorticities in more detail . for the @xmath6 state , shown in fig . [ vcg ] , the vortex sits at the center of the sample which is compatible with the conventional picture . although such result is observed for different parameters of the sample , the state with diagonal location of the vortex can also be found in some cases.@xcite it is clear that the order parameter around the vortex core is suppressed and the profile shows the competition between @xmath5 symmetry and @xmath95 . it means that the vortex has long range ( longer than @xmath13 ) interaction with other objects such as the other vortices and/or boundaries . by increasing the flux to @xmath96 , the ground state shifts from @xmath6 to @xmath8 and two vortices sit along the diagonal . this again coincides with the result from gl theory , and results from the competition between the confinement imposed on vortices by the meissner currents and the vortex - vortex repulsion . in this case , these effects can be clearly seen from the profile of the order parameter , especially from the suppressed area around the vortices . the vortex - vortex interaction suppresses the order parameter mostly in the area between them . this can not be found in gl theory because the order parameter is always smooth and changes slowly in space . the state with @xmath9 shown in fig . [ vcg ] becomes the ground state when the fieldapplied flux is between @xmath97 and @xmath98 . it resembles the multi - vortex state obtained within gl theory where the three vortices are at the apices of a equilateral triangle . however , the perpendicular bisector of the triangle always coincides with one of the diagonals of the square sample in gl theory while it is parallel to one of the edges in our case . this is because the grid - like pattern in the inhomogeneity of the order parameter imposes preferential positions for the vortices inside the square . the state @xmath87 has a similar feature but the configuration is compatible with the gl result . from these states , we conclude that when the gl vortex configuration , which minimizes the vortex - vortex and vortex - boundary interaction , matches the oscillation pattern due to quantum confinement , then the state has a wider flux stability range . two ground states , @xmath99 and @xmath100 , are found for @xmath101 in the flux ranges @xmath102 and @xmath103 , respectively . both of them have a pentagonal vortex configuration . this is because the particular shape resonance at the considered field causes the order parameter to be peaked at the center . therefore , it costs energy for vortices to sit in the center of the sample . for the same reason , the ground states @xmath104 and @xmath105 do not have vortices in the center . moreover , when @xmath106 , vortices start to be compressed in the sample . if they do not form giant vortices , they will be very close to each other and form string - like structures [ see @xmath104 state ] . states with @xmath93 , @xmath107 , @xmath108 and @xmath109 shown in the panels of fig . [ vcg ] have a common feature , as all of them keep the fourfold symmetry . @xmath93 contains a antivortex at the center while @xmath108 and @xmath109 have a single vortex and a giant vortex with @xmath110 at the center , respectively . the state with @xmath93 is the only ground state which contains an antivortex . the antivortex is closely surrounded by four vortices and forms the core structure for @xmath93 . the outer shell is formed by the remaining four individual vortices sitting at four corners . states with @xmath107 and @xmath108 contain vortex dimers , i.e. two vortices close to each other at each corner . the fourfold symmetry makes both former states have a larger ground state flux range . @xmath109 also keeps the @xmath5 symmetry but the energy is sometimes even higher than the state @xmath89 , which has only @xmath111 symmetry . the reason is that the giant vortex costs extra energy . in order to visualize the changes in the ground states when key parameters change , we plot the phase diagram for samples i - iii for @xmath112 and @xmath113 in fig . [ phase ] . different shadings of blocks in fig . [ phase ] indicate different vortex types . the plain white background represents conventional multi - vortex states as found within the gl theory , while the blue background with square grid represents giant vortices , also compatible with the result obtained from gl theory . asymmetric vortex states attained only by bdg theory are represented by yellow background with horizontal grid pattern . states containing parallel vortex chains , represented by orange background with vertical grid pattern , and part of the vortex - antivortex molecules represented by pink(grey ) background , are new compared to gl theory . we can conclude from fig . [ phase ] that the quantum size effect is important not only for the transition field and the stability range of the different vortex states , but also for the vortex configurations . the reason is that the oscillation patterns of the order parameter are very sensitive to @xmath0 and may cause totally different behaviors even for two samples of identical size . as can be seen from fig . [ phase](a ) and ( b ) , the vortex phase transition fields vary greatly with sample size @xmath1 except for the phase boundary between @xmath85 and @xmath6 state . all the phase boundaries oscillate with @xmath1 . we find that some samples favor vortex states with even winding number @xmath7 while other disfavor them . for example , sample i with @xmath58 favors @xmath8 state whereas the one with @xmath114 disfavors @xmath87 to the point of non - existance . when @xmath1 becomes large , the transition fields start to converge and the flux stability range of each @xmath7 state will be roughly one flux quantum . when compared to samples i and ii , sample iii ( with large @xmath0 ) shows a more conventional picture . moreover , the phase transition field increases only slightly with increasing @xmath1 . it means that for large @xmath0 , the quantum size effect on the transition field , at least when winding number @xmath7 is small , can be neglected . nevertheless , a plethora of different vortex configurations is found . for example , an asymmetric @xmath6 state is found in all three samples . moreover , sample i always shows asymmetric states when @xmath115 . one other interesting phenomenon is that , for @xmath8 , sample ii can host all five types of vortex states with @xmath1 increasing . from the phase diagram in fig . [ phase ] , we notice that nanoscale superconductors favor anti - vortices and disfavor giant vortices . for example , the giant vortex state appears for @xmath8 in sample ii with @xmath114 . based on gl theory , only smaller samples will exhibit a giant vortex configuration . however , when @xmath116 , the two individual vortices form an asymmetric vortex state . on the other hand , we find that the probability of forming anti - vortices is much higher than in gl theory . in gl case , antivortex states usually appear for @xmath9 when four vortices are at the four corners and surround a centered antivortex . usually the distance between vortices and the antivortex is small ( less than @xmath4 ) . in the bdg calculation , at least two more antivortex states can be found . one is still the @xmath9 state and the antivortex is still at the center but four vortices are at the edges instead of the corners . this configuration was first presented by us in ref . [ ] where we showed that the size of the vortex / anti - vortex(v - av ) molecule is larger than the one obtained with gl theory . the other antivortex state appears for @xmath8 . in this case , the four vortices still sit at four corners but the centered antivortex carries two flux quanta , e.g. it is a giant antivortex ! due to the strong vortex - antivortex interaction , the size of such v - av molecule is small . in sample i with size @xmath117 ( top panel ) and the corresponding contour plots of the order parameter ( bottom panel ) . note that the contour plot of the ground state @xmath118 is shown in fig . the inset shows the energy difference between the ground state @xmath118 and the metastable states @xmath119 and @xmath120 at higher field.,title="fig : " ] in sample i with size @xmath117 ( top panel ) and the corresponding contour plots of the order parameter ( bottom panel ) . note that the contour plot of the ground state @xmath118 is shown in fig . the inset shows the energy difference between the ground state @xmath118 and the metastable states @xmath119 and @xmath120 at higher field.,title="fig : " ] next we will briefly discuss the metastable states of this system . to do this , we start from sample i with @xmath117 . metastable states are important in the bdg formalism because the energy difference between the ground and the metastable states can be very small . this suggests that these states could be easily found in experiments . alternatively , some metastable states can become ground states as the parameters are changed . all six found metastable states for @xmath8 and their free energy curves are shown in fig . the state @xmath119 is similar to the ground state @xmath118 , but rotated over @xmath121 , hence their free energies are very close to each other . actually , the difference in the orientation of the vortex pattern always results in a small difference in energy . state ( f ) is not obtained within the gl theory . in our case , due to the shape - resonant inhomogeneity of the order parameter , the rotation of the vortex pattern to the ground - state configuration is prevented by the spatial oscillations of the order parameter . the metastable state @xmath120 is only stable at higher field and its free energy is very close to the ground state @xmath118 . therefore we zoomed on the energy difference in the inset of fig . [ fhl2 ] . from the figure , one sees that the energy of the @xmath120 state is lower than the ground state , @xmath118 , when the applied flux is larger than @xmath122 . in fact , the state can exist even up to @xmath123 . from the vortex configuration shown in fig . [ fhl2 ] , we find that this state is a giant vortex . such a state has been predicted by the gl theory because the magnetic field pushes the two vortices towards each other and makes them merge into a giant vortex . usually , the phase transition between the multi vortex state and the giant vortex state is continuous ( second - order ) . however , the barrier induced by the inhomogeneity of the order parameter leads to a first order phase transition in our case . one more difference between the bdg giant vortex and the one in gl theory is its core structure . due to the shape resonances , the contour plot of the core shows a diagonal cross shape while the giant vortex core in the gl case is always circular . furthermore , the giant vortex state in our results has two allotropes : see state @xmath124 , compared to the state @xmath120 . the @xmath120 state exists up to higher field while @xmath124 only exists in lower field . hence , the size of the giant vortex seen in @xmath124 is larger than the one seen in @xmath120 . another difference between them is the orientation of the core . @xmath120 has diagonal cross shape while the state @xmath124 has edge cross shape . in sample i with size @xmath117 and the corresponding contour plots of the order parameter . the contour plot of the ground state @xmath125 is shown in fig . [ vcg].,title="fig : " ] in sample i with size @xmath117 and the corresponding contour plots of the order parameter . the contour plot of the ground state @xmath125 is shown in fig . [ vcg].,title="fig : " ] other three metastable states @xmath126 , @xmath127 and @xmath128 are observed only in lower field . they have in common the fact that at least one vortex is stuck at the boundary since the meissner current pushes the vortex outward at low fields . it is obvious that they have lower energy when the applied field is lower . the energy of state @xmath126 is always lower than the one of state @xmath127 because of the longer distance between the vortices . as is usual in mesoscopic superconductors , vortices in these states avoid to be located at the very corners of the sample , due to strong local superconductivity there . for the @xmath9 metastable states , the results are summarized in fig . the state @xmath129 is a v - av state and exists only in higher field while the giant vortex @xmath130 only exists in lower field . note again that the energy of the giant vortex states is much higher than the other metastable states with three single vortices . @xmath131 has lowest energy for @xmath9 around @xmath132 , when there is one vortex located at the boundary . states @xmath133 and @xmath134 follow when the field decreases and there are two and three vortices stuck at the boundaries , respectively . states @xmath135 and @xmath136 are disfavored and have higher energy due to the close distance between vortices . note again that no vortex sits at the corners in this states . in sample i with size @xmath117 and the contour plots of the order parameter of selected vortex states . inset shows details of the free energy curve around @xmath137 . the contour plot of ground state @xmath138 is shown in fig . [ vcg].,title="fig : " ] in sample i with size @xmath117 and the contour plots of the order parameter of selected vortex states . inset shows details of the free energy curve around @xmath137 . the contour plot of ground state @xmath138 is shown in fig . [ vcg].,title="fig : " ] states with @xmath87 show a wide ground state flux range and @xmath88 different metastable states , which is the largest variety of all @xmath7 vortex states . from fig . [ fhl4 ] , we find that the metastable states concentrate around the applied flux @xmath137 . for low fields , we conclude again that vortices are close to the surface and these states are always the lowest energy state for a given @xmath7 state at low fields . states @xmath139 and @xmath140 have @xmath5 symmetry and all of the four vortices are trapped close to the boundary . please note that in the state @xmath141 four vortices sit at the corners . this kind of state is rare in the bdg results because corners give the highest potential energy contribution for vortices . from the free energy curve of the state @xmath141 , we can find that the slope of the energy curve is opposite to the other @xmath87 states [ such as @xmath139 ] in this field range . this indicates that vortices are repelled by the meissner current in order to balance the inward force the vortices experience from the corner . when the field is too low , a vortex is expelled from the sample and the state jumps to a @xmath9 state . the vortex configuration @xmath143 has been found experimentally in conventional mesoscopic superconductors@xcite but was a result of the presence of pinning sites . this state can not be obtained in plane squares within the gl theory . another state to notice is @xmath144 whose energy curve does not cross any other @xmath87 curve . when the superconductor is in this state and the field is swept down , this state will be the first to jump to the @xmath9 state . this is understandable from the vortex configuration of @xmath144 because the vortex at the corner is easily expelled when field is lowered . at high fields only one metastable state exists @xmath145 . it can be seen as the state obtained after a @xmath146 degrees rotation of the ground state-@xmath138 . this is a consequence of the fact that the inhomogeneous pattern of the order parameter changes with field . at such a high field , the corner vortex position in state @xmath138 becomes unstable , which forces the vortices to sit at the edges , similar to the @xmath145 case . at the same time , the strong field pushes vortices closer together so that the distance between vortices in @xmath145 is shorter than the one in the ground state @xmath138 . so far , all our calculations were done at zero temperature , @xmath49 , where @xmath147 is the bulk critical temperature at zero flux @xmath46 . in what follows , we investigate the effect of temperature on the vortex configuration . first , we show all the vortex states for the flux range @xmath148 $ ] , for sample i with size @xmath117 at @xmath149 where the system is not in the quantum limit since @xmath150 . the corresponding free energy curve as a function of flux is presented in fig . [ hf_t2 ] . contrary to the results for @xmath49 , which were shown in fig . [ fhall ] , the figure looks more conventional ( similar to the results obtained by gl theory in ref . ) and there is only one stable state for each winding number @xmath7 . moreover , only giant vortex states are found in this case for @xmath151 . for the size of the square sample considered here , @xmath152 , gl theory@xcite predicts that multi - vortex states should exist . here we find instead that multi - vortex states are absent since @xmath4 increases as temperature increases . in order to see how temperature affects the coherence and the profile of the superconducting order parameter , we show the order parameter for @xmath153 and @xmath6 in fig . [ xid ] . the diagonal profile of @xmath154 at @xmath49 shows the strongest friedel - like oscillations . as temperature increases to @xmath155 , the profile is similar to the one obtained at zero temperature , but with less oscillations at the vortex core . both cases are in the quantum limit and the order parameter shows rapid variation in the core . when temperature reaches @xmath149 , we find that both the average and the oscillations of the absolute value of the order parameter are suppressed which indicates that the vortex states become more conventional . finally , the order parameter is smooth at @xmath156 and the gl results are approached . and @xmath149 . ] as seen from the fig . [ xid ] , the coherence length , which represents the vortex core radius , increases with increasing temperature . as defined by kramer and pesch@xcite , we calculate the coherence length @xmath157 as @xmath158 where @xmath159 is the distance to the vortex core . we plot in fig . [ xit ] @xmath160 as a function of temperature , @xmath161 . as discussed in ref . [ ] , @xmath157 can be described by @xmath162 when @xmath17 is close to @xmath147 ( @xmath163 in our case ) . in the intermediate temperature regime , there is a substantial suppression of the coherence length because of the bound states . at low temperature , the shrinkage of the coherence length stops and saturates when the system is in the quantum limit . note that @xmath157 at @xmath149 is around three times larger than the one at zero temperature . this explains why only giant vortex states can be found at such temperatures . [ deltat2 ] shows the order parameter for sample i at @xmath149 for @xmath8 in panel ( a ) and @xmath9 in panel ( b ) , respectively . both are giant vortex states and the @xmath5 symmetry grid pattern is strongly suppressed . as can be seen from the figure , the vortex cores show perfect circular symmetry , which is in agreement with the results from gl theory . of course , the size of the vortex core shown in panel ( b ) is larger than the one shown in panel ( a ) because its vorticity is larger . for @xmath153 and @xmath6 at different temperatures . the corresponding contour plots of the order parameter at @xmath49 and @xmath149 are also shown . ] for sample i with size @xmath117 and @xmath153 . ] at @xmath149 for ( a ) @xmath8 and ( b ) @xmath9 . ] finally , we end this section with the ( t-@xmath2 ) phase diagram for lower fields for sample i with @xmath58 . this is shown in fig . [ htphase ] . the thick black curve indicates the phase boundary between the superconducting and the normal state . when the system is in the quantum limit , for these parameters , only unconventional vortex states , such as asymmetric @xmath6 and @xmath9 states and edge - parallel @xmath8 states , are found as ground states . when temperature increases , the vortex states become conventional and the @xmath5 symmetry of the states is always preserved . note that the asymmetric @xmath6 state goes through a continuous phase transition to the symmetric @xmath6 state , which means the vortex moves gradually as the temperature changes . however , for higher winding numbers , the system usually goes through a first order phase transition . for example , the phase transition between the parallel vortex state and the giant vortex state of @xmath8 is of first order . this is different from the gl result , where vortices merge into a giant vortex through a continuous phase transition@xcite . for the @xmath9 state , we note that the ground state flux range for the four - fold symmetric v - av state is larger than the asymmetric one due to the compatibility of its symmetry with the geometry of the sample . concluding this section , higher temperature : 1 ) makes vortex states look more conventional ( closer to the gl results ) ; 2 ) smoothens the order parameter ; 3 ) suppresses the influence of the oscillation of the order parameter and 4 ) increases the superconducting coherence length @xmath4 . as a consequence , the number of metastable states is also lowered . the effect of temperature is very different ( more complex ) from the effect obtained by simply changing the effective size of the sample as is usually done within the gl theory . . the vortex configurations of areas ( i - iii ) are shown as insets in the upper right corner . ] in this section , we discuss the appearance and stability of anti - vortex states in the bdg theory in order to explain the existence of the giant antivortex . actually , such a state was already found in ref . @xcite through the linear gl method by introducing artificial pinning . from the phase diagram shown in fig . [ phase ] , we found that anti - vortex states are surprisingly stable within bdg theory . this is due to the fact that the grid pattern oscillation of the order parameter gives an additional contribution to the symmetry of the vortex states and therefore , in a square sample , the @xmath5 symmetry is enhanced . the other reason to form an anti - vortex is that the oscillations induced by the order parameter are seen in the vortex core where the order parameter is already suppressed . these oscillations can easily lead to a shift in the phase of the order parameter by @xmath164 and , thus , result in the formation of vortex - antivortex molecules . in order to explain this , we first discuss briefly the vortex profile in the gl theory . fig . [ scheme1 ] shows a schematic diagram of the vortex states for different winding number @xmath7 in gl theory . the diagonal profiles of the order parameters vary smoothly in space and the vortex emerges where the order parameter vanishes . note that the phase of the order parameter is adjusted such that along the diagonal the order parameter is real . panel ( a ) from fig . [ scheme1 ] shows the simplest case when only one vortex sits at the center . as can been seen from the figure , the order parameter changes sign , which indicates the @xmath164 phase shift of the order parameter . the profile is an odd function and @xmath165 near the vortex core . panels ( b ) and ( c ) from fig . [ scheme1 ] show the diagonal profiles for @xmath8 . both profiles are even functions due to the @xmath166 phase shift between the opposite corners . the order parameter exhibits @xmath167 property . when there is only one root , as can be seen from panel ( b ) , the vortex is a giant one . when there are two roots , as shown in panel ( c ) , the configurations are multi - vortex states . similarly , the profiles of the order parameter shown in panels ( d ) and ( e ) from fig . [ scheme1 ] for @xmath9 show a @xmath168 spatial dependence . one root means that we have a giant vortex state whereas three roots represent a vortex - antivortex configuration . note that , in order to generate the central anti - vortex , the order parameter has to oscillate around the center of the square . but now for the bdg theory . panels ( a ) and ( b ) show the symmetric and asymmetric @xmath6 vortex states , respectively . panels ( c ) and ( d ) are the giant vortex and vortex - giant antivortex @xmath8 states , respectively . panels ( e ) and ( f ) are the giant vortex and vortex - antivortex @xmath9 configurations , respectively . ] at @xmath169 with winding number @xmath170 . panels ( a ) and ( b ) show contour plots of the order parameter and its phase , respectively . panel ( c ) shows schematically the vortex configuration . dark(blue ) dots and open diamonds indicate vortices and antivortices , respectively . ] now let us move to nano - size superconductors where the bdg theory has to be used and the spatial oscillation of the order parameter can not be neglected . as can be seen from fig . [ scheme2 ] the oscillation plays an important role in generating vortices , especially when the value of the order parameter is comparable to the amplitude of the oscillation . for instance , panels ( a ) and ( b ) from fig . [ scheme2 ] show the symmetric and the asymmetric @xmath6 vortex states . the reason for the appearance of the asymmetric vortex state is the fact that the order parameter has an odd number of oscillations across the diagonal . thus , the vortex can not sit at the center . for the @xmath8 states , panel ( c ) shows a giant vortex state where the sign of the profile of the order parameter is always positive across the diagonal . however , due to the oscillations , the case shown in panel ( d ) of fig . [ scheme2 ] can easily exist and shows a giant anti - vortex ( @xmath50 ) at the center . further , the configurations show a large diversity for a fixed winding number @xmath7 . when @xmath8 , the configuration can be @xmath171 , @xmath172 , @xmath173 , @xmath174 and so on . panels ( e ) and ( f ) from fig . [ scheme2 ] are for @xmath9 states . apparently , they are similar to the gl case shown in fig . [ scheme1 ] , but the probability of the occurrence of the v - av state is much larger than in gl case . the reason is that the result with one root is just a special case while the general case shows oscillations at and around the vortex core . the v - av molecules do not only exist for smaller winding number @xmath7 , but they can also appear for large @xmath7 in the bdg results . [ scheme3 ] shows an example for sample iii with @xmath72=7 at @xmath169 and with a winding number @xmath170 . we find that vortices concentrate in the central dark ( blue ) area where the order parameter is strongly suppressed . from the phase of the order parameter , which is shown in fig . [ scheme3](b ) , the total winding number @xmath170 is found but it is difficult to distinguish each vortex . after a careful analysis , we plot schematic diagram of the vortex configuration in fig . [ scheme3](c ) . the dark(blue ) dots and open diamonds indicate vortices and anti - vortices , respectively . as can be seen , the lattices of vortices and antivortices are nested within each other . since anti - vortices attract vortices , all the vortices ( @xmath175 vortices and @xmath176 antivortices ) can be condensed in the cental area of the sample . this picture becomes more accurate when @xmath0 is large . the stronger the oscillations of the order parameter the more v - av pairs are generated . however , the size of the v - av pair can only be of the order of the fermi wavelength . thus , it will be very hard to detect them in experiments . this is why these states are mostly treated as a giant vortex in conventional superconductors . in other words , the suppressed central area of the order parameter , after coarse graining , will look like a giant vortex with @xmath170 . to summarize , we investigated the vortex states in a nanoscale superconducting square for different sizes @xmath1 , parameters @xmath0 , and temperatures @xmath17 . first , we found that the inhomogeneous pattern of the order parameter in the absence of magnetic field strongly depends on @xmath0 and the size @xmath1 . this oscillation pattern will give an additional contribution to competing effects that determine the vortex configurations when the field is applied . due to the inhomogeneous order parameter induced by the quantum topological confinement , samples with different @xmath0 and @xmath1 will favor different winding numbers @xmath7 . we find unconventional vortex states such as asymmetric , edge - parallel and vortex - antivortex states as the ground state of our nanoscale system . these were never seen in the ginzburg - landau approach . the inhomogeneous pattern of the order parameter , especially the strong oscillation at the boundaries causes additional potential wells for vortices which in turn generates a lot of metastable vortex states . furthermore , in the quantum limit , nano - size superconductors favor vortex - antivortex molecules while disfavoring giant vortex states . we observe that vortex ground states and the phase transition fields are very sensitive to changes in the parameter @xmath0 , size @xmath1 and temperature @xmath17 . this is a direct consequence of the quantum size effect . however , this effect is suppressed when the size @xmath1 is large or when temperature is high . in this case most metastable states become unstable and the ground states become compatible with gl theory . for high magnetic fields , vortex - antivortex pairs can be easily found when @xmath0 is large because the absolute value of the order parameter becomes smaller than the amplitude of its oscillations . however , detection of such states is beyond the current experimental abilities . the peculiar vortex states uncovered in the present work should be observable in superconducting systems where @xmath16 is small . such systems could be high-@xmath147 superconducting nano - grains for which the coherence length is small or cold - atom condensates with small @xmath12 , i.e. large fermi wavelength . of special interest could be hybrid systems made of superconducting substrates and graphene sheets for which the fermi wavelength is highly tunable near the dirac point . future work could also address the fundamental vortex - vortex and vortex - antivortex interactions for systems with a small @xmath16 , for which the oscillations of the order parameter on the order of @xmath10 become important . this work was supported by the flemish science foundation ( fwo - vlaanderen ) and methusalem funding of the flemish government . 99 a. k. geim , i. v. grigorieva , s. v. dubonos , j. g. s. lok , j. c. maan , a. e. filippov , and f. m. peeters , nature ( london ) * 390 * , 259 ( 1997 ) . v. a. schweigert , f. m. peeters , and p. s. deo , phys . rev . lett . * 81 * , 2783 ( 1998 ) . j. j. palacios , phys . lett . * 84 * , 1796 ( 2000 ) . b. j. baelus , f. m. peeters , and v. a. schweigert , phys . b * 63 * , 144517 ( 2001 ) . m. v. miloevi , s. v. yampolskii , and f. m. peeters , phys . rev . b * 66 * , 024515 ( 2002 ) . o. bourgeois , s.e . skipetrov , f. ong , and j. chaussy , phys . lett . * 94 * , 057007 ( 2005 ) . l. f. chibotaru , a. ceulemans , v. bruyndoncx , and v. v. moshchalkov , phys . * 86 * , 1323 ( 2001 ) . v.r . misko , v.m . fomin , j.t . devreese , and v.v . moshchalkov , phys . lett . * 90 * , 147003 ( 2003 ) . h. j. zhao , v. r. misko , f. m. peeters , v. oboznov , s. v. dubonos , and i. v. grigorieva , phys . rev . b * 78 * , 104517 ( 2008 ) . j. bona and v. v. kabanov , phys . rev . b * 65 * , 012509 ( 2001 ) . t. mertelj and v. v. kabanov , phys . b * 67 * , 134527 ( 2003 ) . b. j. baelus , and f. m. peeters , phys . b * 65 * , 104515 ( 2002 ) . a. s. melnikov , i. m. nefedov , d. a. ryzhov , i. a. shereshevskii , v. m. vinokur , and p. p. vysheslavtsev , phys . b * 65 * , 140503 ( 2002 ) . k. tanaka , i. robel , and b. jank , pnas * 99 * , 5233 ( 2002 ) . golubovi , m.v . miloevi , f.m . peeters , and v.v . moshchalkov , phys . b * 71 * , 180502 ( 2005 ) . m.r . connolly , m.v . miloevi , s.j . bending , j.r . clem , and t. tamegai , epl * 85 * , 17008 ( 2009 ) . ong , o. bourgeois , s.e . skipetrov , and j. chaussy , phys . b * 74 * , 140503 ( 2006 ) . pereira , v.v . moshchalkov , and l.f . chibotaru , phys . e * 86 * , 056709 ( 2012 ) . r. geurts , m.v . miloevi , and f.m . peeters , phys . b * 81 * , 214514 ( 2010 ) . a.s . melnikov and v.m . vinokur , nature ( london ) * 415 * , 60 ( 2002 ) . lin , t. nishio , l.n . bulaevskii , m.j . graf , and y. hasegawa , phys . b * 85 * , 134534 ( 2012 ) . t. nishio , t. an , a. nomura , k. miyachi , t. eguchi , h. sakata , s. lin , n. hayashi , n. nakai , m. machida , and y. hasegawa , phys . rev . lett . * 101 * , 167001 ( 2008 ) . t. cren , d. fokin , f. debontridder , v. dubost , and d. roditchev , phys . * 102 * , 127005 ( 2009 ) . j. tbik , v. cambel , and g. karapetrov , phys . b * 86 * , 134433 ( 2012 ) . p.j . pereira , l.f . chibotaru , and v.v . moshchalkov , phys . b * 84 * , 144504 ( 2011 ) . huy , m. kato , and t. ishida , supercond . * 26 * , 065001 ( 2013 ) . x .- hu , a .- c . qiu , w .- m . j. b , * 79 * , 473 ( 2011 ) . t. cren , l. serrier - garcia , f. debontridder , and d. roditchev , phys . * 107 * , 097202 ( 2011 ) . a. kanda , b. j. baelus , f. m. peeters , k. kadowaki , and y. ootuka , phys . * 93 * , 257002 ( 2004 ) . l. f. chibotaru , a. ceulemans , v. bruyndoncx , and v. v. moshchalkov , nature ( london ) * 408 * , 833 ( 2000 ) . r. geurts , m. v. miloevi , and f. m. peeters , phys . lett . * 97 * , 137002 ( 2006 ) . b * 75 * , 184511 ( 2007 ) . p. w. anderson , j. phys . solids * 11 * , 26 ( 1959 ) . a. a. shanenko , m. d. croitoru , and f. m. peeters , phys . b * 75 * , 014519 ( 2007 ) . m. d. croitoru , a. a. shanenko , and f. m. peeters , phys . b * 76 * , 024511 ( 2007 ) . m.d . croitoru , a. vagov , a.a . shanenko , and v.m . axt , supercond . technol . * 25 * , 124001 ( 2012 ) . a. a. shanenko , m. d. croitoru , and f. m. peeters , phys . b * 78 * , 024505 ( 2008 ) . s. bose , a. m. garcia - garcia , m. m. ugeda , j. d. urbina , c. h. michaelis , i. brihuega , and k. kern , nat . * 9 * , 550 ( 2010 ) . m. machida and t. koyama , phys . 90 * , 077003 ( 2003 ) . g. blatter , m. v. feigelman , v. b. geshkenbein , a. i. larkin , and v. m. vinokur , rev . phys . * 66 * , 1125 ( 1994 ) . d. valdez - balderas and d. stroud , phys . b * 76 * , 144506 ( 2007 ) . m. m. doria , a. r. de c. romaguera , and f. m. peeters , phys . 064505 ( 2007 ) . b. xu , m. v. miloevi , and f. m. peeters , phys . b * 77 * , 144509 ( 2008 ) . y. chen , m.m . doria , and f.m . peeters , phys . b * 77 * , 054511(2008 ) . b. xu , m. v. miloevi , and f. m. peeters , new j. phys . * 11 * , 013020 ( 2009 ) . b. xu , m. v. miloevi , and f. m. peeters , phys . b * 82 * , 214501 ( 2010 ) . c .- y . liu , g. r. berdiyorov , and m.v . miloevi , phys . b * 83 * , 104524 ( 2011 ) . m. a. engbarth , s. j. bending , and m. v. miloevi , phys . b * 83 * , 224504 ( 2011 ) . g. r. berdiyorov , b. j. baelus , m. v. miloevi , and f. m. peeters , phys . b * 68 * , 174521 ( 2003 ) . g. r. berdiyorov , m. v. miloevi , and f. m. peeters , phys . b * 74 * , 174512 ( 2006 ) . g. r. berdiyorov , m. v. miloevi , and f. m. peeters , new j. phys . * 11 * , 013025 ( 2009 ) . h. b. heersche , p. jarillo - herrero , j. b. oostinga , l. m. k. vandersypen , and a. f. morpurgo , nature ( london ) * 446 * , 56 ( 2007 ) . h. tomori , a. kanda , h. goto , s. takana , y. ootuka , and k. tsukagoshi , physica c : superconductivity * 470 * , 1492 ( 2010 ) . a. m. black - schaffer and s. doniach , phys . rev . b * 78 * , 024504 ( 2008 ) . l. covaci and f.m . peeters , phys . b * 84 * , 241401 ( 2011 ) . n. b. kopnin , i. m. khaymovich , and a. s. melnikov phys . * 110 * , 027003 ( 2013 ) . s. aubin , s. myrskog , m. h. t. extavour , l. j. leblanc , d. mckay , a. stummer , and j. h. thywissen , nature physics * 2 * , 384 ( 2006 ) . zhang , l. covaci , m.v . miloevi , g.r . berdiyorov , and f.m . peeters , phys . lett . * 109 * , 107001 ( 2012 ) . i. kosztin , s. kos , m. stone , and a. j. leggett , phys . b * 58 * , 9365 ( 1998 ) . a. spuntarelli , p. pieri , and g. c. strinati , physics reports * 488 * , 111 ( 2010 ) . c. caroli , p. g. de gennes , and j. matricon , phys . * 9 * , 307 ( 1964 ) . f. gygi and m. schlter , phys . b * 43 * , 7609 ( 1991 ) . n. hayashi , t. isoshima , m. ichioka , and k. machida , phys . * 80 * , 2921 ( 1998 ) . s. m. m. virtanen and m. m. salomaa , phys . b * 60 * , 14581 ( 1999 ) . c. berthod , phys . b * 71 * , 134513 ( 2005 ) . melnikov and m.a . silaev , jetp lett.*83 * , 578 ( 2006 ) . a. s. melnikov and v. m. vinokur , phys . b * 65 * , 224514 ( 2002 ) . a. s. melnikov , d. a. ryzhov , and m. a. silaev , phys . b * 78 * , 064513 ( 2008 ) . a. s. melnikov , d.a . ryzhov , and m. a. silaev , phys . b * 79 * , 134521 ( 2009 ) . h. suematsu , t. ishida , t. koyama , m. machida , and m. kato , j. phys . jpn . * 79 * , 124704 ( 2010 ) . ben xu , m. v. miloevi , shi - hsin lin , f. m. peeters , and b. janko , phys . 107 * , 057002 ( 2011 ) . l. kramer and w. pesch , z. phys . * 269 * , 59 ( 1974 ) .
we present accurate monte - carlo calculations of ly radiation pressure in a range of models which represent galaxies during various epochs of our universe . we show that the radiation force that ly photons exert on hydrogen gas in the neutral intergalactic medium ( igm ) , that surrounds minihalos that host the first stars , may exceed gravity by orders of magnitude and drive supersonic winds . ly radiation pressure may also dominate over gravity in the neutral igm that surrounds the hii regions produced by the first galaxies . however , the radiation force is likely too weak to result in supersonic outflows in this case . furthermore , we show that ly radiation pressure may drive outflows in the interstellar medium of star forming galaxies that reach hundreds of km s . this mechanism could also operate at lower redshifts , and may have already been indirectly detected in the spectral line shape of observed ly emission lines . [ firstpage ] cosmology : theory galaxies : high redshift radiation mechanisms : general radiative transfer ism : bubbles
0807.2645
hii regions around massive stars convert a significant fraction of the total bolometric luminosity of young galaxies into ly@xmath0 line emission @xcite . this ly@xmath0 radiation can exert a large force on surrounding neutral gas , as the ly@xmath0 transition has a cross - section that is @xmath3 orders of magnitude larger than the thomson cross - section , when averaged over a frequency band as wide as the resonance frequency itself ( e.g. * ? ? ? not surprisingly , the impact of ly@xmath0 radiation pressure on the formation of galaxies has been discussed extensively ( e.g * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but the intricacies of ly@xmath0 radiative transfer in 3d complicated an accurate numerical treatment of its dynamical effect on the gas . nevertheless , an approximate estimate can be obtained from simple energy considerations as shown below . consider a self - gravitating gas cloud of total ( baryons + dark matter ) mass @xmath4 and radius @xmath5 that contains a central ly@xmath0 source . the gravitational binding energy of the baryons inside the cloud , @xmath6 , can be compared to the total energy in the ly@xmath0 radiation field inside the cloud , @xmath7 . here , @xmath8 is the ly@xmath0 luminosity of the central source ( in erg s@xmath1 ) , and @xmath9 is the typical trapping time of ly@xmath0 photons in the cloud owing to scattering on hydrogen atoms . the ly@xmath0 radiation pressure would unbind the baryonic gas from the cloud if @xmath10 , i.e. @xmath11 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this approach , @xmath9 is one of the key parameters in setting the ly@xmath0 radiation pressure . calculations by @xcite imply that @xmath12 for @xmath13 , and @xmath14 otherwise , for a static , uniform , infinite slab of material ( also see fig 1 of * ? ? ? here @xmath15 is the line center optical depth from the center to the edge of the slab , and @xmath16 is the light crossing time if the medium were transparent ( i.e. @xmath17 in the case of the cloud described above ) . note however , that the precise value of @xmath9 depends on other factors including for example , the gas distribution ( clumpiness and geometry ) , the velocity distribution of the gas , and the dust content of the cloud @xcite . the ly@xmath0 radiation pressure becomes comparable to gravity when @xmath18 where @xmath19 erg s@xmath1 , and where we have substituted the virial radius of a galaxy mass @xmath4 , @xmath20 kpc @xmath21@xmath22 , for @xmath5 ( eq . 24 of barkana & loeb 2001 ) . for comparison , a star forming galaxy can generate a ly@xmath0 luminosity of @xmath23 erg s@xmath1 , where the precise conversion factor depends on the gas metallicity and the stellar initial mass function ( e.g. * ? ? ? therefore , a star formation rate of merely sfr@xmath24@xmath25 yr@xmath1 is needed to generate a ly@xmath0 luminosity that is capable of unbinding gas from a halo of mass @xmath26 . halos of @xmath27 are very common at @xmath28 , and have a sufficiently large reservoir of baryons to sustain the above - mentioned star formation rates for a prolonged time . in this paper we provide a more detailed investigation of the magnitude of ly@xmath0 radiation pressure in the environment of high - redshift star forming galaxies . in particular , we use a ly@xmath0 monte - carlo radiative transfer code @xcite to compute ly@xmath0 radiation pressure in a wider range of models . our treatment of radiative transfer and our focus on the environment of high - redshift star forming galaxies , distinguish this paper from previous work . we will show that the radiation force exerted by ly@xmath0 photons on neutral hydrogen gas can exceed the gravitational force that binds the gas to its host galaxy by orders of magnitude , and may drive supersonic outflows of neutral gas both in the intergalactic and the interstellar medium . the outline of this paper is as follows . in [ sec : code ] we describe how ly@xmath0 radiation pressure is computed in the monte - carlo radiative transfer code , and show the tests that are performed to test the accuracy of the code . in [ sec : result ] , we present our numerical results . finally , [ sec : conc ] summarizes the implications of our work and our main conclusions . the cosmological parameter values used throughout our discussion are @xmath29 @xcite . the force @xmath30 experienced by an atom in a direction @xmath31 is related to the flux through a plane normal to @xmath31 , @xmath32 where @xmath33 is the ly@xmath0 absorption cross - section at frequency @xmath34 . the specific flux is given by @xmath35 , where @xmath36 is the specific intensity of the radiation field ( see , e.g. eq . 1.113 in * ? ? ? * ) , and @xmath37 in which @xmath38 denotes the propagation direction of the radiation ( i.e. @xmath39 for radiation propagating perpendicular to the plane ) . the specific intensity obeys the radiative transfer equation , which reads ( in spherical coordinates ) @xmath40 where in this equation @xmath41 , @xmath42 denotes the mean intensity , and @xmath43 the emission function for newly created photons at frequency @xmath34 and radius @xmath44 ( in photons @xmath45 s@xmath1 sr@xmath1 hz@xmath1 , see e.g. loeb & rybicki 1999 ) . furthermore , @xmath46 denotes the opacity at frequency @xmath34 ( e.g. * ? ? ? * ) , where @xmath47 is planck s constant , @xmath48 hz is the ly@xmath0 frequency , @xmath49 is the number density of hydrogen atoms in their electronic ground ( first excited ) state , @xmath50 is the einstein - b coefficient of the @xmath51 transition , @xmath52 is the line profile function ( e.g. rybicki & lightman 1979 , their eq . 1.79 ) , and @xmath53 . here , @xmath54 is the thermal velocity of the hydrogen atoms in the gas , given by @xmath55 , where @xmath56 is the boltzmann constant , @xmath57 the gas temperature , and @xmath58 the proton mass . under the assumption that @xmath36 has only a weak dependence on direction ( which is reasonable given that ly@xmath0 radiation scatters very frequently ) , @xmath36 can be expressed as a first - order taylor expansion in @xmath59 , i.e. @xmath60 . in this so - called `` eddington approximation '' , the expression for flux simplifies to ( e.g. * ? ? ? * their eq . 1.118 ) @xmath61 where we have used the relation , @xmath62 , in which @xmath63 is the specific energy density in the radiation field at a frequency @xmath34 . furthermore , we have decoupled the gas absorption and emission functions from the ly@xmath0 radiation field , and assumed that all neutral hydrogen atoms are in their electronic ground state ( this assumption is justified in more detail in appendix [ app : n2 ] ) , i.e. @xmath64 and @xmath65 . under this assumption , @xmath66 with @xmath67 being the number density of neutral hydrogen atoms . substituting this expression back into eq . ( [ eq : flux1 ] ) yields @xmath68 where we defined @xmath69 . note that the cross - section does not appear in the final expression for the radiation force ) is analogous to the usual pressure gradient force in fluid - dynamics which is not dependent on the scattering cross - section of the fluid particles . ] . in our monte - carlo simulation we sample the gas density and velocity fields with @xmath70 concentric spherical shells . the radius , thickness , and volume of shell @xmath71 are denoted by @xmath72 , @xmath73 , and @xmath74 , respectively . we compute the radiation force using two approaches : * in the first approach , we calculate the energy density ( @xmath75 in eq . [ eq : force ] ) in the ly@xmath0 radiation field as a function of radius : using the monte - carlo simulation we compute the average time that photons spend in shell @xmath71 , which we denote by @xmath76 . the total number of photons that is present in shell @xmath71 at any given time is then given by @xmath77 , where @xmath78 is the rate at which photons are emitted . this yields the energy density , @xmath79 . finally , eq . ( [ eq : force ] ) is used to compute the radiation force on atoms in shell @xmath71 . note that estimators of the energy density in -and the momentum transfer by- a radiation field in a more general context is discussed by e.g. @xcite and @xcite . * in the second approach , we calculate the momentum transfer from a ly@xmath0 photon to an atom in each scattering event , @xmath80 . here , @xmath81 and @xmath82 are the photons wavevectors before and after scattering . we compute the average _ total momentum transfer _ ( i.e. summed over all scattering events ) per photon in shell @xmath71 , @xmath83 , and obtain the total momentum transfer from @xmath84 . the force on an individual atom is obtained by dividing by the total number of hydrogen atoms in shell @xmath71 , i.e @xmath85 . both methods should give identical results , provided that the eddington approximation holds . we begin by considering a ly@xmath0 point source at a redshift @xmath86 embedded in a neutral intergalactic medium ( igm ) that is expanding with the hubble flow . the photons scatter and diffuse away from the source while hubble expansion redshifts the photons away from resonance . in this case , the angle - averaged intensity @xmath87 and its radial dependence can be calculated analytically @xcite . the availability of analytic expressions for @xmath88 , and therefore the radiation force @xmath30 ( through eq . [ eq : force ] ) , makes this a good test case for our code . in figure [ fig : uden ] we plot the radial dependence of the energy density ( in erg @xmath45 ) in the ly@xmath0 radiation field for a model in which the central source is emitting @xmath89 photons s@xmath1 ( where we have introduced the dimensionless quantity @xmath90 photons s@xmath1 ) . this corresponds to a luminosity of @xmath91 erg s@xmath1 , which represents a bright ly@xmath0 emitting galaxy ( e.g. * ? ? ? the _ blue dotted line _ shows the energy density if the igm were fully transparent to ly@xmath0 radiation . in this hypothetical case all photons stream radially outward , and the energy density is given by @xmath92 . the _ red dashed line _ shows the energy density , @xmath93 , derived from the analytic expression for @xmath94 given in loeb & rybicki ( 1999 , their eq . 21 ) , while the _ black histogram _ shows the energy density extracted from the simulation ( [ sec : radpresmc ] ) . clearly , the analytic and monte - carlo calculations yield consistent results . scattering reduces the effective speed at which photons propagate radially outward , which enhances their energy density ( especially at small radii ) relative to the transparent case . at sufficiently large distances however , the photons have redshifted far enough from resonance that they are propagating almost freely to the observer , and the energy density approaches @xmath95 . we note that at a sufficiently high value of @xmath96 , the fraction of hydrogen atoms that populate the 2p ( and 2s ) levels is non - negligible and our assumption that ( almost ) all of the atoms populate their electronic ground state becomes invalid ( so that the solution for @xmath97 in figure [ fig : uden ] breaks down ) . however , as we show in appendix [ app : n2 ] , this only occurs when @xmath98 , well beyond the regime considered in this paper . in figure [ fig : force1 ] we compare the radiation force to the gravitational force on a single hydrogen atom , @xmath99 , where @xmath100 is the total mass enclosed within a radius @xmath44 ) . we plot the ratio @xmath101 scaled by @xmath102 . at @xmath86 is @xmath103 comoving mpc@xmath104 , implying that these rare halos are among the most massive ones in existence at that early cosmic time . ] the _ black dotted line ( grey solid histogram ) _ was calculated by applying eq . ( [ eq : force ] ) to the energy density @xmath97 that was obtained by using the analytic ( monte - carlo ) approach ( also see fig [ fig : uden ] ) . for comparison , the _ black solid histogram _ was obtained by directly computing the momentum transfer rate from photons to atoms as outlined in [ sec : radpresmc ] . figure [ fig : force1 ] shows that the radiation force overwhelms gravity at small radii . the energy density scales approximately as @xmath105 ( fig [ fig : uden ] ) . therefore , @xmath106 and reaches unity at @xmath107 physical kpc . the radiation force increases linearly with @xmath78 while the gravitational force scales as @xmath4 . thus , @xmath101 scales linearly with the ratio @xmath108 . to scale out the dependence on @xmath109 , the vertical axis shows the quantity @xmath110 , where @xmath111 . for example , if @xmath112 then figure [ fig : force1 ] shows that radiation pressure exceeds gravity out to @xmath113 kpc , well beyond the virial radius of a halo of this mass at @xmath114 kpc . most importantly , figure [ fig : force1 ] shows that the two approaches used to compute the radiation force in the simulation yield consistent results , with a noticeable deviation only at the largest radii ( @xmath115 mpc ) . at large radii most photons stream outwards radially and the eddington approximation that was used to derive eq . ( [ eq : flux2 ] ) becomes increasingly unreliable . next , we use the radiative transfer code to explore the magnitude of the ly@xmath0 radiation pressure for a range of models which represent an evolutionary sequence of structure formation in the universe . we focus on the ly@xmath0 radiation pressure on gas surrounding ( _ i _ ) the first stars ( [ sec : firststar ] ) ; ( _ ii _ ) the first galaxies ( [ sec : firstgalaxy ] ) ; and ( _ iii _ ) the interstellar medium of galaxies ( [ sec : wind ] ) . numerical simulations of structure formation suggest the first stars that formed in our universe were massive ( @xmath116 ) , and formed as single objects in dark matter halos with masses of @xmath117 that collapsed at @xmath118 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . here , we focus our attention on a star with a mass @xmath119 that formed at @xmath120 in a dark matter mini - halo of mass @xmath121 . the star emits @xmath122 ionizing photons per second @xcite . we assume that the ionizing flux ionizes all the gas out to the virial radius of the dark matter halo ( @xmath123 kpc ) but not beyond that radius @xcite . hence , the igm gas surrounding this central source is assumed to be neutral ( @xmath124 ) and cold ( @xmath125k , which corresponds to the temperature of the neutral igm at @xmath120 due to x - ray heating , see e.g. fig 1 of pritchard & loeb 2008 ) . recombination following photoionization converts @xmath126 of all ionizing photons into ly@xmath0 photons ( * ? ? ? * p 387 ) . hence , the entire halo is a ly@xmath0 source that is surrounded by neutral intergalactic gas . to determine the radial dependence of the ly@xmath0 production rate ( @xmath127 in eq . [ eq : rt ] ) , we need to specify the gas density profile . we assume that the gas distribution inside the dark matter halo is described by an nfw - profile with a concentration parameter @xmath128 and a thermal core s@xmath1 . the total recombination rate can be increased to balance the photoionization rate by introducing a clumping factor @xmath129 . ] at @xmath130 ( see * ? ? ? we point out however , that our final results are not sensitive to our choice of @xmath43 . once @xmath43 has been determined , we find the radius , @xmath44 , at which a ly@xmath0 photon is generated in the monte - carlo simulation from the relation @xmath131 where @xmath5 is a random number between @xmath132 and @xmath133 , @xmath134 is the total recombination rate inside the dark matter halo , and @xmath135 @xmath136 s@xmath1 is the case - b recombination coefficient at a temperature @xmath137 k ( e.g. * ? ? ? * ) . once the photon is generated , it scatters through the neutral igm until it has redshifted far enough from resonance that it can escape to the observer . in the _ left panel _ of figure [ fig : casei ] we show the energy density ( in erg @xmath45 ) of the ly@xmath0 radiation field as a function of radius . the _ red solid _ line represents a model in which we assumed the igm to follow the mean density and hubble expansion right outside the virial radius . the _ blue dotted line _ shows a more realistic model in which the igm is still overdense near the virial radius , and in which the intergalactic gas is gravitationally pulled towards the minihalo ( see dijkstra et al 2007 for a quantitative description of the density and velocity profiles based on the model of barkana 2004 ) . the _ black dashed line _ shows the same model as the _ red solid line _ but with the neutral fraction increasing linearly between @xmath138 and @xmath139 . this provides a better representation of the fact that the central population iii star emits ionizing photons with energies @xmath140 ev , which can photoionize hydrogen ( and helium ) atoms that lie deeper in the igm . the goal of this model is to investigate whether our results depend sensitively on the presence of a sharp boundary between hi and hii . all models show that the radiation energy density within the fully ionized minihalo ( @xmath141 kpc ) has only a weak dependence on radius , i.e. @xmath142 . naively , this may appear surprising given the fact that within the model , no scattering occurs within the virial radius and one may expect the energy density in the ly@xmath0 radiation field to scale as @xmath143 . however , in reality @xmath75 obtains only a weak radial dependence because the radiation can be scattered back into the ionized minihalo as soon as it hits the wall of neutral igm gas . ly@xmath0 photons are therefore trapped inside the ionized minihalo and their energy density is boosted to a value that is only weakly dependent on radius . on the other hand , for @xmath144 we find that @xmath145 , which is because ly@xmath0 photons are trapped more efficiently near the edge of the hii region , while they stream freely outwards at larger radii ( as in [ sec : test ] and fig [ fig : uden ] ) . figure [ fig : casei ] shows clearly that the radial dependence of the ly@xmath0 energy density is not sensitive to the detailed model assumptions about the gas in the igm . in the _ right panel _ of figure [ fig : casei ] we show the ratio between the radiation force ( eq . [ eq : force ] ) and the gravitational force on a single hydrogen atom ) , @xmath146 , where @xmath100 is the total ( baryons + dark matter ) mass enclosed within a radius @xmath44 . in all models , radiation pressure dominates over gravity by as much as @xmath147 orders of magnitude . the radiation force is largest for the models in which the igm is assumed to be at mean density , because of the @xmath148 factor in the equation for the radiation force ( eq . [ eq : force ] ) . note that the spike near @xmath149 kpc for the other two models is due to an artificial discontinuity in the igm velocity field that exists in this model . ly@xmath0 radiation pressure may operate throughout the lifetime of the central star . over a lifetime of @xmath150 myr ( see table 4 of * ? ? ? * ) , this mechanism is capable of accelerating the gas to velocities of @xmath151 ( @xmath152 ) km s@xmath1 at @xmath153 kpc in the model represented by the _ blue dotted _ ( _ red solid _ ) _ line _ , and to @xmath154 ( @xmath155 ) km s@xmath1 at @xmath156 kpc ( the reason for this large difference is that the edge of the hii region lies at @xmath157 kpc . hence , gas at @xmath153 kpc is separated by 0.04 kpc from this edge , while gas at @xmath156 kpc is separated by a distance that is 3.5 times larger ) . thus , ly@xmath0 radiation pressure can accelerate the gas to velocities that exceed the escape velocity from the dark matter halo ( @xmath158 km s@xmath1 ) as well as the sound speed of the intergalactic medium ( @xmath159 km s@xmath1 ) . note the as the gas is pushed out and its velocity profile changes , the subsequent radiative transfer is altered . for example , we repeated the radiative transfer calculation for models in which gas at @xmath160 was accelerated to velocities in the range @xmath161 km s@xmath1 ( outward ) and found a slightly shallower profile for @xmath97 which lowered the radiation force by a factor of @xmath162 . consequently , the acceleration of the gas decreases with time and the actual velocities reached by the gas are lower than the estimates given above by a factor of a few . nevertheless , the resulting velocities are still substantial . our calculations imply that ly@xmath0 radiation pressure can affect the gas dynamics in the igm surrounding minihalos that contain the first stars . the impact of ly@xmath0 radiation pressure increases with decreasing density of the surrounding gas in the igm . in practice , the distribution of the igm is not spherically symmetric . instead , the density is expected to vary from sightline to sightline ( being large along filaments and small along voids ) . our results imply that ly@xmath0 radiation pressure will be most efficient in blowing out the lower density gas . this conjecture is supported by the tendency of ly@xmath0 photons to preferentially scatter through the low - density gas ; their propagation along the path of least resistance would naturally boost up the ly@xmath0 flux there . this effect will be moderated by the tendency of the hii region around the first stars to extend further into the low density gas ( in butterfly-like patterns , e.g. abel et al , 1999 ) . if the central star dies in a supernova explosion , then the resulting violent outflow could blow most of the baryons out from the minihalo . however , stars with masses in the range @xmath163 and @xmath164 , are not expected to end their lives in a supernova . instead , these stars collapse directly to a black hole @xcite and have weak winds ( because of the lack of heavy elements in their atmosphere ) , so that radiation pressure may be the dominant process that affects their surrounding igm . in summary , ly@xmath0 radiation pressure on the neutral igm around minihalos in which the first stars form , can exceed gravity by orders of magnitude and launch supersonic winds . our limited analysis does not allow a detailed discussion on the consequences of these winds . this requires 3d simulations with cosmological initial conditions that capture the full igm density field around the minihalo and that track the evolution of the shocks that may form in the igm . such simulation are numerically challenging as they require self - consistent treatment of gas dynamics and ly@xmath0 radiative transfer in a moving inhomogeneous medium . our second case concerns a young galaxy that is forming multiple stars in a dark matter halo of mass @xmath165 at @xmath86 . we assume that the galaxy is converting a fraction @xmath166 of its baryons into stars over @xmath167 @xcite , where @xmath168\sim0.49 $ ] gyr , is the age of the universe at @xmath86 . this translates to a star formation rate of @xmath169 yr@xmath1 . for population iii stars forming out of pristine gas , the total emission rate of ionizing photons is @xmath170 s@xmath1 @xcite s@xmath1 @xmath171 in the no - mass - loss model of schaerer ( 2002 ) in which metal - free stars form according to a salpeter imf with @xmath172 and @xmath173 ( his model b ) . ] . if @xmath174 of the ionizing photons escape from the galaxy @xcite , then this translates to a ly@xmath0 luminosity of @xmath175 erg s@xmath1 . furthermore , this galaxy can photoionize a spherical hii region of a radius @xmath176 physical kpc . note however , that other ionizing sources would likely exist within this hii region . indeed , clusters of sources are thought to determine the growth of ionized bubbles during reionization . this results in a characteristic hii region size that is significantly larger than that produced by single source , especially during the later stages of reionization ( e.g. * ? ? ? * ; * ? ? ? * ) . in this framework , our model represents a star forming galaxy during the early stages of reionization or alternatively a galaxy that lies @xmath152 kpc away from the edge of a larger ionized bubble . in this particular case , the majority of all recombination events occur in the central galaxy . thus , we initiate all ly@xmath0 photons at @xmath177 in the monte - carlo simulation . we assume that the gas is completely ionized out to @xmath178 kpc , beyond which it is neutral . as shown in [ sec : firststar ] , this abrupt transition in the ionized fraction of h in the gas does not affect our results . the _ left panel _ of figure [ fig : caseii ] shows the energy density ( in erg @xmath45 ) of the ly@xmath0 radiation field as a function of radius . the _ solid line _ represents the model discussed above . a kink in the energy density is seen at the edge of the hii region ( see [ sec : firststar ] for a more detailed discussion of the profile ) . the _ dotted line _ represents a variant of the model in which we have reduced the size of the hii region to @xmath179 kpc . the _ right panel _ of figure [ fig : caseii ] shows the ratio between the radiation and the gravitational forces on a single hydrogen atom . in our fiducial model , the radiation force does not exceed gravity ; rather , at the edge of the hii region , gravity is @xmath162 times stronger . the radiation force becomes equal to the gravitational force if @xmath180 kpc . this requires an extremely low [ by a factor @xmath181 escape fraction of ionizing photons , @xmath182 . alternatively , radiation pressure _ is _ important when the halo mass of the star forming region is reduced to @xmath183 . halos of this mass are the the most abundant halos at @xmath184 that are capable of cooling via excitation of atomic hydrogen ( i.e. their virial temperature just exceeds @xmath185 k , e.g. barkana & loeb 2001 ) . the total gas reservoir inside these halos is @xmath186 , and so these halos can sustain a star formation rate of @xmath187 yr@xmath1 for up to @xmath188 myr . however , even if the radiation force is allowed to operate for @xmath188 myr , we find that radiation pressure can not accelerate the gas in the igm to velocities that exceed @xmath189 km s@xmath1 . we therefore conclude that although ly@xmath0 pressure may exceed gravity in the neutral igm that surrounds hii regions around @xmath190 halos , the absolute magnitude of the radiation force is too weak to drive the igm to supersonic velocities . in principle , ly@xmath0 radiation pressure can be important when neutral gas exists in close proximity to a luminous ly@xmath0 source . so far , we focused our attention on hi gas in the igm . however , neutral gas in the interstellar medium ( ism ) of the host galaxy is located closer to the ly@xmath0 sources and should be exposed to an even stronger ly@xmath0 radiation pressure . indeed , it has been demonstrated ( e.g. * ? ? ? * ; * ? ? ? * ) that scattering of ly@xmath0 photons by neutral hydrogen atoms in a thin ( with a thickness much smaller than its radius ) , outflowing supershell of hi gas surrounding the star forming regions can naturally explain two observed phenomena : ( _ i _ ) the common shift of the ly@xmath0 emission line towards the red relative to metal absorption lines and the host galaxy s systemic redshift determined from other nebular recombination lines ( e.g. * ? ? ? * ; * ? ? ? * ) ; and ( _ ii _ ) the asymmetry of the ly@xmath0 line with emission extending well into its red wing ( e.g. * ? ? * ; * ? ? ? the existence of thin , outflowing shells of neutral atomic hydrogen around hii regions is confirmed by hi - observations of our own milky - way @xcite and other nearby galaxies ( e.g. * ? ? ? the largest of these shells , so - called supershells , have radii of @xmath191 kpc ( e.g * ? ? ? * ; * ? ? ? * ) and hi column densities in the range @xmath192@xmath193 @xmath194 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? supershells are thought to be generated by stellar winds or supernovae explosions which sweep - up gas into a thin expanding neutral shell ( see e.g. * ? ? ? * for a review ) . the back - scattering mechanism attributes both the redshift and asymmetry of the ly@xmath0 line to the doppler boost that ly@xmath0 photons undergo as they scatter off the outflow on the far side of the galaxy back towards the observer ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? it is interesting to investigate whether ly@xmath0 radiation pressure may provide an alternative mechanism that determines the supershell kinematics . in figure [ fig : caseiii ] we show the energy density ( _ left panel _ ) and the ly@xmath0 radiation force ( _ right panel _ ) for two models . both models assume that : ( _ i _ ) there is a ly@xmath0 source at @xmath177 with a luminosity of @xmath195 erg s@xmath1 ; ( _ ii _ ) the emitted ly@xmath0 spectrum prior to scattering has a gaussian shape as a function of photon frequency with a doppler velocity width of @xmath196 km s@xmath1 ; ( _ iii _ ) the spatial width of the supershell is @xmath197 of its radius ; and ( _ iv _ ) the shell has an outflow velocity of @xmath198 km s@xmath1 . the _ blue dotted ( red solid ) lines _ represent a model in which the supershell has a column density of @xmath199 ( @xmath200 ) @xmath194 and a maximum radius that is @xmath201 kpc ( @xmath202 ) kpc . our calculations assume that there is no neutral gas ( or dust ) interior to the hi supershell . the _ left panel _ of figure [ fig : caseiii ] shows that inside the supershell the energy density decreases more gradually than @xmath203 because of photon trapping ( similarly to the previously discussed cases in [ sec : firststar]- [ sec : firstgalaxy ] ) . the shell with the larger column of hi is more efficient at trapping the ly@xmath0 photons , and thus yields a flatter energy density profile . in both models the energy density drops steeply within the supershell ( the energy density decreases as @xmath203 outside the shell , if no scattering occurs here ) . the _ right panel _ of figure [ fig : caseiii ] shows the ratio between the radiation and gravitational forces . towards the center of the dark matter halo , baryons dominate the mass density and an evaluation of @xmath204 requires assumptions about the radial distribution of the baryons . for simplicity , we consider a fixed total mass interior to the supershell of @xmath205 , so that @xmath206 . note that any assumed mass profile @xmath100 will not affect the results as long as @xmath207 . we find that the radiation force exceeds gravity in both examples under consideration . for @xmath199 @xmath194 and @xmath208 kpc , @xmath209 . thus , radiation pressure would have been important even if we had chosen @xmath210 . in the model with @xmath200 @xmath194 and @xmath202 kpc , @xmath211 , and radiation pressure would have been important even if @xmath212 . hence , our calculations strongly suggest that ly@xmath0 radiation pressure may be dynamically important in the ism of galaxies . the _ total _ ly@xmath0 radiation force is obtained by summing the force over all atoms in the supershell . it is interesting to compare this force to @xmath213 . the latter quantity denotes the total momentum transfer rate ( force ) from the ly@xmath0 radiation field to the supershell under the assumption that each ly@xmath0 photon is re - emitted isotropically after entering the shell ( including multiple scatterings inside the supershell ) . in figure [ fig : fscat ] we plot the quantity @xmath214 which is defined as @xmath215 as a function of the expansion velocity of the shell , @xmath216 for three different values of @xmath217 . figure [ fig : fscat ] shows that @xmath214 , which can be thought of as a force multiplication factor that was introduced by @xcite , as @xmath218 . here , @xmath30 is the total force that radiation exerts on a medium , @xmath219 is the total optical depth to electron scattering through this medium . the function @xmath220 arises because of the contribution of numerous metal absorption lines to the medium s opacity , and can be as large as @xmath221 in the atmospheres of o - stars @xcite . ] , greatly exceeds unity for low shell velocities and large hi column densities . the parameter @xmath214 is related to the mean number of times that a ly@xmath0 photon bounces back and forth between opposite sides of the expanding shell . for example , @xmath222 when all ly@xmath0 photons enter the shell , scatter once , and then escape from the shell in no preferred direction . on the other hand , @xmath223 when all ly@xmath0 photons enter the shell , scatter back towards the opposite direction , and then escape in no preferred direction after scattering in the shell for a second time . a schematic illustration of this argument is provided in figure [ fig : scheme ] . note that when @xmath222 ( @xmath223 ) , each photon spends on average a timescale of @xmath224 ( @xmath225 ) in the bubble enclosed by the shell . in other words , the factor @xmath214 relates to the trapping time , @xmath9 , that denotes the total time over which ly@xmath0 photons are trapped inside the supershell into the shell . if this corresponds to a physical distance that is significantly smaller than the thickness of the shell ( denoted by @xmath226 ) , then only a tiny fraction of the photons will diffuse through the shell . hence , when averaged over these photons , ignoring the time spent inside the supershell itself is justified . alternatively , photons with a mean free path that is at least comparable to the thickness of the shell , only spend a time @xmath227 inside the supershell , which provides a negligible contribution to the trapping time . ] ( see [ sec : intro ] ) through the relation @xmath228 . indeed , when @xmath229 we find that @xmath214 reproduces the value @xmath230 ( indicated by _ horizontal dotted lines _ ) that was found by @xcite and @xcite reasonably well ( keeping in mind that these authors derived their result for a static , uniform , infinite slab of material , and assumed different frequency distributions for the emitted ly@xmath0 photons ) . we have applied an existing monte - carlo ly@xmath0 radiative transfer code ( described and tested extensively in * ? ? ? * ) to the calculation of the pressure that is exerted by ly@xmath0 photons on an optically thick medium . this code enabled us to perform ( the first ) direct , accurate calculations of ly@xmath0 radiation pressure , which distinguishes this work from previous discussions on the importance of ly@xmath0 radiation pressure in various astrophysical environments . we have focused on a range of models which represent galaxies at different cosmological epochs . in [ sec : firststar ] we have shown that the ly@xmath0 radiation pressure exerted on the neutral intergalactic medium ( igm ) surrounding minihalos ( @xmath231 ) in which the first stars form , can exceed gravity by 23 orders of magnitude ( fig [ fig : casei ] ) , and in principle accelerate the gas in the igm to tens of km s@xmath1 . thus , ly@xmath0 radiation pressure can launch supersonic winds in the igm surrounding the first stars . our analysis did not allow a detailed study of the consequences of these winds . a comprehensive study would require numerical simulations that capture the full igm density field around minihalos in 3d and track the evolution of the shocks that may form in the igm together with the ly@xmath0 radiative transfer . in this paper , we have also shown that ly@xmath0 radiation pressure is important in the neutral igm that surrounds the hii regions produced by galaxies with a total halo mass of @xmath190 ( fig [ fig : caseii ] . these are the lowest mass , and hence the most abundant , halos in which gas can cool via atomic line excitation . here , however , the absolute magnitude of the radiation force is too weak to drive the gas to supersonic velocities . finally , we have shown in [ sec : wind ] that the ly@xmath0 radiation pressure exerted on neutral gas in the interstellar medium ( ism ) of a galaxy can also have strong dynamical consequences . in particular , we have found that the ly@xmath0 radiation force exerted on an expanding hi supershell can exceed gravity by orders of magnitude ( fig [ fig : caseiii ] ) , for reasonable assumptions about the gravitational force . it is therefore possible that ly@xmath0 radiation pressure plays an important role in determining the kinematics of hi supershells around starburst galaxies . we have demonstrated that the total ly@xmath0 radiation force on a spherical hi supershell can be written as @xmath232 , where the force - multiplication factor @xmath214 relates to the average trapping time of ly@xmath0 photons in the neutral medium . the factor @xmath214 can greatly exceed unity , as illustrated by fig [ fig : fscat ] . for comparison , the maximum possible radiation force due to continuum radiation ) around late - type stars that may significantly exceed @xmath233 @xcite . this is not because of trapping of continuum photons , but related to the propagation speed of the photons and the wind . we similarly expect the radiative force of trapped ly@xmath0 photons in steady - state outflows to potentially exceed @xmath234 ( and as argued in this paper , it is possible that @xmath235 ) . note though that these steady - state outflows are clearly different from those discussed in [ sec : wind ] , in which a well defined thin shell hi gas is physically separated from the central ly@xmath0 source . ] is @xmath236 , in which @xmath237 is the bolometric luminosity of the central galaxy . for a typical star forming galaxy , @xmath238 ( e.g. * ? ? ? * ) , whereas for a galaxy that contains population iii stars , @xmath239 @xcite . hence , the ly@xmath0 radiation pressure can dominate the maximum possible continuum radiation pressure if @xmath240 ( for a normal stellar population ) , a threshold which is easily exceeded at large column densities of relatively slow - moving hi shells ( see fig [ fig : fscat ] ) . the possibility that ly@xmath0 radiation alone can result in a radiation force that exceeds @xmath236 is important . @xcite have shown that the total momentum carried by radiation from a star - forming region can exceed the total momentum deposited by supernova explosions in it , and so galactic outflows may be driven predominantly by continuum radiation pressure . we have argued that ly@xmath0 radiation pressure may in some cases be even more important than continuum radiation pressure , and thus provide the dominant source of pressure on neutral hydrogen in the ism . the important implication of our last result is that ly@xmath0 radiation pressure may drive outflows of hi gas in the ism . observations of local starburst galaxies have shown that the presence of outflowing hi gas may be required to avoid complete destruction of the ly@xmath0 radiation by dust and to allow its escape from the host galaxies @xcite . at high redshifts , the ly@xmath0 emission line of galaxies is often redshifted relative to other nebular recombination lines ( such as h@xmath0 ) and metal absorption lines ( e.g. * ? ? ? * ; * ? ? ? furthermore , the spectral shape of the ly@xmath0 emission line is typically asymmetric , with emission extending well into the red wing of the line ( e.g. * ? ? ? * ) . both of these observations can be explained simultaneously if the observed ly@xmath0 photons scatter off neutral hydrogen atoms in an outflowing supershell of hi gas that surrounds the star forming regions @xcite . the possibility that ly@xmath0 radiation pressure may be important in determining the properties of expanding supershells is exciting , and is discussed in more detail in a companion paper ( dijkstra & loeb 2008 ) . * acknowledgments * this work is supported by in part by nasa grant nnx08al43 g , by fqxi , and by harvard university funds . we thank christian tapken and an anonymous referee for helpful constructive comments . xx natexlab#1#1 abel , t. , norman , m. l. , & madau , p. 1999 , apj , 523 , 66 abel , t. , bryan , g. l. , & norman , m. l. 2002 , science , 295 , 93 abel , t. , wise , j. h. , & bryan , g. l. 2007 , apjl , 659 , l87 adams , t. f. 1975 , apj , 201 , 350 ahn , s .- h . , & lee , h .- w . 2002 , journal of korean astronomical ahn , s .- h . , lee , h .- w . , & lee , h. m. 2003 , mnras , 340 , 863 ahn , s .- h . 2004 , apjl , 601 , l25 atek , h. , kunth , d. , hayes , m. , ostlin , g. , mas - hesse , j. m. , & . 2008 , arxiv e - prints , 805 , arxiv:0805.3501 barkana , r. , & loeb , a. 2001 , physics reports , 349 , 125 barkana , r. 2004 , mnras , 347 , 59 bithell , m. 1990 , mnras , 244 , 738 bonilha , j. r. m. , ferch , r. , salpeter , e. e. , slater , g. , & noerdlinger , p. d. 1979 , apj , 233 , 649 castor , j. i. , abbott , d. c. , & klein , r. i. , 1975 , apj , 195 , 157 chandrasekhar , s. 1945 , apj , 102 , 402 chen , h .- w . , prochaska , j. x. , & gnedin , n. y. 2007 , apjl , 667 , l125 cox , d. p. 1985 , apj , 288 , 465 dennison , b. , turner , b. e. , & minter , a. h. 2005 , apj , 633 , 309 dijkstra , m. , haiman , z. , & spaans , m. 2006 , apj , 649 , 14 dijkstra , m. , lidz , a. , & wyithe , j. s. b. 2007 , mnras , 377 , 1175 dijkstra , m. , & loeb , a. 2008 , submitted to mnras dijkstra , m. , et al . 2008 , accepted to mnras elitzur , m. , & ferland , g. j. 1986 , apj , 305 , 35 furlanetto , s. r. , zaldarriaga , m. , & hernquist , l. 2004 , apj , 613 , 1 gnedin , n. y. , kravtsov , a. v. , & chen , h .- w . 2008 , apj , 672 , 765 haehnelt , m. g. 1995 , mnras , 273 , 249 haiman , z. , thoul , a. a. , & loeb , a. 1996 , apj , 464 , 523 hayes , m. , ostlin , g. , mas - hesse , j. m. , & kunth , d. 2008 , arxiv e - prints , 803 , arxiv:0803.1176 heger , a. , & woosley , s. e. 2002 , apj , 567 , 532 heiles , c. 1984 , apjs , 55 , 585 hirata , c. m. 2006 , mnras , 367 , 259 hui , l. , & gnedin , n. y. 1997 , mnras , 292 , 27 ivezic , z. , & elitzur , m. 1995 , apj , 445 , 415 kitayama , t. , yoshida , n. , susa , h. , & umemura , m. 2004 , apj , 613 , 631 komatsu , e. , et al . 2008 , arxiv e - prints , 803 , arxiv:0803.0547 kunth , d. , mas - hesse , j. m. , terlevich , e. , terlevich , r. , lequeux , j. , & fall , s. m. 1998 , a&a , 334 , 11 lee , h .- w . , & ahn , s .- h . 1998 , apjl , 504 , l61 lequeux , j. , kunth , d. , mas - hesse , j. m. , & sargent , w. l. w. 1995 , a&a , 301 , 18 loeb , a. , & rybicki , g. b. 1999 , apj , 524 , 527 ( lr99 ) loeb , a. 2001 , apjl , 555 , l1 lucy , l. b. 1999 , a&a , 344 , 282 lucy , l. b. 2007 , a&a , 468 , 649 maller , a. h. , & bullock , j. s. 2004 , mnras , 355 , 694 mckee , c. f. , & tan , j. c. 2008 , apj , 681 , 771 mcclure - griffiths , n. m. , dickey , j. m. , gaensler , b. m. , & green , a. j. 2002 , apj , 578 , 176 mcquinn , m. , lidz , a. , zahn , o. , dutta , s. , hernquist , l. , & zaldarriaga , m. 2007 , mnras , 377 , 1043 murray , n. , quataert , e. , & thompson , t. a. 2005 , apj , 618 , 569 oh , s. p. , & haiman , z. 2002 , apj , 569 , 558 osterbrock , d. e. 1989 , _ astrophysics of gaseous nebulae and active galactic nuclei _ , university of minnesota , et al . mill valley , ca , university science books . ostlin , g. , hayes , m. , kunth , d. , mas - hesse , j. m. , leitherer , c. , petrosian , a. , & atek , h. 2008 , arxiv e - prints , 803 , arxiv:0803.1174 ouchi , m. , et al . 2008 , apjs , 176 , 301 partridge , r. b. , & peebles , p. j. e. 1967 , apj , 147 , 868 pettini , m. , shapley , a. e. , steidel , c. c. , cuby , j .- g . , dickinson , m. , moorwood , a. f. m. , adelberger , k. l. , & giavalisco , m. 2001 , apj , 554 , 981 pritchard , j. r. , & furlanetto , s. r. 2006 , mnras , 367 , 1057 pritchard , j. r. , & loeb , a. 2008 , arxiv e - prints , 802 , arxiv:0802.2102 rybicki , g. b. , & lightman , a. p. 1979 , new york , wiley - interscience , 1979 . 393 p. , ryder , s. d. , staveley - smith , l. , malin , d. , & walsh , w. 1995 , aj , 109 , 1592 salpeter , e. e. 1974 , apj , 193 , 585 schaerer , d. 2002 , a&a , 382 , 28 schaerer , d. 2003 , a&a , 397 , 527 sethi , s. k. , subrahmanyan , r. , & roshi , d. a. 2007 , apj , 664 , 1 shapley , a. e. , steidel , c. c. , pettini , m. , & adelberger , k. l. 2003 , apj , 588 , 65 shimasaku , k. , et al . 2006 , pasj , 58 , 313 tapken , c. , appenzeller , i. , noll , s. , richling , s. , heidt , j. , meinkhn , e. , & mehlert , d. 2007 , a&a , 467 , 63 tenorio - tagle , g. , & bodenheimer , p. 1988 , ara&a , 26 , 145 tenorio - tagle , g. , silich , s. a. , kunth , d. , terlevich , e. , & terlevich , r. 1999 , mnras , 309 , 332 verhamme , a. , schaerer , d. , & maselli , a. 2006 , a&a , 460 , 397 verhamme , a. , schaerer , d. , atek , h. , & tapken , c. 2008 , arxiv e - prints , 805 , arxiv:0805.3601 wyithe , j. s. b. , & loeb , a. 2006 , _ nature _ , 441 , 322 yoshida , n. , omukai , k. , hernquist , l. , & abel , t. 2006 , apj , 652 , 6 processes that populate the @xmath241 level include : ( _ i _ ) collisional excitation of neutral hydrogen atoms in both the @xmath242 and @xmath243 states by electrons and protons , ( _ ii _ ) recombination into the @xmath241 state following photoionization or collisional ionization , ( _ iii _ ) photoexcitation of @xmath241 level , ( _ iv _ ) photoexcitation of @xmath244 ( @xmath245 ) levels , followed by a radiative cascade that passes through the @xmath241 state . processes that _ de - populate _ the @xmath241 level include ( _ v _ ) collisional de - excitation to by electrons , ( _ vi _ ) stimulated ly@xmath0 emission @xmath246 following the absorption of a ly@xmath0 photon , and ( _ vii _ ) spontaneous emission of a ly@xmath0 photon . where @xmath248 s denote collisional excitation rate coefficients from level @xmath249 to @xmath250 ( in @xmath136 s@xmath1 ) , @xmath251 denote number densities is species x ( i.e @xmath252 denotes the electron number density , while @xmath253 denotes the number density of hii ions ) , @xmath254 denote photoexcitation rates to the state @xmath244 ( in s@xmath1 ) , and @xmath255 denotes the probability that photoexcitation of the @xmath244 level results in a radiative cascade that passes through the @xmath241 level . * the einstein - a coefficient of the ly@xmath0 transition is @xmath256 s@xmath1 . that is , spontaneous emission of ly@xmath0 depopulates the @xmath241 state at a rate @xmath256 s@xmath1 . * collisional excitation of neutral hydrogen atoms in @xmath242 level by electrons populates the @xmath241 level at a rate @xmath257 \left[\frac{\omega(1s,2p)}{g_2}\frac{g_{2p}}{g_{1s}}\right]e^{-\chi / kt}~{\rm s^{-1}}$ ] ( e.g. * ? ? ? , @xmath57 denotes the gas temperature in k , @xmath252 is the electron density in @xmath45 , @xmath258 and @xmath259 are the statistical weights of the @xmath242 and @xmath241 levels , @xmath260 ( 5000 k @xmath261 t @xmath261 2@xmath262 k , * ? ? ? * ) , and @xmath263 ev is the energy difference between the @xmath242 and @xmath241 levels . for temperatures @xmath264 k , we find that @xmath265 s@xmath1 . * the collisional excitation rate of neutral hydrogen atoms in @xmath243 level by protons ( which dominate over collisions with electrons by about a factor of @xmath266 ) populates the @xmath241 level at a rate @xmath267 s@xmath1 ( e.g. * ? ? ? * ) . to assess the term @xmath268 requires one to compute @xmath269 the fraction of atoms in the @xmath243 state is determined by rates similar to those mentioned above , except that the @xmath243-state is metastable and its einstein coefficient is @xmath270 s@xmath1 . the @xmath243-state may therefore be overpopulated relative to the @xmath241 state by orders of magnitude ( see e.g. * ? ? ? * and references therein ) . in close proximity to a luminous source , the @xmath243 level is populated mostly via transitions of the form @xmath271 @xcite , and @xmath272 , where @xmath273 is the rate at which ly@xmath274 photons are scattered and the prefactor @xmath275 denotes the probability that absorption of the ly@xmath274 is followed by re - emission of an h@xmath0 photon @xcite . * the recombination rate into the @xmath241 level is given by @xmath276 @xmath277 s@xmath1 ( e.g. * ? ? ? * ) , where @xmath278 is the proton density in @xmath45 . * the rate at which transitions of the form @xmath279 occur by absorbing a photon is given by @xmath280 . assuming for simplicity that @xmath87 does not vary with frequency , i.e. @xmath281 , we have @xmath282 where @xmath283 denotes the oscillator strength of the transition , @xmath284 ( @xmath285 ) the charge ( mass ) of the electron , and @xmath286 denotes the energy difference between the @xmath242 and @xmath244 levels . the oscillator strength decreases rapidly with increasing @xmath287 ( e.g. chapter 10.5 of * ? ? ? * ) , and in practice we can safely ignore all terms with @xmath288 . we then need not worry about the factors @xmath255 ( which have been computed by * ? ? ? * ; * ? ? ? + the ly@xmath0 scattering rate is given by @xmath289 where we replaced @xmath290 ( in erg s@xmath1 hz@xmath1 sr@xmath1 @xmath194 ) with @xmath291 , in which @xmath8 is the ly@xmath0 luminosity of the central source ( in erg s@xmath1 ) , and @xmath292 is the frequency range over which these ly@xmath0 photons have been emitted . the factor @xmath214 takes into account the fact that resonant scattering traps ly@xmath0 photons in an optically thick medium ( see [ sec : wind ] ) . substituting fiducial numbers @xmath293 + the rate at which ly@xmath274 photons scatter can be related to the ly@xmath0 luminosity , @xmath8 , and the equivalent width ( ew ) of the line , if one writes the specific intensity @xmath294 near the ly@xmath274 resonance in terms of the ly@xmath0 luminosity of the central source as @xmath295 ( note that we assumed that the specific intensity of the continuum remains constant between @xmath296 and @xmath297 ) . the rate at which ly@xmath274 photons scatter can then be written as + @xmath298 + equation ( [ eq : lyb ] ) illustrates that it is very difficult to bring the ratio @xmath299 to unity . * collisional deexcitation rates relate to the collisional excitation rates via @xmath300 ( e.g. * ? ? ? combined with the formulas given above , it is straightforward to verify that the collisional de - excitation rates are subdominant relative to the rate at which spontaneous ly@xmath0 emission de - populates the @xmath241 level . * lastly , eq ( [ eq : lya ] ) shows that the stimulated emission rate is @xmath301 . the maximum number density of hydrogen nuclei in this paper is encountered in [ sec : wind ] , for the expanding shell of hi gas with @xmath199 @xmath194 and a thickness @xmath302 kpc , in which @xmath303 @xmath45 . at these column densities , the shell self - shields against ionizing radiation , and is likely mostly neutral . for simplicity , let us assume that @xmath304 @xmath45 . under these conditions : * the collisional excitation rate from @xmath305 is @xmath306 s@xmath1 . * the collisional excitation rate from @xmath307 per atom in the 1s state- is @xmath308 s@xmath1@xmath309 s@xmath1 . the maximum ly@xmath0 luminosity considered in this paper is @xmath310 erg s@xmath1 . let us conservatively assume that ew@xmath311 (rest - frame ) , which corresponds roughly to the detection threshold that exists in narrow - band surveys ( e.g. * ? ? ? * ) . using eq [ eq : lyb ] , we find that @xmath312 s@xmath1 ( for @xmath313 kpc ) , and therefore that @xmath314 s@xmath1 . * the recombination rate is @xmath315 s@xmath1 . * the maximum ly@xmath0 scattering rate is ( substituting @xmath313 kpc , @xmath316 , @xmath317 into eq [ eq : lya ] ) @xmath318 s@xmath1 . by comparing these rates to the rate at which spontaneous emission of ly@xmath0 depopulates the @xmath241 state , @xmath319 s@xmath1 , we find that all excitation rates are @xmath320 orders of magnitude smaller than the de - excitation rate for the wind models discussed in [ sec : wind ] . in equilibrium , hydrogen atoms in their electronic ground ( @xmath242 ) state are therefore @xmath320 orders of magnitude more abundant than hydrogen atoms in their first excited ( @xmath241 ) state . furthermore , as was mentioned above the ratio of atoms in the 2s and 1s levels is given by @xmath321 . since the densities , ly@xmath0 luminosities , and the ly@xmath274 scattering rates are lower , the fraction of hi atoms in their first excited states are even smaller in other sections of the paper . in conclusion , for all applications presented in this paper , no accuracy is lost by assuming that all hydrogen atoms occupy their electronic ground state . finally , in [ sec : test ] we computed solutions for the radial dependence of @xmath97 ( fig [ fig : uden ] ) . the energy density , @xmath97 , was quoted to depend linearly on the luminosity of the central source . this is valid unless ( _ i _ ) the ly@xmath0 scattering rate , @xmath322 s@xmath1 , or ( _ ii _ ) the ly@xmath274 scattering rate exceeds @xmath323 s@xmath1 . in either case , our assumption that all hydrogen atoms populate their electronic ground state breaks down . condition ( _ i _ ) translates to @xmath324 erg s@xmath325 ( eq . [ eq : lya ] ) , while condition ( _ ii _ ) translates to @xmath326 erg s@xmath327 ( eq . [ eq : lyb ] ) . substituting @xmath313 kpc , @xmath316 ( fig [ fig : uden ] shows that resonant scattering enhances the energy density by a factor of @xmath328 relative to @xmath329 ) , @xmath330 ( thermal broadening alone in @xmath137 k gas results in @xmath331 km s@xmath1 , which translates to @xmath332 ) , and @xmath333 , condition ( _ i _ ) translates to @xmath334 erg s@xmath1 , while condition ( _ ii _ ) translates to @xmath335 erg s@xmath1 the more conservative condition ( _ ii _ ) , @xmath335 erg s@xmath1 , translates to @xmath336 , well beyond the regime considered in this paper . [ lastpage ]
it is scientifically and technically possible to build an earth penetrating device that could bury a b61 - 7 warhead 30 meters into concrete , or 150 meters into earth , before detonating it . the device ( based on knowledge and technology that are available since 50 years ) would however be large and cumbersome . better penetrator materials , components able to withstand larger stresses , higher impact velocities , and/or high - explosive driven penetration aids , can only marginally improve the device . it is concluded that the robust nuclear earth penetrator ( rnep ) program may be as much motivated by the development of new technology directly applicable to next generation nuclear weapons , and by the political necessity to periodically reassess the role and utility of nuclear weapons , than by the perceived military need for a weapon able to destroy deeply buried targets . + + +
physics0510052
nuclear weapons are the most destructive instruments of war available to the military . their destructiveness is due to a number of effects mechanical , thermal , and radiological but by far the most important military effect of present - day nuclear weapons is due to the enormous shock - waves generated in the surrounding materials by the energy released from nuclear fission and thermonuclear fusion within the warhead . in an atmospheric explosion the shock - waves generated in air propagate over large distances , destroying every building and laying down every tree within about 3 km from the point of explosion of a 1 _ _ megaton _ _ thermonuclear weapon . it is this air - blast shock - loading effect combined with the incendiary effect of high - yield thermonuclear explosion which is at the origin of the strategy of war deterrence by mutually assured destruction . however , if instead of striking relatively weak structures such as buildings the air - blast strikes the ground , most of the energy in the shock - waves is reflected so that very little energy is in fact transferred to the ground . this very poor `` coupling '' of the energy released by an atmospheric explosion is due to the factor of about 1,000 difference in densities between the air and the soil , so that because of energy and momentum conservation the maximum energy coupling to the ground is at most of 15% @xcite . more precisely , in the case of low - altitude near - surface explosions ( which maximize the energy transfer to the ground ) the typical theoretical energy couplings are between 5 to 10% for chemical explosions , and about 8% for nuclear explosions provided the direct coupling of x - rays ( about 6% ) is added to the mechanical coupling ( about 2% ) @xcite . consequently , if the objective of a weapon is to destroy underground targets by means of shock - waves , the obvious idea is to detonated it under the surface of the ground . this enables a reduction by a factor of about 10 to 15 in the explosive yield of the weapon . indeed , even for a rather shallow burial ( about 0.6 meter ) , the maximum energy coupled to the ground by a 0.5 _ kt _ fission - explosion was computed to be already about 30% of the total yield , an estimate that was confirmed by a nuclear test conducted at the nevada test site @xcite . therefore , nearly 100% coupling is achieved when a warhead is detonated at a few meters below the ground , so that the explosion of an earth penetrating warhead ( epw ) with a yield of about 0.5 _ mt _ can in principle have the same destructive power as a 5 _ mt _ above ground detonation . the military rationale for developing and producing the b61 - 11 ( that is the 11th modified version of the b61 gravity bomb whose development started in 1961 ) was therefore to provide a replacement for the 9 _ megatons _ b53 that was intended to be used for destroying buried command centers and other underground targets . moreover , since the precision and the flexibility of the delivery of the b61 were substantially greater than those of the b53 , it was felt that these advantages would more than compensate for the lower yield of the b61 , which is at most of 340 _ kt_. the b61 - 11 began entering service in 1997 . it was therefore the first `` new '' nuclear weapon to enter into the arsenal of a declared nuclear power after the conclusion ( in 1996 ) of the negotiations of the comprehensive test ban treaty , during which the united states stated that the b61 - 11 would be the last weapon to enter into its nuclear arsenal . so why is it that the united states s administration is now seeking a lifting of the 1993 ban on development of new nuclear weapons ( the spratt - furse law ) ? on 9 may 2003 a committee of the united states senate passed a bill to that effect , calling for the development of a `` bunker buster '' bomb to be called the robust nuclear earth penetrator ( rnep ) . part of the reasons are of a military - technical nature : even though the b61 - 11 can bury itself 36 meters underground before detonation , it can not destroy targets at depths or distances much larger than 30100 meters below or away from the point of explosion . this is because shock - waves propagating through a solid medium are much less destructive that those propagating through air ( * ? ? ? * chap.6 ) . for instance , ordinary buildings and even much stronger structures such as bridges or above - ground military constructions are relatively brittle : the interaction of the blast - wave from an atmospheric explosion with them will be essentially inelastic , resulting in their complete failure at relatively large distances . on the other hand , the interaction of an underground shock wave with a tunnel or a buried bunker will be elastic over a considerable range of overpressure , with the result that well - designed man - made structures will not fail unless the materials themselves fail . for example , the assured destruction range of a shallow - buried explosion of 1 _ mt _ is on the order 150250 m. this corresponds to the onset of gross rock failures for most formations and thus represents the range of survival for the best examples of underground construction @xcite . assuming a scaling with the third root of the yield , this confirms the stated maximum destruction range of the b61 - 11 , i.e. , about 100 meters . however , this maximum destruction range does not take into account the protection measures that can be taken in order to improve the resistance of underground structures to the adverse effect of shock - waves . these include many techniques such as decoupling by means of spaced or multilayered walls made of materials of different densities ; shock absorption , reflection or diffusion ; shock mounts for fragile equipments ; etc . as a result , one can easily estimate that the stated lower destruction range of 30 m is truly a minimum , even for a 340 _ kt _ explosion , because some carefully designed structures may survive even that close to the point of the explosion . it is therefore possible , on the basis of these purely technical facts , and without addressing the related military and political implications , to argue that a warhead that would be able to bury deeper into the ground before exploding would be more effective to destroy underground targets . however , even with perfect accuracy in weapon delivery , this ability to explode at a greater depth will only get the point of explosion closer to the target it will not be able to extend the lethality range of the weapon itself . for example , if the minimum assured lethality range of the b61 - 11 is 30100 m , the corresponding range for a rnep version able to bury 30 m deeper into the ground @xcite will be 60130 m. therefore , contrary to certain claims , it seems impossible that a rnep based on the b61 - 11 could crush targets 300 m underground . this is because extending the penetration range of the b61 - 11 by a factor of 510 , that is from 36 m to about 1530 m in reinforced concrete or solid rock , is already a considerable technical challenge , as will be seen . the paper is organized as follows : in the sec . [ fun:0 ] we review the physics of the penetration of a strong and rigid projectile into a compressible solid target in order to find out whether there is room for improvement in comparison to what is known to be possible with existing materials and techniques . after deriving from first principles the rather simple expression giving the maximum hypothetical penetration depth as a function of the key physical parameters , eq . , it is found by comparisons with a dozen different penetration experiments ( in which a variety of projectile and target materials are used ) that there is very little room for improvement : not more than 30 to 60% . starting form sec . [ pen:0 ] we specialize to a concrete target , for which eq . gives a good estimate of the penetration depth as a function of the length , effective density , and velocity of the penetrator . it is found that the stated penetration capability of the b61 - 11 is consistent with its published physical mensurations and weights . in sec . [ rob:0 ] we give the physical and engineering reasons explaining the b61 - 7 impact resistance ( the b61 - 7 is the warhead of which the b61 - 11 is a ruggedized earth - penetrating version ) , which enables it to withstand the considerable deceleration upon impact with the ground . we also derive simple formulas giving estimates for the maximum and mean decelerations during penetration . in sec . [ sci:0 ] we assess the scientific feasibility of a rnep able to bury the b61 - 7 warhead 30 meters into concrete , or 100 to 150 meters into earth . four general possibilities are analyzed : _ ( i ) _ maximizing the weight and length of the penetrator ; _ ( ii ) _ maximizing the velocity of the penetrator ; _ ( iii ) _ using conventional penetration aids such as a precursor shaped - charge jet ; and _ ( iv ) _ , assuming that radically new types of compact nuclear explosives will become available in the next few decades , speculations are made on hypothetical new designs which could provide a technically more attractive solution than the three previous possibilities . in sec . [ env:0 ] we briefly recall the environmental and political limitations to the use of nuclear weapons , even when detonated underground . in particular , we stress the political and nuclear proliferation problems posed by the `` plutonium mines '' created when an underground plutonium bomb explosion is contained . in the last section , sec . [ dis:0 ] , we conclude that the technology for building a working but cumbersome penetrator suitable to bury the b61 - 7 warhead 30 meters into concrete is available , but that improving that penetrator and trying to reduce its size and/or weight by a factor of two will be very difficult , even if considerable advances are made in the realms of materials science , micro - electromechanical systems s engineering , and nanotechnology . finally , considering these difficulties , as well as the military and political limitations to the battle - field use of existing types of nuclear weapons , we wonder whether apart from keeping the nuclear weapons debate alive , an important reason for considering a new generation of nuclear earth penetrating weapons would not be to develop their associated non - nuclear technologies , which overlap with those required for radically new types of weapons , both conventional and nuclear @xcite . a recurring suggestion in the general scientific literature is that by some new `` trick '' the penetration depth of a kinetic penetrator could be substantially increased ( perhaps by a factor of ten or more ) over what is presently possible with available materials and technologies , e.g. , @xcite . in order to show that this is not possible , we will study a simple model in which everything is idealized except for the most basic processes which enable a high - velocity object to penetrate into a material such as steel , concrete , or earth . moreover , the model will be such that all approximations tend to overestimate the penetration depth , what is for instance the case for a one - dimensional model in comparison to a two- or three - dimensional one . for this model we follow the general idea suggested in ref . @xcite , namely that when a high - velocity projectile penetrates into a material it creates a shock wave which compresses and heats the material ahead of it : the maximum penetration depth is then obtained when the material immediately ahead of the projectile ( i.e. , sufficiently heated to evaporate ) is instantaneously removed by some ideal process , so that the penetrator can continue its course until its kinetic energy is exhausted , or some fundamental limit is reached . in other words , the basic idea is that the kinetic energy of the penetrator is entirely used to melt the material ahead of it , and that all other processes which remove that material are supposed to have no energy cost . similarly , it is supposed that the frictional forces on all sides of the penetrator are negligible , and that all flows of materials in the regions adjacent to the surface of the penetrator are completely free . however , whereas the velocity is supposed to be high , it is not supposed to be so high that one is in the high - velocity hydrodynamic - limit where all materials ( including the penetrator which in the present model is supposed to be perfectly strong and rigid ) behave as ideal incompressible - liquids , and all the kinetic energy is used to move them aside from the path of the penetrator . let us therefore take a penetrator of length @xmath0 , frontal area @xmath1 , density @xmath2 , and very high compressive strength @xmath3 ; and a semi - infinite target of density @xmath4 and compressive strength @xmath5 . , but the `` unconstrained ( or unconfined ) compressive strength . '' for metallic projectiles and cohesive targets it has a magnitude comparable to the yield strength @xmath6 , which is typically in the range of 10 mpa to 2 gpa . for cohesionless or granular targets such as soil or sand , the unconstrained compressive strength in on the order of 0.1 to 10 mpa . ] to begin with we suppose that this penetrator is incompressible and sufficiently strong to remains intact during penetration into the target . according to our model , at any given time during penetration , the velocity of the projectile @xmath7 is such that the compressive strength of the target is exceeded , so that ahead of it the target material is compressed by a shock - wave to a density @xmath8 . as we are considering large penetration velocities , we can assume the strong shock limit , in which case @xmath9 where @xmath10 is the polytrope exponent . the crucial step in the reasoning is now to apply energy - momentum conservation to the shocked target material ahead of the penetrator , which implies @xmath11 where @xmath12 is the distance measured ( for convenience ) from the back of the penetrator and @xmath13 the velocity of the shocked material at any given time . with the boundary condition @xmath14 , this equation is trivially integrated , i.e. , @xmath15 one can therefore calculate the energy distribution in the shocked material @xmath16 where @xmath17 is the kinetic energy of the penetrator at time @xmath18 . by derivating this equation and dividing by @xmath1 , or by directly using , one can also calculate the energy density , i.e. , @xmath19 the last equation can now be used to write down the _ criterion for penetration _ , which is simply the condition that the energy density in front of the penetrator has to be higher than that corresponding to the compressive strength of the target , i.e. , @xmath20 in particular , the _ _ minimum velocity _ _ to the target compressive strength , i.e. , eq . with @xmath21 replaced by @xmath2 . ] of the projectile for penetration to proceed until it has come to a rest is given when @xmath22 , i.e. , @xmath23 equation can also be used to calculate the thickness @xmath24 of the shocked region such that its energy density exceeds @xmath5 , i.e. , @xmath25 in the present model , the energy @xmath26 given by eq . corresponds to the energy used to `` evaporate '' the material in the region of thickness @xmath27 ahead of the penetrator . therefore , since that material is supposed to be instantaneously removed out of its way , the penetrator s linear energy loss is given in first approximation by @xmath28 or , expressing @xmath29 in terms of @xmath30 and rearranging , by @xmath31 where , because the removed material is in gaseous form , we have approximate its adiabatic exponent by that of a monoatomic gas , i.e. , @xmath32 , so that eq . gives @xmath33 . equation is easily integrated by quadrature , giving the penetration depth @xmath34 in function of the initial and final penetrator energies @xmath35 or in function of the initial and final penetrator velocities @xmath36 where the final velocity @xmath37 can been replaced by the velocity given by eq . , i.e. , assuming again @xmath32 , @xmath38 the main result of our model , equation , is remarkable because it explains essentially all the important qualitative features observed in penetration experiments : direct proportionality to the length , density , and velocity of the penetrator ; inverse proportionality to the square root of the density and to that of the compressive strength of the target ; and existence of a minimum velocity . on the other hand , when making quantitative comparisons with the data , it turns out as expected form the discussion in the introduction to this section that eq . tends to overestimate the penetration depth . this can be seen , for example , if one looks at the penetration depth versus striking velocity of steel projectiles into concrete data of ref . @xcite shown in fig . 2 of ref . @xcite . in that case overestimates @xmath39 by a factor of @xmath40 2.8 . but , if one makes the comparison with the data of ref . @xcite shown in fig . 7 of ref . @xcite , the discrepancy is only a factor of @xmath40 1.5 ( see , table [ tbl:1 ] ) . consequently , although the details of the experimental conditions and the exact properties of the materials used are not well known , the `` hypothetical maximum room for improvement '' suggested by these comparisons is small : at most between a factor 1.5 and 3 . to better understand what this means in practice , further comparisons with other experiments using other target and projectile materials are made in table [ tbl:1 ] . & & @xmath4 & @xmath5 & @xmath41 & @xmath42 & @xmath43 & @xmath41 & r ' '' '' + & & [ kg / m@xmath44 & [ mpa ] & ( exp ) & [ s / km ] & [ km / s ] & & + concrete & @xcite & 2310 & 34.7 & 12 & 20 & 0.09 & 18 & 1.5 ' '' '' + concrete & @xcite & 2300 & 62.8 & 6.1 & 15 & 0.11 & 13 & 2.1 ' '' '' + concrete & @xcite & 2340 & 96.7 & 3.7 & 12 & 0.14 & 10 & 2.7 ' '' '' + concrete & @xcite & 2500 & 50 & 5 & 16 & 0.10 & 14 & 2.8 ' '' '' + steel & @xcite & 7900 & 1000 & 1 & 3.8 & 0.25 & 2.8 & 2.8 ' '' '' + titanium & @xcite & 4500 & 800 & 1.2 & 5.5 & 0.30 & 3.9 & 3.2 ' '' '' + lexan & @xcite & 1210 & 60 & 10 & 40 & 0.16 & 34 & 3.4 ' '' '' + aluminum & @xcite & 2700 & 100 & 4 & 21 & 0.14 & 18 & 4.5 ' '' '' + grout & @xcite & 2000 & 21.6 & 7.6 & 27 & 0.07 & 25 & 3.3 ' '' '' + grout & @xcite & 2000 & 13.5 & 8.1 & 34 & 0.06 & 32 & 4.0 ' '' '' + lead & @xcite & 11300 & 17 & 2.5 & 3 & 0.03 & 13 & 5.2 ' '' '' + foam & @xcite & 110 & 20 & 120 & 120 & 0.30 & 85 & 0.7 ' '' '' + let us define the hypothetical room for improvement by the ratio @xmath45 between the hypothetical maximum penetration given by eq . and the corresponding measurement for a given penetrator length @xmath0 at a normalized penetrator velocity of @xmath46 km / s , i.e. , @xmath47 at that velocity . for the sake of comparison , and in order to exhibit some possible trends , the data in table [ tbl:1 ] is listed in approximately increasing order of @xmath45 , except for the last experiment in which the target is a low density foam . the first set of data , from refs . @xcite , is for steel ( i.e. , @xmath48 mg / m@xmath44 ) ordnance - type projectiles into concrete targets . as can be seen , this set corresponds to the smallest room for improvement , possibly because everything feasible has already been done with them in order to facilitate penetration , e.g. , maximal hardness , ogival nose shape , and large length over diameter ratio ( `` long rod '' penetrators ) . the second set of data , from ref . @xcite , is for tungsten carbide ( i.e. , @xmath49 mg / m@xmath44 ) spherical balls into various typical armor materials : rolled homogeneous armor ( rha ) steel , titanium , polycarbonate ( g.e . lexan ) , and 1100-f aluminum . as can be seen @xmath45 is comprised between 2.8 and 4.5 , with a slight trend towards an increase when going from harder to softer materials . compared to the previous set of data , @xmath45 is about 1.6 times larger on average . however , if instead of being spherical the penetrator s shape would have been optimized , and thus its drag coefficient minimized , @xmath45 could have been significantly smaller . this effect is well known and accounts for a factor of about 2 when going from a flat to a conical shape ( see , e.g. , @xcite ) . the third set of data consists of experiments with ogive - nose steel projectiles impacting grout with two different compressive strengths @xcite . the trend for an increase of @xmath45 when going from harder to softer materials is observed again . the last set of data , from fig . 116 of ref . @xcite , is for spherical balls of steel into a soft and heavy material ( lead ) , or a low density foam ( assumed to be polyethylene ) . in the case of lead the hypothetical room for improvement is 5.2 , suggesting that for this material the idealizations leading to eq . are not adequate : in particular , the compressive strength is so small that little energy is needed to crush it , while significant energy is needed to sweep it aside . finally , in the case of the low - density foam , the material can no more be considered as a `` solid , '' even at the relatively high impact velocities considered here : eq . fails and another model should be developed . is not a bad theoretical estimate , the large value of @xmath50 , especially in comparison to that of lead , suggests that a different model is definitely required . in effect , @xmath50 is the minimum velocity for compressive - yielding to be the dominant penetration mechanism , while for a soft material penetration occurs at a much lower velocity . ] consequently , by taking into consideration the most obvious detrimental effects that were neglected in the derivation of eq . ( such as the energy needed to moved the target material out of the way , and the influence of the penetrators s nose shape ) the data in table [ tbl:1 ] show that instead of a room for improvement of about 3 to 4 as suggested by a naive interpretation , the actual margin can not be much more than a factor of about 1.5 , or possibly even less considering the complexity and number of small effects that should be considered at the detailed level . in particular , this implies that processes such as cavitation , which can have a significant impact in aerodynamics and low - velocity hydrodynamics @xcite , will not be able to play any significant role in improving penetration into a solid target @xcite , except possibly when driving into soil or mud . similarly , using for the penetrator a composite or nano - engineered material instead of steel @xcite , or a heavy - material like uranium ( which by adiabatic shearing enables to improve long - rod penetration into a metal by about 10% at high - velocities @xcite ) , will only have a marginal effect on penetration depth apart from that due to their densities . before summarizing , let us recall that the purpose of our model was not to provide an exact fit to the data , but rather to give an estimate for a hypothetical maximum penetration depth assuming that all the kinetic energy of a perfectly hard projectile is used to `` punch '' a hole through the target , i.e. , that none of that energy is used to remove the material evaporated at the head of the penetrator , and that none of that energy is lost as mechanical energy or stored as elastic energy , but rather converted to heat in order to vaporize the material ahead of the penetrator . this means that no conceivable physical process can possibly beat the maximum penetration given by eq . . moreover , if the projectile itself can not be considered as perfectly hard , eq . will provide an even larger upper limit to the actual maximum penetration depth . this happens when the velocity of the projectile @xmath29 is so large that the ram pressure exceeds its compressive strength @xmath3 , and that it begins to erode and loose rigidity . an estimate for that _ maximum velocity _ is provided by going to the projectile rest frame , where the ram pressure is @xmath51 , giving the expression @xmath52 then , inserting eq . in eq . , we can derive a simple formula for the maximum possible penetration at maximum projectile velocity @xmath53 which like eq . will of course overestimate the actual maximum possible penetration depth . , while eq . gives about 11 , i.e. , a factor of 2.4 more , as could be expected from the value of @xmath45 given in table [ tbl:1 ] . ] in conclusion , even if for some improbable reason the maximum penetration depths could be larger than given by eqs . or say by a factor of two or three the maximum hypothetically possible penetration depth would still not be larger than the experimentally measured and theoretically known values by one order of magnitude . implicitly introduced through eq . . if a reasonable value different from 5/3 is taken for this parameter , the depths given by eqs . or will be at most different by a factor of two . this has , however , no influence on what is physically possible in order to improve penetration . ] therefore , any conceivable improvement in projectile technology can not increase penetration by a factor of be more than about 1.5 . finally , for the penetration depth into a solid to be maximum , the projectile velocity has to be comprised between the two values approximately given by eqs . and . below @xmath54 penetration is not possible , and above @xmath55 penetration will decrease because of projectile erosion until reaching the hydrodynamic limit , given by the well known formula @xmath56 which always gives a smaller penetration depth than eq . , even when modified to take compressibility effects into account , e.g. , @xcite . the present earth penetrating version of the b61 , i.e. , the b61 - 11 , has a length @xmath57 m , and a maximum diameter @xmath58 m. its total weight is @xmath59 550 kg , about 200 kg more than the b61 - 7 , the model of which the b61 - 11 is a modified earth - penetrating version . further characteristics of the b61 - 7 and b61 - 11 are listed in table [ tbl:2 ] , and additional information can be found in references @xcite . assuming that the overall shape of the b61 - 11 can be approximated by an ovoid , its volume will be @xmath60 m@xmath44 , which means that the effective density of the b61 - 11 is @xmath61 mg / m@xmath44 about three times less than if it was made out of solid steel . let us suppose that the target material consists of concrete , for which we take the characteristics listed in the third line of table [ tbl:1 ] . for the sake of the argument , let us also suppose that eq . , which gives the hypothetical maximum penetration depth , is overestimating the actual penetration depth by a factor of @xmath62 , i.e. , the average value of @xmath45 listed in the last column of table [ tbl:1 ] . taking this factor into account , we therefore get the following expression @xmath63 } - 0.1),\ ] ] which agrees well with the data and the two empirical formulas given in ref . @xcite when @xmath64 . in eq . the penetrator s material enters by only one of its physical properties : the effective density @xmath65 . this is an elementary but important point which , by going through the physical reasoning leading to eq . , comes from the fact that for a perfectly hard and strong projectile the only thing that matters for penetration is its total kinetic energy . another trivial but important feature of eq . is that it contains the _ product _ of @xmath65 by @xmath0 , a quantity called `` opacity '' that appears in many physical problems . consequently the penetration is the same if the density of a projectile is doubled at the same time as its length is halfed , provided the penetration is measured from the surface of the target to the tip of the penetrator when it comes to a rest . this implies that is not necessary to know how the mass is distributed within the penetrator . for example , in the case of the b61 - 11 , one can suppose that its 550 kg are uniformly distributed over its 3.7 m of length , or else concentrated in the first half ( which is actually not far from being the case , as can be seen in table [ tbl:2 ] ) and assume that its length is only 1.85 m. the notion that the _ effective opacity _ @xmath66 is the decisive projectile - dependent parameter defining its penetration capability implies that knowledge of the construction details of the b61 - 11 , or any future rnep , is not necessary . for example , looking at the numbers in table [ tbl:2 ] , we see that out of the 203 kg mass difference between the b61 - 7 and b6 - 11 , 133 kg are in the front . to account for this difference several hypothesis can be made . for example , if we approximate the nose by a cone with base diameter @xmath67 , and assume that its made out of steel with density 7.9 mg / m@xmath44 , it would weight 146 kg . on the other hand , if we approximate the new penetration aid subassembly by a cylinder of diameter @xmath67 , and assume that its made out of a tungsten or depleted uranium alloy of density 18 mg / m@xmath44 , it would weight 153 kg . after properly adjusting the numbers , either option would lead to the same effective opacity , and thus the same penetration . however , for many reasons , such as flight stability and terminal guidance , neither of these options will be optimum . & * b61 - 7 * & * b61 - 11 * & * b61 - 7 * & * b61 - 11 * ' '' '' + subassembly & mass & mass & length & length + & [ kg ] & [ kg ] & [ m ] & [ m ] + nose & & & 0.65 & 0.65 + center & & & 1.15 & 1.15 + penetration aid & & & & 0.10 + nose + center + penetrator & 243 & 377 & 1.80 & 1.90 + preflight controller & 23 & 23 & 0.30 & 0.30 + tail & 80 & 150 & 1.50 & 1.50 + * total * & * 346 * & * 549 * & * 3.60 * & * 3.70 * + let us now consider the penetration capability of the b61 - 11 . if it is dropped from a height of @xmath68 m , as in typical drop tests @xcite , its maximum ground impact velocity neglecting air drag would be @xmath69 where @xmath70 is the acceleration of earth s gravity . consequently , at a terminal velocity of 0.5 km / s , eq . gives a penetration into concrete of 2.4 m , i.e. , @xmath71 , . ] in agreement with the advertised capability of the b61 - 11 . how could this penetration capability be increased ? obviously , one could increase the impact velocity by fitting the bomb with a rocket engine , and possibly reach the maximum set by the strength of the bomb casing , i.e. , eq . where one has to divided by 3 to be consistent with eq . for example , if the bomb s casing is made out of steel so that @xmath72 gpa , gpa . ] that maximum would be @xmath73 , i.e. , approximately @xmath74 m in concrete quite far away from the goal of 30 m penetration in concrete or hard rock mentioned for the rnep in the general scientific literature @xcite . before discussing a number of possible designs which could lead to greater penetration depths , it is useful to make a few comments based on elementary physical considerations , which explain how the b61 - 7 warhead contained in the b61 - 11 can withstand the enormous deceleration of impact and penetration . this is particularly important since any possible future rnep will most probably also use this warhead , because in the absence of nuclear tests an entirely new fission - fusion warhead can not easily be developed , and would not have the same deterrence value as a fully tested warhead . in order to have a quantitative characterization of the robustness of the b61 - 7 , it is necessary to find an estimate of the magnitude of deceleration during impact . a first approximation to this is obtained by assuming that after free - fall from a height @xmath75 a perfectly strong penetrator of mass @xmath76 and length @xmath0 is decelerated by a constant acceleration @xmath77 . the equation @xmath78 gives then the estimate @xmath79 for @xmath68 m and @xmath80 m this gives a mean deceleration @xmath81 . is it possible that a nuclear warhead can withstand such a large mean acceleration ? the answer is certainly yes since earth penetrating warheads and nuclear artillery shells have been designed and deployed in the 1950s already . this gives an opportunity to make an estimate for the maximum acceleration during impact , and to compare that estimate with what is known about the robustness of nuclear artillery shells . suppose therefore that the ram pressure of a projectile striking a target ( both assumed to be perfectly strong ) is @xmath82 , and that the pressure at impact can be written as @xmath83 where @xmath1 is the projectile s cross - sectional area . in the case of an artillery shell elementary ballistic physics shows that the projectile s velocity is related to the maximum range by the relation @xmath84 . thus @xmath85 where @xmath0 is now the length of the shell assumed to be a cylinder . if we take for example the w82 warhead , i.e. , the `` neutron bomb '' artillery shell , @xmath86 m and @xmath87 km , giving a maximum acceleration @xmath88 at impact . in fact , by time symmetry , the argument leading to that estimate can be reversed and interpreted as the maximum acceleration during launching of the shell by an artillery gun , which in the case of the w82 is known to be 15,000 @xmath70 @xcite . acceleration at launching '' @xcite . ] our estimate is therefore reasonable . how can a nuclear warhead survive an acceleration or deceleration on the order of 15,000 @xmath70 ? clearly , the warhead should contain a minimum of moving parts and all components must be mounted securely to resist shocks . therefore all free spaces have to be filled by a strong light - weight material such as polyurethane or beryllium , of such a density that it is compatible with the functioning of the warhead . moreover , since the crushing pressures are longitudinal , the warhead s design should if possible be cylindrical rather than spherical . finally , if the warhead is enclosed in a heavy penetrator , shock absorbers can be used to decouple it form the penetrator , and thus to reduce the stresses . for example , the early version of the b61 shown in fig . [ fig:2 ] has a levitated fissile pit that must be precisely positioned at the center of the reflector and high - explosive shells surrounding it . because of the relatively high densities of uranium or plutonium , such a design can not resist very large shocks , which is why most versions of the b61 are `` lay - down bombs , '' i.e. , bombs that are slowed - down by a parachute and ment to explode once they are at rest on the ground . one the other hand , the w48 artillery shell developed between 1957 and 1963 shown in fig . [ fig:3 ] is based on a hybrid gun - assembly / implosion design with a diameter of less than 15 cm . this cylindrical design , code - named `` manticore '' in ref . @xcite , was first tested in 1955 and opened the way to the development of cylindrical primaries for two - stage thermonuclear weapons . it is therefore most probable that cylindrical primaries are used in the warheads of the w82 and b61 - 7 , and therefore in the b61 - 11 . a french experiment related to such primaries is shown in fig . [ fig:4 ] . the robust nuclear earth penetrator ( rnep ) program is an engineering feasibility study under way since may 2003 in the united states , with the objective to determine if a more effective earth penetrating weapon could be designed using major components of existing nuclear weapons . this is a quite ambitious objective because the engineering aspects of the problem are very complex . in this section we focus on a less difficult objective , i.e. , to determine if it is scientifically possible to envisage a rnep capable of penetrating 30 m of concrete @xcite . specifically we will consider four possibilities , and discuss them in order of increasing sophistication , which in the medium to long term may correspond to increasing order of plausibility . according to eq . , the most simple approach to increasing penetration is to maximize the effective opacity because when dropped from a given height the terminal velocity does not depend on the penetrator s mass or size ( as recalled by eq . ) . this approach was actually taken in 1991 at the end of the gulf war to destroy a well - protected bunker north of baghdad after repeated directed hits . in that case an earth penetrating bomb , named gbu-28 , was hastly created using a 3.9 m - long , 36 cm - diameter , surplus army eight - inch - bore gun tube filled with high - explosives @xcite . by taking densities of 7.9 and 1.9 g / cm@xmath44 for the steel and the high - explosives , respectively , the weight of the gbu-28 comes out to be about 2,300 kg , and its effective density approximately 6,000 kg / m@xmath44 . for a length of 3.9 m and a terminal velocity of 0.5 km / s the penetration in concrete given by eq . is 5.9 m , in good agreement with the improved version , gbu-28/blu-113 , which is claimed to `` break through 7 meters of concrete or 30 meters of earth '' @xcite . moreover , the same approach was already taken in the 1950s for designing earth - penetrating nuclear weapons @xcite . for instance , `` the uranium gun - type mark 8 bomb ( nicknamed ` elsie ' for lc , or light case ) was almost 3 meter long , 36 cm in diameter , 1,500 kg , and had a yield of approximately 25 _ kt_. ( ... ) the mark 11 was an improved version of the mark 8 , slightly heavier , and according to the national atomic museum , ` able to penetrate up to 6.6 m of reinforced concrete , 27 m of hard sand , 36 m of clay , or 13 cm of armor plate , ' and fuzed to detonate 90120 seconds after penetration '' @xcite . in other words , the characteristics of these 1950s - bombs are very similar to those of the gbu-28 , except for the nuclear warhead . using the same method one can easily propose a design for a weapon able to break through 30 meters of concrete , i.e. , five times more . first , one can replace the steel tube by one made out of a strong tungsten or uranium alloy with densities of @xmath90 g / cm@xmath44 and get this way an effective density possibly as large as 15,000 kg / m@xmath44 . ( for this purpose a tungsten heavy alloy will most probably be preferred because it has similar density and yield strength , but a three times larger elastic modulus @xmath91 gpa , than the best uranium alloy @xcite . ) second , in order to get the desired factor of five increase in effective opacity , one can double the length of the penetrator so that @xmath92 m , and the total weight becomes @xmath93 kg . will such a design work ? most probably yes , provided the 8-meter - long , 10-ton - heavy penetrator can be delivered and properly guided to the target , which seems possible since its aspect ratio of @xmath94 is similar to that of modern kinetic - energy antitank penetrators @xcite . but what about the deceleration on impact will it differ from that of b61 - 11 for which eq . gave an average value of 6,000 g assuming a penetration @xmath80 m ? as a matter of fact , no : for a penetration of 30 m eq . actually gives a mean deceleration of only 400 @xmath70 . indeed , the maximum impact pressure , which can be estimated by eq . , does not depend on the penetration depth or the penetrator s weight , only on its length @xmath0 so that the mean deceleration is smaller for a greater penetration depth . moreover , by going to the penetrator s rest - frame , one sees that the target is always impacting the penetrator with the ram pressure @xmath95 , which does not depend on any of the penetrator s characteristics except velocity . thus , a warhead that survived penetration after free - fall from a given height , will also work for a heavier and longer penetrator able to penetrate deeper into the ground after falling from the same height . in summary , by simply increasing the effective opacity of the penetrator one can in principle achieve any desired penetration , and the existing b61 - 7 warhead can be used as the explosive charge with only minor modifications . however , the actual delivery , guidance , and successful penetration of the resulting very - heavy and very - long penetrator will pose considerable engineering problems , which are out of the scope of the present paper . as shown by eqs . or , the second general possibility to increase penetration is to augment the impact velocity . this can be done for example with a rocket - engine fitted to the bomb , which has the advantage that the weapon could be released by a low - flying aircraft rather than dropped from a stratospheric bomber . however , it is not possible to increase the impact velocity beyond the material limit set by the condition that the penetrator has to remain intact upon impact and during penetration , a limit that is defined by the yield ( or compressive ) strength of the penetrator casing . moreover , as will be shown below , increased velocity also means increased maximal impact deceleration , which results in severe constraints on the nuclear warhead and its ancillary components . for these reasons , a realistic design will most probably be a compromise in which the impact velocity is increased at the same time as the effective opacity of the penetrator . however , in order to simplify the argument , we will assume in this subsection that the primary free parameter is the impact velocity , and that everything is done to make its increase possible . let us rewrite eq . in the form @xmath96 where the ` proportional to ' symbol has been introduced in order to stress that what we are interested in is the scaling of the penetration depth with the materials s properties , and where the maximum initial velocity consistent with the compressive strength of the penetrator is approximately given by eq . , i.e. , @xmath97 let us consider again a concrete target , and suppose that we accept as a starting point a 4 m - long steel penetrator similar to the gbu-28 . since it is capable of penetrating 6 m of concrete at an impact velocity of 0.5 km / s , we ask for a five time increase in velocity to @xmath98 km / s in order to reach @xmath99 m according to eq . . it then follows from eq . that the compressive strength of the casing should be at least equal to 8 gpa , i.e. , about three times larger than the strongest steel alloy available , maraging steel ( mar 350 ) which has a yield strength of 2.7 gpa . we have therefore to consider a material different from steel for the casing ( or at least for its outer skin ) . if for example we take a ceramic of the type used in tank armor ( * ? ? ? * table 2 ) , we find that either boron carbide ( b@xmath100c , @xmath101 , @xmath102 gpa , @xmath103 ) or silicon carbide ( sic , @xmath104 , @xmath105 gpa , @xmath106 ) could be used provided their lower density is compensated for by replacing some of the steel in the penetrator by tungsten or uranium in order to keep the same effective density . therefore , provided the penetrator could be jacketed with a ceramic or some other superhard and elastic material , it may survive the factor of @xmath107 increased impact pressure implied by a 5 times larger penetration depth . but what about the the nuclear warhead and its ancillary components ? let us rewrite eq . which gives an estimate of the maximum deceleration of a perfectly hard and rigid projectile in terms of @xmath108 instead of the ballistic range , i.e. , @xmath109 with @xmath110 m and @xmath111 km / s , this is 31 km / s@xmath112 , a deceleration smaller than the 15,000 @xmath70 that we can assume the b61 - 7 can survive . however , at a velocity @xmath113 km / s , the maximum deceleration becomes 80,000 @xmath70 , i.e , significantly larger than this value . returning to the analysis made in sec . [ rob:0 ] in which we concluded that all the major nuclear components in the b61 - 7 are axially symmetric and strongly held in position , we can infer that at least all its metallic ( uranium , plutonium , aluminum , etc . ) and homogeneous ( beryllium , lithium - deuteride , etc . ) components may be able to survive a deceleration of possibly up to 100,000 @xmath70 . on the other hand , the high - explosives ( which have a heterogeneous and fragile constitution ) are known to become brittle and to crack at sustained accelerations greater than 30,000 @xmath70 @xcite . this can cause significant problems such as premature ignition and non - perfectly - symmetrical implosion . moreover , as is well known , the many ancillary components that are necessary for the proper functioning of a nuclear weapon are in fact much more fragile than its main nuclear components . these include the safety and security devices , the neutron generators , fuzes , firing systems , high - explosive detonators , etc . it is primarily the resistance to failure of these key components that has to be improved to make them compatible with the increased stresses stemming from a twenty - five - fold increase in impact stress . at that point we see that if deeper penetration has to succeed by going to higher impact velocities , the primary implication is that considerable progress has to be made in material science ( i.e. , to produce new superhard materials suitable for strengthening the weapon s casing , and new high - explosives than can sustain greater accelerations ) and in micro - electromechanical engineering ( i.e. , to build stronger ancillary components ) . in both of these engineering domains , the current `` buzz word '' is nanotechnology @xcite , which is a term emphasizing that improving materials is mainly to be able to design them by controlling their structure at the atomic or molecular level ( better alloys , composite materials , and chemical explosives ) ; and to be able to reduce as much as possible the linear size of all moving components because ( at a given yield stress ) this is the easiest way to improve resistance to mechanical stress . indeed , if using elementary elasticity theory we calculate the bending of rod ( fixed at one extremity ) of length @xmath114 and density @xmath115 under the effect of an acceleration @xmath116 one finds the relation @xmath117 which implies that smaller objects have a larger resistance to stress than larger ones . in summary , by increasing the impact velocity by a factor of five one is reaching a number of material and engineering limits which are at the forefront of contemporary research and development . if a compromise is made and the earth penetrating weapon is made somewhat heavier and longer than the circa 2,000 kg and 4 m we took as a starting point , the resulting weapon could not be much lighter and smaller than the design considered in the previous subsection , especially if we take into account the weight and volume of the rocket motor required to increase impact velocity . after having considered the possibility of higher velocities and `` wrapping the weapon in some superhard new nanomaterial , '' the third general possibility to increase penetration is to use `` repeated explosions to clear the path '' through the ground @xcite . this possibility is based on the idea that weakening a target before impact could increase penetration depth . in conventional weaponry this is the rationale of `` tandem '' or `` two - stage '' warheads in which a precursor shaped - charge jet is followed by a kinetic - energy projectile containing a high - explosive filling @xcite . to evaluate the influence of a precursor hole , the best is to first assume that the hole has been `` pre - drilled , '' i.e. , that some target material has simply been removed before impact of the penetrator , but that otherwise the target material has not been weakened . in that case the theory developed in sec . [ fun:0 ] still applies , provided one corrects the penetration given by eq . in inverse proportion to the ratio of the area of the circle defined by the projectile radius @xmath116 , and that of the annulus of outer radius @xmath116 and inner radius @xmath118 , the pre - drilled hole s radius @xcite , @xmath119 this expression has been found to be in reasonable agreement with various experiments @xcite , which showed that it generally tends to underestimate the actual penetration depth when @xmath120 , the relative radius , is larger than @xmath121 , i.e. , when the pre - drilled hole radius is not much smaller than the penetrators s radius . thus , as the increased penetration given by eq . depends on the square of the relative radius , a small pre - drilled hole has only a very small influence on penetration depth . it is therefore important to evaluate the influence of the target s material weakening when , for instance , the hole is made by means of a precursor shaped - charged and the material is damaged several crater radii away from axis during penetration of the jet . this can be done by integrated experiments ( provided they can be made ) , by separated experiments in which the shaped charge and the kinetic penetrator strike the target in different experiments , or by computer simulations calibrated with data from such experiments . for example , in one such simulation backed by experiment @xcite , it was found that whereas the penetration depth of a caliber 60 mm projectile impacting a concrete target at a velocity of 0.51 km / s is 115 cm , the same projectile striking the target pre - drilled and weakened by a shaped charged charge jet to a depth of 60 cm will exit a 140 cm thick target with a residual velocity of 0.23 km / s . it is therefore possible to aid kinetic - energy penetration by means of a precursor shaped charge , but the improvement is not dramatic . in particular , leaving aside the complex interactions ( such as slowing - down effects ) of the precursor charge on the follow - through projectile , one may estimate that a `` tandem '' penetrator may have twice the penetration capability of a kinetic - energy penetrator on its own . if now the process is repeated and a hypothetical second shaped charge is detonated underground ( once a first shaped charge followed by a first penetrator have opened the way to some depth ) one could go further down but at the expense of great complexity and a very uncertain design . finally , if we consider a single tandem such that a shaped charge would aid the penetration of a weapon containing a nuclear warhead like the b61 - 7 , which has a minimum diameter of about 30 cm , the precursor hole should have a diameter of at least 5 to 10 cm over a length of more than 5 m to have some positive effect . this implies a very large and powerful shaped charge jet , which is very difficult to create using conventional means , because tens ( or possibly a few hundreds ) of kilograms of high - explosives may be required to form and drive such a jet . the fourth general possibility to increase penetration is to look at the contemporary trend in nuclear weapons development , and to see whether some radically new technology could be on the design horizon that would complete change the rules of the game . as a matter of fact , since many years and more so since the conclusion of the comprehensive test ban treaty , nuclear weapons s testing has moved into laboratories where huge laser facilities are used to better understand the ignition and burn of small thermonuclear explosions . the two largest such facilities are under construction in the united states ( the national ignition facility , nif ) and in france ( the laser mgajoule , lmj ) . but quite powerful facilities are already or will be soon available in japan , england , china , russia , germany , etc . the full understanding of the ignition and burn of small pellets of fusion material ( the so - call inertial confinement fusion , icf , route to thermonuclear energy ) will open the way to a radically new generation of nuclear weapons in which there will be no , or very little , fissile material @xcite . these `` fourth - generation nuclear weapons '' ( fgnw ) are a great concern for nuclear proliferation because they can potentially be built by all technologically advanced countries , including non - nuclear - weapon states , and because they are militarily `` very sweet '' as they are very powerful and politically usable since they do not have some of the major drawbacks of existing types of nuclear weapons : radioactive fallout and too much yield . for the purpose of the present paper we will assume that such fourth - generation nuclear explosives can be built , and that such explosives with yields in the range of 1 to 1000 _ tons _ of high - explosive equivalent could become available at most a few decades from now . we will also simplify the discussion by assuming that these explosives can be approximated by a point source of high - energy ( i.e. , 14 mev ) neutrons , even though they will most probably consist of spherical or cylindrical devices with a radius of at least a few centimeters . the availability of such powerful and compact sources of neutrons leads to many military applications other than just explosives . this is because the neutrons can be used to work on some material which can be accelerated ( as by a rocket engine ) or forged and projected ( as by a shaped charge ) @xcite . however , because of their small size ( say a diameter of 5 cm or less instead of the 15 cm of the smallest nuclear artillery shells ) , fourth - generation nuclear warheads could already take more easily advantage of a precursor chemical - explosive shaped - charge in penetrating and defeating a bunker or buried target than current nuclear warheads . in the context of the present problem of driving a powerful explosive deep into the ground before detonating it , fgnw technology gives the possibility to envisage active earth - penetrating devices that could be both compact and lightweight . although the technological and engineering hurdles could be formidable , it is essential not to be shy because the consequences can be far - reaching , and further nuclear weapons proliferation is at stakes . for example , one can think of a tandem weapon in which a front pure - neutron explosive of 10 _ tons _ would create a powerful shaped charge jet , and a back pure - neutron explosive of 100 _ tons _ would drive a package containing the main nuclear explosive charge into the hole formed by the jet . the development of such a complex device would require very powerful computer simulation capabilities , and considerable interaction with reduced scale experiments at icf facilities . in fact , the development of fgnws such as the earth - penetrating device just described might be a far more credible justification than `` stockpile stewardship '' for the enormous computer facilities made available to the nuclear weapons laboratories ... in summary , new types of nuclear explosives in which the main output of energy is in the form high - energy neutrons may lead to entirely new solutions to the deep earth penetration problem . however , for these solutions to become a reality , considerable research and development is required to improve the non - nuclear components of the penetrators as well , just like for the more conventional approaches to increased penetration , as was seen in the previous subsections . this implies that the rnep program is consistent with the development of fgnws and their use in advanced earth penetrating weapons . the environmental consequencies of shallow buried nuclear explosions have been addressed in a number of recent studies , and it has been convincingly argued that insufficiently deep burial of an earth penetrating warhead will lead to extensive radioactive pollution of the environment @xcite . for example , the minimum depth of burial required to fully contain a 1 _ kt _ nuclear explosion at the nevada test site is about 100 m. the implication for the case under consideration in this paper , i.e. , a penetration depth of about 30 m in concrete or rock , or about 100 to 150 m in sand or earth , is that the nuclear yield should be less than 1 _ kt _ if radioactive fallout is to be avoided . therefore , if a buried target can only be defeated by an explosive yield on the order of the maximum that can delivered by the b61 - 7 , i.e. , 340 _ kt _ , one can not expect that this explosion will be contained . the use of a rnep will thus have the same environmental and political consequences as any atmospheric nuclear explosion , and will not in any way be comparable to a kind of `` nuclear test explosion on enemy territory in which there would be no visible mushroom cloud . '' another seldom mentioned political limitation to the use of earth penetrating nuclear weapons arises even if the explosions are fully contained . this is because any nuclear weapon detonated below the surface of the earth creates a radioactive hot spot , containing the highly - radioactive fission - products , as well as the fissile materials that have not been fissioned during the explosion that is more than 90% of the initial fissile - materials content of the warhead . this leads to numerous problems , especially if the weapon contained plutonium . indeed , most of the un - fissioned plutonium left in the cavity created by an underground nuclear explosion can later be recovered , possibly by terrorists , and reused to manufacture nuclear weapons . this is the problem of _ `` plutonium mines , '' _ which are a major nuclear proliferation concern in countries such as kazakhstan @xcite and algeria , where numerous plutonium - based nuclear explosives were detonated underground . consequently , earth penetrating warheads like the b61 - 7 use highly - enriched uranium as fissile material , because in that case the un - fissioned u-235 can not be recovered after the explosion as it will be intimately mixed with the u-238 from the non - enriched uranium components of the weapon . the only plutonium will be the small amount bred by neutron capture in u-238 . on the other hand , the un - burnt fraction of the tritium used for boosting is much less a problem . first because tritium is very volatile and easily drifts away from the point of explosion , second because its half - life is only twelve years , and third because its radio - toxicity is relatively small . for these reasons , in the case of fgnws , the related problem of `` tritium mines '' is potentially much less serious than the problem of `` plutonium mines . '' the paper was organized to begin with a derivation , from first principles , of the fundamental equation giving to first order the penetration depth of a rigid penetrator into a solid material . this derivation was necessary in order to confirm that in the parameter regime of interest to defeat well - protected underground targets there is very little room for improvement outside the main parameters which appear explicitly in eq . . in particular , fine tuning parameters such as the shape of the penetrator , or its surface texture , will only have second order effects on penetration depth , at the level of 10% or less . we have then reviewed the physical and engineering reasons which explain the penetration depth of the main earth - penetrating nuclear weapon currently available in the united states s arsenal , the b61 - 11 , and derived crude estimates in terms of maximum sustainable decelerations of the ultimate impact - resistance of the b61 - 7 , the warhead enclosed in the b61 - 11 penetrator . this has led us to consider four different possibilities for designing a rnep able to bury the b61 - 7 warhead 30 meters into concrete , or 100 to 150 meters into earth . the first solution is simply to build a 8-meter - long , 360-mm - caliber , 10-ton - heavy penetrator made out of a good tungsten alloy . this would be a `` 1950s generation '' style weapon , albeit fitted with a modern two - stage thermonuclear warhead , and a gps guidance system allowing for high precision delivery under all weather conditions . little sophisticated engineering development would be required to produced the penetrator , and no change would have to be made in the b61 - 7 warhead : the version carried by the b61 - 11 could be used without modification . however , considering the weight and size of that first solution , we have investigated the possibility of augmenting the impact velocity by means of a rocket engine . this immediately led to contemplate considerable engineering difficulties , because while the major nuclear components could probably resist the increased stresses arising from the greater impact and penetration decelerations ( with the exception of the chemical explosives which may have to be improved ) , many of the more fragile ancillary non - nuclear components of the weapon would have to be redesigned , using new materials and techniques which are not yet readily available . moreover , even by making a compromise such that the penetrator would not be so large and heavy as the one considered in the first possibility , fitting it with a suitable rocket engine would not make it much less cumbersome . we have therefore turned to the possibilities of aiding penetration by means of a precursor shaped - charge . here we have met again with considerable engineering difficulties , especially since designing a shaped charge powerful enough to clear the way to a circa 30 cm diameter penetrator containing a b61 - 7 warhead looks very difficult , and this only to augment its penetration capability by possibly 50% under the best conditions . the last possibility we have considered is more speculative : assuming that radically new types of compact nuclear explosives will become available in the next few decades ( i.e. , low - yield pure - fusion thermonuclear explosives derived from icf pellets which are extensively studied at present in many laboratories ) speculations have been made on hypothetical new designs which could provide a technically more attractive solution than the three previous possibilities . in particular we have taken the example of an earth - penetrating weapon which would be equipped with a powerful thermonuclear - driven precursor shaped - charge , followed by a main charge that would be accelerated and driven into the ground by the momentum supplied by a 100 _ tons _ thermonuclear explosion at the back of it . the main conclusion stemming from considering this more speculative possibility is that the ancillary technological advances required in order to make it feasible are similar to those implied by the second possibility , i.e. , considerable advances in the realms of materials science , micro - electromechanical systems s engineering , and nanotechnology . since these potential advances significantly overlap with the requirements of many other future weapons , both conventional and nuclear , it is therefore possible that an important reason for considering an improved version of the b61 - 11 is simply to develop these technologies as they will be needed , for instance , to weaponize fourth generation nuclear explosives . finally , we have briefly looked at some major environmental and political implications of the actual use of earth - penetrating nuclear weapons . since these aspects have been discussed at length in many recent publications therefore reviving the still open debate on the arguable military utility and political roles of nuclear weapons we have not gone further than mentioning that even if an underground nuclear explosion is fully contained , the site of the explosion will in due time have to be clean - up , in order to avoid long - term contamination and further nuclear proliferation problems if someone tries to recover the un - burnt fissile materials remaining after the bomb s explosion . the author is indebted to david hambling for stimulating correspondence , to john coster - mullen for several contributions to the iconography , and to dr . jean - pierre hurni and many others for their help , ideas , and criticism . a reviewer has asked why there was no explicit reference in this paper to the `` young / sandia penetration equations '' which are widely used in the u.s . weapons community , i.e. , ref . @xcite and references therein . the main reason for this was that these equations are in fact empirical engineering formulas providing a good estimate for target penetrability in materials for which measurements have been made using projectiles that are typical of those in existence in u.s . arsenals that is , regularly updated formulas providing a simple and good fit to an expanding data base . for these reasons the `` young / sandia penetration equations '' can not reliably be used outside of the range of materials , projectiles , and parameters corresponding to this data base . nevertheless it is useful to compare the approximate but physically well understood formula derived in sec . [ fun:0 ] , i.e. , eq . in the case of a typical concrete target , to the main `` young / sandia penetration equation , '' i.e. , eq . ( a-3.2 ) on page a-2 of @xcite . this equation can be written @xmath122 } - 0.0305),\ ] ] where for the sake of comparison the velocity is measured in km / s as in eq . . in this formula , the exponent of the projectile s opacity ( `` weight to area ratio '' in engineering language ) has evolved for 0.5 in the 1967 version to 1.0 in the 1988 version , see ref . @xcite and eq . ( 2 ) in ref . @xcite , to finally become 0.7 in the current version , i.e. , eq . . however , assuming for example a `` nose performance factor '' of @xmath123 for a moderately sharp nose , and a `` s - factor '' of @xmath123 for a relatively hard concrete , we get @xmath124 as in our eq . . therefore , in that case , for a projectile of length on the order of a fraction or a few meters , eq . and eq . will given comparable estimates for penetration depth . it can therefore be concluded that eqs . and will give comparable results for typical standard projectiles and target materials , but that the `` young / sandia penetration equation '' will typically be more accurate in these cases . however , if one considers much larger velocities or projectile lengths , or else very untypical materials , eq . may give more reliable estimates , especially if the goal is to compare relative penetration performances rather than absolute penetration depth estimates . finally , reference @xcite and the `` s - factor '' are quite useful in providing first order estimates for the typical relative penetration depths into various natural materials such as rock , sand , clay , soil , or ice . in particular , if a moderately soft concrete can be characterized by @xmath125 , ice and frozen soil may be characterized by @xmath126 , hard soil by @xmath127 , and soil fill ( as above a buried bunker ) by @xmath128 , etc . d.l . orphal , _ calculations of explosion cratering ( i ) the shallow - buried nuclear detonation johnie boy _ , in d.j . roddy et al . , eds . , impact and explosion cratering ( pergamon press , new york , 1977 ) 897906 . canfield and i.g . clator , _ development of a scaling law and techniques to investigate penetration into concrete _ , nwl report no . 2057 ( u.s . naval weapons laboratory , dahlgreen , va , august 1966 ) . cited in fig . 7 of ref . @xcite . r. continliano and c. donaldson , _ the development of a theory for the design of lightweight armor _ , report affdl - tr-77 - 114 , document ada054046 ( aeronautical research associates of princeton , princeton , nj , 1977 ) 80 pp . a. gsponer , _ depleted - uranium weapons : the why s and wherefores _ , in a. gut , b. vitale and p. low , depleted uranium ( spokesman , nottingham , 2003 , isbn 0 - 85124 - 685 - 0 ) 126158 , e - print arxiv : physics/0301059 available in pdf format at . the information division , albuquerque operations office , _ developing and producing the b61 _ , video tape 0800072 ( us atomic energy commission , dated `` 1970s , '' sanitized and declassified march 1998 ) 26:29 minutes . p. de brem , _ airix : radiographier la matire en un clair _ , les dfis du cea , no.96 ( juin - aot 2003 ) 1819 ; see also , j .- leyrat and x. clment , _ le programme simulation : les diagnostiques ultra - rapides , cls des exprience de dtonique _ , les clfs du cea , no.47 ( hiver 2002 - 2003 ) 6263 . p. lundberg , interface defeat and penetration : two modes of interaction between metallic projectiles and ceramic targets , dissertation no . 1033 , isbn 91 - 554 - 6077 - 1 ( uppsala university , se-751 20 uppsala , sweden , 2004 ) 54 pp . available at . magness , and t. g. farrand , _ deformation behavior and its relationship to the penetration performance of high - density kinetic energy penetrator materials _ , proceedings of the 1990 army science conference ( durham , nc , may 1990 ) 465479 . w. lanz , w. odermatt , dr . g. weihrauch , _ kinetic energy projectiles : development history , state of the art , trends _ , 19th international symposium on ballistics ( interlaken , switzerland , 711 may 2001 ) 11911197 . available in pdf at . n. heider and s. hiermaier , _ numerical simulation of the performance of tandem warheads _ , 19th international symposium on ballistics ( interlaken , switzerland , 711 may 2001 ) 14931499 . available in pdf at . teland , _ cavity expansion theory applied to penetration of targets with pre - drilled cavities _ , 19th international symposium on ballistics ( interlaken , switzerland , 711 may 2001 ) 13291325 . available in pdf at . a. gsponer and j .- p . hurni , the physical principles of thermonuclear explosives , inertial confinement fusion , and the quest for fourth generation nuclear weapons , inesap technical report no.1 , presented at the 1997 inesap conference , shanghai , china , 8 - 10 september 1997 , seventh edition , isbn : 3 - 9333071 - 02-x ( september 2000 ) 195 pp . abstract available at . karraker and w.c . perkins , _ recovery of actinides produced by nuclear explosions in salt formations _ , report dp-956 ( savannah river laboratory , june 1965 ) 47 pp ; _ recovery of actinides produced by nuclear explosions in volcanic tuff or desert alluvium _ , report dp-1055 ( savannah river laboratory , december 1966 ) 18 pp . siddiqui and h. abbas , _ reliability analysis of projectile penetration in geological targets _ , transactions , smirt 16 ( washington dc , august 2001 ) 8 pp . available at . g. valentine , _ defeating underground targets _ , los alamos science , no . 29 ( los alamos national laboratory , 2005 ) 68 .
using hydrodynamical zoom simulations in the standardcdm cosmology , we investigate the evolution of the distribution of baryons ( gas and stars ) in a local group - type universe . + first , with standard star formation and supernova feedback prescriptions , we find that the mean baryonic fraction value estimated at the virial radius of the two main central objects ( i.e. the milky way and andromeda ) is decreasing over time , and is 10 - 15% lower than the universal value , 0.166 , at . this decrease is mainly due to the fact that the amount of accretion of dissipative gas onto the halo , especially at low redshift , is in general much lower than that of the dissipationless dark matter . indeed , a significant part of the baryons does not collapse onto the haloes and remains in their outskirts , mainly in the form of warm - hot intergalactic medium ( whim ) . moreover , during the formation of each object , some dark matter and baryons are also expelled through merger events via tidal disruption . in contrast to baryons , expelled dark matter can be more efficiently re - accreted onto the halo , enhancing both the reduction of inside , and the increase of the mass of whim outside . varying the efficiency of supernovae feedback at low redshift does not seem to significantly affect these trends . + alternatively , when a significant fraction of the initial gas in the main objects is released at high redshifts by more powerful sources of feedback , such as agn from intermediate mass black holes in lower mass galaxies , the baryonic fraction at the virial radius can have a lower value ( ) at low redshift . hence physical mechanisms able to drive the gas out of the virial radius at high redshifts will have a stronger impact on the deficit of baryons in the mass budget of milky way type - galaxies at present times than those that expel the gas in the longer , late phases of galaxy formation . galaxies : local group galaxies : haloes dark matter methods : n - body simulations
1205.4694
the distribution of galaxies in the universe on large scales , as revealed by sky surveys such as the cfa redshift survey ( geller & huchra 1989 ) , the 2df galaxy redshift survey ( colless et al . 2001 ) or the sloan digital sky survey ( gott et al . 2005 ) is structured into a filamentary web in which the intersections correspond to the most massive objects ( i.e. galaxy clusters , bond et al . this specific distribution is now commonly and easily reproduced by cosmological hydrodynamical n - body simulations of the @xmath0cdm cosmology which also indicate that the distribution of dark matter , the dominant mass component of the universe ( komatsu et al . 2011 ) , follows the same trend . indeed , since in the traditional picture of galaxy formation , galaxies are supposed to form when baryonic gas falls into the gravitational potential of their host dark matter halo , the distribution of dark matter is therefore expected to faithfully trace that of the baryons . however on galactic scales , observations tend to suggest that the spatial distributions of dark matter and baryons ( especially in the form of gas ) may display some substantial differences . in particular , it has been shown that galaxies are missing most of their baryons , most galaxies are severely baryons - depleted relative to the cosmological fraction ( see for instance bell et al . 2003 ; hoekstra et al . 2005 ; mcgaugh 2010 ) . this so - called missing baryons problem " ( see bregman 2007 for a complete review ) , if _ real _ , calls for two alternative scenarios . either a significant part of the gas never collapsed into the gravitational potential wells of protogalaxies in the first place , or some of the gas has been expelled by galaxy formation feedback processes such as supernova winds . hence solving the missing baryon problem may prove to be central in order to constrain galaxy formation models . past studies based on hydrodynamical simulations indicate that most of the `` missing '' baryons might lie in a gaseous phase ( the so - called warm - hot intergalactic medium , hereafter whim ) in the temperature range @xmath5 @xmath6 and at moderate over - density ( cen & ostriker 1999 ; dav et al . 2001 ; cen & ostriker 2006 ) . according to these studies , the whim is primarily shock - heated during the formation of large - scale structures , while feedback mechanisms associated with star formation should also have an additional impact on this phase during galaxy formation . for instance , for low mass systems , the cosmological uv - background is supposed to reduce both the star formation and baryon content ( thoul & weinberg 1996 ; bullock , kravtsov & weinberg 2000 ; gnedin 2000 ; somerville 2002 ; ricotti , gnedin & shull 2002 ; benson et al . 2002 ; read , pontzen & viel 2006 ; hoeft et al . 2006 ; okamoto , gao & theuns 2008 ; peirani 2010 ; nickerson et al . 2011 , hambrick et al . supernova feedback is also expected to expel some gas , especially for low mass systems , as suggested by various numerical investigations ( scannapieco et al . 2008 , 2009 ; faucher - gigure , kere & ma 2011 and references therein ) . other mechanisms have been proposed , such as turbulence in the baryonic intergalactic medium ( zhu , feng & fang 2011 ) or pre - heating by pre - virialisation ( mo et al . 2005 ) , although crain et al . ( 2007 ) claimed that this process is unable to prevent the collapse of gas ( instead , according to these authors , non - gravitational feedback is required in order to reduce the efficiency of gas cooling and star formation in dwarf galaxies ) . another less investigated potential mechanism , which may also affect the evolution of the distribution of baryons on small scales , is the effect of accretion and merger events ( and in particular their associated tidal disruption ) that are expected to be frequent in the framework of the hierarchical model . for instance , in an early paper , navarro & white ( 1993 ) have shown that during a merger involving dark matter and adiabatic gas , there is a transfer of energy between these two components that leads to a situation where the gas is less tightly bound . more recently , sinha & holley - bockelmann ( 2010 , 2011 ) found that a few percent of gas can be driven into the intergalactic medium ( igm ) by galaxy mergers , using either merger trees calculated from the press - schechter formalism or idealized simulations of mergers of galaxy clusters . using hydrodynamical simulations , nickerson et al . ( 2011 ) have also concluded that tidal forces may cause significant mass loss from satellites of all masses . in the present work , we make use of cosmological `` zoom '' simulations with an extended treatment of the physics of baryons to study the formation of milky way - like galaxies . our aim is to characterise the relative role of supernova feedback to accretion and mergers in the evolution of the distribution of baryons for objects of such masses . we will also test scenarios in which a significant fraction of gas in progenitors is expelled at high redshift by more powerful sources of feedback , such as agn associated with massive black holes . this will allow us to quantify two distinct processes which may allow us to address the so - called missing baryon problem , should it persist . this paper is organised as follows . in section 2 , we summarize the numerical modelling , and section 3 presents our main results on the evolution of the distribution of baryons in our simulated local group universes . we give our main conclusions in the last section . the numerical methodology used in the present paper is described in detail in peirani ( 2010 ) to which we refer the reader for more information . for the sake of clarity , we summarize the main steps below . we analyse three cosmological zoom simulations for a @xmath0cdm universe using wmap5 parameters ( komatsu et al . 2009 ) , namely @xmath7 , @xmath8 , @xmath9 , @xmath10 km / s / mpc , @xmath11 and @xmath12 . each simulation was performed in a periodic box of side @xmath13 mpc with @xmath14 effective dark matter and gas particles in the highest resolution region ( a sphere of @xmath15 mpc of radius ) . in this region of interest , the mass resolution are @xmath16 and @xmath17 . the plummer - equivalent force softening adopted for the high mass resolution particles were 1 and 0.5 @xmath18 kpc for dark matter and gas particles respectively and were kept constant in comoving units . initial conditions have been generated from the mpgrafic code ( prunet et al . 2008 ) and the simulations were performed using gadget2 ( springel 2005 ) , with added prescriptions for metal - dependent cooling , star formation ( in this work we remind that the star formation efficiency is @xmath19=0.02 ) , feedback from type ia and ii supernovae ( sn ) , uv background ( starting at @xmath20 ) and metal enrichment . the three simulations have common initial conditions whose phases are consistent with the local group but essentially differ in the quantity of energy released by sn derived from star particles @xmath21 ( @xmath22 ) . as mentioned in peirani ( 2010 ) , we consider that a fraction @xmath23 of this energy is deposited in the @xmath24 neighbour gas particle by applying a radial kick to its velocity with a magnitude @xmath25 , where @xmath26 is the weighting based on the smoothing kernel and @xmath27 is the mass of gas particle @xmath28 . the first simulation @xmath29 uses the standard value of @xmath30 while in the second one , @xmath31 , we have considered a higher efficiency @xmath32 in order to investigate how our results would be affected . in the third simulation ( @xmath33 ) , our aim is to study the effects of earlier high energy ejection to the ism induced either by intermediate mass black holes or other high energy processes such as hypernovae events . intermediate mass black holes are the likely missing link between population iii and the supermassive black holes in quasars and plausibly as part of the hierarchy of structure formation . hypernovae are likely to be more prevalent than normal supernovae in the earliest phases of star formation : for example , the population iii imf is most likely top - heavy , and this may also be the norm for the precursors of spheroids , as evidenced for example by the radial distribution of sne in disturbed ( including recently merged ) galaxies ( habergham , anderson & james 2010 ) and the frequency of low mass x - ray binaries in ultracompact dwarf galaxies ( dabringhausen et al . . a higher frequency of hypernovae could plausibly provide an order - of - magnitude higher feedback efficiency than supernovae . the case of agn is intriguing , as even higher feedback is required in deep potential wells , where supernovae are relatively inefficient , in order to account for the correlation between black hole mass and spheroid velocity dispersion ( silk & nusser 2010 ; debuhr et al . moreover , the recent detection of a massive gas outflow in a distant quasar ( @xmath34 ) strongly suggests that a strong quasar activity is already in place at very early times ( maiolino et al . 2012 ) . for this purpose , a simple modelling was used in which a much higher efficiency ( @xmath35 ) was considered during a very short ( @xmath36myr ) at earlier times ( @xmath37 ) and @xmath30 otherwise . we also justify the choice of this high efficiency by simple energetics and momentum comparisons between supernovae and agn . for instance , for a @xmath38m@xmath39 black hole along with its @xmath40m@xmath39 in stars , the number of type ii sne produced in star formation phase ( @xmath41m@xmath39 per sn for a chabrier imf ) is @xmath42 . thus , the total sne energy injected is @xmath43ergs over a dynamical time , say @xmath44yr , or @xmath45ergs / s . since the eddington luminosity is @xmath46ergs / s energetics favour agn . furthermore , for supernovae momentum conservation starts at shell velocity @xmath47km / s . since the ejection velocity is @xmath48km / s , the momentum injected is @xmath49 of initial momentum namely @xmath50gmcm / s or @xmath51dynes . for agn , momentum injected is @xmath52 or @xmath53dynes . thus agn injects 100 times more momentum flux than supernovae . finally , there is another factor of 10 - 100 that favours agn , the so - called mechanical advantage factor due to ram pressure on the expanding bubble as suggested by recent hydrodynamical simulations ( wagner , bicknell & umemura ) . therefore the value of @xmath35 seems to be reasonable in order to account for the accumulation of the different potential sources of feedback at high redshifts . the feedback parameters are summarized in table [ table1 ] . .fractions @xmath23 of the energy released by supernovae [ cols="<,<,<,<",options="header " , ] [ table2 ] the time evolution of the virial , dark matter ( dm ) , gas and stellar masses of each object at the virial radius and for the three simulations is shown in figure[mass ] . first , we notice that there is no particular difference between the evolution of dark matter halo masses ( and hence viral masses ) between @xmath29 and @xmath31 . however , the baryonic compositions are quite different . indeed , stellar masses are significantly reduced in @xmath31 due to higher amounts of energy released by the sn . the mass of the gas component is therefore higher in @xmath31 , as expected . for @xmath33 , the stellar masses are even more reduced . this is mainly due to the fact that a significant fraction of the gas has been expelled at high redshifts . indeed , while the dm masses are quite similar between the 3 runs , the virial masses for objects in @xmath33 tend to be lower at high redshift , which indicate that a fraction of gas expelled at high z does not recollapse at some later stage . interestingly , @xmath54 undergoes a minor merger event at @xmath55 , which corresponds for instance to a sudden variation in the evolution of its stellar mass . this same accretion event takes place at @xmath56 in @xmath33 . moreover , we can also identify clearly in @xmath33 another minor merger event at @xmath57 that has not occurred ( yet ? ) in @xmath29 and @xmath31 . this explains why the dark matter mass ( and therefore the virial mass ) of @xmath54 is higher in @xmath33 than @xmath29 or @xmath31 . this also suggests that small perturbations in the properties of haloes at earlier times can have some significant impact on the whole accretion history ( see for instance , thiebaut et al . it is also worth mentioning that although the two main central objects have the same virial mass at @xmath1 , they have undergone different mass accretion histories . indeed , while @xmath54 grows regularly through smooth mass accretion , the evolution of the mass of @xmath58 is more sporadic thanks to several major merger events . it is therefore instructive to see to which extent these two different mass evolutions could affect the evolution of the distribution of baryons in the vicinity of these two objects . it is also worth studying how the stellar masses found in the different simulations are related to the mass of their host dark matter haloes . from table [ table2 ] , the mass of dark matter haloes varies from @xmath59 to @xmath60 while according to observational and numerical analysis , the expected stellar mass range inside those haloes is @xmath61\times10^{10 } m_\odot$ ] ( guo et al . 2010 ; moster et al . 2010 ; brook et al . in order to make a suitable comparison , instead of considering the stellar mass inside the virial radius @xmath3 , we define the galaxy radius as @xmath62 and all stars inside this radius contribute to the galaxy stellar mass @xmath63 . by doing this , we only select stars which belong to the central main galaxy while excluding stars of accreted satellites inside @xmath3 . in figure[mass ] , the evolution of each stellar mass inside @xmath64 is shown in dashed line . one can see that at @xmath1 , galaxy stellar mass ranges are @xmath65\times10^{11 } m_\odot$ ] , @xmath66\times10^{11 } m_\odot$ ] and @xmath67\times10^{11 } m_\odot$ ] for @xmath29 , @xmath31 and @xmath33 respectively . thus , each simulation tends to produce too many stars in the central region of haloes and the trend is more severe in @xmath29 and @xmath31 relative to @xmath33 . this problem was already pointed out from other hydrodynamical simulations which indicate that low feedbacks may result in more than an order - of - magnitude too many stars ( see for instance piontek & steinmetz 2011 ; brook et al . 2012b ) . as a first conclusion , stronger and earlier feedbacks seem to be required in order to improve the stellar mass - halo mass relation ( also recently suggested by stinson et al . 2012 ) and , in the present study , one probably also needs the action of additional sources of feedback in the late phase of galaxy formation such as agn to get even closer to the expected galaxy stellar masses . moreover the fact that our model leads to too high central stellar masses suggests that the effect of all different sources of feedback has been underestimated . thus in the following , baryonic fraction values found in our simulations may be slightly overestimated . finally , figure[fig_sfr ] shows the evolution of the star formation rate ( sfr ) for each object and simulation . at @xmath1 , sfr values are around @xmath68 $ ] @xmath69yr which is consistent with observational values derived by robitaille & whitney ( 2010 ) using spitzer data . note that these cosmic evolutions are similar for the two haloes : at high redshifts ( @xmath70 ) the lower the sn efficiency the higher the stellar mass . and the higher the mass of formed stars at high redshift , the higher the subsequent sn feedback . this explains why , at low redshifts , a higher fraction of stars is produced in @xmath33 relative to @xmath29 and @xmath31 . fig.[distribution1 ] shows the projected distribution of baryons in our lg type universes derived from @xmath31 and @xmath33 and at four specific epochs ( i.e. @xmath71 and @xmath72 ) . we clearly see that the evolution of the distributions of baryons and dark matter do not follow the same trend . indeed , the regions of the universe in red and dark blue correspond to regions where the baryonic fraction is lower than the universal value . and those specific regions are mainly located around the forming protogalaxies or galaxies at each redshift , while in the very inner part of dark matter haloes , @xmath2 is higher . this is an expected result : due to cooling , the gas collapses to the center of haloes where stars can be formed . however , note that the size of these `` red '' cavities increases over time which suggests that the fraction of gas that has collapsed to the center of the halo is not immediately replaced by some fresh gas from its vicinity . note also , at high and low redshift , the existence of relative high baryon content regions which are located either in the filaments or beyond the virial radii . such anisotropic distribution seems to be more pronounced in @xmath33 relative to @xmath31 . if the high baryon fractions in filaments can be understood by the dissipative nature of gas , allowing it to cool to the dense filaments , it is crucial to characterise the mechanisms that drive high baryon content regions beyond the virial radii . 0.5 cm let us first only consider the role of supernovae feedback in the evolution of the mass budget of milky way - type galaxies . we therefore focus on both @xmath29 and @xmath31 in this sub - section . the cosmic evolution of the baryonic fraction @xmath2 estimated at the virial radius for the two main haloes is shown in figure[fb_evolution ] . in each of these two simulations , they follow the same trend . at high redshift , @xmath2 is close to the universal value and sometimes is slightly higher . cold flows provides gas to form stars . note that the virial radius is estimated according to the local density , which is dominated by the newly formed stars . since the resolution in these simulations is limited , especially for low mass systems , the virial radius may be underestimated , and subsequently this baryonic fraction may be in fact overestimated at high redshifts . from @xmath73 , @xmath2 is decreasing with cosmic time until it reaches the values of @xmath74 and @xmath75 at @xmath1 for @xmath58 and @xmath54 respectively . these latter values are quite close to the universal value which is not surprising since similar trends have been already found from other hydrodynamical simulations using weak or no supernovae feedback , for objects with a mass of @xmath76 at @xmath1 ( crain et al . 2007 ; faucher - gigure , kere & ma 2011 ) . more interestingly , no particular differences are seen in the evolution of @xmath2 between @xmath29 and @xmath31 . this suggests that higher sn feedback can reduce the star formation rate ( and therefore the final stellar mass ) but is rather inefficient in expelling the gas outside the virial radius at high and low redshifts for massive galaxies . this results are in agreement with stinson et al . ( 2011 ) who show that increased feedback in haloes of such mass scale affects the star formation more than baryon content in the circumgalactic medium ( cgm ) . moreover , they show that higher feedback models can account for sufficient ovi and hi gas in the cgm compatible with the observed distributions . therefore , in the present scenario , the relatively small differences in the baryon fraction between @xmath77 and @xmath78 could make supernovae feedback prescriptions difficult to distinguish and one probably needs to focus on stellar mass and/or properties of the intergalactic medium in order to have better constraints ( see also dav , oppenheimer & sivanandam 2008 ; shen , wadsley & stinson 2010 ; stinson et al . 2011,2012 ; scannapieco et al . 2012 ; brook et al . in particular , it seems that higher feedback of @xmath78 compared to @xmath77 may be necessary in reproducing values closer to the correct stellar masses and properties of intergalactic medium as we will see in section 3.3 . we have also studied the distribution of the integral @xmath79 and differential @xmath80 as respect to the radius @xmath81 at three different redshifts @xmath82 , @xmath83 and @xmath72 in figure[fb_dis ] . only results from @xmath54 are shown since similar trends are obtained for @xmath58 . at the center of each halo , @xmath2 is high due to the presence of the galaxy . at larger radii , @xmath84 , @xmath79 tends towards the universal value as expected . interestingly , for high redshifts ( @xmath85 ) , @xmath79 converges to the universal value from above , while this is not the case at lower redshifts , which suggest , as mentioned above , that the gas infall is not recurrently replaced by some fresh gas from the outskirt of the haloes . also , an excess of baryons at @xmath86 can be clearly seen in the variations of the differential @xmath80 for different cosmic times , which correspond to regions in yellow in fig.[distribution1 ] around haloes . similar plots were recently derived from observations using chandra and suzaku facilities for an isolated elliptical galaxy with a @xmath87 milky way mass ( humphrey et al . 2011 ) and a fossil group ( humphrey et al . 2012 ) . in @xmath33 however , due to earlier and important eviction of gas from other potential sources of feedback ( agn , hypernovae , ... ) @xmath2 reaches its lowest value @xmath88 at @xmath89 , then increases until @xmath90 and becomes nearly constant ( @xmath4 ) until the present time . this strongly suggests that sources of feedback acting at high redshift , even for a very short period , can have a stronger impact on the final mass budget of massive galaxies than those acting in the later and longer phases of galaxy formation . one can also notice from figure[fig_sfr ] that the gas outflow taking place at high redshift lead to the quenching of the star formation during a few million years ( @xmath91 ) . this is in nice agreement with conclusions of valiante , schneider & maiolino ( 2012 ) using semi - analytical models of the formation and evolution of high - redshift quasars . indeed , according to their model , the observed outflow of a distant quasar at @xmath34 is dominated by quasar feedback which _ `` has considerably depleted the gas content of the host galaxy , leading to a down - turn in the star formation rate at @xmath92''_. nevertheless , future observational proofs of star formation quenching at these redshifts are required in order to confirm such statements . the time evolution of the baryonic fraction inside the virial radius depends on the competition between the evolution of the mass that arrives within @xmath3 , and the mass that leaves @xmath3 . it raises the following question : are the low values of @xmath2 found at @xmath1 due to the fact that more baryons have been expelled out of the virial radius or because more dark matter has been accreted inside @xmath3 ? to answer this question , instead of studying the evolution of the mass that enter or leave the virial radius at each subsequent step , we focused on the location at a given redshift @xmath93 of all particles that have been accreted up to this specific redshift . to do this , we have first identified all the particles that have been accreted at each snapshot of the simulation from @xmath94 to `` @xmath93 '' . then , we have computed the fraction of those particles that are inside or outside @xmath3 at the considered redshift . by doing this , the interpretation of our findings will be easier since each accreted particle contributes only once in the final result . indeed , certain particles can be accreted , ejected and accreted again during the whole evolution of these haloes . figure[mass_acc2 ] shows the evolution of the total accreted dm and baryon mass from @xmath94 to a given redshift , @xmath93 , for both haloes , and for both @xmath31 and @xmath33 ( results derived from @xmath29 are not shown since they are similar to those from @xmath31 ) . we also plot the fraction of these accreted particles that are located inside or outside @xmath3 at this given redshift @xmath93 . first , we look at the evolution relative to @xmath58 . one can see that a higher fraction of dark matter mass has been accreted from @xmath95 to @xmath1 relative to the baryon component ( solid lines ) . moreover , a similar fraction of dark matter and baryons are located outside the virial radius at @xmath1 . that means that the fraction of baryons released during the formation of @xmath58 is not higher than that of dark matter , which demonstrate that the low value of @xmath2 at @xmath1 follows from the fact that more dark matter particles has been accreted over cosmic time . this can be understood given the difference in nature between the dissipative gas and the non - dissipative dark matter . in particular , the gas component can be shock - heated at the virial radius which considerably affects and slows down its further accretion on to the halo as already shown in hydrodynamical simulations using no feedback ( see for instance oorbe et al . in contrast , especially after an important episode of accretion or merger event , the dark matter component can `` damp out '' its orbital motion until it completely relaxes . for @xmath54 the same trends are observed but the mass of baryons outside the virial radius turns out to be higher that that of dark matter . however this does not necessarily mean that a higher fraction of baryons has been expelled through physical mechanisms such as feedback or merger events . indeed , if we represent the fraction of baryons and dm released from @xmath96 to @xmath1 ( see figure[mass_acc2 ] ) , one can see that same amounts of mass have been expelled at @xmath1 by the two components . it implies in fact that a significant fraction of dark matter which has been expelled between @xmath95 and @xmath96 has been re - accreted between @xmath96 and @xmath1 . this is not the case for baryons which are generally heated through the expelling process and therefore remain outside the virial radius afterwards . in order to illustrate this , figure[illustration ] presents the evolution of the distribution of baryons and dark matter for 1 . from @xmath97 , the main progenitor undergoes two mergers with mass ratios of 1:1.8 and 1:5 . here , we have selected all particles inside the virial radius of each of the 3 objects . note that the orbits of the two mergers are quite radial . after the two merger events , the spatial distributions of baryons and dark matter are quite different . most of the dark matter particles are located inside the virial radius , though a small part has been expelled in the same orbital direction . the baryons tends to either collapse toward the center or are expelled all around . this phenomenon is observed at @xmath1 where a non - negligible part of baryons have been expelled and remain outside the virial radius . figure[illustration ] also shows the evolution of the mass of dm and baryons located outside the virial radius . in the case of the dark matter component , there are 3 distinct phases . first , from the beginning of the experiment until @xmath98 gyr , both satellites are accreted inside @xmath3 . then in the second phase ( from @xmath98 to @xmath99 gyr ) some of the dark matter is expelled via tidal disruption . finally from @xmath99 gyr some of the dark matter particles are re - accreted . note that evolution from @xmath100 gyr is probably due to the effect of another merger event . however , the baryon component follows the two first phases but not the third one : once baryons are expelled , most of them remain outside the virial radius . in this specific example , we found that 2.1% and 5.8% of dark matter and baryon masses respectively are released at @xmath1 from the three main objects . this is in good agreement with results obtained from idealized simulations ( sinha & holley - bockelmann 2009 ) . it is also interesting to see that due to mergers , unbound gas diffuses far away from the galaxy ; thus mergers represent a potential mechanism for enriching the igm with metals . note that this enrichment must occur early , as in our model , to avoid disruption of the cold and weakly enriched lyman alpha forest . it is also worth mentioning that during a merger event , gas can be expelled via both tidal disruption and outflows from feedback due to the merger driven starburst . however , since there is no significant difference in the evolution of @xmath2 within the virial radius between @xmath77 and @xmath78 , this suggests that supernovae feedbacks have a lower contribution relative to tidal disruption for milky way mass galaxies . as observed in the evolution of the dark matter component , tidal disruption should guarantee escape whereas a starburst driven outflow is much more problematic ( see for instance powell , slyz and devriendt 2011 ) . finally , it is interesting to point out that @xmath58 undergoes major merger events between @xmath101 and @xmath102 , leading to important mass outflows from @xmath102 and @xmath1 . in particular , figure[mass_acc2 ] indicates that more dark matter particles are released than baryons after the merger events . most of this dark matter is not yet re - accreted , which explains why @xmath58 tends to have values of @xmath2 higher than @xmath54 at @xmath1 . this also explains why in figure[distribution1 ] the high @xmath2 value regions around @xmath58 seem to be less pronounced in comparison to those around @xmath54 . thus , the evolution of the mean baryonic fraction value @xmath2 at the virial radius is essentially governed by the relative efficiency at which the dark matter and baryons are accreted . indeed , the evolution of the accreted mass of both dark matter and baryons follows two different regimes : a rapid growth at high redshifts ( @xmath103 ) and a slower one at lower redshifts . these evolutions are directly related to the total mass evolutions in figure[mass_acc2 ] which also presents these two regimes . this behaviour compares well to the evolution of individual dark matter haloes ( see for instance wechsler et al . 2002 ; vitvitska et al . 2002 , peirani , mohayaee & pacheco 2004 , mcbride et al . 2009 and tillson , miller & devriendt 2011 ) , indicating a fast growth of the mass for @xmath104 , followed by a phase where the accretion rate is slower . hydrodynamical simulations have also suggested that galaxy formation presents two phases : a rapid early phase ( @xmath103 ) during which stars mainly form within the galaxy from infalling cold gas ( katz et al . 2003 ; kere et al . 2005 ; ocvirk , pichon & teyssier 2008 ; dekel et al . 2009 ) and a second one ( @xmath105 ) during which stars are accreted ( oser et al . 2010 ; hirschmann et al . this latter phase is dominated by the accretion of diffuse matter and small satellites ( fakhouri & ma 2010 ; genel et al . 2010 ; wang et al . 2011 ; van de voort et al . 2011 ) . hence the high @xmath2 values at high redshift derived from @xmath29 and @xmath31 can be explained by the accretion of high gas - to - dark matter ratio from dense region such as filament via cold flows . but at lower redshift , @xmath2 is decreasing due to the accretion of low gas - to - dark matter ratio material , in particular from the diffuse region . indeed , we found that the baryonic fraction of the diffuse accreted matter ( namely @xmath106 ) is on the average much lower than the universal value : as the haloes become more massive , the temperature at the virial radius increases and the gas is shock - heated and this process slows down its accretion onto the halo . for @xmath33 , @xmath2 reaches its lowest value ( @xmath88 ) at @xmath107 right after some significant expulsion of gas . after that , there is a short phase where @xmath2 increases because the accreted matter has a higher baryonic content ( but not necessarily higher than the universal value ) . then , in the late phase of evolution of each halo , @xmath2 tends to be constant ( @xmath4 ) because the average accreted matter tends to have the same baryonic fraction value as the mean value inside virial radii . the whim has proven to be difficult to detect and represents a serious candidate for the missing baryons . it is therefore instructive to estimate the fraction of the whim in our simulated local universes . for this purpose , we have divided the inter - galactic medium ( igm ) into four components : ( 1 ) stars , ( 2 ) cold gas ( @xmath108 ) , ( 3 ) whim ( @xmath109 ) and hot gas ( @xmath110 ) . figure[whim ] shows the time evolution of three baryon components ( stars , cold gas and whim ) within a sphere centred on the local universe - type haloes and of comoving radius 1.5 mpc / h . the evolution of the hot gas ( @xmath111k ) is not shown in figure[whim ] since this component represents less than @xmath112 of the total baryon mass budget in our selected region , as expected since we are not in a cluster environment . it shows that the whim represents @xmath113% of the baryon budget at @xmath1 , in good agreement with past work based on hydrodynamical simulations ( cen & ostriker 1999 ; dav et al . 2001 ; cen & ostriker 2006 ; rasera & teyssier 2006 ; tornatore et al . 2010 ) when agn feedback is not taken into account . we also found that the warm and hot gas becomes the dominant baryonic component at large radii ( @xmath114 kpc ) from each main halo , in good agreement with previous observational analysis of massive galaxies ( humphrey et al . 2011 , 2012 ) or with previous numerical simulations ( van de voort et al . 2011 ; van de voort & schaye 2012 ) . however , when the contribution of the whim is not taken into account in the estimation of the baryonic fraction at the virial radius , namely when only the cold gas ( @xmath115k ) and stars are considered , @xmath2 values obtained at @xmath1 only decrease by @xmath116% . this follows from the fact that stellar masses obtained in our experiments have a large contribution in the final @xmath2 values and thus we probably need the additional effect of agn feedback to regulate the sfr and also increase the mass of whim inside the virial radius . nevertheless , the distribution of baryonic fraction seen in figure[distribution1 ] is significantly affected when the whim is not taken into account . indeed , figure[whim2 ] shows that in this latter case , the baryonic fraction is lower than @xmath117 in the direct vicinity of the virial radius . outside the virial radii , our analysis in the previous section has suggested that the igm is composed of gas that either never collapses onto haloes or is released by mechanisms such as feedback or merger events . but what are the relative proportions ? from figure[mass_acc2 ] we can estimate that @xmath118 of the total accreted baryons is expelled from @xmath95 to @xmath1 for @xmath58 and @xmath54 , which is quite significant . these results are also in good agreement with sinha & holley - bockelmann ( 2010 ) who found from semi - analytic techniques that up to @xmath87 25% of gas can be unbind over the course of galaxy assembly . from our analysis , we found that the gas that has been released after being accreted on to haloes is mainly located between @xmath3 and @xmath119 at @xmath1 . if we focus on @xmath58 which is quite isolated , we found that the gas component between @xmath3 and @xmath119 consists of 55.9% , 58.6% and 60.5% of gas that has been expelled from @xmath58 from @xmath29 , @xmath31 and @xmath33 respectively . in other words @xmath87 40% of this gas has never collapsed onto @xmath58 . it is also worth mentioning that @xmath120 of the gas that has been released through merger events or feedback is in the form of whim at @xmath1 but only @xmath87 65% in the case the gas that has never collapsed into the halo . hence , the gas that has been ejected through different mechanisms during the galaxy formation and evolution significantly increases the mass of the whim . in this paper , we have analysed hydrodynamical simulations with an extensive treatment of the physics of baryons in order to study the evolution of the distribution of baryons in a local group - type object . in particular , we have investigated the impact of i ) supernova feedback , ii ) more powerful sources of feedback at high redshift such as agn from intermediate mass black holes or hypernovae and iii ) accretion / ejection and merger events in the evolution of the mass content of milky - way type galaxies . first , when no high early feedback is considered as is the case in @xmath77 and @xmath78 , we found that the evolution of the mean baryonic fraction value @xmath2 estimated at the virial radius of the two main central objects is essentially governed by the relative efficiency at which dark matter and baryons are accreted . indeed , in the first and early phase of galaxy formation ( the cold mode of accretion @xmath103 ) during which stars mainly form within the galaxy from infalling cold gas ( katz et al . 2003 ; kere et al . 2005 ; ocvirk , pichon & teyssier 2008 ; dekel et al . 2009 ) @xmath2 is in general higher than the universal value . but in the second phase ( the hot mode of accretion ) , dominated by the accretion of diffuse matter or small satellites , @xmath2 tends to decrease and is 10 - 15% lower than the universal value , 0.166 , at @xmath1 . this decrease is not due to the fact that more baryons ( relative to dm ) are expelled beyond the virial radius by physical mechanisms such as feedback or merger events . rather this essentially follows from the relative nature of dissipative gas and non - dissipative dark matter . specifically , the gas component can be shock - heated at the virial radius , which considerably slows down its further accretion into the halo ( see for instance oorbe et al . however , shock - heating will presumably not be a large factor in low mass galaxies , but as mentioned in the introduction , feedbacks such as uv - background and supernovae winds are supposed to play a major role in the effective reduction of the baryon content of these galaxies . additionally , we found that accretion and particularly merger events can release a fraction of both dark matter and baryons . but again , dark matter that has been expelled can be re - accreted more efficiently than the corresponding baryons , hereby enhancing the decrease of @xmath2 inside the virial radius . consequently , our simulations suggest that the gas located beyond the virial radius is mainly in the form of whim that comes from either gas that never collapsed prior to the formation of the proto - galaxy ( @xmath121 ) or gas expelled by feedback and merger processes ( @xmath122 ) . our comparison of 1a and 1b also suggest that globally increasing somewhat the efficiency of supernovae does not significantly affect the evolution of @xmath2 inside the virial radius . in other words , sn feedback is not powerful enough to expel the baryons out of the gravitational potential of dark matter haloes of such mass scales . this also means that , if future observations indicate that @xmath2 inside the virial radius of milky way mass galaxy at @xmath1 is relatively close to the universal value , then in order to make clear a distinction between the different feedback models , the stellar mass and the properties of the intergalactic medium may provide better constraints than baryon fractions , as suggested by recent numerical works ( dav , oppenheimer & sivanandam 2008 ; shen , wadsley & stinson 2010 ; stinson et al . 2011,2012 ; scannapieco et al . 2012 ; brook et al . but at high redshift the situation is quite different since haloes are less massive and thus feedback mechanisms have a more effective contribution . indeed for our other scenario ( for which a higher fraction of gas is expelled at high redshift ( @xmath94 ) by more powerful source of feedback such as agn from intermediate mass black holes or hypernovae implemented in 2 ) the accretion of gas is strongly affected at high redshift , and drives the mean baryonic value inside the virial radius to a lower values at @xmath1 . this is one important conclusion of this paper : to get a low value of @xmath2 at present time , physical mechanisms able to expel the gas _ at high redshift _ will have a stronger impact on the deficit of baryons in the mass budget of milky way type - galaxy today than those that expel the gas in the longer , late phases of galaxy formation . thus , if future observations favour a relatively low value of @xmath2 at @xmath1 , then this high redshift feedback " scenario is the most probable . however , it this worth mentioning that the effective contribution of non - gravitational heating such as galactic winds from supernovae in the evolution of @xmath2 within the virial radius of milky way type galaxy is still a controversial and a debated topic in the recent literature . from our modelling and experiments , we found that increasing the efficiency of supernova feedback leads to the reduction of the stellar mass but has no particular impact on the evolution of @xmath123 for system of such mass scales ( cf powell , slyz & devriendt 2011 ) . such statements were also advocated by anderson & bregman ( 2010 ) from observational constraints on the density of hot gas around the milky way . moreover , using cosmological simulations , faucher - gigure , kere & ma ( 2011 ) have investigated the net baryonic accretion rates through @xmath124 . although the net baryonic accretion rate is sensitive to galactic outflows , especially for low mass systems , the baryon mass fractions at @xmath1 for objects with a mass higher than @xmath125 and for different sn feedback prescriptions have similar values to those derived in the present study . however , other hydrodynamical simulations from scannapieco et al . ( 2008 , 2009 ) using higher sfr efficiency ( @xmath126 ) found that at @xmath1 , the baryon fractions inside the virial radius are @xmath127 within the virial radius _ `` indicating that a significant amount of baryons has been lost through winds''_. agn feedback is supposed to play an important role in the formation of massive objects such as giant ellipticals in galaxy groups ( tabor & binney 1993 ; ciotti & ostriker 1997 ; silk & rees 1998 ) . from previous numerical studies , the inclusion of agn is expected to reduce the stellar mass ( see for instance sijacki et al . 2007 ; dubois et al . 2010 , 2012 ; hambrick et al . 2011b ) and also increase the mass of whim ( cen & ostriker 2005 ; tornatore et al . 2010 ; durier & pacheco 2011 ; roncarelli et al . our study does not include the effect of agn feedback at low redshift , but we think that such sporadic events will not have a major contribution in the baryonic fraction value inside the virial radius at present time . indeed , from observationally motivated constraints , agn and sn at low to moderate redshift seem not to produce the expected correlations with the baryonic tully - fisher relationship ( anderson & bregman 2010 ) . moreover , as mentioned above , observations of massive objects reported by humphrey et al . ( 2011 , 2012 ) seem to suggest that agn feedback has no particular impact on the baryonic fraction values within the virial radius . also , recent hydrodynamical simulations show that agn and sn feedback may mutually weaken one another s effect by up to an order of magnitude in haloes in the mass range of @xmath128 ( booth & schaye 2012 ) . at high redshift , the effects of galactic winds on the igm are also controversial . in the present paper , our simple minded modelling indicates that galactic winds can expel some gas outside the virial radius and affect significantly the evolution of the baryon content of massive galaxies . similar conclusions have been recently drawn by roncarelli et al . ( 2012 ) from hydrodynamical simulations suggesting that galactic winds acting at earlier epochs can prevent the igm to collapse in dense structures . also , as mentioned above , semi - analytical model from valiante , schneider & maiolino ( 2012 ) indicates that the observed outflow of a distant quasar at @xmath34 ( maiolino et al . 2012 ) is expected to have considerably depleted the gas content of the host galaxy . however , simulations conducted by di matteo et al . ( 2011 ) or shen et al . ( 2012 ) found that galactic outflows at high redshift may not be very effective at stopping the cold gas from penetrating the central regions . in summary , our study indicates that in order to reach lower @xmath2 values at @xmath1 for milky way type galaxies , the eviction of cold gas by feedbacks during the first phase of galaxy formation at high redshifts proves to be crucial . our study also suggests that these high redshift feedback mechanisms more efficiently reduce galaxy stellar mass in the central part of haloes and thus show better agreement to galaxy observable such as the stellar mass - halo mass relation . moreover , efficient high redshift feedback processes seem also to be required in order to slow down the growth of galaxies at high redshift and thus to reproduce the observed number density evolution as recently pointed out by weinmann et al . the presence of strong early feedbacks is also motivated in disk galaxy formation theory . indeed , it has been suggested that early outflows could eject low angular momentum gas allowing disk galaxies to form ( binney , gerhard and silk ( 2001 ) . such hypothesis was confirmed by hydrodynamical simulations which show that outflows can indeed remove preferentially low angular momentum material which can be re - accreted latter with higher angular momentum and forms a disk ( brook et al . 2011 , 2012a ) . if such statements are correct , numerical and observational efforts have to be focused towards characterising the respective role of each feedback process on the igm at high redshift . in particular , are the baryons expelled from protogalaxies or do they never collapse into the gravitational potential well of these objects ? in the present paper , we only focus on the first mechanism , but it is not excluded that processes such as photoionization play a major role especially at high redshifts for low mass systems . in this regard , it is interesting to compare our results with those of guedes et al . ( 2011 ) who have studied the formation of late - type spirals from the eris simulation . specifically , they claimed that the collapse of baryons is heavily suppressed at high redshift by the uv background . and it is particularly instructive to see that from their highest mass resolution simulation , the evolution of the baryonic fraction ( see their figure 6 in guedes et al . 2011 ) is very similar to those we obtained from @xmath33 . they also found a similar value at @xmath1 namely @xmath129 . our simulations do include uv background but our mass resolution may be too small to account for this process properly . more detailed and well - resolved hydrodynamical simulations are required to investigate the evolution of the mass content of galaxies at high and low redshifts . but in order to achieve this goal , one must first improve the constraints of the expected energy injection to the igm from sn , agn , uv background ( etc ... ) via observations . . acknowledges support from the global internship program by the national research foundation of korea ( nrf ) . we warmly thank the referee for an insightful review that considerably improved the quality of the original manuscript . we warmly thank d.le borgne , j.devriendt , y.dubois , r.gavazzi , t.kimm , g.mamon , j.a.de freitas pacheco and t.sousbie for interesting discussions . we also thank d.munro for freely distributing his yorick programming language ( available at ` http://yorick.sourceforge.net/ ` ) which was used during the course of this work . this work was carried within the framework of the horizon project ( ` http://www.projet-horizon.fr ` ) . anderson , m. e. , & bregman , j. n. 2010 , apj , 714 , 320 bell , e. f. , mcintosh , d. h. , katz , n. , & weinberg , m. d. 2003 , apjs , 149 , 289 benson , a. j. , frenk , c. s. , lacey , c. g. , baugh , c. m. , & cole , s. 2002 , mnras , 333 , 177 bregman , j. n. 2007 , ara&a , 45 , 221 brook , c. b. , governato , f. , rokar , r. , et al . 2011 , mnras , 415 , 1051 brook , c. b. , stinson , g. , gibson , b. k. , et al . 2011 , mnras , 419 , 771 brook , c. b. , stinson , g. , gibson , b. k. , wadsley , j. , & quinn , t. 2012b , mnras , 424 , 1275 bond , j. r. , kofman , l. , pogosyan , d. , 1996 , nature , 380 , 603 booth , c. m. , & schaye , j. 2012 , arxiv:1203.3802 bullock , j. s. , kravtsov , a. v. , & weinberg , d. h. 2000 , apj , 539 , 517 cen , r. , & ostriker , j. p. 1999 , apj , 514 , 1 cen , r. , & ostriker , j. p. 2006 , apj , 650 , 560 ciotti , l. , & ostriker , j. p. 1997 , apjl , 487 , l105 colless , m. , dalton , g. , maddox , s. , et al . 2001 , mnras , 328 , 1039 crain , r. a. , eke , v. r. , frenk , c. s. , et al . 2007 , mnras , 377 , 41 dabringhausen , j. , kroupa , p. , pflamm - altenburg , j. , & mieske , s. 2012 , apj , 747 , 72 dav , r. , cen , r. , ostriker , j. p. , et al . 2001 , apj , 552 , 473 debuhr , j. , quataert , e. , & ma , c .- p . 2012 , mnras , 420 , 2221 dav , r. , oppenheimer , b. d. , & sivanandam , s. 2008 , mnras , 391 , 110 dekel , a. , birnboim , y. , engel , g. , et al . 2009 , nature , 457 , 451 di matteo , t. , khandai , n. , degraf , c. , et al . 2012 , apjl , 745 , l29 dubois , y. , devriendt , j. , slyz , a. , & teyssier , r. 2010 , mnras , 409 , 985 dubois , y. , devriendt , j. , slyz , a. , & teyssier , r. 2012 , mnras , 420 , 2662 durier , f. , & de freitas pacheco , j. a. 2011 , international journal of modern physics e , 20 , 44 fakhouri , o. , & ma , c .- p . 2010 , mnras , 401 , 2245 faucher - gigure , c .- a . , kere , d. , & ma , c .- p . 2011 , mnras , 417 , 2982 geller , m. j. , & huchra , j. p. 1989 , science , 246 , 897 genel , s. , bouch , n. , naab , t. , sternberg , a. , & genzel , r. 2010 , apj , 719 , 229 gnedin , n. y. 2000 , apj , 542 , 535 gott , j. r. , iii , juri , m. , schlegel , d. , et al . 2005 , apj , 624 , 463 guedes , j. , callegari , s. , madau , p. , & mayer , l. 2011 , apj , 742 , 76 guo , q. , white , s. , li , c. , & boylan - kolchin , m. 2010 , mnras , 404 , 1111 habergham , s. m. , anderson , j. p. , & james , p. a. 2010 , apj , 717 , 342 hambrick , d. c. , ostriker , j. p. , johansson , p. h. , & naab , t. 2011a , mnras , 413 , 2421 hambrick , d. c. , ostriker , j. p. , naab , t. , & johansson , p. h. 2011b , apj , 738 , 16 hirschmann , m. , naab , t. , somerville , r. s. , burkert , a. , & oser , l. 2012 , mnras , 419 , 3200 hoeft , m. , yepes , g. , gottlber , s. , & springel , v. 2006 , mnras , 371 , 401 hoekstra , h. , hsieh , b. c. , yee , h. k. c. , lin , h. , & gladders , m. d. 2005 , apj , 635 , 73 humphrey , p. j. , buote , d. a. , canizares , c. r. , fabian , a. c. , & miller , j. m. 2011 , apj , 729 , 53 humphrey , p. j. , buote , d. a. , brighenti , f. , et al . 2012 , apj , 748 , 11 katz , n. , keres , d. , dave , r. , & weinberg , d. h. 2003 , the igm / galaxy connection . the distribution of baryons at z=0 , 281 , 185 kere , d. , katz , n. , weinberg , d. h. , & dav , r. 2005 , mnras , 363 , 2 komatsu , e. , dunkley , j. , nolta , m. r. , et al . 2009 , apjs , 180 , 330 komatsu , e. , smith , k. m. , dunkley , j. , et al . 2011 , apjs , 192 , 18 maiolino , r. , gallerani , s. , neri , r. , et al . 2012 , mnras , 425 , l66 mcbride , j. , fakhouri , o. , & ma , c .- p . 2009 , mnras , 398 , 1858 mcgaugh , s. s. , schombert , j. m. , de blok , w. j. g. , & zagursky , m. j. 2010 , apjl , 708 , l14 mo , h. j. , yang , x. , van den bosch , f. c. , & katz , n. 2005 , mnras , 363 , 1155 moster , b. p. , somerville , r. s. , maulbetsch , c. , et al . 2010 , apj , 710 , 903 navarro , j. f. , & white , s. d. m. 1993 , mnras , 265 , 271 nickerson , s. , stinson , g. , couchman , h. m. p. , bailin , j. , & wadsley , j. 2011 , mnras , 415 , 257 ocvirk , p. , pichon , c. , & teyssier , r. 2008 , mnras , 390 , 1326 okamoto , t. , gao , l. , & theuns , t. 2008 , mnras , 390 , 920 oorbe , j. , domnguez - tenreiro , r. , siz , a. , & serna , a. 2007 , mnras , 376 , 39 oser , l. , ostriker , j. p. , naab , t. , johansson , p. h. , & burkert , a. 2010 , apj , 725 , 2312 peirani , s. , mohayaee , r. , & de freitas pacheco , j. a. 2004 , mnras , 348 , 921 peirani , s. 2010 , mnras , 407 , 1487 piontek , f. , & steinmetz , m. 2011 , mnras , 410 , 2625 powell , l. c. , slyz , a. , & devriendt , j. 2011 , mnras , 660 prunet , s. , pichon , c. , aubert , d. , pogosyan , d. , teyssier , r. , & gottloeber , s. 2008 , apjs , 178 , 179 rasera , y. , & teyssier , r. 2006 , a&a , 445 , 1 read , j. i. , pontzen , a. p. , & viel , m. 2006 , mnras , 371 , 885 ricotti , m. , gnedin , n. y. , & shull , j. m. 2002 , apj , 575 , 49 robitaille , t. p. , & whitney , b. a. 2010 , apjl , 710 , l11 roncarelli , m. , cappelluti , n. , borgani , s. , branchini , e. , & moscardini , l. 2012 , mnras , 424 , 1012 scannapieco , c. , tissera , p. b. , white , s. d. m. , & springel , v. 2008 , mnras , 389 , 1137 scannapieco , c. , white , s. d. m. , springel , v. , & tissera , p. b. 2009 , mnras , 396 , 696 scannapieco , c. , wadepuhl , m. , parry , o. h. , et al . 2012 , mnras , 423 , 1726 shen , s. , wadsley , j. , & stinson , g. 2010 , mnras , 407 , 1581 shen , s. , madau , p. , guedes , j. , mayer , l. , & prochaska , j. x. 2012 , arxiv:1205.0270 sijacki , d. , springel , v. , di matteo , t. , & hernquist , l. 2007 , mnras , 380 , 877 silk , j. , & rees , m. j. 1998 , a&a , 331 , l1 binney , j. , gerhard , o. , & silk , j. 2001 , mnras , 321 , 471 silk , j. , & nusser , a. 2010 , apj , 725 , 556 sinha , m. , & holley - bockelmann , k. 2010 , mnras , 405 , l31 sinha , m. , & holley - bockelmann , k. 2009 , mnras , 397 , 190 somerville , r. s. 2002 , apjl , 572 , l23 springel , v. 2005 , mnras , 364 , 1105 stinson , g. s. , brook , c. , prochaska , j. x. , et al . 2012 , mnras , 425 , 1270 stinson , g. , brook , c. , macci , a. v. , et al . 2012 , arxiv:1208.0002 tabor , g. , & binney , j. 1993 , mnras , 263 , 323 thibaut , j. , pichon , c. , sousbie , t. , prunet , s. , & pogosyan , d. 2008 , mnras , 387 , 397 thoul , a. a. , & weinberg , d. h. 1996 , apj , 465 , 608 tillson , h. , miller , l. , & devriendt , j. 2011 , mnras , 417 , 666 tornatore , l. , borgani , s. , viel , m. , & springel , v. 2010 , mnras , 402 , 1911 valiante , r. , schneider , r. , & maiolino , r. 2012 , arxiv:1205.3488 van de voort , f. , schaye , j. , booth , c. m. , haas , m. r. , & dalla vecchia , c. 2011 , mnras , 414 , 2458 van de voort , f. , & schaye , j. 2012 , mnras , 2882 vitvitska m. , klypin a. , kravtsov a. v. , wechsler r. h. , primack j.r . & bullock j. s. , 2002 , apj 581 , 799 wagner , a. y. , bicknell , g. v. , & umemura , m. 2012 , arxiv:1205.0542 wang , j. , navarro , j. f. , frenk , c. s. , et al . 2011 , mnras , 413 , 1373 wechsler , r. h. , bullock , j. s. , primack , j. r. , kravtsov , a. v. , & dekel , a. 2002 , apj , 568 , 52 weinmann , s. m. , pasquali , a. , oppenheimer , b. d. , et al . 2012 , arxiv:1204.4184 zhu , w. , feng , l .- l . , & fang , l .- z . 2011 , mnras , 415 , 1093
long wavelength oscillations ( tkachenko waves ) of the triangular lattice of quantized vortices in superfluid neutron stars have been suggested as one of the possible explanations for the timing noise observed in many radio pulsars , in particular for the 100 - 1000 day variations in the spin of psr 1828 - 11 . most studies to date have , however , been based on the hydrodynamics developed for superfluid helium . in this paper we extend the formulation to a two fluid neutron and proton system , relevant for neutron star interiors and include the effect of chemical coupling , compressibility and mutual friction between the components . in particular we find that chemical coupling and compressibility can have a drastic effect on the mode structure . however , for the slower pulsars rotating at 1 - 10 hz ( such as psr b1828 - 11 ) , most choices of parameters in the equation of state lead to tkachenko oscillations with frequencies in the correct range to explain the timing noise . we also investigate the case of more rapidly rotating pulsars ( above 100 hz ) for which we find that there is a vast portion of parameter space in which there are no tkachenko modes , but only modified sound waves at much higher frequencies . 0.5 truein v s
1011.1180
a growing number of radio pulsars have now been observed for over a decade ( some for more than 30 years ) and are , in general , very stable rotators . however , many pulsars also exhibit timing irregularities , such as `` glitches '' , which are sudden increases in the rotation rate , and `` timing noise '' , a general term which refers to low frequency quasi - periodic structures that appear in the timing residuals , once the `` regularly '' pulsating solution has been removed . in particular , while the irregularities in younger pulsars are dominated by the recovery from glitch events , for a handful of older pulsars there is growing evidence for long - period ( @xmath0 100 - 1000 days ) oscillations in the timing residuals @xcite . in some cases these periodicities , and the correlated pulse shape changes , can be partially explained by neutron star free - precession , and one of the best examples of this is psr b1828 - 11 which shows significant periodicity at @xmath1 days and @xmath2 days @xcite . however there are theoretical arguments stating that mutual friction between the interior superfluid components of the star would damp out any precessional motion on a short timescale @xcite ( although glampedakis et al . @xcite have shown that short wave length instabilities in the pinned superfluid could cast an element of doubt on such conclusions ) . furthermore recent work shows that some pulsars may be switching abruptly between two different states with different spin - down rates , thus giving rise to the observed timing behaviour @xcite . noronha and sedrakian @xcite , following earlier suggestions by ruderman @xcite , indicated that an alternative explanation for the observed long term periodicity could be the propagation of tkachenko waves in the star . in fact , it has been suggested that tkachenko waves excited by glitches may be driving precession in one of the x - ray dim isolated neutron stars ( xdins ) , rx j0720.4 - 3125 @xcite . neutron star interiors are expected to contain charge neutral superfluids that rotate by forming an array of quantized vortices . in their lowest energy state the vortices form a two - dimensional triangular lattice that can support elastic oscillations , tkachenko waves @xcite , that have been studied extensively , both theoretically and experimentally , in superfluid @xmath3he ( see e.g. @xcite for a review ) and recently in bose - einstein condensates ( becs ) ( see e.g. @xcite and @xcite ) . the undamped propagation of tkachenko waves in a neutron star would lead to periodic variations in the angular momentum of the superfluid which , due to coupling to the crust , would lead to variations in the observed rotation rate . in order to ascertain if this is a viable hypothesis it is crucial to understand how the detailed microphysics of neutron star interiors affects the propagation of the modes . most studies to date have been based on the hydrodynamical theory of tkachenko waves developed by baym and chandler @xcite for superfluid @xmath3he . in this case the fluids can be treated , to a good degree of approximation , as incompressible , given that the rotation rate is always well below the sound wave frequency ( note , however , that in becs , compressibility has a strong effect , due to the interactions being much weaker and the sound speed much lower than in helium ) , and the system can be described as a condensate ( the `` superfluid '' ) coupled to a `` normal '' fluid which consists , loosely speaking , of the thermal excitations of the system . in a realistic neutron star , on the other hand , one must take into account not only the effects of rapid rotation , but also the presence of several massive fluids , describing the flow of electrons , protons , superfluid neutrons ( and their excitations at finite temperature ) and possibly exotic particles such as hyperons or deconfined quarks , which can not , in general , be assumed to be incompressible . furthermore , one must consider various dissipative processes that damp out the oscillations . it is well known that in a multifluid system there will , in general , be many more dissipation channels than in a simple one - fluid flow described by the navier - stokes equation @xcite . solving the full problem is clearly a daunting task , so in this paper we shall make a series of simplifying assumptions . we take a system of two fluids : the superfluid neutrons and a charged component of protons and electrons , which are locked by the coulomb interaction on a much shorter timescale than the dynamical timescales considered here . this assumption is justified as the frequencies of the modes we calculate ( tkachenko waves , sound waves and inertial waves ) are always significantly lower than the electron - proton cyclotron and plasma frequencies , i.e. lower than @xmath4 hz ( @xcite ) . both fluids are assumed to be compressible , and we shall assume some simplified analytic models for the equation of state , derived from haskell et al . furthermore we will only consider the damping due to superfluid mutual friction , which has been found to have a significant effect on the mode propagation @xcite . the paper is structured as follows . in section ii we present the formalism for studying the oscillations of a two - fluid neutron star with mutual friction and vortex lattice elasticity . in section iii we perform a plane wave analysis of the oscillation spectrum in the incompressible case and in section iv we present the more realistic compressible case . as we shall see compressibility and chemical coupling between the components can profoundly alter the nature of the tkachenko waves , leading in some cases to much shorter periods of oscillation , close to the rotation period of the star , which are not consistent with the observed periodicities in pulsar timing residuals . finally in section v we outline our conclusions . our starting points will be the multi - fluid formulation of superfluid hydrodynamics of andersson and comer @xcite , and the baym - chandler formalism for including the effects of vortex lattice elasticity in the study of superfluid @xmath3he @xcite . let us consider a two fluid system of neutrons and protons ( which we assume locked to the electrons @xcite ) and write the euler equations for the neutrons , in a frame rotating with the star at fixed angular velocity @xmath5 and in the absence of external forces and mutual friction . following @xcite this takes the form : ( _ t + v^_j^j)(v^_i+_w^_i)+2_ijk^j v_^k + _ i ( _ + ) + _ w^j__i v^_j = 0[eulern ] where @xmath6 is the neutron velocity , @xmath7 , with @xmath8 the proton velocity , @xmath9 is the entrainment parameter for the neutrons @xcite , @xmath10 is the chemical potential per unit mass of the neutrons and @xmath11 is the gravitational potential . note that we assume summation of repeated indices and assume the neutron and proton masses equal , @xmath12 . in the above equation we have not yet imposed that the neutrons be superfluid . to do this we must require that the fluid rotates by forming an array of singly quantized vortices , and that averaging over a large number of such vortices gives rise to the macroscopic vorticity @xmath13 of the fluid via the relation : ^i = n_v^i = ^ijk_j ( v^_k+^w^_k ) [ circolo ] where @xmath14 is a unit vector along the direction of the vortex array , @xmath15 @xmath16 s@xmath17 is the quantum of circulation and @xmath18 is the number of vortices threading a unit surface . it is important to remark here that the quantization condition on the circulation is a condition on the momentum of the neutron fluid @xmath19 , and not on its velocity @xmath20 ( which will not in general be aligned with @xmath21 due to the entrainment ) . from equation ( [ circolo ] ) one can derive the equation of motion for the circulation _ t _ i+_ijk^klm^j _ l v^_m=0[circolo1 ] and a conservation equation for the vortex number _ ( n_v^i _ ) = 0[circolo2 ] with @xmath22 the macroscopically averaged vortex velocity . it is possible to show ( @xcite ) that , in order for equations ( [ circolo1 ] ) and ( [ circolo2 ] ) to be staisfied it is necessary to add a `` magnus force '' term to the right hand side of equation ( [ eulern ] ) , which thus takes the form ( _ t + v^_j^j)(v^_i+_w^_i)+2_ijk^j v_^k + _ i ( _ + ) + _ n w^j__i v^_j = n_v_ijk ^j ( v_^k - v_v^k)[eulern2 ] it is clear from equation ( [ eulern2 ] ) that in the absence of other forces the vortices will be forced to move with the superfluid neutron condensate . the presence of vortices will , however , also affect the proton fluid , which will experience a drag force of the form @xmath23 , where the exact nature of the process giving rise to the drag is encoded in the dimensionelss parameter @xmath24 . in a neutron star there are , in fact , a variety of mechanisms that can produce a dissipative drag : scattering of electrons off vortex cores is likely to be the dominant process in the core @xcite , while in the crust the main contribution is due to interactions with the lattice phonons @xcite and the excitation of vortex kelvin waves @xcite . the diverse nature of these processes leads to the drag parameter spanning several orders of magnitude in the different regions of a neutron star interior ( from as low as @xmath25 to @xmath26 ) . we shall thus treat @xmath27 as a free parameter and investigate how its variations affect the modes . the euler equations for the proton fluid take the form : ( _ t + v^_j^j)(v^_i-_w^_i)+2_ijk^j v_^k + _ i ( _ + ) + _ w^j__i v^_j = n_v ( v^_i - v^_i)[eulerp ] where @xmath28 . @xmath29 and @xmath30 are now the chemical potential per unit mass and entrainment parameter of the protons , such that @xmath31 @xcite . we also need an equation of motion for the vortex lines which , if we assume that they have negligible inertia , takes the form of a force balance between the magnus force , the drag force and the elastic force exerted by the lattice @xcite : _ n_v_ijk^j ( v_v^k - v_^k)+_n_v(v^_i - v^_i)-__i=0[forcev ] where @xmath32 represents the contribution due to lattice elasticity and takes the form : _ i = where @xmath33 is the displacement of the vortex line from its equilibrium position , @xmath34 is the gradient perpendicular to the direction of the array and @xmath35 is the shear modulus of a triangular vortex lattice @xcite . note that the above expression only describes the linear order corrections in the lattice displacements , which are assumed to be small . furthermore we are neglecting the contribution of vortex bending , which would give rise to kelvin waves propagating along the vortex lines . note that this could be accounted for by including a vortex `` tension '' term in @xmath32 , which we denote @xmath36 , of the form : ^t_i=- ( ) where @xmath37 is the vortex core radius , @xmath38 is the inter - vortex spacing for a triangular lattice and the @xmath39 axis is taken along the rotation axis of the star . the continuity equations for neutrons and protons take the form & & _ t _ + ^i(_v^_i)=0 + & & _ t _ + ^i(_v^_i)=0 and the gravitational potential obeys the poisson equation ^2 = 4 g ( _ + _ ) where @xmath40 is the gravitational constant . finally to solve the problem we need to supply an equation of state for the system . as we shall examine different cases we delay the discussion of the equation of state to the following sections and move on to discussing perturbations of the multi - fluid equations of motion presented above . in order to keep the problem tractable we shall consider linear perturbations of a background in which the two fluids rotate together with uniform angular velocity @xmath5 . for such a background equation ( [ circolo ] ) takes the form : n_=2and the perturbed euler equations can be written , in a frame co - rotating with the star , as : & & _ t ( v_i^+ _ w_i^ ) + 2_ijk ^j v_^k + _ i _ = -2 ( v^_i - v^_i)-_i + & & _ t ( v_i^- _ w_i^ ) + 2_ijk ^j v_^k + _ i _ = 2 ( v^_i - v^_i)[eulers ] where we have made the cowling approximation , i.e. neglected the perturbations of the gravitational potential @xmath41 . note that as a consequence of the extra elastic term in the force balance equation for the vortices ( [ balo ] ) , the forces on the right hand side of the euler equations are no longer symmetric and the vortex elasticity term only acts on the neutron superfluid . the elastic force @xmath32 can be written as _ i = c_t^2where we have defined the tkachenko wave speed @xmath42 and are assuming the vortices to be in equilibrium in the background , such that @xmath43 . we assume all vortex displacements to be perturbed quantities , and write @xmath33 in place of @xmath44 , unless otherwise specified , and thus consider @xmath32 to also be a perturbed quantity . as we are dealing with linearised elasticity and the displacement vectors @xmath33 it would be natural to consider lagrangian perturbations of the two - fluid equations of motion , given that in general one would have that @xmath45 . however , given that we are working in a rotating frame , and have assumed that the fluids ( and thus the vortices ) are moving together in the background , one has that @xmath46 . we can thus continue to work with eulerian perturbations , which simplifies somewhat the problem . the equation of force balance for the vortices ( [ forcev ] ) can be cast in the form : v_i^=v_i^+_ijk^j w^k_-_i+(w_j^^j)-- [ balo]and the perturbed continuity equations , in the absence of reactions , take the form & & _ t _ + ^i(_v^_i)=0 + & & _ t _ + ^i(_v^_i)=0[continuo ] following @xcite we can combine equations ( [ eulers ] ) to obtain an euler equation for the `` total '' velocity @xmath47 : _ ip-_i p+2_ijk^jv^k=-(1-x_)_i [ tutto1]and one for @xmath48 : ( 1-|)_t w_i^+_i=-2^_ijk^jw_^k+2_ijk^j^klm_lw_m^+_i [ tutto2]where we have defined the total pressure , such that @xmath49 , the entrainment parameter @xmath50 , @xmath51 and the mutual friction parameters @xmath52 $ ] and @xmath53 $ ] . the perturbed continuity equations ( [ continuo ] ) can be cast in the form : & & _ t + _ i ( v^i)=0 + & & _ tx_+_j[x_(1-x_)w^j]+v^j _ j x_=0 [ tutto3 ] as we shall see in the following , this formulation can be advantageous when discussing the compressible problem . in order to make contact with previous results , let us consider first of all the case of incompressible fluids , such that @xmath54 and the continuity equations reduce to ^i v_i^=^i v_i^=0 we consider plane waves , such that a perturbed quantity @xmath55 takes the form @xmath56 , with @xmath57 a constant amplitude . without loss of generality we choose our coordinate system such that the @xmath39 axis points along the rotation axis and such that the wave vector @xmath58 lies in the @xmath59 plane , i.e. @xmath60 . the equations of motion can thus be written as : & & -i|v_i^(1- _ ) -i_|v_i^ + 2_ijk ^j |v_^k + i k_i | _ = 2 ( i|_i+|v^_i)-_i[eulersk1 ] + & & -i|v_i^(1- _ ) -i_|v_i^ + 2_ijk ^j |v_^k + i k_i | _ = -2 ( i|_i+|v^_i ) + & & -i|_i-|v_i^-_ijk^jw^k_+_i - w_j^^j++=0[forceb ] + & & k_j v^j_=k_j v^j_=0 [ eulersk ] where we have defined @xmath61 , @xmath62 and to simplify notation we have defined @xmath63 as the amplitude of @xmath64 , with @xmath65 . finally the vortex elasticity contribution takes the form : = = ( -(c_t k)^2,(c_t k)^2,0 ) in order to obtain the dispersion relation for the modes of the system we thus need to solve the characteristic equation det@xmath66=0 , where @xmath67 follows from equations ( [ eulersk1]-[eulersk ] ) and is given in equation ( [ kappa1 ] ) . in the undamped case , neglecting the effect of entrainment ( @xmath68 ) one obtains , as expected , two families of modes , the inertial modes ^2=4 ^ 2 ( ) ^2 and the tkachenko waves ^2=4 ^ 2()^2+c_t^2 k^2()^4-()^44 ^ 2||^2+c_t^2 k^2()^4 [ mod1]where we are assuming that @xmath69 . this will always be the case if we consider typical pulsar spin rates from a few hz to a few hundred hz and long wavelength oscillations across the whole superfluid region , such that @xmath70 cm@xmath17 . for propagation perpendicular to the rotation axis ( @xmath72 ) one then obtains the well known tkachenko wave dispersion relation = c_t k [ mod2 ] let us now still consider undamped propagation of the modes , but include the effect of entrainment . clearly introducing coupling between the two fluids profoundly alters the nature of the modes and leads , in the @xmath69 limit , to two families of mixed inertial - tkachenko waves : ^2&&4 ^ 2()^2+(1-x_)c_t^2k^2()^4 + ^2&&4 ^ 2()^2()^2+c_t^2k^2()^4 in the limit @xmath73 and @xmath74 one has again @xmath75 from the relation @xmath31 , the two fluids decouple and we have the two separate families of modes of equations ( [ mod1 ] ) and ( [ mod2 ] ) . we now consider the dissipative terms due to mutual friction , i.e. to the drag parameter @xmath24 . in order to keep the results tractable we take @xmath68 . it is still however impractical to consider the whole solution , so let us first of all consider modes propagating along the @xmath39 axis ( @xmath76 ) . in this case one has two families of inertial modes , one which is undamped with dispersion relation = 2[inertia1 ] and one which is affected by mutual friction : = 2^-i2 [ inertia2 ] where we remind the reader that @xmath77 $ ] and @xmath78 $ ] . the results in ( [ inertia1 ] ) and ( [ inertia2 ] ) agree well with those of @xcite , in which the authors show that there is one class of inertial modes that corresponds to the fluids co - moving and is undamped ( in the absence of chemical coupling ) and another class of counter - moving modes that is rapidly damped by mutual friction . let us now examine the case of tkachenko waves propagating perpendicular to the rotation axis ( @xmath72 ) . in figure [ res1 ] we plot the frequency of the modes for @xmath79 and @xmath80 hz , as a function of @xmath24 in the weak drag regime . for large values of the proton fraction @xmath81 we recover the solution of @xcite , in which the real part of the frequency vanishes and the damping becomes large for values of the drag parameter such that the mutual friction damping timescale @xmath82 is approximately equal to the the tkachenko wave period @xmath83 with @xmath84 . in fact close to this value the mode has an extra purely imaginary root , as found by @xcite . for more realistic values of the proton fraction we see , however , that the pathological behaviour disappears and even though the damping is stronger when the mutual friction timescale is close to the period of the modes , the real part does not vanish , the mode is always oscillatory , and there are always only two purely imaginary roots . = 0.96 and @xmath81=0.1 . we take @xmath79 . for @xmath85 we recover the results of @xcite : the real part of the mode vanishes when the mutual friction damping timescale is close to the mode period , and there is an extra purely imaginary root . for the more realistic , but still large , value of @xmath81=0.1 we see that , on the other hand , the mode is always oscillatory , although the imaginary part is larger when the damping timescale and mode period are similar . for higher values of the drag the frequency of the mode is reduced to @xmath86 of the original value.,title="fig:",height=207]=0.96 and @xmath81=0.1 . we take @xmath79 . for @xmath85 we recover the results of @xcite : the real part of the mode vanishes when the mutual friction damping timescale is close to the mode period , and there is an extra purely imaginary root . for the more realistic , but still large , value of @xmath81=0.1 we see that , on the other hand , the mode is always oscillatory , although the imaginary part is larger when the damping timescale and mode period are similar . for higher values of the drag the frequency of the mode is reduced to @xmath86 of the original value.,title="fig:",height=207 ] up to now we have assumed that the vortex lines are free to move and experience a drag force as they do so . however it is commonly believed that vortex lines can interact strongly with lattice points in the neutron star crust and pin to them , in such a way that they are forced to move with the charged components of the star @xcite . the nature of such a pinning force is well beyond the scope of this paper , but to study the propagation of tkachenko waves in this scenario it is sufficient to consider an unspecified force @xmath87 acting on the vortices such that they are forced to move with the proton fluid , i.e. such that v_v^i = v_^i [ velo ] in this case the equation of force balance for vortices takes the form : _ n_v_ijk^j(v_^k - v_^k)-_+f^=0[pinnedv ] where there is no drag force acting , as the vortex lines flow with the protons . if we now consider the neutron and proton fluid there will be a reaction force @xmath88 acting on a the protons and the magnus force acting on the neutrons . making use of equation ( [ pinnedv ] ) we can thus cast the euler equations in the form : & & _ t ( v_i^+ _ w_i^ ) + 2_ijk ^j v_^k + _ i _ = -2_ijk ^j(v_^k - v_^k ) + & & _ t ( v_i^- _ w_i^ ) + 2_ijk ^j v_^k + _ i _ = 2_ijk^j(v_^k - v_^k)-_i [ eulersp ] which lead to : & & -i|v_i^(1 - _ ) -i_barv_i^+ 2_ijk ^j |v_^k + i k_i | _ = -2_ijk ^j(|v_^k-|v_^k ) + & & -i|v_i^(1 - _ ) -i_|v_i^+ 2_ijk ^j |v_^k + i k_i | _ = 2_ijk^j(|v_^k-|v_^k)-_i [ eulersp2 ] the characteristic equation given by the equations in ( [ eulersp2 ] ) , together with the condition in ( [ velo ] ) can be obtained by calculating the determinant of the matrix @xmath67 given in ( [ ap2]-[ap22 ] ) . this leads to two families of modes : ^2&=&=4 ^ 2()^2()^2+c_t^2k^2()^4 + ^2&=&=4 ^ 2()^2+c_t^2k^2()^4(1-x _ ) these are once again mixed inertial - tkachenko waves , but we see that in the limit @xmath73 the tkachenko waves disappear and we are left with only one family of inertial modes . this resembles the situation in superfluid @xmath3he , in which even a small amount of pinning swamps the contribution due to lattice elasticity and transforms the tkachenko waves into inertial waves @xcite . it is well known from the study of superfluid @xmath3he that compressibility can have a drastic effect on the mode structure @xcite . including compressibility in the equations of motion for the superfluid leads to a dispersion relation of the form @xcite : ^2= where @xmath89 is the sound speed . in the long wavelength limit ( @xmath90 ) the nature of the mode is thus profoundly altered and the dispersion relation is no longer linear in @xmath91 , but rather parabolic , leading to the so - called `` soft '' tkachenko wave frequency : k^2 in the study of @xmath3he the long wavelength limit is , however , mainly of theoretical interest , as one would need containers of several hundreds of meters in diameter to explore it experimentally . the situation is very different for becs as , in contrast with a strongly interacting bose liquid such as @xmath3he , they are weakly interacting bose gases with low sound speeds for which the effect of compressibility is important at high rotation rates . for becs the compressible tkachenko wave spectrum has thus been studied both theoretically @xcite and experimentally @xcite . let us now consider a realistic neutron star . the situation is clearly quite complex as not only can we be in the long wavelength limit ( @xmath92 ) for the more rapidly rotating pulsars , but one also has to account for multi - fluid effects and chemical coupling between the different constituents via the equation of state . one can not , in general , assume incompressibility for the proton and neutron fluids and it is clearly of great interest to adapt our formalism to include the effects of compressibility and chemical coupling . to study this problem it is now advantageous to write the perturbation equations in the form of equations ( [ tutto1])-([tutto3 ] ) . in the plane wave approximation the euler equations take the form ( in the cowling approximation ) : & & -i|v_i + i |p-_i p + 2_ijk^j|v^k=-(1-x_)_i + & & -i(1-|)|w_i+i k_i|+2_ijk^j|w^k=-2 ( i|_i+|v_i+(1-x_)|w_i)+_i [ eulkt1 ] and the continuity equations can be written as : & & -i|+ik_j|v^j+|v^j_j=0 + & & -i|x_+i x_(1-x _ ) k_j|w^j+|w^j_j[x_(1-x_)]=0,[contkt1 ] while the equation of force balance still takes the form in ( [ forceb ] ) . as we are now considering compressible matter , we will also need an equation of state for the perturbations . choosing to work with the density ( @xmath93 ) and proton fraction ( @xmath94 ) perturbations , one can write : thermodynamical variables in equation ( [ eos ] ) should be derived from a fully consistent multi - parameter equation of state , which should also allow us to calculate the entrainment parameters and the superfluid gaps for neutrons and protons . however , not only is such an equation of state not currently available , but its use would also be beyond the scope of our simplified plane wave analysis . in order to keep the problem tractable we shall assume that our background model is described by an @xmath95 polytrope and use for the perturbations two simplified analytic equations of state , that are essentially extensions of a single - fluid @xmath95 polytrope . first of all we shall consider the equation of state of @xcite , which we refer to as model a. in this case we have : & & ( ) = c_s^2,()= + & & ( ) = , ( ) = where @xmath89 is the sound speed of the background we shall then consider a second model , in order to understand the importance of the chemical coupling on the mode structure . this model , which we refer to as model b , is essentially a re - parametrisation of the model ii equation of state in @xcite ( also used in @xcite where it is denoted as model b0 ) and takes the form : & & ( ) = c_s^2,()= + & & ( ) = , ( ) = we can thus study the behaviour of the solutions to are problem as we vary the parameters @xmath96 and @xmath97 . clearly model a corresponds to the case @xmath98 , @xmath99 . the sound speed in the background , for an @xmath95 polytrope , takes the form c_s=2k where @xmath100 depends only on the stellar radius . however in our plane wave approximation we shall assume that the background quantities vary over a length - scale greater than that of the oscillations , and thus take them to be constant and neglect their gradients . this approximation is not necessarily justified , as in the crust the density and pressure vary by several orders of magnitude over a length - scale of approximately 1 km , which is comparable with the longest wavelengths we consider for our tkachenko waves . it is , however , a reasonable approximation for shorter wavelengths and in the neutron star core . the sound speed will thus be a constant in our formulation and specifically we take @xmath101 cm s@xmath17 . we also take the proton fraction , which in a rigorous description should also be derived from the equation of state , as a constant and will study the effect that varying it can have on the modes . needless to say , future work should aim to relax this approximation and consider a fully stratified neutron star . finally let us remark that for simplicity we take @xmath102 in the following discussion . we have experimented with varying the parameter @xmath103 between -0.8 and 0.8 , but it is found to have very little effect on the dispersion relation . let us consider , first of all , the undamped propagation of waves in a neutron star . we thus take @xmath104 and , to keep the problem tractable , @xmath102 , and solve the characteristic equation obtained form the determinant of the matrix @xmath67 given in ( [ ap3 ] ) . as a first step we focus on model a. the results are two families of modes , such that : ^2&=&4 ^ 2- [ cacio1 ] + ^2&=&(4 ^ 2 + k^2 c_s^2)+ as we can see we obtain sound waves and rotationally corrected tkachenko waves , for which the main contribution to the frequency is now given by the stellar rotation frequency . this result is somewhat surprising , as in this limit the classical tkachenko waves no longer exist and the vortex elasticity simply provides a small correction to what is , in essence , the frequency of an inertial wave . in this case , for a pulsar rotating at @xmath105hz , the frequency of the tkachenko waves would be much too high to explain the observed periodicities of @xmath106 days observed in the timing residuals . such a drastic modification in the dispersion relation clearly needs to be investigated in more detail . we have , after all , used a simplified version of the equation of state , so let us turn our attention to model b in order to understand how varying the parameters ( and thus the coupling between the components ) can affect the mode structure and whether there is a reasonable set of parameters for which one can still obtain the usual tkachenko waves . in the case of model b the extra parameters make it necessary to solve the characteristic equation numerically , but they also allow us extra freedom to explore different regimes for the equation of state . first of all let us eliminate all chemical coupling by setting @xmath107 . in figure [ variob ] we plot the mode frequency for varying @xmath97 at different rotation rates . we see that in this case there is a mode with the frequency of the classical tkachenko mode and furthermore varying @xmath97 has very little effect on its frequency . if we now keep @xmath97 fixed and vary the rotation rate of the star we see in figure [ varioom ] that for low rotation rates the frequency of the tkachenko mode is the classical one . for higher rotation rates the classical frequency increases but the frequency of the tkachenko mode tends to that of the so called `` soft '' tkachenko mode @xmath108 . the result is thus that , for longer wave - length of order the stellar radius , there is always a tkachenko mode with periods consistent with the @xmath109 day variability typical of pulsar timing noise . this picture is clearly very different from that of the previous section , in which the tkachenko waves had essentially disappeared , so let us investigate how re - instating the chemical coupling and varying the parameter @xmath96 can affect the mode structure . , whilst keeping @xmath107 . we take @xmath110 . we can see that we have two families of high frequency sound waves and then the tkachenko waves , the frequency of which is shifted from the classical value at higher rotation rates . this is expected as for higher rotation rates the effects of compressibility become more important . furthermore it is clear from the graph that , although varying @xmath97 has a small effect on the frequency of the sound waves , it has no effect on the frequency of the tkachenko waves . in these plots we have set @xmath102 , @xmath79 and taken @xmath111.,title="fig:",height=207 ] , whilst keeping @xmath107 . we take @xmath110 . we can see that we have two families of high frequency sound waves and then the tkachenko waves , the frequency of which is shifted from the classical value at higher rotation rates . this is expected as for higher rotation rates the effects of compressibility become more important . furthermore it is clear from the graph that , although varying @xmath97 has a small effect on the frequency of the sound waves , it has no effect on the frequency of the tkachenko waves . in these plots we have set @xmath102 , @xmath79 and taken @xmath111.,title="fig:",height=207 ] and to the `` soft '' tkachenko wave frequency @xmath108 . as we can see the frequency tends to that of the `` soft '' mode for higher rotation frequencies ( in the millisecond range , which is that of the fastest known pulsars ) and is slightly modified by multi - fluid effects for high values of @xmath81 . in these plots we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] and to the `` soft '' tkachenko wave frequency @xmath108 . as we can see the frequency tends to that of the `` soft '' mode for higher rotation frequencies ( in the millisecond range , which is that of the fastest known pulsars ) and is slightly modified by multi - fluid effects for high values of @xmath81 . in these plots we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] in figure [ variotutto ] we plot the mode frequency for a stellar rotation frequency of 10 hz varying @xmath96 whilst keeping @xmath97 fixed . the result is now much more intriguing as one still has one family of sound waves , but there is then an avoided crossing between the second family of sound waves and the tkachenko waves , with the frequency of the sound wave becoming that of a classical tkachenko wave as we vary @xmath96 . the real part of frequency of the tkachenko wave , on the other hand , vanishes after the avoided crossing and one obtains two purely imaginary roots . summarising there is a vast portion of parameter space in which one has a mode close to the frequency of a classical tkachenko wave but there exists a small region ( which thus includes model a for which @xmath112 ) where the avoided crossing occurs , in which the tkachenko mode is not oscillatory in nature and the frequency of the sound waves is too high to account for the slow variability of pulsar timing residuals . if we now increase the stellar rotation rate we can see from figure ( [ varioomt ] ) that the region in which the tkachenko mode is not oscillatory becomes even larger , allowing for vast portions of parameter space in which there is no mode close to the tkachenko frequency . however we have not yet considered the impact of mutual friction damping , which could potentially limit even more the range of parameters for which one has long lived tkachenko oscillations . let us thus move on to consider the full problem . . we see that there is still a family of sound waves ( indicated as @xmath113 ) , but there is now an avoided crossing between the second family of sound waves ( @xmath114 ) and the tkachenko waves ( @xmath115 ) , with the frequency of the sound wave becoming that of a classical tkachenko wave . there is thus a vast portion of parameter space in which one has a mode close to the frequency of a classical tkachenko wave , but there exists a small region where the mode crossing occurs , in which the tkachenko mode is not oscillatory in nature and the frequency of the sound waves is too high to account for the slow variability of pulsar timing residuals . once again we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] . we see that there is still a family of sound waves ( indicated as @xmath113 ) , but there is now an avoided crossing between the second family of sound waves ( @xmath114 ) and the tkachenko waves ( @xmath115 ) , with the frequency of the sound wave becoming that of a classical tkachenko wave . there is thus a vast portion of parameter space in which one has a mode close to the frequency of a classical tkachenko wave , but there exists a small region where the mode crossing occurs , in which the tkachenko mode is not oscillatory in nature and the frequency of the sound waves is too high to account for the slow variability of pulsar timing residuals . once again we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] and @xmath114 and tkachenko waves @xmath115 ) , normalised to the classical tkachenko wave frequency , for two different rotation rates of the star and for a varying parameter @xmath96 , whilst keeping @xmath116 . we take @xmath110 . we can see that as the rotation rate increases not only does the tkachenko wave frequency decrease as expected , but there is also a vast region of parameter space in which the tkachenko mode dissapears . in these plots again we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] and @xmath114 and tkachenko waves @xmath115 ) , normalised to the classical tkachenko wave frequency , for two different rotation rates of the star and for a varying parameter @xmath96 , whilst keeping @xmath116 . we take @xmath110 . we can see that as the rotation rate increases not only does the tkachenko wave frequency decrease as expected , but there is also a vast region of parameter space in which the tkachenko mode dissapears . in these plots again we have set @xmath102 and taken @xmath79.,title="fig:",height=207 ] the inclusion of mutual friction makes the problem considerably more complicated , and once again the characteristic equation must be solved numerically . the picture that emerges is however not very different from that of the previous section . one still finds a high frequency family of sound waves , largely unaffected by mutual friction , then a second family of sound waves at lower frequency and a family of tkachenko waves . in addition one has two purely imaginary roots to the characteristic equation . let us focus on the tkachenko waves and on the lower frequency sound waves . as we can see in figure ( [ comparo ] ) the behaviour of the modes depends strongly on the chemical coupling . for @xmath117 and slow rotation of the star one has tkachenko waves close to the classical frequency , while for @xmath118 one has an avoided crossing between the second family of sound waves and the tkachenko waves . in both cases the waves oscillating close to tkachenko frequency are strongly damped by mutual friction in a narrow range of the parameter @xmath24 for large values of @xmath81 , exactly as in the incompressible case . for more realistic values of @xmath81 we find that , as expected , the damping is negligible . hz , for @xmath79 , @xmath102 and @xmath72 . in the top panel we take @xmath119 and @xmath120 , in the bottom panel we take @xmath119 and @xmath120.for @xmath117 and slow rotation of the star one has tkachenko waves close to the classical tkachenko frequency , while for @xmath118 one has an avoided crossing between the second family of sound waves and the tkachenko waves . in both cases the waves oscillating close to tkachenko frequency are strongly damped by mutual friction in a narrow range of the parameter @xmath24 and for large values of @xmath81 . as expected the mutual friction damping is weak in both cases for lower , more realistic values of @xmath81 . , title="fig:",height=207 ] hz , for @xmath79 , @xmath102 and @xmath72 . in the top panel we take @xmath119 and @xmath120 , in the bottom panel we take @xmath119 and @xmath120.for @xmath117 and slow rotation of the star one has tkachenko waves close to the classical tkachenko frequency , while for @xmath118 one has an avoided crossing between the second family of sound waves and the tkachenko waves . in both cases the waves oscillating close to tkachenko frequency are strongly damped by mutual friction in a narrow range of the parameter @xmath24 and for large values of @xmath81 . as expected the mutual friction damping is weak in both cases for lower , more realistic values of @xmath81 . , title="fig:",height=207 ] hz , for @xmath79 , @xmath102 and @xmath72 . in the top panel we take @xmath119 and @xmath120 , in the bottom panel we take @xmath119 and @xmath120.for @xmath117 and slow rotation of the star one has tkachenko waves close to the classical tkachenko frequency , while for @xmath118 one has an avoided crossing between the second family of sound waves and the tkachenko waves . in both cases the waves oscillating close to tkachenko frequency are strongly damped by mutual friction in a narrow range of the parameter @xmath24 and for large values of @xmath81 . as expected the mutual friction damping is weak in both cases for lower , more realistic values of @xmath81 . , title="fig:",height=207 ] hz , for @xmath79 , @xmath102 and @xmath72 . in the top panel we take @xmath119 and @xmath120 , in the bottom panel we take @xmath119 and @xmath120.for @xmath117 and slow rotation of the star one has tkachenko waves close to the classical tkachenko frequency , while for @xmath118 one has an avoided crossing between the second family of sound waves and the tkachenko waves . in both cases the waves oscillating close to tkachenko frequency are strongly damped by mutual friction in a narrow range of the parameter @xmath24 and for large values of @xmath81 . as expected the mutual friction damping is weak in both cases for lower , more realistic values of @xmath81 . , title="fig:",height=207 ] if we now increase the rotation rate in the case @xmath117 one finds that the frequency of the tkachenko waves approaches that of the `` soft '' mode and mutual friction only weakly damps the mode for realistic values of @xmath81 . the picture is considerably different if we take @xmath118 , as can be seen from figure ( [ compres2 ] ) . for a rotation rate of 60 hz one has a sound wave close to the classical tkachenko wave frequency and a highly damped `` soft '' tkachenko wave for low values of @xmath24 . however , if we increase the rotation rate to 100 hz , the frequency of the `` soft '' mode becomes purely imaginary and the sound waves return to being high frequency modes , weakly damped by mutual friction . the picture that emerges is thus that , while for low rotation rates of order of a few hz one has long lived tkachenko oscillations for a vast range of plausible parameters ( except for the particular case of @xmath121 as in model a ) , for higher rotation rates , approaching 100 hz , the situation is radically different and for several choices of parameters there are no tkachenko modes at all . and @xmath122 . for a rotation rate of 60 hz one has a sound wave close to the classical tkachenko wave frequency and a highly damped `` soft '' tkachenko wave for low values of @xmath24 . at 100 hz , the frequency of the `` soft '' mode becomes purely imaginary and the sound waves return to being high frequency modes , weakly damped by mutual friction.,title="fig:",height=207 ] and @xmath122 . for a rotation rate of 60 hz one has a sound wave close to the classical tkachenko wave frequency and a highly damped `` soft '' tkachenko wave for low values of @xmath24 . at 100 hz , the frequency of the `` soft '' mode becomes purely imaginary and the sound waves return to being high frequency modes , weakly damped by mutual friction.,title="fig:",height=207 ] finally we examine , as in section iii c , the case of perfect pinning . the euler equations can be written as : & & -i|v_i + i |p-_i p + 2_ijk^j|v^k=-(1-x_)_i + & & -i(1-|)|w_i+i k_i|-2_ijk^j|w^k=-_i [ eulkpin ] together with the pinning condition @xmath123 . in the limit of no entrainment ( @xmath102 ) for purely transverse propagation ( @xmath72 ) , one finds that , as in the incompressible case , the spectrum depends heavily on the value of the proton fraction @xmath81 . for slow rotation rates ( of the order @xmath124 hz ) one always has a tkachenko mode for low values of the proton fraction ( less than @xmath125 ) . however the situation changes for higher rotation rates . for a stellar rotation rate of 100 hz , one can see from figure ( [ pin ] ) that for the , possibly more realistic , case of small values of @xmath81 one has a tkachenko mode only for @xmath126 . however for larger values of @xmath81 the opposite is true and the tkachenko mode only exists in the limit @xmath127 . if @xmath128 ( model a ) one has , as expected , no tkachenko waves for any value of the proton fraction . in the case of perfect pinning . we consider purely transverse propagation , @xmath79 and @xmath102 . for the , possibly more realistic , case of small values of @xmath81 one has a tkachenko mode only for @xmath126 . however for larger values of @xmath81 the opposite is true and the tkachenko mode only exists in the limit @xmath127 , but its frequency vanishes for @xmath129.,title="fig:",height=207 ] in the case of perfect pinning . we consider purely transverse propagation , @xmath79 and @xmath102 . for the , possibly more realistic , case of small values of @xmath81 one has a tkachenko mode only for @xmath126 . however for larger values of @xmath81 the opposite is true and the tkachenko mode only exists in the limit @xmath127 , but its frequency vanishes for @xmath129.,title="fig:",height=207 ] in this paper we present a formalism for the inclusion of vortex lattice elasticity in the multi - fluid hydrodynamics of superfluid neutron stars . as a first step we consider incompressible neutron and proton matter , in order to make contact with the well known results for superfluid @xmath3 he . we obtain the standard dispersion relation for tkachenko waves and find that , in the limit in which @xmath130 , mutual friction completely damps out the oscillations in less than a period when the damping timescale due to the drag is close to the mode period ( as found by @xcite ) . however in a neutron star one will , in general , have that @xmath131 , and in this case we always find weakly damped oscillatory solutions . it should be stressed that a more realistic model should include other sources of damping , such as shear and bulk viscosity , and the calculation should be performed for global modes in spherical symmetry , without assuming constant background quantities . the main focus of this work is on the effect of compressibility and chemical coupling on the mode spectrum . we find that for slow rotation rates ( a few hz ) one can , in general , obtain solutions that correspond to low frequency tkachenko waves and could explain the timing noise in older pulsars . however , for particular choices of the eos parameters , such as model a , there are no tkachenko waves , but only modified inertial waves and sound waves , that oscillate at frequencies that are to high to have any connection with the timing noise . finally the situation is considerably more complicated for more rapidly roatating nss ( above 100 hz ) for which there is now a large portion of parameter space for which there are no propagating tkachenko waves . it is thus clearly imperative to obtain more stringent constraints on the eos from nuclear physics , in order to understand if the regions of parameter space in which one has no tkachenko waves are of physical significance or not . in the light of these uncertainties , it is still very much an open question whether or not the period of tkachenko waves in a realistic neutron star could explain the observed periodicity of @xmath132 days in the timing residuals of psr b1828 - 11 ( which is rotating at @xmath133 hz ) and the timing noise in other pulsars , or if it is likely to power low frequency precessional motion of the star . our results , however , indicate that for a large range of physical parameters long period tkachenko waves can in fact propagate in a ns interior and are likely to play a role in the dynamics of the system . finally let us remark that we have considered perturbations of a co - moving background . while this is not a bad approximation in many situations , it is possible that if the vortices are pinned to the crust a significant lag could build up between the charged component and the superfluid neutrons . this can lead to a series of short wave - length instabilities ( @xcite,@xcite ) that are likely to have an impact on pulsar glitches and on the stellar response to external torques , such as those experienced by neutron stars in accreting systems . we plan to relax the assumption of a co - moving background and explore the consequences on these physical scenarios in future work . i wish to thank n.andersson and d.i.jones for useful discussions . i acknowledge support from stfc via grant number pp / ee001025/1 and via a eu marie - curie intra - european fellowship , project number 252470 , amxp dynamics . this work was partially supported by compstar , a research training network programme of the european science foundation for the incompressible case the equations of motion ( [ eulersk1]-[eulersk ] ) can be cast in the form : k_ij= ( c c c c c c c c c c & ^_ij & & & & ^_ij & & & ^_ij & + & & & & & & & & & + & ^_ij & & & & ^_ij & & & ^_ij & + & & & & & & & & & + & ^_ij & & & & ^_ij & & & _ ij ) ( c^j _ + + ^j ) [ kappa1 ] where ^j_=(|v^j _ , | _ ) ^j_=(|v^j _ , | _ ) and we recall that the displacement @xmath134 only has components in the plane perpendicular to the vortices , i.e. the @xmath135 plane in our formulation . the components of the matrix @xmath67 take the explicit form : ^_ij= ( c c c c c c c -i(1- _ ) & & -2 & & 0 & & i k + & & & & & & + 2 & & -i(1- _ ) & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i(1- _ ) & & i k + & & & & & & + k & & 0 & & k & & 0 ) , ^_ij= ( c c c c c c c -i(1-_)+2 ( ) & & -2 & & 0 & & i k + & & & & & & + 2 & & -i(1-_)+2 ( ) & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i(1-_)+2 ( ) & & i k + & & & & & & + k & & 0 & & k & & 0 ) , ^_ij= ( c c c c c c c -i_-2 & & 0 & & 0 & & 0 + & & & & & & + 0 & & -i_-2 & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i_-2 & & 0 + & & & & & & + 0 & & 0 & & 0 & & 0 ) , ^_ij= ( c c c c c c c -i _ & & 0 & & 0 & & 0 + & & & & & & + 0 & & -i _ & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i _ & & 0 + & & & & & & + 0 & & 0 & & 0 & & 0 ) , ^_ij= ( c c c c c - & & - & & 0 + & & & & + & & - & & 0 ) , ^_ij= ( c c c c c - & & & & 0 + & & & & + - & & - & & 0 ) , ^_ij= ( c c c -c_t^2(k)^2 - 2i & & 0 + & & + 0 & & c_t^2(k)^2 - 2i ) , ^_ij= ( c c c 2i & & 0 + & & + 0 & & 2i ) , _ ij= ( c c c -i- & & - + & & + - & & -i+ ) . where we recall the definition @xmath61 . the characteristic equation then follows from the determinant of the matrix @xmath67 . in the case of perfect pinning the elements of the matrix @xmath67 are modified in the following way : ^_ij= ( c c c c c c c -i(1- _ ) & & 0 & & 0 & & i k + & & & & & & + 0 & & -i(1- _ ) & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i(1- _ ) & & i k + & & & & & & + k & & 0 & & k & & 0 ) , [ ap2 ] ^_ij= ( c c c c c c c -i(1- _ ) & & 2 ( ) & & 0 & & i k + & & & & & & + -2 ( ) & & -i(1- _ ) & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i(1- _ ) & & i k + & & & & & & + k & & 0 & & k & & 0 ) , ^_ij= ( c c c c c c c -i _ & & -2 & & 0 & & 0 + & & & & & & + 2 & & -i _ & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i _ & & 0 + & & & & & & + 0 & & 0 & & 0 & & 0 ) , ^_ij= ( c c c c c c c -i _ & & -2 ( ) & & 0 & & 0 + & & & & & & + 2 ( ) & & -i _ & & 0 & & 0 + & & & & & & + 0 & & 0 & & -i _ & & 0 + & & & & & & + 0 & & 0 & & 0 & & 0 ) , ^_ij= ( c c c c c 1 & & 0 & & 0 + & & & & + 0 & & 1 & & 0 ) , _ ij= ( c c c i & & 0 + & & + 0 & & i ) . ^_ij=0 ^_ij=0 , ^_ij= ( c c c -c_t^2(k)^2 & & 0 + & & + 0 & & c_t^2(k)^2 ) , [ ap22 ] in the compressible case we need to cast the equations of motion ( [ eulersk1]-[eulersk ] ) in the form : k_ij= ( g.hobbs , a. g. lyne , m. kramer , mnras * 402 * , 1027 ( 2010 ) . i. stairs , a. g. lyne , s. shemar , nature * 406 * , 484 ( 2000 ) . j. shaham , apj * 214 * , 251 ( 1977 ) . a. sedrakian , i. wasserman , j. m. cordes , apj * 524 * , 341 ( 1991 ) . k. glampedakis , n. anderssson , d. i. jones , phys . lett . * 100*(8 ) , 081101 ( 2008 ) . a. lyne , g. hobbs , m. kramer , i. stairs , b. stappers , science * 329 * , 408 ( 2010 ) . j. noronha , a. sedrakian , phs . d * 77 * , 023008 ( 2008 ) . m. ruderman , nature * 225 * 619 ( 1970 ) s. b. popov , astrophys . space sci . * 317 * , 175 ( 2008 ) v. k. tkachenko , sov . phys .- jetp * 23 * 1049(1966 ) c. d. andereck & w. i. glaberson , jltp * 48 * 257 ( 1982 ) g. baym , phys . letters * 91 * 110402 ( 2003 ) e.b . sonin , phys . a * 71 * , 011603 ( 2005 ) g. baym , e. chandler , jltp * 50 * , 57 ( 1983 ) e. chandler , g. baym , jltp * 62 * , 119 ( 1986 ) i. easson , c. j. pethick , apj * 333 * , 880 ( 1979 ) n. andersson , g. l. comer , cqg * 23 * , 5503 ( 2006 ) g. mendell , apj * 380 * , 515 ( 1991 ) b. haskell , n. andersson , a. passamonti , mnras * 397 * , 1464 ( 2009 ) r. prix , phys . d * 69 * , 43001 ( 2004 ) k. glampedakis , n. andersson , l. samuelsson , arxiv:1001.4046v1 h. e. hall , w. f. vinen , proc . a * 238 * , 215 ( 1956 ) n. andersson , t. sidery , g. l. comer , mnras * 368 * , 162 ( 2006 ) p.b . jones , mnras * 243 * , 257 ( 1990 ) r. i. epstein , g. baym , apj * 387 * , 276 ( 1992 ) p.b . jones mnras * 257 * , 501 ( 1992 ) n. andersson , k. glampedakis , b. haskell , phys . rev . d * 79 * , 103009 ( 2009 ) p.w anderson , n. itoh , nature * 256 * , 25 ( 1975 ) m.a . alpar , apj * 213 * , 527 ( 1977 ) r.i . epstein , g. baym , apj * 328 * , 680 ( 1988 ) p.m. pizzochero , l. viverit , r.a . broglia , phys . . lett . * 79 * , 3347 ( 1997 ) p. donati , p.m. pizzochero , nucl . a * 742 * , 363 ( 2004 ) e.b . sonin , rev . phys . * 59 * , 87 ( 1987 ) m. cozzini , l.p . pitaevskii , s. stringari , phys , rev . lett . * 92 * , 220401 ( 2004 ) i. coddington , p. engels , v. schweikhard , e.a . cornell , phys . lett . * 91 * , 100402 ( 2003 ) r. reatto , phys . d * 167 * , 191 ( 1968 ) r. prix , m. rieutord , a&a * 393 * , 949 ( 2002 ) a. passamonti , b. haskell , n. andersson , mnras * 396 * , 951 ( 2009 ) t. sidery , n. andersson , g. l. comer , mnras * 385 * , 335 ( 2008 ) k. glampedakis , n. andersson , phys . * 102 * , 141101 ( 2009 )
a generalized version of the coupled coherent states method for coherent states of arbitrary lie groups is developed . in contrast to the original formulation , which is restricted to frozen - gaussian basis sets , the extended method is suitable for propagating quantum states of systems featuring diversified physical properties , such as spin degrees of freedom or particle indistinguishability . the approach is illustrated with simple models for interacting bosons trapped in double- and triple - well potentials , most adequately described in terms of and bosonic coherent states , respectively .
1511.02728
a vast number of physical systems exhibit the property that some of their parts behave in a sort of classical way , meaning that quantum effects play only a minor role in the description of those parts . this distinctive classical character of specific degrees of freedom is a much welcomed attribute , for it makes possible the development of tractable computational approaches capable of carrying out the time - evolution of complex quantum systems , being thus the fundamental property upon which time - dependent trajectory - guided methods are based . in this kind of technique quantum states are represented in terms of time - dependent basis functions or ` configurations ' . within a single configuration , those degrees of freedom in which quantum effects are negligible are evolved according to classical equations of motion . this classical dynamics may be prescribed in a number of different ways and different choices correspond to different propagation schemes . in spite of the fact that individual configurations have some of their parts bound to obey classical laws , a complete quantum solution is in principle attainable by combining many configurations . the key idea behind such ` multiconfigurational ' approaches is that trajectory - guided basis functions are more likely to remain in the important regions of the hilbert space , thus being more efficient at representing the quantum state in the sense that a reduced number of basis elements is required in order to achieve an accurate description . and it is precisely through a significant reduction in the number of basis functions needed to propagate the system that one hopes to escape the exponential scaling of basis - set size with dimensionality typical of standard static - basis formulations . this ` mixed quantum - classical ' picture is adopted in many methods of quantum chemistry . @xcite a recurrent theme in this field is the development of techniques which , by means of equally simple recipes to guide the basis functions , would be readily applicable to systems presenting authentically non - classical qualities , such as spin degrees of freedom or particle exchange symmetry . several works have been directed to that purpose , most often aiming at a time - dependent description of the electronic structure of molecules during non - adiabatic processes . one particular example of such a recipe is the classical model for electronic degrees of freedom proposed by miller and white @xcite in which a second - quantized fermionic hamiltonian is properly reduced to a classical function wherein number and phase variables play the role of generalized coordinates . in contrast , a more ` mechanistic ' approach to fermion dynamics is found on the multiconfigurational formula proposed by kirrander and shalashilin @xcite in which the basis functions consist of antisymmetrized frozen gaussians @xcite guided by fermionic molecular dynamics . @xcite yet if one seeks to describe non - classical degrees of freedom by means of classical - like variables , then generalized coherent states defined in the group - theoretical sense are indisputably the most appropriate tools to be employed . there are many reasons supporting this assertion . first of all , coherent states are defined in terms of non - redundant parameters and equations of motion for these parameters can be readily obtained from the time - dependent variational principle . @xcite in this way an optimized time - evolution can be assigned in an unambiguous manner . moreover , they are naturally able to capture the desired symmetries of the system which are maintained during propagation . furthermore , the coherent - state parameters evolve in a classical phase space in the strict sense of the word , hence we automatically have at our disposal the wealth of analytical techniques applicable to hamiltonian systems . at the same time , through this intimate connection to classical dynamics , coherent states provide a compelling classical interpretation to quantum phenomena , in so far as individual configurations are chosen to represent familiar objects i.e. in such a way that it is meaningful to discuss the dynamics of the system in terms of their trajectories . to this extent , coherent states which are also minimum uncertainty states ( provided a proper meaning is assigned to the term ` uncertainty ' ) @xcite are valuable tools in enhancing our comprehension with respect to the semiclassical features of the quantum system under investigation . in addition , the group - theoretical formalism secures a well - defined integral form for the coherent - state closure relation @xcite a crucial element to the developments hereby presented . this list of virtues is not exhausted and other advantages of a generalized coherent - state representation will be pointed out throughout the paper . along these lines , van voorhis and reichman @xcite have considered a number of alternative representations of electronic structure making use of different coherent - state parametrizations and also examined their adequacy to a variety of systems . @xcite within the context of non - adiabatic molecular dynamics , a particularly interesting coherent - state representation @xcite is found in the simplest and most throughly investigated version of the electron - nuclear dynamics theory , developed by deumens , hrn and collaborators . @xcite the same kind of coherent state has been discussed in detail , within the field of nuclear physics , by suzuki and kuratsuji . @xcite turning to bosonic dynamics , a semiclassical trajectory - based formula in the @xmath2 coherent - state representation has been recently derived and successfully applied to a model of trapped bosons . @xcite these methods are representative of the kind of technique one has in mind when a description of intrinsically quantum degrees of freedom in terms of classical - like variables is desired . however , they either constitute approximate single - configuration approaches @xcite or involve complicated trajectories that live in a duplicated phase space , @xcite sometimes relying on sophisticated root - search techniques in order to determine them . @xcite it seems that a multiconfigurational , generalized coherent - state approach , based on simple as opposed to duplicated phase - space trajectories would be more in the spirit of the familiar time - dependent guided - basis methods of quantum chemistry . @xcite this is precisely the direction we take here . in this work a quantum initial value representation method which employs a generalized coherent - state basis set guided by classical trajectories is formulated . the resulting propagation scheme is regarded as a natural extension of the coupled coherent states technique of shalashilin and child @xcite in so far as ( i ) basis - set elements represent localized quantum states ; ( ii ) each element evolves independently in a generalized classical phase space and carries an action phase ; and ( iii ) the quantum amplitudes associated with individual elements obey fully coupled equations of motion which present a number of attractive qualities . we begin in [ sec : cs - formalism ] with an overview on generalized coherent states , deliberately avoiding the underlying group - theoretical formalism associated with their construction . in particular , we demonstrate how the time - dependent variational principle leads to classical equations of motion for the coherent - state parameters in a curved phase space . this preliminary discussion is illustrated with specific examples . next , in [ sec : ccs - generalized ] , we set forth to derive the working equations of the generalized coupled coherent states method , first in continuum form and then in terms of a discrete basis set . in the latter case , both unitary and non - unitary versions of the formulas are devised . a primitive method , resulting from propagation of a single configuration , is also discussed . in [ sec : ccs - application ] we use our approach to study model hamiltonians and compare our results against exact quantum data . we also take the opportunity to expose certain particular aspects of the generalized coherent - state formulation . finally , conclusions are drawn in [ sec : conclusion ] . coherent states are most elegantly discussed within the context of group theory and this is the point of view adopted here . we shall not venture into the group - theoretical formalism itself though on that subject the reader is referred to the instructive review by zhang , feng and gilmore @xcite or to a recent paper by the present authors . @xcite here , we will rather follow a more pragmatic approach according to which a coherent state is given in the form of an expansion in a proper set of orthonormal states and work its geometrical properties thereon . coherent states are hilbert space vectors labeled by a complex vector @xmath3 . @xcite they can be understood as the result of a lie - group operator acting on a reference state , which is recovered by setting all entries of the vector @xmath4 to zero . we shall denote a non - normalized coherent state by @xmath5 . these curly _ ket _ states are analytical in @xmath4 , while the _ bra _ states @xmath6{| z \}}$ ] are analytical in the complex conjugate variable , denoted by @xmath7 ; the normalized state @xmath8 is not analytical in @xmath4 for it depends on @xmath7 through the normalization factor @xmath9 . the length @xmath10 of @xmath4 will be identified as the number of degrees of freedom of the classical phase space associated with the dynamics of @xmath4 . coherent states of different groups are characterized by their distinct geometrical properties which , in turn , are described in terms of a function @xmath11 related to the scalar product between two non - normalized coherent states : @xmath12 the _ classical phase - space metric _ @xmath13 is a hermitian matrix obtained by taking the cross derivatives of the real function @xmath14 with respect to its complex arguments , treating @xmath4 and @xmath7 as independent variables : @xcite @xmath15 the ( non - orthogonal ) coherent states span an overcomplete basis of the corresponding hilbert space and a closure relation holds : @xmath16 where the integration domain depends on the specific type of coherent state being considered for semisimple compact lie groups or the heisenberg - weyl group , for example , it extends over the entire @xmath10-dimensional complex plane . in , the _ integration measure _ @xmath17 is defined as below : @xmath18 \prod^{d}_{\alpha=1 } \frac{{\mathrm{d}}^2 z_\alpha}{\pi},\ ] ] where @xmath19 and the constant @xmath20 is determined by normalization of e.g. by setting the expectation value of in the reference state to unity and therefore it depends on the quantum numbers that characterize the particular hilbert space that carries the coherent - state representation ( some examples are found below in [ sub : cs - examples ] ) . in order to shorten the notation , we shall write simply @xmath21 , but keeping in mind that the measure is a real function of both @xmath4 and @xmath7 . the norm - invariant lagrangian @xcite that gives rise to the schrdinger equation is @xmath22 -\frac{{\langle \psi | } \hat{h } { | \psi \rangle}}{{\langle \psi | \psi \rangle}},\ ] ] where @xmath23 is an arbitrary quantum state and @xmath24 represents the system s hamiltonian operator . the time - dependent variational principle ( tdvp ) states that , given a parametrization of @xmath25 in terms of some set of variables , the euler - lagrange equations obtained from translate into equations of motion for those same variables . if the parametrization is flexible enough an exact quantum solution is achieved . on the other hand , if the parametrization contains less than the number of variables needed to span the associated hilbert space , the resulting dynamics will be only approximate . we shall _ define _ the _ classical equations of motion _ for @xmath4 as those equations obtained from the tdvp when a trial state @xmath26 is substituted in that is , a trial state whose dynamics is restricted to the nonlinear subspace consisting only of coherent states . in this case , equation takes the form : @xmath27 -h(z^ { { * } } , z),\ ] ] where the _ classical hamiltonian _ is the diagonal element of the operator @xmath24 in the coherent - state representation : @xmath28{| z \ } } \hat{h } \ketn{z}}{{\ { z | z \ } } } = { \langle z | } \hat{h } { | z \rangle}.\ ] ] by means of the euler - lagrange equations one immediately finds that the dynamics of @xmath4 obeys : @xmath29 and that an equivalent ( complex conjugate ) equation holds for @xmath7 . notice how the coherent - state geometry introduces a curvature in phase space by means of the metric @xmath30 . one now can distinguish between two kinds of coupling between the components of the vector @xmath4 : a _ dynamical coupling _ via @xmath31 , and a _ geometrical coupling _ induced by @xmath30 . the group - theoretical formalism assures us that eq . ( [ eqn : z - dot-1 ] ) describes a hamiltonian system in the most strict sense : the phase space exhibits a symplectic structure , since a nondegenerate poisson bracket can always be established . @xcite furthermore , the measure is invariant under the classical dynamics given by equation , that is , @xmath32 , for any two instants @xmath33 and @xmath34 a property that we recognize as a generalized form of the liouville theorem and that remains valid even when the system s hamiltonian has an explicit time dependence . @xcite finally , we define the _ complex action _ @xmath35 : @xmath36,\end{gathered}\ ] ] where the first term is the time integral : @xmath37 with the lagrangian @xmath38 , given by equation , being evaluated over a classical orbit , satisfying . as can be seen in , the complex action @xmath35 carries imaginary surface terms , which ensure that this is a well - defined analytical function on both its complex arguments : @xmath39 and @xmath40 , and also the time @xmath41 . the derivatives with respect to each of these variables are : [ eqn : action - derivatives ] @xmath42 the above relations are recognized as the signature of a properly defined classical action integral . @xcite yet it is in terms of the real quantity @xmath43 of eq . that our results are most conveniently expressed . @xcite therefore we shall denominate @xmath43 the _ action_. in order to illustrate the formalism presented above , we consider simple examples of coherent states and evaluate some of their geometrical elements , such as the metric matrix @xmath30 and integration measure @xmath44 . canonical coherent states have their functional definition given in terms of a superposition of bosonic fock states with _ unrestricted _ occupation numbers ; if the hilbert space comprises @xmath45 modes , then the non - normalized canonical coherent state is defined by : @xmath46 { | m_1 , \ldots , m_n \rangle},\ ] ] where the vacuum @xmath47 is the reference state and the length of the vector @xmath4 equals the number of bosonic modes : @xmath48 . since the modes are assumed to be orthonormal , it follows immediately from that the overlap between canonical coherent states is @xmath49 thus , using , we identify the function @xmath11 as : @xmath50 from , the phase - space metric matrix @xmath30 is simply the identity matrix : @xmath51 which means that canonical coherent states give rise to a flat phase space and therefore the degrees of freedom are not geometrically coupled . the normalized measure , as defined in , is then trivial : @xmath52 in what concerns semiclassical trajectory - based methods , canonical coherent states are undoubtedly the most widely used type of coherent state . this is so because of the following well - known homomorphism connecting the ladder operators @xmath53 of each bosonic mode in with the position and momentum operators @xmath54 : @xmath55 with @xmath56 = \delta_{\alpha\beta}$ ] and where @xmath57 is an arbitrary constant that sets the appropriate length scale in each mode . the position representation of @xmath8 is then found to be a multidimensional gaussian wavepacket , whose mean position @xmath58 and mean momentum @xmath59 are related to the real and imaginary parts of the complex vector @xmath4 , respectively . furthermore , the dynamics of @xmath60 , as obtained from the time - dependent variational principle , is simply given by hamilton s classical equations of motion in canonical form , the hamiltonian being the mean value given by equation . this so - called ` frozen - gaussian representation ' provides an obviously suitable framework for semiclassical applications . @xcite the @xmath2 bosonic coherent states are suitable for describing systems in a fock space comprising @xmath45 modes and a _ fixed _ total particle number @xmath61 . their non - normalized form is @xmath62 { | m_1 , \ldots , m_n \rangle},\ ] ] where the reference state is @xmath63 . in , the primed summation symbol means that the set of occupation numbers @xmath64 must satisfy the condition @xmath65 . because of this constraint , the number of entries of the vector @xmath4 is one less than the number of modes : @xmath66 . the overlap is easily evaluated from with the help of the multinomial theorem : @xmath67^n;\ ] ] hence , by means of , all geometrical aspects of these coherent states are codified in the function @xmath11 given by : @xmath68.\ ] ] the metric matrix , according to , is : @xmath69 where @xmath70 . clearly , the fixed particle number condition translates into a geometrical coupling among the components of the vector @xmath4 . despite the complications introduced by the curved geometry , the metric s determinant can be evaluated and the integration measure , defined in , is found to be : @xmath71 recently , semiclassical methods employing @xmath2 bosonic coherent states , including an initial value representation based on classical trajectories in a duplicated phase space , have been developed and tested with an su(3 ) model hamiltonian @xcite in [ sub : su3-triple - well ] we shall have the opportunity to revisit that same problem . a particularly interesting @xmath2 coherent state originates when the bosonic fock space has only @xmath72 modes . the @xmath73 states @xmath74 can be put into one - to - one correspondence with the well - known simultaneous eigenstates @xmath75 of the angular momentum operators @xmath76 for a fixed total angular momentum @xmath77 . in this way , the non - normalized @xmath0 coherent states @xcite can be expressed as : @xmath78 which are especially designated as _ atomic _ or _ spin coherent states_. notice that , in this specific case , the complex vector label @xmath4 has dimension @xmath79 and @xmath80 is the reference state . the overlap @xmath81 is easily evaluated using the orthogonality of the @xmath75 states . alternatively , we can simply set @xmath72 and @xmath82 in equation , thus obtaining : @xmath83 which leads to @xmath84 according to equation , it follows that the phase - space metric @xmath30 ( in this case , a scalar ) is simply : @xmath85 the normalized measure is then found to be @xmath86 the natural topology of the spin coherent state is that of the surface of a sphere . in practical applications , one typically writes @xmath4 in terms of angles @xmath87 and @xmath88 : @xmath89 where @xmath90 $ ] and @xmath91 . in these coordinates the integration measure reads : @xmath92 spin coherent states are discussed in more detail in [ sub : su2-double - well ] where we investigate a test model consisting of an @xmath0 hamiltonian . fermionic coherent states of the special unitary group are suitable for describing a number - conserving system of @xmath61 fermions which are allowed to occupy a set of @xmath45 orthonormal single - particle states ( @xmath93 ) . while ultimately arbitrary , these underlying single - particle states are often taken to be eigenstates of the non - interacting part of the hamiltonian or a set of hartree - fock spin orbitals . @xcite a reference state @xmath94 is specified by constructing a slater determinant out of @xmath61 of such spin orbitals ( e.g. the ones having lowest energies ) . these are denoted by @xmath95 with @xmath96 , and are said to belong to the _ occupied _ space . the remaining @xmath97 spin orbitals , denoted by @xmath98 with @xmath99 , are said to belong to the _ virtual _ space . then , the non - normalized fermionic coherent state can written as : @xmath100,\ ] ] where the symbol @xmath101 instructs anti - symmetrization of the direct product of the @xmath61 single - particle states @xmath102 , which are sometimes called _ dynamical orbitals_. @xcite in the context of quantum chemistry the coherent state is designated as the _ thouless parametrization _ of a slater determinant . @xcite here , the label @xmath4 is understood as an @xmath103 matrix , the number of degrees of freedom of the corresponding phase space being @xmath104 . the elements @xmath105 describe the mixing between occupied and virtual spin orbitals in such a way that any single - determinantal state not orthogonal to @xmath106 can be represented by @xmath5 . @xcite the overlap between two distinct thouless - parame - trized slater determinants is readily found to be : @xmath107 where @xmath108 is the identity matrix of size @xmath109 . following the definitions given in [ sec : cs - formalism ] , geometrical properties are determined from the function : @xmath110.\ ] ] from , one finds ( through elementary determinant and matrix identities ) that the phase - space metric can be expressed as : @xmath111_{\nu\mu}[(i_n+z^\dagger z)^{-1}]_{\alpha\beta},\ ] ] where @xmath112 is the @xmath113 identity matrix and the derivatives of @xmath11 were taken with respect to @xmath105 and @xmath114 . despite the complicated outlook of the classical equations of motion for the thouless parameters @xmath4 , as governed by a standard many - body hamiltonian consisting of one- and two - body terms , display a simple structure see , for instance , section iiia in ref . . finally , the integration measure appearing in the closure relation is found to be : @xmath115^{-n } \prod^{n}_{\alpha=1 } \prod^{m}_{\mu=1 } \frac{{\mathrm{d}}^2 z_{\mu\alpha}}{\pi}.\ ] ] the normalization constant can be computed by evaluating the required phase - space integral through a recurrence relation method ; the result is @xmath116 . as one might infer from the above , fermionic coherent states of this kind are somewhat more intricate than the ones discussed in the previous examples . and , although the content of the present paper encompasses all basic ingredients needed to implement the proposed method in terms of these states , a proper treatment would nevertheless require introduction of additional concepts . therefore , we do not pursue applications involving this particular set of coherent states in this paper . the coupled coherent states ( ccs ) method , as originally developed by shalashilin and child @xcite using canonical coherent states , belongs to the family of multiconfigurational guided - basis methods . its characteristic attributes are the non - orthogonality of the basis set and the use of simple classical mechanics @xcite to guide the basis elements , as opposed to more complicated full - variational approaches like the gaussian - based version of the multiconfigurational time - dependent hartree ( g - mctdh ) method . @xcite formulation of the method for gaussian states is fairly straightforward and the same is true in the generalized context . in order to better appreciate the additional features that arise in the latter case , we will first present the method in its continuum form ; the discrete version is developed subsequently . we begin by considering the coherent - state decomposition of an arbitrary quantum state , @xmath117 which follows from . it is assumed that @xmath118 is bound to obey the classical equations of motion . by virtue of phase - space volume conservation , we are allowed to transfer the integration measure to the initial instant and conveniently integrate over initial conditions @xmath119 , as indicated in the second equality in . the derivation of the ccs equations amounts to finding a solution of the schrdinger equation @xmath120 for @xmath25 in the form given by with the _ ansatz _ : @xmath121 where @xmath122 is the action defined in . in other words , we seek an equation of motion for the time - dependent amplitude @xmath123 that solves . let us make a few observations regarding this particular choice of solution . first , all quantities that specify @xmath25 i.e. @xmath8 , @xmath123 and @xmath122 are to be regarded as functions of the initial conditions @xmath124 . thus , the method is conceived as an initial value representation from its onset . second , we note that it follows from that @xmath123 depends on the initial state @xmath125 through the relation @xmath126 . in numerical applications , the integral in has to be approximated somehow ; the typical procedure is to sample initial conditions @xmath124 in phase space with the overlap modulus @xmath127 playing the role of a weight function . once the @xmath124 s have been properly sampled , the values of the corresponding @xmath128 s are uniquely defined . third , the motivation behind the factorization of @xmath129 into a complex amplitude times an action exponential comes from a general result of semiclassical theory , according to which the classical action provides a first - order approximation to the phase of the quantum state . since this phase accounts for most of the wavefunction s oscillatory behavior , @xmath123 is expected to present a rather smooth time dependence , thus facilitating numerical treatment . we now proceed to look for a differential equation for @xmath123 . taking the time derivative of and making use of the schrdinger equation , we find ( after rearranging terms ) : @xmath130e^{-{\frac{{i}}{\hbar}}s(z)}.\ ] ] next , we factor out @xmath25 by separating the scalar products on the right - hand side of the equation with the help of the closure relation @xmath131 , with @xmath132 , which leads to @xmath133}.\ ] ] here we have already shifted the integration measure to the initial instant [ @xmath134 and replaced the @xmath135 that appeared under the integral sign for @xmath136 . the coupling @xmath137 in is given by @xmath138 where the non - diagonal matrix element @xmath139{| z \}}\hat{h}\ketn{z'}}{{\ { z | z ' \ } } } = \frac{{\langle z |}\hat{h}{| z ' \rangle}}{{\langle z | z ' \rangle}},\ ] ] is an analytical function of @xmath7 and @xmath140 that can be directly obtained by analytical continuation of the classical hamiltonian . as a last step , we express the first term in as a function of readily computable quantities . since @xmath141{| z \}}$ ] we observe that : @xmath142 the total time derivative of @xmath143 is simply @xmath144,\ ] ] while the remaining term involving @xmath145{| \dot{z } \}}$ ] can be written as @xmath146 owing to the the analyticity of @xmath6{| z \}}$ ] on @xmath7 . hence , collecting together the above results and making the necessary substitutions in , we find that the coupling may be expressed as : @xmath147 \dot{z}^{{*}}_\alpha;\end{gathered}\ ] ] which is an analytic function on @xmath140 . by integrating the equation of motion the amplitudes at time @xmath41 can be determined from their initial values . once the amplitudes are known , we can reconstruct the quantum state with , reproduced below in terms of @xmath123 : @xmath148 the integro - differential equation with @xmath137 given by relates directly to the canonical coherent states version of the ccs method and shares its attractive characteristics , namely : ( i ) in the semiclassical regime , according to the reasons mentioned earlier , the amplitude @xmath123 is expected to have a smooth time dependence ; ( ii ) because of the coherent - state overlap @xmath149 , the @xmath140 integral is localized around @xmath4 ; @xcite and ( iii ) the coupling between amplitudes of different basis elements is not only sparse but also non - diagonal , since the integrand is identically zero when @xmath150 . as a final remark , we should note that , if one performs a series expansion of @xmath137 for small @xmath151 , one finds that this series begins with a second - order term . in the generalized coherent state case however , unlike the specific situation for canonical coherent states , this does not coincide with the second- and higher - order terms in the taylor series of @xmath152 . here we re - derive the ccs equations using a discrete set of coherent states as basis , as opposed to the continuous set employed in the previous section . we shall find that the discrete formulas do not differ from their analogue expressions in the canonical coherent state case . this is due to the fact that all information concerning distinct coherent - state geometries is codified in a number of key quantities , namely : the overlap , the phase - space metric and the classical equations of motion . each of these quantities participates in the same way in the working equations , regardless of the particular type of coherent state being used incidentally , a most desirable feature for programming purposes , for it means that the core subroutines of the method are essentially independent of geometry . nevertheless , for the sake of consistency , we must review the discrete formulas , for they are the ones actually used in practice . the first step towards a discrete unitary formulation of the generalized ccs method is the assumption that one is able to write down a closure relation by employing a finite number of basis elements as below : @xmath153 where @xmath154 is the size of the basis set . @xcite it suffices that this closure relation represents the identity operator only on the dynamically accessible part of the phase space and that it holds only during the time interval upon which the propagation takes place . in order to properly represent the identity , the matrix @xmath155 in must satisfy the relation : @xmath156 where @xmath157 is the _ overlap matrix_. this guarantees that we have a well - defined discrete coherent - state decomposition : @xmath158 in other words , @xmath159 must be sufficiently well - conditioned so that expressions involving its inverse @xmath155 remain numerically stable during the entire propagation . therefore , the basis - set initial conditions must be sampled in such a way as to ensure that this requirement is fulfilled . one such procedure , that results in a well - conditioned overlap matrix ( at initial time ) , is described in appendix [ sec : sampling ] . yet , nothing prevents that an initially well - conditioned overlap matrix becomes singular at some later time a notable weakness of time - dependent methods formulated with non - orthogonal basis sets.@xcite in the event that @xmath159 becomes singular , one should take appropriate measures before resuming the propagation . in this regard , a particularly interesting protocol has been proposed by habershon.@xcite the ` singularity problem ' , though , did not occur in the simple applications considered in this paper . upon these considerations , we introduce a discrete set of @xmath154 amplitudes @xmath160 as well as their corresponding action phases @xmath161 , writing : @xmath162 next , we proceed exactly as in [ sub : ccs - continuum ] . the equation of motion in the discrete unitary case is then readily found to be : @xmath163 with coupling matrix given by : @xmath164 \dot{z}^{{*}}_{j\alpha}.\end{gathered}\ ] ] in practice , matrix @xmath155 is never explicitly constructed ; rather , one introduces a set of auxiliary amplitudes @xmath165 which are related to the coefficients @xmath166 according to @xmath167 thus , at every time step @xmath168 is obtained from @xmath169 by means of the above intermediate equation an operation that requires solving a linear system of size @xmath154 . then , the equation of motion can be recast as : @xmath170 d_k,\ ] ] while the quantum state is expressed in terms of amplitudes @xmath168 as : @xmath171 the propagation scheme comprised by eqs . , and , together with eqs . , and , represents the standard form of the generalized coherent - state method proposed here . it can be shown to preserve normalization given by the sum @xmath172 as long as the overlap matrix remains sufficiently well - conditioned . in addition , it preserves total energy ( for time - independent hamiltonians ) as long as the identity operator can be resolved in terms of the basis - set elements , though this situation is hardly achieved in multidimensional problems . it has been pointed out that energy conservation is closely related to the accuracy of the ccs method . @xcite thus , by monitoring total energy , one can make an ` on - the - fly ' diagnosis as regards to the quality of the ccs results ; indeed we observe in our simulations that deviations from the exact quantum solution are accompanied by fluctuations in total energy . it may be convenient particularly when the system under study has only one or two degrees of freedom to attempt a more straightforward discretization of the closure relation , as below : @xmath173 with @xmath174 approximating the integration measure @xmath175 at each phase - space point . the equation of motion for @xmath169 in this case can be obtained at once from by setting @xmath176 : @xmath177 c_k.\ ] ] similarly , the quantum state in this case is given by : @xmath178 this propagation scheme is computationally less demanding than the one discussed in the previous section if the basis - set size is kept the same , since there is no need to solve a linear system at each time step . on the other hand , a larger basis set ( usually constructed as a grid in phase space ) may be necessary to converge the results if is employed . moreover , as discussed by shalashilin and child , @xcite the direct discretization of equation does not preserve the unitarity of an exact quantum time evolution , that is , the norm of the propagated quantum state is not automatically conserved , meaning that results must be normalized on output . to end this section , we discuss the particular situation whereupon a single coherent - state basis element is used to describe the system : @xmath179 as can be seen from the above equation , this scheme only applies if the quantum state to be propagated is a coherent state , that is : @xmath180 . the form of the approximated @xmath25 , with an action phase , can be derived from the working equations of the previous sections by setting the basis - set size @xmath181 , in which case we find that @xmath182 and hence @xmath183 . we shall denominate the primitive method defined by , together with , and , as the _ classical propagation scheme _ , in view of the fact that only classical ingredients are present in its formulation . it serves as a reference method , against which more sophisticated approaches , such as those described earlier , may be confronted which is useful , for example , in order to identify non - classical behavior ( defined in this sense ) , as in [ sub : su3-triple - well ] . it should be pointed out that if the hamiltonian of the system is such that application of the time - evolution operator maps one coherent state onto another or , more formally , when the hamiltonian is an element of the lie algebra associated with the set of coherent states under consideration , then the classical propagation scheme actually gives the exact solution . in other situations it may provide a reasonable approximation for very short times . as a first application we consider a simplified model for the dynamics of an @xmath61-particle bose - einstein condensate trapped in a double - well potential , where individual bosons interact through contact forces , that is , with an interacting potential @xmath184 . this model has been discussed in detail in several works ; @xcite here we briefly sketch its main ideas . in the two - mode approximation , it is assumed that only the single - particle ground state and first excited state of the double - well potential , as obtained from first - order perturbation theory , have a significant occupation , so that the dynamics of the system is restricted to these two levels . this should be a good approximation if one seeks to describe the low temperature regime , wherein most of the particles are expected to be occupying the ground state see ref . for a discussion on the validity of the two - mode approximation to the double - well problem . since the number of bosons is preserved , the particle number operator @xmath185 is a constant of the motion . by means of a well - known homomorphism between the algebra @xmath186 and the bosonic creation and annihilation operators , the bosonic dynamics can be described in terms of three independent angular momentum operators schwinger s pseudo - spin operators @xcite with total angular momentum @xmath77 . in terms of these operators the hamiltonian of the model is : @xcite @xmath187 where the tunneling rate @xmath159 equals the energy difference between the two occupied single - particle states and the self - collision parameter @xmath188 is proportional to the interaction strength of bosons located in the same potential well . applying definitions and to , the classical hamiltonian becomes @xcite @xmath189 in which we discarded , without any loss , a constant term . from identity , the equation of motion for @xmath190 is : @xmath191 notice that the @xmath192 scaling of the self - collision parameter in was specifically chosen so that the equation of motion is independent of particle number @xmath61 , therefore representing a well - defined classical limit of the system when @xmath193 . this also means that , by employing the ccs method with a fixed set of classical trajectories , we can obtain quantum solutions for different particle number regimes . we shall examine the quantum dynamics of the hamiltonian from the perspective of a non - unitary implementation of the ccs method . following the procedure outlined in [ subsub : ccs - discrete - non - unitary ] , a direct discretization of expressions and is performed by replacing the integrals over phase - space variables at @xmath194 by finite summations over a discrete set of initial conditions . the accuracy of this straightforward discretization procedure is strongly dependent on the selection of initial values over the phase space . in the particular case of a regular grid , results are found to be extremely sensitive to characteristics such as spacing , placement , number of points and , most importantly , the choice of grid variables . for the subsequent numerical calculations , we consider a regular grid in the phase - space coordinates @xmath195 , which are related to the @xmath0 angular coordinates @xmath196 defined in equation , according to : @xmath197 this choice aims at reducing the inherent difficulties imposed by the spherical topology of spin coherent states at @xmath198 . in terms of this new set of coordinates , the integration measure takes a simpler form : @xmath199 which , unlike expressions and , is independent of the integration variables . hence , the @xmath174 s that appear in the non - unitary evolution equation and also in are all equal to the same time - independent @xmath200 given by : @xmath201 where @xmath202 and @xmath203 are the grid spacings in the @xmath204 and @xmath205 directions , respectively . upper panel : regular grid of classical initial conditions with approximately three hundred initial values for the case of @xmath206 particles . the red dot indicates the center of the grid , whose angular coordinates are @xmath207 . this phase - space point represents the label of the initial coherent state @xmath208 . bottom panel : some examples of classical trajectories over the spherical phase space . the red orbit corresponds to the initial value @xmath140 . the cartesian coordinates used to plot these trajectories are related to the coherent - state angular variables according to @xmath209.,title="fig:",scaledwidth=48.0% ] upper panel : regular grid of classical initial conditions with approximately three hundred initial values for the case of @xmath206 particles . the red dot indicates the center of the grid , whose angular coordinates are @xmath207 . this phase - space point represents the label of the initial coherent state @xmath208 . bottom panel : some examples of classical trajectories over the spherical phase space . the red orbit corresponds to the initial value @xmath140 . the cartesian coordinates used to plot these trajectories are related to the coherent - state angular variables according to @xmath209.,title="fig:",scaledwidth=48.0% ] the initial quantum state to be propagated is chosen to be the spin coherent state @xmath210 with @xmath211 specified by angular coordinates @xmath212 . as it is usually assumed in most guided - basis methods , the most important dynamical contributions at least for short - time propagation are expected to arise from classical trajectories initially located at the same phase - space region occupied by @xmath213 . therefore , the grid of initial conditions is intentionally centered at the point @xmath214 , as illustrated on the upper panel of fig . [ fig : initial - grid ] for the particular case of @xmath206 particles the bottom panel portrays the classical dynamics in phase space . notice that the classical trajectory with initial value @xmath140 is located close to a stable equilibrium point and within the boundaries of a separatrix of motion . with the purpose of quantifying the agreement between the ccs - propagated quantum state @xmath215 and the exact result @xmath216 obtained via diagonalization of in the angular momentum basis we shall compute the _ fidelity _ : @xmath217 the fidelity reflects the physical similarity between two states : its values are restricted to the interval @xmath218 , with the maximum value corresponding to physically identical quantum states . we wish to analyze the behavior of @xmath219 as the number of particles in the system is changed . there is a subtlety involved , though : due to its fundamental property of minimal uncertainty , @xcite the coherent state @xmath8 represents a quantum state with maximal localization in phase space around the point @xmath4 . moreover , for spin coherent states , the linear dimensions of the phase - space region effectively occupied by @xmath8 decrease proportionally to @xmath220 . @xcite thus , in order to make a meaningful comparison amongst trajectory - based propagations carried out at different particle number regimes , one must account for the ` shrinking ' of the initial state @xmath208 as @xmath61 grows larger . therefore , in our simulations , the grid spacing was reduced proportionally to @xmath220 whilst the number of initial conditions in each run was roughly unchanged . fidelity as a function of the dimensionless time @xmath221 with parameters @xmath222 , @xmath223 for different particle number regimes . each run employed roughly three hundred trajectories the number of points varies slightly for different runs due to the cropping of grid borders.,scaledwidth=48.0% ] fig . [ fig : fidelity ] shows @xmath224 as a function of time for various values of @xmath61 . notice that , even for long times , the non - unitary results remain accurate ( @xmath225 ) . on the other hand , it is clear that the fidelity tends to decrease over time , evidencing the accumulation of numerical errors in the ccs dynamics . in part , this inaccuracy stems from the non - unitary nature of the discretization procedure . however , to a large extent , numerical error arises due to dispersion of the classical trajectories over phase space , resulting in an incomplete description of the quantum system . also note that the fidelity is consistently higher for larger particle numbers . since the number of classical trajectories was kept nearly constant for all the runs , and the range of the initial condition grid was adjusted so as to reflect the width of the initial quantum state , we may conclude that the ccs method is better suited for describing the dynamics in the _ many - particle regime _ , which we recognize as the _ semiclassical regime _ for this problem . indeed , it is well known that the limiting case @xmath226 ( or @xmath227 , as it can be interpreted in this problem ) coincides with the classical limit of quantum mechanics . @xcite in this sense , the ccs method , though in principle formulated as an exact method , when numerically implemented and therefore subjected to inevitable practical limitations should be regarded more as a semiclassical technique than a genuine quantum approach , an assertion that can be made with respect to both its non - unitary and unitary versions . finally , we point out that , for the system under consideration , the computational cost of the method is insensitive to particle number . hence , for very large values of @xmath61 , the computational resources needed for evaluating the exact quantum dynamics by diagonalization of the hamiltonian will certainly exceed the analogous requirements of the non - unitary ccs method . next , we consider a simplified model describing an @xmath61-particle bose - einstein condensate trapped in a symmetric triple - well potential where individual bosons are once more assumed to interact by contact forces ; the main ideas involved are as follows : the triple - well trapping potential , under suitable conditions , can be approximated by an harmonic expansion around each of its three ( symmetrically located ) minima . the three - fold degenerate fundamental states of this approximated problem can be determined without difficulty . it is then assumed that the dynamical regime is such that the energy eigenspace spanned by these three local modes is sufficiently isolated from the rest of the single - particle spectrum , so that at low temperatures they alone provide an adequate description of the system . for more details on the derivation and particularities of this model , see refs . . let @xmath228 , @xmath229 and @xmath230 denote the annihilation operators associated with the aforementioned fundamental single - particle modes related to the locally approximated wells . in terms of these bosonic operators , the ` three - mode approximation ' @xcite to the hamiltonian is : @xmath231 where @xmath159 is the tunneling rate , describing hops between adjacent wells , and @xmath188 is the collision parameter , that controls the strength of two - body interactions within the same well . @xcite owing to particle number conservation , this system is suitably described in terms of @xmath1 bosonic coherent states @xmath232 , which represent a particular case of the coherent states discussed in [ subsub : sun - cs ] . using definition together with , we find from that the classical hamiltonian is @xmath233 from it follows that the equations of motion are : [ eqn:3-well - zdot ] @xmath234 similarly to the double - well model , we have deliberately tuned the collision parameter @xmath188 with a @xmath192 factor , thereby making the classical dynamics independent of particle number ; in this way the classical system is well - defined in the limit @xmath235 . the classical system defined in has three dynamically equivalent invariant subspaces , specified by the constraints : @xmath236 , @xmath237 and @xmath238 . let us concentrate on the first of these ( @xmath236 ) and refer to it as the @xmath239 subspace . now , consider the set of operators @xmath240 , @xmath241 and @xmath242 , defined by the canonical transformation : [ eqn:3-well - b ] @xmath243 it can be demonstrated that @xmath239 is an @xmath0 subspace whose two single - particle modes are associated with operators @xmath240 and @xmath241 . @xcite as a consequence , under the _ classical approximation _ described in [ sub : classical - propagation ] , any @xmath1 coherent state initially located in @xmath239 will always display zero occupation of the @xmath242 mode ; as a matter of fact , the expectation value @xmath244 is identically null when @xmath236 . this conclusion , however , does not apply to the actual quantum problem : even if the initial state @xmath213 has null occupation in the @xmath245 mode , this situation will not be preserved as the system evolves in time . it is precisely this ` non - classical behavior ' that we wish to describe using the @xmath1 ccs method . let us then consider the following initial state @xmath246 , with @xmath247 and thus located on the classical invariant subspace @xmath239 . we shall propagate this state employing the unitary method developed in [ subsub : ccs - discrete - unitary ] and compute the occupation @xmath248 of the single - particle mode associated with @xmath242 . following , this function will be calculated according to : @xmath249{| z_j \ } } b^\dagger_3 b_3 \ketn{z_k}}{{\ { z_j | z_k \ } } } \right ] e^{{i}(s_k -s_j)},\ ] ] where the non - diagonal matrix elements between square brackets can be obtained by analytic continuation of . we shall also monitor the total energy of the system , @xmath250 as a means to probe the quality of our results . in order to construct the basis set , it is necessary to choose adequate sampling variables . in the present case we opt for angular variables @xmath251 defined by : @xmath252 the initial conditions @xmath40 are then randomly sampled around @xmath253 from gaussian probability distributions expressed in terms of these angular variables , with each angle being individually selected : for example , @xmath254 and @xmath255 are selected according to @xmath256 ^ 2 / 2 \sigma_{\theta_1 } } ; \ \sim e^{-[\phi_1(0 ) -\phi'_1]^2 / 2 \sigma_{\phi_1}}.\ ] ] notice that the widths of these distributions are adjustable parameters of the method . the actual sampling procedure which also comprises specific criteria for accepting and neglecting candidate basis elements is somewhat involved and further details are reserved to appendix [ sec : sampling ] . once the basis - set initial conditions are known , the amplitudes @xmath169 and @xmath168 of eqs . and can be initialized and propagation of may be started . mode as a function of dimensionless time @xmath257 for @xmath258 , @xmath259 , and @xmath260 and different basis - set sizes @xmath154 ; inset : zoom on the time interval @xmath261 . bottom panel : energy fluctuation as a function of @xmath257 for the same runs shown on the upper panel ; inset : basis - set conditioning factor @xmath262 for the corresponding curves.,title="fig:",scaledwidth=48.0% ] mode as a function of dimensionless time @xmath257 for @xmath258 , @xmath259 , and @xmath260 and different basis - set sizes @xmath154 ; inset : zoom on the time interval @xmath261 . bottom panel : energy fluctuation as a function of @xmath257 for the same runs shown on the upper panel ; inset : basis - set conditioning factor @xmath262 for the corresponding curves.,title="fig:",scaledwidth=48.0% ] we studied the case of @xmath258 trapped bosons , with tunneling rate @xmath263 , and collision parameter @xmath264 . the widths of the gaussian distributions exemplified in were set to @xmath265 for both @xmath266 and @xmath267 and to @xmath268 for both @xmath269 and @xmath270 . in order to visualize whether results converge as the basis - set size @xmath154 is increased , simulations have been performed with different number of orbits while maintaining all other parameters including sampling widths and random number generator seed @xcite fixed . results obtained with the unitary @xmath1 ccs method and by a direct diagonalization of the hamiltonian in the bosonic fock basis are compared in fig . [ fig : b - occupation ] . on the upper panel , the occupation of the classically inaccessible @xmath242 mode is displayed for different basis - set sizes . on the bottom panel , the total energy expectation value for each run is shown as well as the corresponding behavior of basis - set conditioning factors ( see appendix [ sec : sampling ] ) . it is found that ccs results improve as more basis elements are included in the propagation , as expected although there is some anomalous behavior , with the @xmath271 run performing better than the @xmath272 case . this is most likely due to the basis - set sizes involved , which are probably too small for a uniform convergence to be observed . we also note that the quality of ccs results is intimately connected to conservation of total energy : the more accurate runs are also those in which energy is better conserved . the most computationally expensive run , with @xmath273 basis elements , led to an initial basis - set conditioning factor @xmath274 , a value that diminished during the propagation . this behavior was observed in all cases as shown in the inset of the bottom panel of fig . [ fig : b - occupation ] , indicating spread of the classical trajectories . for this particular run ( @xmath273 ) , excellent agreement between the ccs and exact solutions is found , except for slight discrepancies at @xmath275 . it should be noted that the propagation with @xmath273 which is quite a modest number of basis elements for a problem with two degrees of freedom , not only proved to be accurate @xcite but also considerably faster than solving the 5151-dimensional quantum eigensystem . this gain in efficiency , at least for the model discussed here , drastically increases as more particles are added to the system , since the dimension of the @xmath1 bosonic fock space , given by @xmath276 , grows rapidly with @xmath61 . at the same time , the system becomes more classical as more particles participate in the dynamics , therefore making a trajectory - based approach more inviting . yet , despite these advantages , a reliable and positive way to check for the convergence of the ccs method still remains to be developed the criterion of energy non - conservation being only indicative of accuracy loss during propagation . this represents a critical shortcoming when studying more complex systems for which there are no exact numerical results or experimental data available for comparison ; for in those cases no precise statements about the system could be made based on ccs results alone . in this work we have formulated a multiconfigurational , trajectory - guided quantum propagation method whose distinctive quality consists in employing generalized coherent states as basis elements . in this sense , the technique is seen as a natural extension of the coupled coherent states method of shalashilin and child @xcite whereupon frozen gaussians are replaced by more general configurations ; at the same time , the main features of the original ccs are retained : quantum amplitudes obey an integro - differential equation with sparse coupling and present a smooth time dependence , owing to their oscillatory behavior being partially compensated by the classical motion of the basis elements and their action phases . as pointed out in [ sec : cs - formalism ] , no deep understanding of group - theory concepts is necessary , neither to derive the basic equations of the method nor to implement it numerically in fact , we have seen that all geometrical quantities that enter the formulas can be straightforwardly evaluated from the coherent - state overlap function alone and that the discrete versions of the working equations do not differ in overall structure from their analogue expressions in the original ccs approach . in [ sec : ccs - generalized ] , three versions of the method have been devised : continuum , non - unitary and unitary . the continuum version most evidently displays the novel elements due to the non - euclidean geometry associated with the generalized coherent - states and it serves primarily as a starting point for a number of possible analytical approximations . the non - unitary version , in turn , might be understood as a direct attempt to reproduce the continuum formulas by reducing phase - space integrals into finite sums . the unitary version , meanwhile , is the standard form of the method , being the most adequate for the majority of practical applications . we have also briefly discussed one weakness of this last version of the method which is the ill - conditioning of the overlap matrix . in that respect , we note that the basis - set sampling procedure that we propose detailed in appendix [ sec : sampling ] is more appealing than the techniques used in previous ccs applications , for it assures an initially well - conditioned overlap matrix and hence stability of the amplitude equation , at least for short time propagation . in [ sec : ccs - application ] , we have illustrated the general aspects of the proposed approach with applications to simple models of many - boson systems , described in terms of @xmath0 and @xmath1 bosonic coherent states . one particular aspect , namely , the choice of appropriate phase - space variables for either sampling of initial conditions or grid construction , is found to be of great importance coordinates chosen for these purposes must somehow reflect the intrinsic geometry of the coherent state under consideration , otherwise only a poor representation of the system is obtained . moreover , the accuracy and efficiency of the method was established by comparing results against exact quantum calculations . excellent agreement was observed and , in the triple - well system examined in [ sub : su3-triple - well ] , small discrepancies in the results were found to be correlated with energy non - conservation . these tests also allowed us to gain some insight regarding the domain of applicability of our formulas . a careful analysis of the fidelity in the double - well system studied in [ sub : su2-double - well ] revealed that the ccs approach is best suited for describing the regime of large particle number , also identified as the semiclassical regime of that problem . this conclusion can in fact be extended for @xmath2 bosonic systems in general . the same goes for the observations concerning the computational advantages of the method ( over standard matrix diagonalization ) when the number of particles is large ; assertions which were made within the context of the triple - well problem . though we have exemplified the use of the method with @xmath0 and @xmath1 bosonic hamiltonians , we emphasize that the formulation presented is by no means limited to these types of systems in addition to the @xmath277 fermionic coherent states discussed in [ subsub : thouless - cs ] , other possible coherent - state parametrizations in terms of which the method could be readily implemented were referenced throughout [ sec : introduction ] . furthermore , the conclusion put forth in [ sec : ccs - application ] , namely , that propagation of quantum states with the help of classically guided basis sets is better suited for describing systems in their semiclassical regime , is certainly expected to hold for any coherent - state representation that might be employed provided one correctly interprets what ` semiclassical regime ' means in each case . a few words regarding the purely computational aspects of the method are in place . in the numerical calculations reported in this paper we have adopted the simplest possible strategy of implementation , which consists in propagating simultaneously the coherent - state trajectories and their corresponding amplitudes . but since trajectories evolve independently , an alternative approach would be to propagate them separately , storing the coherent - state coordinates at pre - determined time intervals , and use this information afterwards in order to propagate the fully coupled amplitude equation perhaps employing interpolation algorithms if coordinates are needed at intermediate instants . this preliminary propagation of classical orbits could , of course , take full advantage of parallelization , especially if single basis elements are propagated with ease . it is however in those cases in which the very propagation of individual trajectories is a computationally demanding task either because the system has an extremely large number of degrees of freedom or because no analytical expression for the hamiltonian is available that this ` two - stage ' procedure would most definitely be the strategy of choice . in addition , it would allow for more sophisticated sampling techniques , since it would then be possible to select trajectories based on knowledge of their entire story , and not relying just on their initial proximity to the quantum state . thus , for instance , one could identify orbits which , though unimportant at early stages of the propagation , give a significant contribution later on . as a final and interesting remark , we note that the multiconfigurational ehrenfest method , @xcite as specifically designed for ` on - the - fly ' non - adiabatic dynamics , can be obtained at once from the presented formalism as the particular case wherein each basis - set element is taken to be a composite coherent state consisting of a canonical part and an @xmath2 part with the particle - number parameter @xmath61 set to unity . @xcite in this picture , different ` diabatic ' potential surfaces would be represented by the @xmath45 single - particle states composing the @xmath2 coherent state ; further extension to born - oppenheimer ` adiabatic ' energy surfaces could be achieved without difficulty . this work was supported by fapesp under project grants 2008/09491 - 9 , 2011/20065 - 4 , 2012/20452 - 0 , and 2014/04036 - 2 . in addition , mama acknowledges support from cnpq . numerical calculations made use of routines from the gnu scientific library . the basis set sampling procedure that we propose for the unitary ccs method ( [ subsub : ccs - discrete - unitary ] ) is outlined here . it applies to any type of coherent state once two geometry - dependent ingredients are provided : adequate sampling coordinates @xmath278 with a known inverse @xmath279 and a weight distribution function @xmath280 according to which these coordinates are to be randomly selected . the procedure assumes that the initial state is a coherent state , i.e. @xmath208 , in which case the initial coherent - state sampling coordinates @xmath281 must also be supplied . \(4 ) if @xmath288 is less than some threshold value @xmath289 , accept @xmath284 permanently adding it to the basis set , whose size increases by 1 . else , discard the selected basis element , in which case the basis - set size does not change . in either case , return to step ( 2 ) ; the above procedure is then iterated until either a desired basis - set size is achieved or _ saturation _ occurs , meaning that the algorithm is unable to select a new @xmath284 that satisfy the @xmath288 threshold condition ( a certain maximum number of attempts may be specified ) . how fast saturation takes place will depend upon the system s dimensionality , the threshold value @xmath289 , the coherent - state parameters and the details of the sampling distribution @xmath280 . typically , we take @xmath290 , and use a predetermined basis - set size below saturation , thus ensuring a reasonably well - conditioned overlap matrix at initial time and hence the stability of the propagation ( at least for sufficiently short times ) . this sampling protocol requires the eigenvalues of the overlap matrix to be computed at every iteration . we point out , however , that this does not compromise the method s efficiency since the sampling is performed only once , before the actual propagation . moreover , the overlap matrix typically does not grow too large ; this assertion holds even for multidimensional systems , as long as the sampling distribution is kept sufficiently localized around the initial - state coordinate @xmath140 from where the most relevant contributions to the initial value representation formula are expected to originate . finally , note also that the initial state @xmath282 is always included in the basis set ; this is crucial for accuracy of short - time results and also secures that the initial norm is unity , regardless of how the remaining basis elements are distributed in phase space . recently , grossmann _ et . al . _ [ f. grossmann , m. buchholz , e. pollak , and m. nest . . , 89(3):032104 , 2014 . ( http://dx.doi.org/10.1103/physreva.89.032104 ) ] have investigated , in a semiclassical context , whether propagation with antisymmetrized basis states is essential for the description of electron scattering . we note that the approximations to the generalized coherent - state path integral considered by kuratsuji and suzuki @xcite as well as specific formulations for slater determinants @xcite are very much akin to the techniques develop in this paper . t. f. viscondi . phd thesis , universidade estadual de campinas , instituto de fsica gleb wataghin , campinas , so paulo , brazil , 2013 . available in portuguese http://www.bibliotecadigital.unicamp.br/document/?code=000904028&opt=4&lg[here . ] in the majority of applications involving coherent states and classical trajectories , the imaginary surface terms in @xmath35 are canceled out by normalization factors . however , when _ duplicated phase - space _ trajectories are considered i.e. when @xmath4 and @xmath7 are analytically continued in such a way that they become _ completely independent _ variables ( thereby doubling the number of degrees of freedom in the system ) , the complex action @xmath35 plays a prominent role , for its analytical properties are of the uttermost importance in those cases . @xcite owing to the finite width of the gaussian functions employed as basis , ccs trajectories actually evolve under a smoothed - out classical hamiltonian whose width - dependent terms are usually referred to as ` simple quantum corrections ' . su(n ) fermionic coherent states should be contrasted with the antisymmetrized gaussian wavepackets mentioned in [ sec : introduction ] . while the former are defined in terms of a finite set of underlying molecular spin - orbitals , thus providing a convenient description of the electronic structures of chemical systems , the latter are more adequate for representing otherwise localized fermions , such as protons and neutrons in the atomic nucleus , @xcite or perhaps outer shell electrons in highly excited atoms . @xcite j. schwinger . on angular momentum . in l. c. biedenharn and h. van dam , editors , _ quantum theory of angular momentum : a collection of reprints and original papers _ , page 229 . academic press , london , 1965 . we note that in the triple - well model , the energy difference between the ground state and doubly degenerate excited states of the one - body hamiltonian is @xmath295 within the three - mode approximation these stationary states span the same eigenspace as the local modes associated with operators @xmath228 , @xmath296 and @xmath230 . @xcite also , in eq . , we have made an additional simplification by excluding cross - collision terms , which arise from the interaction between bosons in different wells . as can be seen from eq . , for the particular case of @xmath297 , the bosonic @xmath2 coherent state reduces to @xmath298 , where the occupation number representation has been replaced by ` first - quantized ' notation , with @xmath299 being the single - particle orbital associated with the @xmath300-th mode . this is nothing but a standard decomposition of a quantum state in a finite basis of size @xmath45 , only written without redundant parameters . the overlap matrix is hermitian and positive - definite , meaning that its eigenvalues are real and positive , though numerical diagonalization may produce null or very small negative eigenvalues . alternatively , one could employ a singular value decomposition and carry on the sampling procedure using the singular values rather than the eigenvalues .
we calculate the josephson critical current across in - plane ( 001 ) tilt grain boundary junctions of high temperature superconductors . we solve for the electronic states corresponding to the electron - doped cuprates , two slightly different hole - doped cuprates , and an extremely underdoped hole - doped cuprate in each half - space , and weakly connect the two half - spaces by either specular or random josephson tunneling . we treat symmetric , straight , and fully asymmetric junctions with- , extended- , or-wave order parameters . for symmetric junctions with random grain boundary tunneling , our results are generally in agreement with the sigrist - rice form for ideal junctions that has been used to interpret `` phase - sensitive '' experiments consisting of such in - plane grain boundary junctions . for specular grain boundary tunneling across symmetric junctions , our results depend upon the fermi surface topology , but are usually rather consistent with the random facet model of tsuei _ et al . _ [ phys . rev . lett . * 73 * , 593 ( 1994 ) ] . our results for asymmetric junctions of electron - doped cuprates are in agreement with the sigrist - rice form . however , our results for asymmetric junctions of hole - doped cuprates show that the details of the fermi surface topology and of the tunneling processes are both very important , so that the `` phase - sensitive '' experiments based upon in - plane josephson junctions are less definitive than has generally been thought .
cond-mat0411069
because of its relation to the mechanism for superconductivity in the high temperature superconducting compounds ( htsc ) , there has long been a raging debate regarding their orbital symmetry of the superconducting order parameter ( op).@xcite only `` phase - sensitive '' experiments involving josephson tunneling can distinguish the op from the non - superconducting pseudogap.@xcite although the first in - plane phase - sensitive experiment on yba@xmath3cu@xmath4o@xmath5 ( ybco ) suggested a dominant @xmath1-wave op,@xcite experiments using tricrystal films of ybco , bi@xmath3sr@xmath3cacu@xmath3o@xmath6 ( bi2212 ) , and nd@xmath7ce@xmath8cuo@xmath9 ( ncco ) , and tetracrystal films of ybco and la@xmath10ce@xmath11cuo@xmath12 ( lcco ) suggested that these materials had a dominant @xmath2-wave op.@xcite very different results were obtained for @xmath13-axis junctions . low temperature @xmath14 @xmath13-axis josephson junctions between pb and ybco , bi2212 , and ncco suggested varying amounts of an @xmath1-wave op component.@xcite more recently , three bi2212 @xmath13-axis twist josephson junction experiments showed that the op has at least a substantial , and possibly a dominant @xmath1-wave component for @xmath14 up to the transition temperature @xmath15.@xcite here we search for a reason for these qualitative differences.@xcite to date , most theoretical treatments of in - plane ( 001 ) tilt grain boundary ( gb ) junctions either used the ginzburg - landau ( gl ) approach,@xcite or assumed a circular in - plane fermi surface ( fs ) cross - section , and that the fs - restricted @xmath2-wave op @xmath16.@xcite most of those treatments focussed upon the single quasiparticle density of states on the surface , rather than the critical current across the misalignment grain boundary junctions . the only previous treatment that included a tight - binding fs was that of shirai _ et al . _ , for which the fs consisted of very small pockets centered at the corners of the first brillouin zone ( bz).@xcite their chosen fs was extremely different from any that have been extracted from angle - resolved photoemission spectroscopy ( arpes ) experiments.@xcite shirai _ _ studied both the single quasiparticle and supercurrent properties of gb junctions constructed from four specific surfaces.@xcite here we explicitly take account of the tight - binding hole and electron - doped fs s observed using arpes and of the surface boundary conditions ( bc s ) at the interfaces . we find that a @xmath2-wave op can be consistent with the tricrystal experiments for electron - doped cuprates , for which the two - dimensional ( 2d ) fs cross - section is nearly isotropic . for hole - doped cuprates , the situation is found to be inconclusive , as one fs cross - section is consistent with the tricrystal experiments for random gb tunneling , but not for specular gb tunneling . another , slightly different hole - doped fs , on the other hand , is consistent with the tricrystal experiments for specular tunneling , but not for random tunneling . the fs of shirai _ et al . _ is inconsistent with most experiments on the hole - doped cuprates , including the tricrystal experiments , for both specular and random gb tunneling . these combined results suggest that defects and small changes in the fs topology play essential roles in interpreting the in - plane gb junction experimental results.@xcite an important modification to the tetracrystal experiment is also warranted.@xcite for a @xmath2-wave op with straight ( dotted green ) , symmetric ( dotted green ) and asymmetric ( solid blue ) gb junctions in the ideal gl model , and straight ( dot - dashed black ) , symmetric ( dashed red ) and asymmetric ( solid blue ) gb junctions in the faceted gl model . inset : sketch of a ( 001 ) tilt gb junction.,scaledwidth=45.0% ] we let @xmath17 for @xmath18 be the angles the @xmath19 axes on the left ( l ) and right ( r ) sides make with the normal to a straight gb interface , fig . gl treatments of the josephson current @xmath20 for a @xmath2-wave op yielded for ideal ( non - reconstructed ) and faceted gbs,@xcite @xmath21\sin(\delta\phi)/2,\label{randomd}\end{aligned}\ ] ] where @xmath22 is the op phase difference across the gb . these have the form @xmath23 , where @xmath24 is the critical current . @xmath25 and @xmath26 are the `` 0-junction '' and `` @xmath27-junction '' cases , as @xmath28 in the latter . there are three types of junctions we shall consider . these are asymmetric , symmetric , and straight junctions . we note that other workers have used the terminology `` mirror '' and `` parallel '' for symmetric and straight junctions , respectively.@xcite for asymmetric junctions , @xmath29 , @xmath30 . for symmetric junctions with @xmath31 and straight junctions with @xmath32 , @xmath33 . for faceted symmetric and straight junctions , @xmath34 , respectively . these ideal and faceted gl results shown in fig . 1 are in qualitative agreement with our results for electron - doped cuprates with random and specular tunneling , respectively , but differ qualitatively from our microscopic results for hole - doped cuprates . to calculate @xmath20 across an in - plane gb , we assume the fourier transform of the quasiparticle green function matrix in the bulk of the @xmath35th superconductor is@xcite @xmath36,\label{fermi}\end{aligned}\ ] ] where @xmath37 is a matsubara frequency , @xmath38 is the wave vector on the @xmath35th gb side , @xmath39 is the respective two - dimensional ( 2d ) gap function , @xmath40 is the tetragonal in - plane lattice constant , @xmath41 , @xmath42 , and @xmath43 are parameters defining the chemical potential , and in - plane bandwidth and dispersion , respectively , and the @xmath44 are the pauli matrices . @xmath45 is the usual rank 2 matrix with elements @xmath46 , and @xmath47 that describes both the single quasiparticle and mean - field pair excitations . we take @xmath48 $ ] , where @xmath1 is the @xmath13-axis repeat distance . here we are mainly interested in the 2d limit @xmath49 . for one tight - binding 2d fs appropriate for bi2212 , fs2 , we take @xmath50 mev , @xmath51 , and @xmath52 , and set @xmath53.@xcite a slightly different fs , denoted arpes , that most closely resembles the fs of bi2212 as observed in arpes experiments , is obtained with @xmath54 mev , @xmath55 , and @xmath56.@xcite a nearly circular fs , fs3 , is obtained with @xmath57 and @xmath58.@xcite finally , we also studied the extreme hole - doped near - neighbor tight - binding fs used by shirai _ et al._@xcite taking the maximum value of the gap to be 20 mev , this fs is governed by the parameters @xmath59 mev , @xmath58 , and @xmath60 . these four fs s are pictured in fig . 2 . next , we construct the half - space green function matrices on the @xmath35th side of a straight ( 001 ) tilt gb . we let @xmath61 and @xmath62 be unit vectors normal and parallel to the gb satisfying @xmath63 , as sketched in fig . 1 . letting @xmath64 and @xmath65 , we set @xmath66 , @xmath67 , and @xmath68 . the lack of translational invariance requires a discrete indexing of the lattice planes along @xmath61 , as in our previous technique for @xmath13-axis tunneling.@xcite . thus , for integers @xmath69 , @xmath70 where @xmath71 and the @xmath72 for @xmath73 , @xmath74 for @xmath75 are the respective lattice plane separations along @xmath61 . we choose the gb interface to be between lattice planes 1 and 0 and set @xmath76 , @xmath77 for @xmath78 , respectively . suppressing the @xmath79 and @xmath37 dependencies , we construct each half - space @xmath80 from a combination of the @xmath81 that obeys the free surface boundary conditions ( bc s ) @xmath82 for @xmath76 and @xmath83 for @xmath77 , obtaining@xcite @xmath84 we then paste together the two half - space layered superconductors . the supercurrent @xmath20 across the gb between the two half - spaces is given for a general junction hopping matrix element @xmath85 by @xmath86,\label{iexact}\end{aligned}\ ] ] where @xmath87 is the full green s function matrix for the combined two half spaces coupled via tunneling across the ( 001 ) tilt gb.@xcite an example of the resulting interface is pictured in fig . 3 . in principle , it is straightforward to evaluate @xmath20 to all orders in @xmath88 , as was done for the case of coherent tunneling across a @xmath13-axis twist junction.@xcite as in that case , however , the most relevant case , especially for gb misalignment angles @xmath89 sufficiently large , is for weak tunneling across the gb.@xcite junction . the bottom rows of the left @xmath90 lattice ( open red circles ) and right @xmath91 ( filled blue circles ) lattice are aligned . an incident quasiparticle can hop along the path @xmath35 , and be specularly reflected along the path @xmath92 , but tunnels non - specularly into the @xmath91 lattice via the two primary initial paths @xmath93 and @xmath94 , as indicated.,scaledwidth=45.0% ] we set @xmath95 and define @xmath96 to be the @xmath97th matrix element of the rank 2 matrix @xmath80 with @xmath98 . to leading order in the general tunneling matrix element @xmath99 , we obtain @xmath100 where @xmath101 is the junction area , @xmath102 is the electron charge , @xmath22 , @xmath103 , @xmath104 , and the @xmath105 are the lattice constants along @xmath62 . we note that @xmath106 and @xmath107 are the @xmath108 and @xmath47 pair functions for real op s on opposite sides of the junction , respectively . we note that although eq . ( [ current ] ) is proportional to @xmath109 since it is obtained in the limit of weak gb tunneling , our exact expression for the supercurrent , eq . ( [ iexact ] ) , also contains the full temperature dependence of all of the harmonics proportional to @xmath110 for @xmath111 . experimentally for large symmetric tilt angles @xmath112 , @xmath113,@xcite justifying our weak tunneling assumption , eq . ( [ current ] ) , although the approximation is less accurate for @xmath114 . others working on ( 001 ) tilt grain boundary junctions of htsc have also made this assumption,@xcite although they used a circular fs and did not necessarily take account of the surface boundary conditions.@xcite since the effects induced by the gb to the real - space @xmath2-wave pairing interaction are formidable , here we only treat the surface bc effects , and discuss the effects of self - consistency elsewhere.@xcite we note that a self - consistent calculation by tanaka and kashiwaya for an op of mixed symmetry suggested no qualitative changes upon the imposition of self - consistency.@xcite we include the limiting cases of specular ( or coherent ) tunneling , where @xmath115 , and random ( or incoherent ) tunneling , where @xmath116 and @xmath117 are independent , writing@xcite @xmath118 we remark that the specular ( coherent ) and random ( incoherent ) tunneling processes are microscopic processes that take place on an atomic scale . for specular tunneling to occur , the component of the wave vector of the quasiparticle parallel to the junction does not change at each atomic site upon tunneling . with random tunneling , the components of the quasiparticle wave vectors parallel to the junction on the two sides of the junction are completely independent variables . for example , let us consider a tight - binding model of near - neighbor hopping in a tetragonal lattice , as in eq . ( 4 ) with @xmath58 . in this model , a quasiparticle far from the junction hops from one site to an adjacent site . in a defect - free straight ( or symmetric ) @xmath119 junction , the lattices on both sides of the junction are identical . thus , a quasiparticle can propagate from any site on one side of the junction to the nearest neighbor site on the other side without changing its direction . a quasiparticle can also backscatter coherently at the interface , preserving the component of the wave vector parallel to the junction , as in other treatments.@xcite for instance , when the interface is a ( 110 ) surface , the surface is theoretically atomically flat , allowing a substantial amount of coherent reflection , as sketched in fig . although with the usual vapor phase epitaxy , such @xmath120 junctions are far from perfect,@xcite when they were constructed using the far superior liquid phase epitaxy , an excellent fit to the fraunhofer pattern was obtained by application of a magnetic field parallel to the junction , without any evidence of a zero - bias conductance peak at the interface,@xcite at least in some cases . such behavior is consistent with the incoherent tunneling model of ambegaokar and baratoff ( ab ) for an @xmath1-wave superconductor . in earlier vapor deposition of the somewhat different @xmath121 gb s , sometimes a rough approximation to the fraunhofer pattern was observed , with a peak at low fields , and sometimes a very different pattern with a dip at zero field , possibly consistent with a @xmath122-wave op,@xcite was observed.@xcite in both of these experiments , high resolution transmission electron microscopy ( hrtem ) demonstrated the quality of the junctions.@xcite however , at interfaces such as those containing a ( 130 ) surface , fig . 4 , the amount of coherent reflection off the surface is likely to depend strongly upon the incident quasiparticle wave vector direction , as the surface is a periodic , but uneven zigzag . to date , such junctions have never been prepared with the atomic precision of the above two cases.@xcite axis ) , where @xmath123 . the solid black and dashed blue arrows indicate selected incident and reflected quasiparticle paths involving only near - neighbor and next - nearest neighbor hopping , respectively.,scaledwidth=45.0% ] on the other hand , for an asymmetric @xmath120 junction , fig . 3 , the ratio of the parallel lattice spacings on the two sides of the junction is @xmath124 , an irrational number . although specular reflection at the interface by a quasiparticle is entirely possible , in this case , even if the junction were atomically perfect , the proportion of atomic sites at which coherent tunneling across the junction can occur is vanishingly small . that is , in tunneling to the nearest site across the junction , nearly all quasiparticles will change the component of their direction parallel to the junction . one example of this is sketched in fig . 3 . when one includes the dominant junction tunneling from all sites along the junction , the tunneling component of the quasiparticle wave vector parallel to the junction is effectively random . in general , one expects that atomically perfect straight junctions can exhibit specular , or coherent tunneling . defects would lead to a component of the tunneling that is random , or incoherent . asymmetric junctions and symmetric junctions that can not also be described as straight junctions ( such as the @xmath119 junction ) intrinsically lead to predominantly random , or incoherent tunneling . thus , atomically perfect @xmath125 asymmetric or @xmath126 symmetric junctions for @xmath127 ( with @xmath128 ) are expected to exhibit primarily random , or incoherent microscopic tunneling processes . facets , atomic disorder , stoichiometry variations , etc . , at the junction contribute further to the randomness of the tunneling processes.@xcite this microscopic ( atomic scale ) coherence or incoherence of the tunneling process is completely unrelated to the macroscopic coherence of the superconducting wave function , which can occur over the entire area of the junction , provided that it is prepared sufficiently uniformly . for example , the ab microscopic model of josephson tunneling,@xcite which averages independently over the parallel components of the quasiparticle wave vector on each side of the junction , assumes uniformly random , or incoherent , microscopic quasiparticle tunneling only . however , that model leads to macroscopic coherence of the superconducting wave function on each side of the junction . note that although most experiments are consistent with bi2212 having strongly incoherent intrinsic @xmath13-axis tunneling between the double layers , the product of the critical current and the normal state resistance @xmath129 is about 20% of that expected for a pure @xmath1-wave superconductor with perfect interfaces , the ab value.@xcite this occurs both for intrinsic and for @xmath13-axis twist junctions several @xmath41 m in cross - section.@xcite for asymmetric junctions with @xmath130 , @xmath131 , for @xmath1- ( thin blue ) , extended-@xmath1- ( intermediate thickness red ) , and @xmath2-wave ( thick black ) superconductors , for specular ( solid ) and random ( dashed ) gb tunneling and the tight - binding hole - doped fs2 . the @xmath122-wave curves satisfy @xmath132 , but the @xmath1- and extended-@xmath1-wave curves satisfy @xmath133 . see text.,scaledwidth=45.0% ] for symmetric junctions with @xmath134 , for @xmath1- ( thin blue ) , extended-@xmath1- ( intermediate thickness red ) , and @xmath2-wave ( thick black ) op s for specular ( solid ) and random ( dashed ) gb tunneling with the tight - binding , hole - doped fs2 . results for straight junctions with @xmath32 and a @xmath2-wave op for specular ( dotted green ) and random ( also dashed black ) gb tunneling are also shown . all curves satisfy @xmath133.,scaledwidth=45.0% ] in fig . 5 , we present our results obtained from eq . ( [ current ] ) for the low-@xmath14 @xmath0 across asymmetric junctions , @xmath130 , @xmath131 , including the tight - binding fs2 and the surface bc effects . in all of our calculations , we assume the @xmath2- and extended-@xmath1-wave op s have the bulk real - space pairing forms @xmath135 $ ] and @xmath136 , respectively . the ordinary-@xmath1-wave op is @xmath137 , independent of @xmath138.@xcite we take @xmath139 mev , which implies a maximum @xmath122-wave op of 20 mev . in fig . 4 , we plotted @xmath140 , where @xmath24 and @xmath141 , 4e|j_j^r|^2/[a^4s^2(0.1j_{||})^2]$ ] , respectively.@xcite the solid ( dashed ) curves are the results for specular ( random ) tunneling , respectively . the specular @xmath122-wave gb junction behaves as a @xmath27-junction only for @xmath142 and 75.1@xmath143 , qualitatively different from the gl results pictured in fig . 1.@xcite note that the @xmath122-wave @xmath0 for random gb tunneling is small but non - vanishing due to the surface bc s . for this and all other cases of random tunneling across gb s between @xmath2-wave superconductors , neglect of the surface bc s would lead necessarily to @xmath144 . in the 2d limit with random gb tunneling , however , the asymmetric gb results are qualitatively in agreement with the gl results , except for the overall magnitude of @xmath0 , which is greatly reduced from that obtained for specular gb tunneling . we also investigated the role of @xmath145 for the @xmath122-wave case . for random gb tunneling , as @xmath145 increases to 40 mev , the broad peak in @xmath146 becomes much flatter , and for much larger @xmath145 , @xmath146 non - monotonically approaches the gl result @xmath147 . for specular tunneling , @xmath148 mev yields @xmath122-wave results nearly identical to those pictured for @xmath49 . increasing @xmath145 beyond @xmath14940 mev eliminates the regions within @xmath150 of @xmath27-junctions , so that the overall region of @xmath27-junctions is for @xmath151 , as for the gl case . hence , for @xmath152 , the specular , tight - binding asymmetric gb @xmath122-wave results with fs2 are qualitatively different than the gl ones . in fig . 6 , we present the analogous results for symmetric junctions with fs2 . we also plotted our results for straight gb junctions with a @xmath2-wave op . for random gb tunneling , the straight gb junction results are identical to the dashed black curve for random gb tunneling with symmetric junctions . for specular gb tunneling , the straight junction results are shown as the dotted green curve . note that for specular tunneling , straight and symmetric junctions have opposite @xmath0 values at @xmath153 . for symmetric gb junctions , the specular @xmath122-wave junction behaves as a @xmath27-junction for @xmath154 , but behaves as a 0-junction otherwise . for random tunneling , the symmetric @xmath122-wave gb junction behaves similarly to the ideal gl model , eq . ( [ dgl ] ) , as both have @xmath155 but never behave as @xmath27-junctions . however , the @xmath122-wave gl facet model , eq . ( 2 ) , has @xmath156 , which is qualitatively different.@xcite we note from figs . 5 , 6 that for small angle ( @xmath157 ) specular gb junctions , the extended-@xmath1 and @xmath2-wave op s lead to indistinguishable @xmath158 . however , for random gb tunneling , the @xmath122-wave @xmath158 is much smaller than the extended-@xmath1-wave @xmath158 , and would vanish in the absence of the surface bc s . except for the broad peak and the greatly reduced magnitude of @xmath0 , the symmetric gb junction across @xmath2-wave superconductors with fs2 and random gb tunneling is qualitatively in agreement with the gl result of fig . 1.@xcite for specular gb tunneling , however , our microscopic @xmath2 results differ qualitatively from the gl forms.@xcite for symmetric junctions with specular ( thick solid red ) and random ( thick dashed black ) tunneling , and asymmetric junctions with specular ( thin solid blue ) and random ( thin dashed purple ) tunneling , for a @xmath2-wave op with the arpes fs.@xcite the straight and symmetric gb junctions satisfy @xmath159 , but the asymmetric gb junctions satisfy @xmath160.,scaledwidth=45.0% ] we remark that the ordinary-@xmath1- and extended-@xmath1-wave op s lead to @xmath161 values that are comparable to each other for both gb tunneling mechanisms . however , for the @xmath2-wave op , random gb tunneling with fs2 leads to a suppression of @xmath161 by a factor of 461 from that of the ordinary-@xmath1-wave op . this suggests that @xmath161 for the @xmath2-wave op with weak , random gb tunneling is unlikely to be consistent with the large values observed experimentally across low - angle ( 001 ) tilt junctions.@xcite the broad peak in @xmath146 for the @xmath122-wave op with random gb tunneling for both asymmetric and symmetric junctions is also inconsistent with most experimental data . on the other hand , the @xmath122-wave op with specular gb tunneling across asymmetric junctions does show a rapid decrease in @xmath162 with increasing @xmath163 , in agreement with low - angle experiments,@xcite but vanishes at 14.9@xmath164 and 39.4@xmath164 for fs2 , respectively , unlike the experiments.@xcite in figs . 7 - 9 , we compare our @xmath165 results for @xmath2-wave superconductors across symmetric and asymmetric gb junctions with three different fs s . for each fs , @xmath0 results for straight and symmetric gb junctions with random gb tunneling are identical , and for specular gb tunneling , the @xmath166 s are opposite in sign and equal in magnitude . in figs . 7 - 9 , the results for symmetric gb junctions with specular ( random ) gb tunneling are shown as the thick solid red ( dashed black ) curves , respectively , and results for asymmetric gb junctions with specular ( random ) gb tunneling are shown as the thin solid blue ( dashed purple ) curves , respectively . results for straight and symmetric gb junctions with random tunneling are identical , and the results for straight gb junctions with specular gb tunneling are shown as the dotted green curves . in fig . 7 we plotted @xmath165 versus @xmath163 for symmetric and asymmetric gb junctions for a @xmath2-wave superconductor with the arpes fs . note that for symmetric junctions , a @xmath27-junction occurs with specular gb tunneling for @xmath167 , very similar to the region obtained for fs2 pictured in fig . 5 . in addition , for random gb tunneling across symmetric junctions , no @xmath27-junctions are obtained . however , for asymmetric junctions , the results differ substantially from those obtained for a @xmath2-wave superconductor with fs2 , pictured in fig . in that ( fs2 ) case , the asymmetric gb junction was a @xmath27-junction for specular gb tunneling with @xmath168 and @xmath169 , but was a 0-junction otherwise . for random gb tunneling , it was always a 0-junction for @xmath150 and a @xmath27-junction for @xmath151 . for this slightly different ( see fig . 2 ) arpes fs , however , nearly the opposite behavior was found . asymmetric gb junctions with the arpes fs with specular gb tunneling are 0-junctions for @xmath170 and @xmath27-junctions for @xmath171 , but for random tunneling , a @xmath27-junction appears for @xmath172 ( and also for @xmath173 ) , and a 0-junction otherwise . hence , these slightly different hole - doped fs s lead to qualitatively different @xmath0 behavior for @xmath2-wave superconductors . for @xmath1- and extended-@xmath1-wave superconductors , on the other hand , such strong fs dependence of @xmath158 does not occur . although for clarity in presenting the @xmath122-wave results , we have not shown our @xmath1- and extended-@xmath1-wave data for the arpes fs , our calculations confirm the close similarity of those results to the analogous ones pictured in figs . 5 and 6 , confirming the reliability of our results . with the arpes fs , asymmetric gb junctions with specular tunneling and for symmetric gb junctions with random tunneling are qualitatively consistent with the ideal gl model,@xcite and symmetric gb junctions with specular tunneling are qualitatively consistent with the gl facet model,@xcite except that the position of @xmath174 has shifted from @xmath175 to @xmath176 . however , asymmetric gb junctions with random tunneling and the arpes fs are completely unlike either gl model . next , we calculated @xmath146 at low @xmath14 across gb s with the nearly circular fs3 , appropriate for the electron - doped cuprates such as ncco , and presented our results in fig . 8 . for asymmetric gb s , our @xmath122-wave results agree with the gl model , @xmath177 , for both specular and random gb tunneling , but for the latter , @xmath178 is greatly reduced in magnitude for a @xmath122-wave op . our results for symmetric or straight gb s with random gb tunneling coincide with each other , and @xmath179 , as in the gl model for ideal symmetric gb junctions ( fig . 1 ) . for specular gb tunneling across symmetric gb junctions , the results are qualitatively similar to the gl facet model , eq . ( 2 ) except that the @xmath163 value for the crossover from @xmath180- to @xmath27-junction behavior has shifted from @xmath175 to 29.8@xmath164 . with that modification , all of our @xmath122-wave results for the electron - doped fs3 are qualitatively in agreement with the gl facet model for specular gb tunneling,@xcite and with the gl ideal model for random gb tunneling between @xmath2-wave superconductors.@xcite this conclusion is somewhat unexpected , as one navely might have expected the specular and random results with fs3 to agree with the ideal and facet gl models , respectively . our low-@xmath14 symmetric junction results are also qualitatively in agreement with those obtained at low @xmath14 for weak tunneling with a perfectly circular fs and a fs - restricted @xmath122-wave op with different gb roughness,@xcite and our straight gb junction results with random gb tunneling are qualitatively consistent with previous straight , dirty-@xmath181 , sns junction results.@xcite for symmetric junctions with specular ( thick solid red ) and random ( thick dashed black ) tunneling , and asymmetric junctions with specular ( thin solid blue ) and random ( thin dashed purple ) tunneling , for a @xmath2-wave op with the nearly isotropic , `` electron - doped '' fs3 . also shown is the specular @xmath122-wave curve ( dotted green ) for straight junctions . the straight and symmetric junctions satisfy @xmath159 , but the asymmetric junctions satisfy @xmath160.,scaledwidth=45.0% ] in order to compare the results obtained using our technique of imposing the surface bc s upon the green function matrix to that obtained using different techniques by others , we performed explicit calculations for these purposes . according to the formulas in their paper , shirai _ _ used the fs pictured by the small green pockets near the corners of the first bz pictured in fig . 2.@xcite those authors constructed half - spaces exhibiting ( 100 ) , ( 130 ) , ( 120 ) , and ( 110 ) surfaces , and pasted them together with specular tunneling in either the straight or symmetric configurations ( which they denoted as parallel and mirror , respectively ) . their discrete quantization axis was always along the @xmath182 direction parallel to the cuo bonds , and hence only normal to the surface for the ( 100 ) case . the number of rows in the @xmath183 direction varied from 1 for the @xmath91 and @xmath90 surfaces to 3 for the @xmath184 surface . unlike our quantization axis that was defined to be parallel and normal to the surface as pictured in figs . 3 and 4 , upon fourier transformation in the @xmath183 direction , their non - normal quantization lattice basis set could lead to unintended hopping across the gb with the strong , intrinsic tight - binding near - neighbor hopping matrix element . in addition , shirai _ et al . _ included tunneling from a few rows other than the surface row of idealized atomic sites , by assuming that the specular gb tunneling matrix elements decreased inversely instead of exponentially with the tunneling distance , taking the maximum hopping strength to be 0.05@xmath185 , or 6.2% of the effective near - neighbor hopping strength , @xmath186.@xcite nevertheless , they calculated the single quasiparticle densities of states and @xmath0 across the straight and symmetric gb junctions with the above form of specular tunneling , for @xmath187 , and @xmath188 , respectively . some of their low-@xmath14 limits of @xmath0 can be compared with ours using their unusual fs.@xcite although they did not mention it explicitly , shirai _ et al . _ found the straight @xmath189 and @xmath190 , junctions to be 0-junctions . however , they found strong zero - bias conductance peaks at all junction sites for the straight @xmath188 junction , and at two of the three distinct junction sites for their @xmath191 junction , suggesting that both of these junctions behave as @xmath27-junctions . for the symmetric junctions , they found the low-@xmath14 limits of the symmetric 18.4@xmath164 and 45@xmath164 junctions to be @xmath27-junctions , but the low-@xmath14 limits of the @xmath192 and @xmath190 symmetric gb junctions to be 0-junctions.@xcite of these , @xmath193 for the 18.4@xmath164 symmetric junction changed sign at a finite @xmath14 value . our low-@xmath14 results for @xmath165 versus @xmath163 for symmetric , asymmetric , and straight junctions with the shirai fs are pictured in fig . 9 . in all cases , even for @xmath194 , the extremely small overlap of the fermi surfaces for specular gb tunneling leads to an extremely small @xmath0 . our results for specular gb tunneling with symmetric junctions are shown as the thick solid red curves with three alternating regions of @xmath27- and 0-junctions for @xmath150 . in the domain @xmath195 , @xmath27-junctions are found in the three regions @xmath196 , @xmath197 , and @xmath198 . the behavior of the domain @xmath151 is obtained from @xmath159 . random tunneling across either straight or symmetric gb junctions leads to just a 0-junction for @xmath199 , although there are strong dips in @xmath0 at @xmath200 and @xmath201 ) ( not pictured ) and @xmath155 , as indicated by the thick dashed black curve . for asymmetric gb junctions with specular gb tunneling , the thin solid blue curve shows a @xmath27-junction for @xmath202 ( and from @xmath203 ) , and the thin dashed purple curve for random gb tunneling indicates a @xmath27-junction for @xmath204 ( and from @xmath205 ) . in addition , the dotted green curve represents straight junctions with specular tunneling , which are always 0-junctions , as in the ideal and faceted gl models pictured in fig . 1 , and in all other fs s we studied ( e. g. , figs . 6 and 8) . for symmetric junctions with specular ( thick solid red ) and random ( thick dashed black ) tunneling , and asymmetric junctions with specular ( thin solid blue ) and random ( thin dashed purple ) tunneling , for a @xmath2-wave op with the extreme hole - doped fs of shirai _ et al._@xcite also shown is the specular @xmath122-wave curve ( dotted green ) for straight junctions . the straight and symmetric junctions satisfy @xmath159 , but the asymmetric junctions satisfy @xmath160 . the thick dashed black curve represents 0-junctions only.,scaledwidth=45.0% ] some of our results and those of shirai can be directly compared . both procedures lead to @xmath206 @xmath27-junctions for symmetric gb junctions with specular gb tunneling with @xmath207 , but 0-junctions at @xmath208 . as for straight junctions , our results for all four fs s we studied always lead to 0-junctions , as for both the ideal and faceted gl models . for straight junctions at the two angles @xmath192 and @xmath190 , shirai _ et al . _ also obtained 0-junctions for all @xmath209 . however , as mentioned above , shirai _ et al . _ appear to have obtained @xmath27-junctions for the straight junctions with @xmath210 and 45@xmath164 . in particular , we find their apparent @xmath27-junctions for the two straight junctions , @xmath210 , @xmath188 to be difficult to understand . suppose shirai _ _ were to increase their dominant junction coupling strength from 0.05@xmath185 to their bulk hopping strength , @xmath186 . then , the junction would be invisible to the quasiparticles , and one would certainly not expect any surface effects , such as in the straight 0@xmath164 junction they studied.@xcite although shirai _ et al . _ attributed their peculiar results to the oddness of the number of site rows in their minicell , we suspect that their results may be artifacts of their peculiar minicell basis , which was not defined to be parallel and perpendicular to the gb , as in fig . 4.@xcite finally , in fig . 10 we compare our specular @xmath162 @xmath122-wave results with the surface bc s with those obtained from the bulk green functions . we compare those results obtained with the tight - binding fs2 for asymmetric ( fig . 5 ) and symmetric ( fig . 6 ) junctions , and also our results for symmetric junctions obtained with the nearly circular fs3 ( fig . 8) . for both symmetric gb cases , there are no regions in @xmath195 of @xmath27-junctions for the bulk calculations . including the surface effects , these symmetric gb cases have @xmath27-junctions for @xmath211 and @xmath212 for fs2 and fs3 , respectively . for asymmetric junctions with fs2 , the bulk calculation leads to weak @xmath27-junctions for @xmath213 with a negative slope at @xmath188 , yielding additional @xmath27-junctions for @xmath214 and @xmath215 . the surface bc s greatly enhance the magnitude and range of such @xmath27-junctions to @xmath168 and @xmath169 , and change the sign of the slope at 45@xmath164 . although not pictured in fig . 10 , the surface bc s play an essential role in the case of random tunneling across gb s between @xmath122-wave superconductors . in the absence of the surface bc s , @xmath144 in such cases . thus , we conclude that the surface bc effects can not be ignored , especially with regard to @xmath2-wave superconductors . for the @xmath2-wave op with specular tunneling , with ( solid ) and without ( dashed ) the surface boundary conditions , for symmetric gb junctions with the nearly circular fs3 ( thick black ) and the tight - binding fs2 ( thin red ) and , and for asymmetric gb junctions with fs2 ( intermediate thickness blue ) . the symmetric curves satisfy @xmath159 , and the antisymmetric curves satisfy @xmath160.,scaledwidth=45.0% ] tricrystal experiments were performed for a variety of cuprate superconductors.@xcite tricrystal ( a ) consists of one asymmetric 0@xmath216 junction and two 0@xmath217 junctions . tricrystal ( b ) consists of two 0@xmath218 junctions and one symmetric 18.4@xmath219 junction . tricrystal ( c ) consists of two asymmetric 0@xmath220 junctions and an asymmetric 2@xmath221 junction.@xcite it was claimed that an odd number of @xmath27-junctions was observed only for tricrystal ( a).@xcite in the case of bi2212 , only results for tricrystal ( a ) were shown.@xcite first of all , our results for the electron - doped fs3 with either specular or random gb tunneling are in complete agreement with a predominant @xmath2-wave op as determined from the tricrystal experiments on ncco.@xcite that is , one would only expect an odd number of @xmath27-junctions only for tricrystal ( a ) , for both random and specular gb tunneling . however , for hole - doped cuprates , the agreement with the tricrystal experiments is less robust . assuming only a @xmath2-wave op,@xcite our fs2 results pictured in figs . 3 , 4 show that tricrystal ( a ) would have two @xmath27-junctions if the tunneling were specular , and one @xmath27-junction if it were random . tricrystal ( b ) should have no @xmath27-junctions , as claimed . tricrystal ( c ) , if the 2@xmath221 junction can be approximated by an asymmetric 0@xmath222 junction , would have one @xmath27-junction for specular tunneling , and two @xmath27-junctions for random tunneling . hence , our fs2 results are only consistent with the observation claims if the tunneling is non - specular . on the other hand , the arpes fs , while only slightly different from fs2 , leads to very different conclusions with regard to the agreement with the tricrystal experiments . for specular gb tunneling , the arpes fs is in complete agreement with the observations of the three tricrystal experiments . however , for random gb tunneling , tricrystal ( a ) is expected to have a very small overall @xmath0 , as @xmath223 for all three asymmetric gb junctions in that tricrystal . although tricrystal ( b ) would be expected to exhibit an even number of @xmath27-junctions , as claimed to be observed , tricrystal ( c ) would be expected to exhibit one @xmath27-junction for random gb tunneling with the arpes fs . hence , the arpes fs can only be fully consistent with the tricrystal experiments on hole - doped cuprates for non - random gb tunneling . one interpretation is that if the quasiparticle tunneling across the relevant grain boundaries is random and the op were pure @xmath122-wave , then the fs would be expected to be more like fs2 . in the unlikely event that the grain boundary tunneling were coherent , then a pure @xmath122-wave op would have a fs more like the arpes fs . such a scenario might be used to argue that the fs of bi2212 , usually thought to have the arpes fs form , is actually closer to fs2 . however , we caution that fs2 and the arpes fs are one - band approximations to the fs of bi2212 , which contains two cuo@xmath3 layers thought by many to be responsible for the superconductivity , and that no tricrystal ( c ) experiments were performed on bi2212 . in addition , this scenario ignores the @xmath13-axis twist experiments , which indicate that essentially all of the @xmath13-axis tunneling is in the @xmath1-wave channel.@xcite if those three experiments are correct , then bi2212 contains an op with an @xmath1-wave component that is at least 20% of the total . although not appreciated by tsuei _ et al._,@xcite in the presence of nanoscale disorder that has recently been observed in stm experiments,@xcite a mixed @xmath224 op is indeed allowed.@xcite thus , it is possible that bi2212 could have a dominant @xmath122-wave op and still have all of the intrinsic @xmath13-axis tunneling be in the @xmath1-wave op channel . if this scenario proves to be true , then further calculations with a mixed @xmath1- and @xmath122-wave op would be warranted.@xcite with regard to the most - studied material , ybco , in addition to the two cuo@xmath3 layers , there is also a conducting cuo chain layer , which is responsible for the orthorhombicity and the strong twinning that occurs in thin films . the strong twinning makes ( 100 ) and ( 010 ) films indistinguishable , which can complicate the analysis . second , unless strongly underdoped , ybco appears to be more three - dimensional than are most other cuprates , so that @xmath225 might be large enough to cause an important modification to our results . third , it is clear from many other experiments that ybco also has a substantial @xmath1-wave component,@xcite which has often been ignored by many workers.@xcite in each twin domain , a purported dominant @xmath122-wave op would have either a @xmath224 or a @xmath226 op mix , and if the twin areas were equal , then one might expect the @xmath1-wave op component to make a negligible contribution to the measured critical currents . however , in @xmath13-axis ybco / pb josephson junctions made with highly twinned ybco thin films , quantitative agreement to the fraunhofer pattern was obtained , although @xmath129 was considerably less than the ab value.@xcite thus , the different results of fs2 and the arpes fs indicate that the tricrystal experiments are less robust determinations of the symmetry of the order parameter in hole - doped cuprates than has been generally thought . small variations in the fs topology and the non - specularity of the gb tunneling processes play a crucial role that can not be anticipated _ a priori_. to the extent that defects play a role in the tunneling processes , it might be possible to understand the experimental observations with a substantial @xmath1-wave op component . especially since the ( 001 ) tilt gb junctions are known to be increasingly underdoped with increasing gb misalignment,@xcite it is possible that @xmath27-junctions arising from the local magnetism of the cu@xmath227 spins formed from oxygen loss at the gb junctions could be important . all of our calculations using the shirai fs are in complete disagreement with both the tricrystal and tetracrystal experiments , regardless of the degree of randomness in the gb tunneling matrix elements . in addition , the shirai _ results for specular symmetric 18.4@xmath219 gb s are in direct conflict with all experiments using tricrystal ( b).@xcite taking all of our results for the various fs s into account , it appears that there is a topological crossover relevant for the agreement of the critical currents with the gl results,@xcite with full agreement for the electron - doped fs s centered about the @xmath228 point in the first bz , and complete disagreement with the extreme shirai fs . the cross - over appears to occur for fs s intermediate to the arpes fs and fs2 . hence , for the hole - doped cuprates , not only are the details of the tunneling important , but the topological structure of the hole - doped fs is also very relevant , and minute details can change the interpretation . tetracrystals consisting of highly twinned thin films of the hole - doped ybco and electron - doped lcco containing a symmetric @xmath229 junction and a 45@xmath230 junction , plus two symmetric @xmath231 junctions were made.@xcite for comparison , they also made bicrystals with two symmetric 22.5@xmath232 junctions . we also calculated the straight gb junction case @xmath32 for a @xmath2-wave op , and our results for fs2 and fs3 are shown by the green dotted lines in figs . 6 and 8 , and at 45@xmath164 , are opposite in sign to the respective symmetric junction cases with specular gb tunneling . if the tunneling were non - specular , as required for the @xmath122-wave interpretation with fs2 of the tricrystal experiments , then the 45@xmath230 junction would have the smallest ( or vanishing , for random tunneling ) @xmath0 value for a @xmath122-wave op . if instead the tunneling were specular , as required for the arpes fs @xmath122-wave interpretation of the tricrystal experiments on hole - doped cuprates , then that junction would allow a small but significant @xmath0 value , but would behave either as a @xmath27- or 0-junction , respectively , for a @xmath122-wave op , depending upon the particular distribution of twin domains present at the junction . thus , this tetracrystal would have an indeterminate number of @xmath27-junctions for specular tunneling and a @xmath122-wave op , but no supercurrent for random tunneling . similar arguments apply to the lcco case . for a highly twinned thin film junction , both specular and random tunneling across the 45@xmath230 gb junction would lead to @xmath223 for a @xmath2-wave op , contrary to the claims in the experiment.@xcite the other two junctions , symmetric 22.5@xmath232 junctions , would not be @xmath27-junctions , unless the tunneling were specular and the fs were different from any that we studied . hence , the interpretation put forward by those authors , which assumed the 45@xmath230 junctions were straight , should be reexamined,@xcite as the @xmath27-junction could arise extrinsically from defects in this very imperfect 45@xmath230 junction.@xcite we urge that the experiment be redesigned in the form of a tricrystal , omitting the 45@xmath230 junction . recently , there have been zigzag in - plane ybco / nb experiments.@xcite in these experiments , a ybco film is angularly cut from above with a ramping angle @xmath233 into a regular zigzag pattern of nominally exposed ( 100 ) and ( 010 ) edges , each edge of equal length @xmath234 m . after deposition of a thick layer of au , a thicker layer of nb was applied . this experiment is a variation of the earlier @xmath235 ybco / pb superconducting quantum interference device ( squid ) experiment of mathai _ et al._,@xcite which contained ybco / pb junctions with a thin ag interstitial layer along the ramped nominally ( 100 ) and ( 010 ) edges of ybco , respectively . in those experiments , one of the junctions functioned as a 0-junction , and the other functioned as a @xmath27-junction.@xcite the beauty of the recent experiment is that the zigzag array of alternating 0-/@xmath27-junctions was observed by a squid microscope to exhibit an ordered array of fractional vortices , the magnitude of which was within the experimental error of @xmath236 , where @xmath237 is the flux quantum . although no microscopic analysis of the edges of the ybco was presented , the ion milling technique employed necessarily leaves a rough surface on a scale of at least 1 - 5 nm , so that the tunneling across each of these edges is most likely incoherent . the thick au layer could modify the overall tunneling process , however . the much longer length scales @xmath238 of the josephson vortices , which is only 1 - 2 orders of magnitude smaller than the edge dimensions , allows the vortices to weakly interact , forming a regular array . these experiments are suggestive of a dominant @xmath122-wave op . however , we recall that gim _ studied ybco / pb squid s prepared with a much wider selection of squid angles @xmath163.@xcite in those experiments , which did not include @xmath235 , the results were inconsistent with either a predominant @xmath122- or @xmath1-wave op component,@xcite but were instead consistent with a @xmath239-wave polar state . apparently , the preparation of a @xmath180-junction or a @xmath27-junction depends upon the ramping angle @xmath240 of the bombarding ions , so that many pairs of squid s with angles @xmath163 and @xmath241 contained one 0-junction and one @xmath27-junction.@xcite although the nice zigzag experiments do demonstrate the reproducibility of the 0-junction and @xmath27-junction preparation at fixed @xmath240 and @xmath235 with ybco / nb junctions , those authors did not report the results of @xmath163 and @xmath240 variation studies.@xcite we are presently engaged in calculating the single quasiparticle and pair excitations at low @xmath14 , both of which are contained in the green function matrix given by eqs . ( [ gmnp ] ) and ( [ gmnm ] ) . for brevity we have here only presented the josephson critical current across the grain boundaries , the quasiparticle densities of states at the sites on the rows adjacent to or near to the junction can be evaluated from the imaginary part of the single particle @xmath242 for @xmath243 although the most commonly used procedure for obtaining the zero - bias conductance peak on a surface is to use the bogoliubov - degennes technique,@xcite our green function technique with the appropriate surface boundary conditions is equally valid,@xcite but is considerably more compact and powerful . with our green function technique , we are able to evaluate the critical current at arbitrary grain boundary angles , and are not restricted to specific interfaces such as @xmath244 for integral @xmath245 . it is also much easier to generalize the tunneling forms , although we have limited our treatment to the most commonly studied specular and random cases . although it would be rather easy to generalize the tunneling matrix elements to forms intermediate to these limiting cases for symmetric and straight junctions,@xcite for asymmetric junctions , such a generalization is more difficult . we have compared our results with those obtained by shirai _ _ using a purely numerical technique , and some of their numerical results at low @xmath14 are in qualitative agreement with our results obtained from eq . ( [ current ] ) . our results with the nearly circular fs3 are in fairly good agreement with those obtained with the gl model and with microscopic calculations using a circular fs.@xcite the main drawback with the green function technique is the procedure for making the calculations self - consistent . however , as noted by tanaka and kashiwaya,@xcite making the results self - consistent does not usually result in qualitative changes . we are presently engaged in self - consistent calculations for various grain boundaries of the @xmath244 types , as was done by shirai _ et al._@xcite but we are using the near - neighbor and next - nearest neighbor single quasiparticle matrix elements appropriate for either the arpes or fs2 fs s on bi2212 , and will present our results elsewhere.@xcite we will also present the quasiparticle densities of states for these more realistic fs models elsewhere.@xcite finally , we remark that recent ybco thin film structures have been studied using @xmath246 thin film boundaries with 45@xmath164 in - plane misalignment angles , and the results interpreted using the sigrist - rice formula . @xcite we have not yet analyzed such grain boundaries , because the fermi surface of ybco is particularly complicated , and our fs2 and arpes fs s are likely to be inadequate to do so reliably . nevertheless , from the results we obtained with these two fs s , it is evident that the results obtained from such devices , the grain boundaries of which have not been analyzed with atomic scale precision as was done with the bi2212 @xmath13-axis twist experiments,@xcite might also have a more conventional explanation.@xcite it has been suggested that the experimental @xmath158 observed in @xmath247 tilt grain boundary junctions can only be understood in terms of interface defects , regardless of the op.@xcite of course , in bi2212 , it has long been established that a significant amount of static defects exist not only on the top cleaved surface,@xcite but throughout even the best single crystals.@xcite such defects combined with significant oxygen loss at the grain boundaries may be the most important source of the observed @xmath27-junctions in the tricrystal and tetracrystal experiments , accounting for the apparent inconsistency between those and the @xmath13-axis bicrystal and cross - whisker experiments.@xcite finally , we recall that @xmath27-junctions are now routinely produced using junctions of the sns and sinis varieties constructed from conventional superconductors,@xcite so that a close reexamination of the tri- and tetracrystal experiments is warranted . this is especially true in the case of the hole - doped cuprates . we note that not only are @xmath27-junctions readily obtained with nb , al and their oxides,@xcite but the associated zero - bias conductance peaks and higher harmonic order josephson phase relations in such conventional junctions have also been observed,@xcite as in ybco.@xcite we calculated the critical current @xmath0 versus ( 001 ) in - plane tilt angle @xmath163 for symmetric and asymmetric grain boundary junctions , assuming the order parameter has either the @xmath1- , extended-@xmath1- , or @xmath2-wave form . we used two fermi surfaces appropriate for hole - doped cuprates , one appropriate for electron - doped cuprates , and an extreme fermi surface for comparison purposes , and took account of the surface boundary conditions appropriate for the interfaces . we studied both tunneling limits of specular and random tunneling . an important feature of our results is that the presence of the surface boundary conditions allows for a finite @xmath122-wave critical current across junctions in which the microscopic tunneling is random . this is especially important because the tunneling across nearly all in - plane ( 001 ) tilt junctions is expected to be random on an atomic scale , even within each reconstructed facet . although our results for electron - doped cuprates are in complete agreement with the interpretation of the observation of the tricrystal experiments on those compounds , our results for hole - doped cuprates differ qualitatively with those obtained from ginzburg - landau models,@xcite and indicate that the interpretation of the tricrystal experiments is complicated , depending upon strong assumptions about the details of the fermi surface topology and the nature of the tunneling processes . in addition , corresponding problems in the interpretation of the tetracrystal experiments on hole - doped cuprates persist , leading to expectations of an @xmath158 inconsistent with many experiments.@xcite we urge further in - plane @xmath247 tilt grain boundary experiments using films deposited by liquid phase epitaxy to be made , and for high resolution transmission electron microscopy studies of the pertinent grain boundaries to be presented.@xcite the authors would like to thank t. claeson , s. e. shafranjuk , and a. yurgens for useful discussions . 99 c. c. tsuei and j. r. kirtley , rev . phys . * 72 * , 969 ( 2000 ) . h. hilgenkamp and j. mannhart , rev . mod . phys . * 74 * , 485 ( 2002 ) . k. a. mller , phil . mag . lett . * 82 * , 279 ( 2002 ) . r. a. klemm , phil . mag . * 85 * , 801 ( 2005 ) ; int . j. mod b * 12 * , 2920 ( 1998 ) . r. a. klemm in _ nonequilibrium physics at short times scales . formation of correlations _ , edited by k. morawetz ( springer , berlin , 2004 ) , pp . 381 - 400 . p. chaudhari and s. y. lin , phys . lett . * 72 * , 1084 ( 1994 ) . c. c. tsuei , j. r. kirtley , c. c. chi , l. s. yu - jahnes , a. gupta , t. shaw , j. z. sun , and m. b. ketchen , phys . 73 * , 593 ( 1994 ) . j. r. kirtley , c. c. tsuei , h. raffy , z. z. li , a. gupta , j. z. sun , and s. megtert , europhys . lett . * 36 * , 707 ( 1996 ) . c. c. tsuei , j. r. kirtley , g. hammer , j. mannhart , h. raffy , and z. z. li , phys . . lett . * 93 * , 187004 ( 2004 ) . c. c. tsuei and j. r. kirtley , phys . lett . * 85 * , 182 ( 2000 ) . r. r. schulz , b. chesca , b. goetz , c. w. schneider , a. schmall , h. bielefeldt , h. hilgenkamp , j. mannhart , and c. c. tsuei , appl . * 76 * , 912 ( 2000 ) . b. chesca , k. ehrhardt , m. mle , r. straub , d. koelle , r. kleiner , and a. tsukada , phys . lett . * 90 * , 057004 ( 2003 ) . a. g. sun , d. a. gajewski , m. b. maple , and r. c. dynes , phys . lett . * 72 * , 2267 ( 1994 ) ; a. g. sun , a. truscott , a. s. katz , r. c. dynes , b. w. veal , and c. gu , phys . rev . b * 54 * , 6734 ( 1996 ) . r. kleiner , a. s. katz , a. g. sun , r. summer , d. a. gajewski , s. h. han , s. i. woods , e. dantsker , b. chen , k. char , m. b. maple , r. c. dynes , and j. clarke , phys . * 76 * , 2161 ( 1996 ) . a. s. katz , a. g. sun , r. c. dynes , and k. char , appl . lett . * 66 * , 105 ( 1995 ) . e. ilichev , m. grajcar , r. hlubina , r. p. j. ijsselsteijn , h. e. hoenig , h .- meyer , a. golubov , m. h. s. amin , a. m. zagoskin , a. n. omelyanchouk , and m. yu . kupriyanov , phys . rev . lett . * 86 * , 5369 ( 2001 ) . s. shirai , h. tsuchiura , y. asano , y. tanaka , j. inoue , y. tanuma , and s. kashiwaya , j. phys . soc . jpn . * 72 * , 2299 ( 2003 ) . in their eq . ( 15 ) , these authors defined their quasiparticle dispersion to be @xmath248 , with @xmath249 , their eq . ( 17 ) , in units of the lattice constant @xmath250 . following their eq . ( 21 ) , the definitions @xmath251 , @xmath252 , @xmath253 , @xmath254 , and the hole density @xmath255 were given , where the summation is only over the two spin states . putting these numbers into the above dispersion , one obtains @xmath256 , which never vanishes , and hence does not exhibit a fermi surface . however , the @xmath257-independent part of this expression is inconsistent with their eq . ( 4 ) for the effective chemical potential , @xmath258 , where the summation is over the four near - neighbor in - plane sites denoted by @xmath259 and the two spin states . hence , the 8 in their eq . ( 15 ) should be replaced with the number 4 to avoid overcounting the two spin states . with this modification , the correct dispersion of shirai _ et al . _ becomes @xmath260 , which has the fermi surface at small pockets in the corners of the first brillouin zone shown in fig . although shirai _ _ may have intended to include this factor of 4 in their definition of @xmath261 , which would have led to a fermi surface resembling fs2 in fig . 2 , from the equations in their manuscript that they very likely used the above dispersion corresponding in our notation to @xmath59 mev , @xmath58 , and @xmath60 , where @xmath42 can be determined from their value of @xmath262 , which we set equal to 10 mev . however , their strange results for the symmetric ( mirror ) and especially the straight ( parallel ) 18.4@xmath164 junctions do not appear to correspond to any results by other authors for any fs studied . y. gao , p. coppens , d. e. cox , and a. r. moodenbaugh , acta cryst . a * 49 * , 141 ( 1993 ) ; p. coppens , p. lee , g. yang , and h. s. sheu , j. phys . solids * 52 * , 1267 ( 1991 ) ; v. petricek , y. gao , p. lee , and p. coppens , phys . b * 42 * , 387 ( 1990 ) . j. j. a. baselmans , a. f. morpurgo , b. j. van wees , and t. m. klapwijk , nature ( london ) * 397 * , 43 ( 1999 ) ; j. j. a. baselmans , b. j. van wees , and t. m. klapwijk , appl . phys . lett . * 79 * , 2940 ( 2001 ) ; j. j. a. baselmans , t. t. heikkil , b. j. van wees , and t. m. klapwijk , phys . lett . * 89 * , 207002 ( 2002 ) .
cloud storage services , like dropbox and google drive , have growing ecosystems of 3rd party apps that are designed to work with users cloud files . such apps often request full access to users files , including files shared with collaborators . hence , whenever a user grants access to a new vendor , she is inflicting a privacy loss on herself and on her collaborators too . based on analyzing a real dataset of 183 google drive users and 131 third party apps , we discover that collaborators inflict a privacy loss which is at least 39% higher than what users themselves cause . we take a step toward minimizing this loss by introducing the concept of _ history - based decisions_. simply put , users are informed at decision time about the vendors which have been previously granted access to their data . thus , they can reduce their privacy loss by not installing apps from new vendors whenever possible . next , we realize this concept by introducing a new privacy indicator , which can be integrated within the cloud apps authorization interface . via a web experiment with 141 participants recruited from crowdflower , we show that our privacy indicator can significantly increase the user s likelihood of choosing the app that minimizes her privacy loss . finally , we explore the network effect of history - based decisions via a simulation on top of large collaboration networks . we demonstrate that adopting such a decision - making process is capable of reducing the growth of users privacy loss by 70% in a google drive - based network and by 40% in an author collaboration network . this is despite the fact that we neither assume that users cooperate nor that they exhibit altruistic behavior . to our knowledge , our work is the first to provide quantifiable evidence of the privacy risk that collaborators pose in cloud apps . we are also the first to mitigate this problem via a usable privacy approach . < ccs2012 > < concept > < concept_id>10002978.10003029.10011703</concept_id > < concept_desc > security and privacy usability in security and privacy</concept_desc > < concept_significance>500</concept_significance > < /concept > < concept > < concept_id>10002978.10003018.10003021</concept_id > < concept_desc > security and privacy information accountability and usage control</concept_desc > < concept_significance>300</concept_significance > < /concept > < concept > < concept_id>10002978.10003029.10011150</concept_id > < concept_desc > security and privacy privacy protections</concept_desc > < concept_significance>300</concept_significance > < /concept > < concept > < concept_id>10003120.10003121.10003122.10010854</concept_id > < concept_desc > human - centered computing usability testing</concept_desc > < concept_significance>300</concept_significance > < /concept > < /ccs2012 >
1702.08234
* the rise of cloud apps : * + the popularity of consumer cloud storage providers ( csps ) over the previous decade has been on a roll . dropbox , google drive , and one drive have each amassed hundreds of millions of users . in order to further appeal to their users , the csps have been transitioning from being pure _ service providers _ to becoming _ app ecosystems_. hence , they now offer apis for developers to import and process users files stored in the cloud . consider , for example , a web app called https://pandadoc.com[pandadoc ] , which allows creating , editing , and signing documents online . when a user uses pandadoc from her laptop browser , she can import files stored in her google drive instead of her hard drive . such a pattern is increasingly more prevalent with the growing number of 3rd party cloud apps ( or 3pc apps ) that are tightly integrated with cloud storage services . dropbox alone claims that hundreds of thousands of apps have been integrated with its platform . even in the enterprise setting , 3rd party cloud apps are on the rise . this is first because companies are officially adopting the likes of _ dropbox business _ , _ onedrive for business _ , and _ google drive for work_. second , it is due to employees utilizing their personal cloud accounts to share company s files ( a.k.a shadow it ) . various reports from cloud application security providers state that organizations use from 10 to 20 times more cloud apps than their it department thinks @xcite . * risks in 3rd party cloud apps : * + however , in our previous work , we have shown that 76% of the 3rd party google drive apps featured on google chrome store request full access to users google drive data @xcite . around 64% of these apps are _ over - privileged _ : they require more permissions than are needed for them to function . accordingly , users are now faced with a new kind of privacy adversary : the 3rd party app vendors . with every app authorization decision that users make , they are trusting a new vendor with their data and increasing the potential attack surface . elastica , the cloud application security provider , estimates that the average financial impact on a company as a result of a cloud - storage data breach is $ 13.85 m , including remediation costs @xcite . in 2015 , the data breach at anthem , a us insurance company , has reportedly cost more than $ 100 m , with 80 m unencrypted health records leaked . this was a result of an exfiltration exploit leveraging a popular public cloud storage application @xcite . even on the personal level , the risk extends from breaches exposing financial information and health records to unnoticeable , continuous profiling based on stored files . * exposure through collaboration : * + an additional intricacy is that when users grant access to a 3rd party cloud app , they are not only sharing their personal data but also others data . this is because cloud storage providers are inherently collaborative platforms where users share and cooperate on common files . hence , protecting these files is not solely in the hands of the user . skyhigh networks , another provider of cloud security software , reports that 37.2% of documents ( across 23 million users ) are shared with at least one other user . in organizations , documents are shared , on average , with accounts from 849 external domains @xcite . moreover , around 23% of cloud documents were found by elastica to be `` broadly shared '' , which means that they are shared ( a ) among all employees , ( b ) with external partners and clients , or ( c ) with the public @xcite . interestingly , 12% of those documents contained compliance - related or confidential data . this further highlights what has been termed as the _ interdependent privacy problem _ @xcite , where the decisions of friends can affect the user s privacy and vice - versa . this concept was initially proposed in the context of third - party social networking apps , such as facebook . however , while 1.92% of facebook apps request friends personal information , this is much more pronounced in 3rd party cloud apps , where all apps accessing one s files get access to the part which is shared too . moreover , unlike facebook apps , due to the collaborative nature of cloud apps , the csps do not provide an option for users to control whether their collaborators apps can get access to data they own . * research questions : * + so far , the main approach to reducing the risk of 3pc apps has been focused on discovering over - privileged apps and deterring users from installing them @xcite . even then , a lot of users would still install such apps as they prioritize short - term utility over long - term risk aversion or due to the absence of alternatives . furthermore , that approach relies on manually inspecting each app by experts and on applying a plethora of machine learning algorithms to visualize the various risks for users . these issues could present a hurdle towards a wide - scale deployment by csps . in this work , we address the wider problem of minimizing the risk of all 3pc apps , regardless of whether they are over - privileged or least - privileged . we are further driven by the rationale that users will inevitably continue to install apps to achieve various services . hence , instead of stopping them , we aim to lead them to select apps from vendors in a way that minimizes their privacy risk . we achieve this by leading users to take what we term as _ history - based decisions_. such decisions account for the vendors who previously obtained access to the user s data , whether directly ( with her consent ) or via her collaborators . our strategy consists of introducing privacy indicators to the current permissions interfaces that help users minimize the number of vendors with access to their data . our `` usable privacy '' approach is guided via a data - driven study and is evaluated via a data - driven simulation . in essence , we tackle the following research questions : * from a practical perspective , are the collaborators decisions significant enough to be accounted for in users app adoption decisions ? * do users already account for entities with access to their data ? if not , to what extent can the usage of privacy indicators lead to users taking history - based decisions ? * how significant is the effect of adopting these privacy indicators in the case of large networks of users and teams ? * contributions : * towards addressing these questions , we make the following contributions : * in section [ sec : dataset ] , we analyze a real - world dataset of google drive users , and we show that the median privacy loss that collaborators cause by installing apps can be much higher than that inflicted by the user s own app adoption decisions ( 39% higher with 5% of shared files and 523% higher with 60% of shared files ) . to our knowledge , this is the first usage of a real - world dataset to give a concrete evaluation of interdependent privacy in any ecosystem . * driven by the significant impact of collaborators , we design new privacy indicators for helping users mitigate the privacy risk via history - based decisions ( cf . section [ sec : study ] ) . we assess these indicators via a web experiment with 141 users . we show that they significantly increase the likelihood that users choose the option with minimal privacy loss , even if not all of these users are motivated by privacy per se . to the best of our knowledge , this is also the first work to investigate a usable privacy approach to mitigating the problem of interdependent privacy . the few studies on this problem have mainly approached it from a theoretical perspective , such as developing game - theoretic or economic models @xcite or from a behavioral perspective , such as studying the factors affecting real users monetary valuation of others privacy @xcite . * we explore the potential of history - based decisions by performing a simulation on two large user networks . we show that the network - effects of our approach result in curtailing the growth privacy loss by 70% in a synthetic google drive - based collaboration network and by 40% in a real author collaboration network . we also simulate the effect of such decisions in a teams network . we demonstrate that teams can reduce the privacy loss by up to 45% by solely accounting for team members decisions ( cf . section [ sec : simulation ] ) . there are four main entities that interact in the third - party cloud app system : 1 . a _ user _ @xmath0 who uses that app for achieving a certain service 2 . a _ cloud storage provider ( csp ) _ hosting the user s _ data _ 3 . a _ data subject _ to whom the files belong and whose privacy is being considered . we further define two levels of data subject granularity : * _ individual - level granularity _ : i.e. , the user herself is interested in guarding her own data privacy , * _ team - level granularity _ : i.e. , a group of users are interested in guarding the privacy of team - owned data ( e.g. , using an enterprise version of cloud storage services ) 4 . a _ vendor _ @xmath1 that is responsible for programming and managing a 3rd party cloud app ( or shortly a cloud app or a 3pc app ) . these vendors register their apps with the csps . the apps themselves are hosted on any website the vendors choose ( i.e. , not hosted by the csp itself ) . each user has access to a set @xmath2 of files stored at the csp . a subset of these files is owned exclusively by the data subject while the other subset is composed of files that are each shared with at least one other _ collaborator_. we denote the set of all collaborators of user @xmath0 by @xmath3 . for simplicity reasons , we will assume throughout this work that the files of all data subjects , as well as the collaborators for each file , are all fixed from a reference step @xmath4 . using the csp s api , the vendor @xmath1 can get access , at step @xmath5 , to the subject s data upon _ user authorization _ , which consists of @xmath0 accepting a list of _ permissions_. we will alternatively refer to this as _ app installation _ , and we will assume that exactly one app is installed at each step @xmath6 . permissions are named differently across various providers , but , in general , we can categorize them into three categories : * * per - file access * : where the user has to authorize the vendor for each file access individually . this is typically done via a file picker provided by the csp itself . * * full - access * : where the vendor gets access to all users data . in the interface , this is worded , for instance , as `` view the files in your google drive '' or `` access to the files and folders in your dropbox '' . * * per - type access * : where the vendor gets access to all files of a specific type . for example , dropbox words it as `` access to images in your dropbox '' . some platforms , like google drive , do not provide app developers with such fine - grained options . the authorization can also give @xmath1 access to files shared with the collaborators of @xmath0 . similarly , collaborators of @xmath0 can install apps that expose files shared with @xmath0 to new vendors . we denote the set of files of @xmath0 accessible by vendor @xmath1 at step @xmath6 as @xmath7 . .summary of notations used [ cols="<,<",options="header " , ] a user is further assumed to be _ self - interested _ , i.e. , only caring about optimizing the privacy of the data subject ( a.k.a . , privacy egoist ) , and _ non - cooperative _ , i.e. , does not coordinate her decisions with others . we do not assume that the risks of installing each app are known to the users or calculated a priori . in fact , unlike other 3rd party app ecosystems , the risk of each cloud app can not be automatically estimated based on techniques such as taint tracking @xcite or code analysis @xcite because the main app s functionality is typically implemented on the server side ( which can not be accessed by external entities ) . such assumptions constitute the _ worst case _ in the scenarios we consider , and further privacy optimizations can be obtained by relaxing them . we also assume that the mental model for privacy - concerned users matches the possible permission granularities they are given . accordingly , privacy - concerned users can have one of the following privacy - goal granularities : * * per - type privacy goal * : where users aim to optimize their privacy independently for different file types . for example , in an ecosystem like dropbox , where per - type access is an option , users might follow the separation - of - concerns principle . hence , they might install photo - related apps from a set of vendors that is different from the set authorized for document processing . * * all - files privacy goal * : where users aim to reduce the privacy risk for their entire set of files . this can be in the case of ecosystems which do not have the option of per - type access , like google drive . it can be also the case that a user of dropbox has this goal in mind despite being presented with finer - grained app permissions . we consider the 3rd party app vendors as the adversary ( and not the csp ) . the privacy indicator we introduce is best implemented by the csp , which already has access to the users and collaborators data . alternatively , this can be a feature within cloud access security brokers ( e.g. , skyhigh networks , netskope , etc . ) , which are already trusted by thousands of enterprises to protect their cloud data against other 3rd parties . moreover , we consider the protection against over - privileged apps as an orthogonal problem , which we have considered in @xcite . we rather focus on the interdependent privacy problem , which covers all vendors with full access and is an issue in least - privileged apps too . in order to quantify the privacy loss that a user incurs with time , we introduce now the _ vendors file coverage ( @xmath8 ) _ metric . consider a user @xmath0 and a set @xmath9 of vendors at a certain time step . for notation simplicity , we will omit the time step henceforth . @xmath10 is computed as the summation of the files fractions shared with each of these vendors : @xmath11 intuitively , @xmath10 increases as vendors in @xmath9 get access to more files of @xmath0 . it has the range @xmath12.$ ] by @xmath13 as multiple vendors with access to all the user s files induce a higher privacy loss than one vendor with such access . ] if we consider the set @xmath14 of vendors explicitly authorized by user @xmath0 , we can define the _ self - vendors file coverage _ as : @xmath15 similarly , if we consider the set @xmath16 of vendors authorized by the collaborators @xmath3 of @xmath0 , we can define the _ collaborators - vendors file coverage _ as : @xmath17 finally , the _ aggregate @xmath18 _ for a user @xmath0 is that due to all vendors authorized by @xmath0 or its collaborators : @xmath19[equation : aggvfc ] throughout this work , we will use the terms _ privacy loss _ and @xmath8 interchangeably . as will become evident in section [ sec : study ] , this metric choice allows relaying a message that is simple enough for users to grasp , yet powerful enough to capture a significant part of the privacy loss . obviously , one can resort to a deeper inspection of content or metadata sensitivity ( as in @xcite ) had the purpose been finding the best privacy model in general . however , for instigating a behavioral change , telling users that a company has 30% of their files is more concrete than a black - box description informing them that the calculated loss is 30% and constitutes less information - overload than presenting them with detailed loss metrics at this point , we are in a position to handle the first research question on the extent of collaborators contribution to a user s privacy loss . hence , we want to test the following hypothesis : _ h1 : the collaborators app adoption decisions have a significant impact on the user s privacy loss . _ if this hypothesis is valid in practice , it provides a strong motivation for designing privacy notices that aid users in accounting for their collaborators decisions , which is what we will study in section [ sec : study ] . towards that , we will be dissecting the privacy loss , quantified by @xmath8 , that users incur in a realistic 3rd party cloud apps dataset . to study the problem in a realistic context , we will be taking google drive as a case study in this work , given that it has one of the most popular 3rd party ecosystems . nevertheless , the insights gained from our work are applicable to other cloud platforms as well . the main ( content - related ) google drive permissions that 3pc apps vendors can request are presented table [ tab : defaultpermissions ] , along with the google - provided description for each . this short description is also presented to the user when installing an app ( see figure [ fig : baseline ] for an example app ) . the user can click on the info button next to each permission to read additional explanations in a popup . the user has to accept all permissions in order to utilize the app . these apps can be found on google chrome web store ( and other google stores ) , where users can rate and review them . in this work , we will focus on content - related permissions . hence , as discussed in section [ sec : models ] , we differentiate between two levels of access : ( 1 ) full access , which includes the drive_readonly and drive permissions and ( 2 ) per - file access that includes the drive_file permission . google drive does not offer the per - type permissions option . l4.5cmcl * permission * & & * short name * + + view the * files * in your google drive . & & drive_readonly + + view and manage the * files * in your google drive . & & drive + + view and manage google drive files that you have * opened or created with this app*. & & drive_file + + view your google drive apps . & & drive_apps_readonly + [ tab : defaultpermissions ] one of the main challenges when studying the privacy loss in 3rd party cloud apps is the absence of public datasets with realistic file distributions , collaborator distributions , sharing patterns , 3rd party app installations , etc . we benefit in this section from a dataset that we have collected in a previous work via the privyseal service @xcite . we build our analysis on it in order to evaluate the @xmath8 of users in a realistic context . privyseal is a web service for for assisting users in avoiding google drive apps that are over - privileged . it deters users by showing them the far - reaching insights that such apps can glean from their data ( e.g. , their topics of interest , collaboration and activity patterns , etc . ) . there are currently over 1500 registered users in privyseal , and we refer the interested reader for our previous work for more details @xcite . the dataset , henceforth referred to as the _ privyseal dataset _ , was anonymized and contained metadata - only information . it included a subset of the files metadata of 183 privyseal users in addition to the google drive apps installed by those users prior to authorizing privyseal s app ( the drive_apps_readonly permission was requested by privyseal ) . each user had a minimum of @xmath20 files in total and at least @xmath21 of files that are shared . the dataset specifically contained : * list of user ids ( anonymized via a one - way hash function ) ; * ids of files in each user s google drive , * list of anonymized collaborators ids for each file i d ; * list of apps with full - access installed by each user ; * the vendor of each app . in total , the number of users in addition to collaborators was 3422 . overall , these users had installed 131 distinct google drive apps from 99 distinct vendors . figure [ fig : datasetstats ] characterizes the privyseal dataset . particularly , it displays 4 distributions in this dataset , which realistically model the system under study : * number of files per user , which follows a skewed distribution with a median of 67 files * sharing pattern : percentage of shared files out of all user files , which also follows a skewed distribution with a median around 18% * number of collaborators across all user files ( a.k.a . , the degree of the user node in the collaboration network ) : where 75% of the users had less than 23 collaborators * number of vendors authorized per user : also follows a skewed distribution with a median of 1 vendor per user we computed the _ self-@xmath8 _ , the _ collaborators-@xmath8 _ , and the _ aggregate-@xmath8 _ ( as defined in section [ sec : privloss ] ) for users in the privyseal dataset _ but not in computing _ collaborators-@xmath8_. ] . as we did not have the actual number of apps for each collaborator of users in the dataset , we assigned to these collaborators a set of apps from a random user of the dataset . we show in figure [ fig : coverage_growth_withsharing ] how these metrics evolve as we gradually consider populations that collaborate more frequently . with @xmath21 , we had a median of 1.39 for _ collaborators-@xmath8 _ , which was 39% higher than a median of 1.00 for _ self-@xmath8_. the significance of the median difference is evidenced by the non - overlapping box - plot notches . this difference became much larger when we considered users that share more files . we had a 100% median difference at @xmath22 and 523% median difference at @xmath23 . such results indicate that : * the collaborators app adoption decisions contribute a core component to the user s privacy loss , thus confirming our hypothesis @xmath24 . * the higher the number of collaborators is , the higher the magnitude of loss these collaborators can potentially inflict . both conclusions motivate the need for taking collaborators decisions into account when designing privacy indicators for cloud apps , which is what we will embark on next . 0.5 , @xmath25 ) . the numeric labels denote the corresponding number of users in the dataset.,title="fig : " ] 0.5 , @xmath25 ) . the numeric labels denote the corresponding number of users in the dataset.,title="fig : " ] 0.5 , @xmath25 ) . the numeric labels denote the corresponding number of users in the dataset.,title="fig : " ] 0.45 , @xmath25 ) . the numeric labels denote the corresponding number of users in the dataset.,title="fig : " ] [ fig : datasetstats ] , @xmath25 ) . the numeric labels denote the corresponding number of users in the dataset.,scaledwidth=80.0% ] up till now , we have confirmed that , if users want to minimize their privacy loss , they are better off not ignoring the app installation decisions of collaborators . in this section , we tackle the next research question , where we investigate the potential of privacy indicators in leading users to minimize their exposure to 3pc app vendors . we show first our design methodology for the privacy indicators , and we follow that by a web experiment that investigates the efficacy of these indicators in realistic scenarios . we call our proposed privacy indicators _ `` history - based insights '' ( _ hb _ insights ) _ as they allow users to account for the previous decisions taken by them or by their collaborators . we continue to consider google drive as a case study , and we show this indicator in the context of google drive apps permissions in figure [ fig : history_interface ] . compared to the current interface provided by google ( figure [ fig : baseline ] ) , we added a new part to highlight the percentage of user files readily accessible by the vendor ( computed based on @xmath26 for each vendor @xmath1 ) . as we prove in appendix [ sec : optimal ] , selecting the vendor that already has the largest percentage of user files is the optimal strategy to minimize the privacy loss in our context . we denote this strategy as _ `` history - based decisions''_. following the best practices in privacy indicators design @xcite , our indicator was multilayered , with both textual and visual components . the wording of the main textual part was brief and general enough to hold for both the data percentage exposed by friends and that exposed by the user . we used a percentage value rather than a qualitative measure to facilitate making comparisons among apps based on this value . the visual part showed the percentage as a progress bar with a neutral violet color . the bottom textual part was added in a smaller font to provide further explanation for those interested . we used the term `` company '' in our interface instead of `` vendor '' as it is more commonly understood by the general audience . in order to evaluate the new permissions interface , we performed an online web experiment ( rather than a lab study ) as we were mainly motivated by obtaining a large sample of users that is also geographically and culturally diverse . the hypothesis we wanted to test is : _ h2 : introducing the new privacy indicator significantly increases the probability that users take history - based decisions . _ in addition , the study allowed us to build a realistic user decision model based on the choices taken by participants in different conditions . we will utilize this model in section [ sec : simulation ] to simulate the app choices in a large user network and to study the effect on the overall @xmath8 in the network . we structured our study to have ( 1 ) an introductory survey , ( 2 ) a series of app installation tasks , and ( 3 ) a concluding survey . * user recruitment : * we recruited users via crowdflower s crowdsourcing platform . in our study , we restricted participation , via the platform s filtering system , to the highest quality contributors ( performance level 3 ) . we also geographically targeted countries where english is a main language as our interface was only in english . in order to further guarantee quality responses , each user was rewarded a small amount of @xmath27 for merely completing the study and an additional amount of @xmath28 that was manually bonused for those who did not enter irrelevant text in the free - text fields . * instructions : * participants were first presented with introductory instructions that explained the context of the study ( i.e. , cloud storage services and 3rd party apps that can be connected to them ) . they were asked to only continue if they had good familiarity with cloud storage services ( e.g. , google drive , dropbox , etc . ) . we did not explicitly require that participants have experience with 3rd party cloud apps . however , we educated them about such apps throughout the instructions , particularly showing them two examples of 3rd party apps in action ( pandadoc for signing documents and iloveimg for cropping photos ) . these apps were displayed via animated gifs that play automatically and do not rely on the user clicking . we used limited deception by neither mentioning the focus of the study on participants privacy nor giving hints about selecting apps based on the installation history . the advertised purpose was to `` check how people make decisions when they install 3rd party apps . '' * introductory survey : * after checking the instructions , users were presented with an introductory survey , where they first entered general demographic information . this survey was also front - loaded with questions about cloud storage services ( several of which required free - text input ) in order to discourage users who had not used these services from continuing to the actual study . next , users could proceed to the study page . we used a split - plot design in the study . participants were randomly assigned to one of two groups : 1 . * baseline group ( _ bl _ ) * : where the permissions interface used is that currently provided by google drive ( figure [ fig : baseline ] ) . * history - based group ( _ hb _ ) * : where the _ history - based insights_permissions interface ( figure [ fig : history_interface ] ) is used . in each group , the study consisted of 3 modules , which cover the main conditions that can occur when users desire to install a cloud app . on a high level , the modules investigate the following questions : 1 . * module 1 : * are users likely to select apps from the same vendor they installed from before ? 2 . * module 2 : * are users likely to select apps from vendors that her collaborators have used before ? 3 . * module 3 : * do users consider the differences in access levels obtained by vendors that collaborators installed ? in all modules , whenever the user was asked to _ choose _ an app , she was presented with a list of 12 apps ( figure [ fig : app_info ] shows an example app ) . only two of these apps were relevant to the task purpose , and they were placed on top of the list ( randomly positioned as first or second ) . with this setup , we wanted to mimic the realistic setup of app browsing while not squandering the user s effort on finding apps . all apps had the same full access permissions too ( namely drive permission ) . unlike in chrome store , we removed elements such as ratings , user reviews , and screenshots and kept a minimal interface . this is all in order to reduce the distractions from factors outside the study . we refer the user to the work of kelly et al . , @xcite who investigated the effects of those elements on users decisions for android apps . in order to account for fatigue and learning effects , modules 1 , 2 , and 3 were presented in a random order for users . we piloted our experimental setup in two stages : with colleagues and with online users from the crowdflower community itself . for reviewing the online pilot testers work , we embedded a javascript code for session recording in our study s web page , which allowed us to view the user s mouse and keyboard actions on our side . * demographics : * we had 157 users who completed the study . based on manually reviewing the users inputs , we removed 16 users who were inputting irrelevant free - text in the survey in the study . we thus report the results of 141 users , 72 of which were in the _ bl _ group and 69 in the _ hb _ group . in table [ tab : demographics ] , we describe the participants demographics based on the introductory survey . of these participants , @xmath29 were males and @xmath30 were females . they were between 18 and 62 years old , with a median of 31 . moreover , 42.3% of the participants had worked or studied in it before . participants were mostly from india ( 37% ) , usa ( 35% ) , britain ( 7% ) , germany ( 7% ) , and canada ( 7% ) . crowdflower presents the users with an optional satisfaction survey after completing the study , and 49 users took this survey . on average , the study received 4.2/5 for instructions clarity , 3.8/5 for questions fairness , 3.8/5 for ease of job , 3.6/5 for pay sufficiency ( before the bonus was rewarded ) . this ensures that participants behavior has not been affected by either a lack of time to complete the task or the task design in general . r l l + * age * & 18 - 62 & ( median 31 years ) + & & + * gender * & 35.5% & female + & 64.5% & male + + * occupation * & 59.6% & full - time employees + & 14.2% & student + & 6.4% & part - time worker + & 8.5% & self - employed + & 5.0% & homemaker + & 6.4% & unemployed / retired + + * it experience * & 41.8% & have worked or studied in it + + * degree * & 19.1% & high school + & 7.1% & trade / tech./vocational training + & 51.1% & associate or bachelor s degree + & 22.7% & post graduate degree + + * countries * & 35.0% & usa + & 37.5% & ind + & 7.5% & gbr + & 6.9% & deu + & 6.9% & can + & 7.4% & aus+irl+ nld + pak + [ fig : instructions ] + [ fig : exp-1-install ] [ fig : exp-1-choice ] + [ fig : exp-2-intro ] [ fig : exp-2-install ] + [ fig : exp-3-intro ] + we now move to the detailed description of the modules and the results obtained . these modules are summarized in figure [ fig : modules ] , to which we refer henceforth . we also show sample screenshots from the online study in figure [ fig : screenshots ] . the results are also presented in table [ tab : studyresults ] . * module 1 ( self - history scenario ) : * tests whether the user is more likely to select an app from the same vendor she has just installed from before . in step ( a ) , the user is made aware the she installed an app from a specific ( figure [ fig : exp-1-install ] ) . in step ( b ) , she is asked to install an app that satisfies the given purpose ( figure [ fig : exp-1-choice ] ) among a list of apps . two of the listed apps were relevant , and one of them was from vendor @xmath1 itself . despite the participants being informed one step earlier that they installed an app from `` thetimetube.com '' , that did not make a difference in the _ bl _ case : half of the users still chose the app from the new vendor `` nitrosafe.org '' ( cf . table [ tab : studyresults ] . in the absence of traditional signals that users follow for deciding on apps ( reviews , ratings , permissions ) , participants apparently made decisions that cancelled out , making the two apps equally favored across participants . * the vast majority of users were not approaching the installation from the angle of keeping their data with fewer shareholders*. based on their provided justifications , they rather looked for other cues , such as selecting the app that , in their opinion , has a more comprehensive description , a more professional logo , a better sounding name , or a more trustable url . still , 12 users have explicitly mentioned in their text input that they chose an app _ because _ it is from the same vendor they have dealt with earlier . even then , neither of them has alluded to a privacy motivation behind the choice . these 12 participants mainly provided cross - app compatibility , interface familiarity , and satisfaction with the previous vendor as justifications . for example , one participant wrote : `` _ _ i favoured malware scanner due to the fact that the name ` thetimetube.com ' was in the last app installed , and i tend to install apps from the same company due to cross - app compatibility usually found in apps by the same company . _ _ '' interestingly , two users justified their installation of the app from the new vendor ( nitrosafe.org ) by writing that they had just installed an app from the same company before . this indicates that , * even when users try to account for previous decisions , they might find it difficult to remember the previous app vendors*. given that our study had a short time span separating the current from the previous installation , we expect that such mistakes would be even more common in real scenarios when app installation instances are separated by longer time spans . the _ hb _ group witnessed a much larger proportion of users who favored the option with less privacy loss . 72.2% of the participants selected the app from the `` thetimetube.com '' ( the vendor which already has access ) . the difference of 22.8% compared to the _ bl _ group is statistically significant ( fisher s exact test , @xmath31 ) . many of the participants who chose the app from `` thetimetube.com '' reported that they were motivated by the 100% access that the app already has . we counted around 40 such users ( i.e. , 57% of the _ hb _ some of them went further and explicitly mentioned that their selection was motivated by giving data to fewer data owners ( i.e. , more privacy ) . for example , one user wrote : `` _ _ this company has access to all my files , so i would choose them as i do nt want to have 2 companies with full access to my files _ _ '' . in a nutshell , we were able to verify our hypothesis in this scenario : * the new privacy indicator leads users to more frequently choose the app from a vendor they already authorized*. furthermore , we have discovered that the _ hb _ insights interface has indirectly made users think about various positive effects brought by using apps from the same vendor . this eventually lead them to make more privacy - preserving decisions . @rcccccccccc@ & & & & & & * @xmath32 * & & * @xmath33-value * + ( lr)3 - 4 ( lr)6 - 7 & & _ vwa _ & _ nv _ & & _ vwa _ & _ nv _ & & & & + + * self history * & & 50.0% & 50.0% & & 75.4% & 24.6% & & 25.4% & & 0.003 + * collaborator s app * & & 52.8% & 47.2% & & 88.4% & 11.6% & & 35.6% & & < 0.001 + * collaborator s vendor * & & 58.3% & 41.7% & & 82.6% & 17.4% & & 24.3% & & 0.002 + * multiple collaborators * & & 44.4% & 55.6% & & 82.6% & 17.4% & & 38.2% & & < 0.001 + * module 2.1 ( collaborator s app scenario ) : * tests the likelihood that the participant selects the same app that her collaborator had used . in step ( a ) , the participant is made aware that she had shared all her photos with a friend @xmath34 ( figure [ fig : exp-2-intro ] ) . for more familiarity , we also added a picture for each of the two fictitious friends throughout the study . in step ( b ) , the user is made aware that her friend @xmath34 has installed an app @xmath35 ( figure [ fig : exp-2-install ] ) from vendor @xmath1 . she is asked to type the name of the app s vendor ( `` paste '' option was disabled in the input field to further ensure the participant is aware of the vendor ) . in step ( c ) , the user is asked to install an app with a certain purpose ( similar to figure [ fig : exp-1-choice ] ) . one of the two matching apps is app @xmath35 . similar to the previous module , the _ bl _ group witnessed an almost even split between `` online player '' , installed previously by the friend , and `` enjoy music player '' , from a new vendor ( cf . table [ tab : studyresults ] . we also noticed that 20 participants in this group justified their decision by mentioning that their friend has used the app . still , neither of them alluded at privacy reasons in their justifications . instead , the two most prevalent motivations were ( 1 ) considering the friend s use of the app as a _ recommendation _ or ( 2 ) achieving _ compatibility _ with their friends app , which facilitates data sharing within the app itself . quoting one user : `` _ _ this is the same app my friend is using so it should be quite compatible for us to both share . _ _ '' in addition to having a significant 35.6% difference in the case of the _ hb _ group , we noticed that 32 users mentioned the existing data access as a reason for choosing the app `` online player '' . also , 26 users referred to the fact that the friend has installed this app before ( including those who mentioned both of the previous reasons ) . unlike the _ bl _ group s justifications though , where the friend s recommendation and the app s compatibility prevailed , the privacy issue was explicitly brought up by at least 10 users . one participant put it as follows : `` _ _ thanks to john , they have already access to 70% of my data . sharing the last 30% is nt as bad as sharing 100% of my data with driveplayer.com.__ '' * module 2.2 ( collaborator s vendor scenario ) : * we proceed in steps ( d ) and ( e ) as in the previous scenario s steps ( b ) and ( c ) , with the difference that a _ new _ app from @xmath1 is included among the options in step ( e ) instead of the exact same app @xmath35 . one interesting insight from this scenario is that * the line between the company and the app is blurred in the minds of several users * who used the two entities interchangeably . in fact , 3 users in the _ bl _ group and 7 participants in the _ hb _ group justified their choices by mentioning that their friend installed the _ same app _ before , which was not the case . for example , one user wrote : `` _ _ this app already has access to my files , and i do nt want to install any new app . _ _ '' * module 3 ( multiple collaborators scenario ) : * given collaborators @xmath36 and @xmath37 , where the user shares much more data with @xmath36 , this scenario checks the likelihood of the participant authorizing an app that @xmath36 has installed . in step ( a ) , the participant is made aware that @xmath36 has access to more data than @xmath37 ( figure [ fig : exp-3-intro ] ) . in steps ( b ) and ( c ) , the participant is made familiar with the apps each of the friends installed ( similar to figure [ fig : exp-2-install ] ) . in step ( d ) , the user is asked to select an app with a specific purpose . the two friends apps are the only ones matching , and the choice is to be made between them ( similar to figure [ fig : exp-1-choice ] ) . in the _ bl _ group , we had 44.4% of the participants choosing the app installed by @xmath36 . still this percentage is relatively close to an equal split between the two apps . out of this percentage , 13 users justified their choice by mentioning that they were encouraged to follow the choice of friend @xmath36 . even though they did not mention privacy , the larger number of files shared with @xmath36 was often used as a justification . for example , one participant wrote : `` _ _ this is the app that john already uses , and he has access to all of my files . the pdf mergy app is used by lisa , but she only has access to part of my files . _ _ '' in the _ hb _ group , around 82.6% chose the app previously installed by the friend @xmath36 , which is significantly more than those in the _ bl _ case ( fisher s exact test , @xmath38 ) . looking at the justifications , around 37 users explicitly mentioned the higher access level that this app already possesses as a reason for their choice . privacy was additionally mentioned by 8 of these users . quoting one of them : `` _ _ pdf mergy already has access to 70% of my files . using pdf files merger would unnecessarily increase third party app access to my files . _ _ '' however , we still had 2 users who went for the app with less existing access , with one of them saying he favors the app that only `` _ _ had accessed 30% of files before installation _ _ '' . what was interesting though is that * almost all users who mentioned friends were actually making a comparison between the two friends existing access level * , regardless of their final choice . at the end of the user study , users were presented with a final set of questions . we asked them whether they would like to be notified when a friend installs an app that gets access to their shared files . around 92% of users in the _ bl _ group and 90% of users in the _ hb _ group agreed . we further asked the participants whether they are fine with a collaborator being notified when they install applications that access files shared with that collaborator . the percentage of people who agreed dropped to 75% in the _ bl _ and 78% in the _ hb _ group . the relatively small difference between the answers to these two questions highlights that * only a minority of users is not willing to make the trade - off of contributing to the overall system . * such users can be given the option to not use privacy indicators based on their friends decisions . next , users were asked the following question _ `` assume you have installed an application called youmusic from a company called musicana and gave it access to all your files on google drive . now you are considering installing an application called youvideo from the same company . how do you think that this application will affect your privacy:''_. only 11% of each group replied by _ `` negatively''_. the vast majority in both groups either perceived the avoidance of a new vendor as a positive outcome or considered that the privacy loss will remain the same . interestingly , the users in the _ bl _ showed a similar reasoning in justifying their choices as the _ hb _ group although the latter were primed about these aspects via the privacy indicators . this indicates that * the privacy indicators actually match the first intuition for a large fraction of users*. overall , we found out that , in the three modules , participants in the _ hb _ group were significantly more likely to install the app with less privacy loss ( i.e. , the app from the vendor with the largest share of the user s files ) than those in the _ bl _ group . despite showing the efficacy of history - based insights , our study still has its limitations . in order to get a large , diverse sample size , we resorted to a web experiment based on role - playing with hypothetical data . it would be interesting to see how such results extrapolate to the case where users own data is in question . moreover , in our design , we have abstracted several factors ( e.g. , ratings and reviews ) , which have been previously studied in similar ecosystems @xcite , in order to focus on one factor . these factors might have diluted the effect of the privacy indicator . still , we conjecture that , although the absolute values of our findings might not strictly apply , the differences between the two groups will still be practically significant . additionally , in this paper , we have investigated only one type of history - based privacy indicators . evidently , such indicators can be integrated at different stages of the app installation process . for example , they can be part of the recommendation strategy for suggesting alternative apps . they can also be included in the apps search interface . apps can also be labelled as `` privacy preserving '' in the web store based on this metric . it is also possible that the privacy indicator is only shown when the vendor has existing access to the user s data . this might serve to reduce the habituation effect and the information overload . the best choice among these deployment scenarios needs further investigation . furthermore , it is important to note that , although our experimental interface mentions the collaborators name in the explanation under the progress bar , this does not have to be the case in actual deployments . we hypothesize that removing the name will not have a significant impact on the results as it was not highlighted in the interface . this allows the csp to relay such information to the users without exposing sensitive data about particular collaborators . the csp can resort to more sophisticated anonymization methodologies , such as showing a non - exact percentage that can be mapped to multiple collaborators . exploring the impact of these techniques is left for a future work . moreover , we note that this anonymization might not be needed at all in the enterprise settings , where apps installed by team members are supposed to be visible for the administrators . as we show in section [ sec : teams ] , a significant reduction in privacy loss can be achieved without even accounting for decisions by users external to the team . finally , the privacy indicator in our study has addressed two granularity levels : full and per - file access . however , the same indicator can be extrapolated to the case of per - type access . for example , the interface can say : `` the app s company already has access to 70% of your _ photos _ '' ( instead of _ files _ ) . in the previous section , we showed the significant change that our privacy indicator can effect through encouraging users to make history - based decisions . we will tackle the next research question , where we investigate the impact of adopting such privacy indicators on the privacy risk in realistic scenarios with large user networks . as we are not in the position of the csp to study an actual implementation of the _ hb _ insights interface over time , we will perform a simulation of potential users installation behavior . we will base this on both the crowdsourced decision model inferred from the user study and on new collaboration networks that we construct . * collaboration networks : * for the purposes of this simulation , we constructed the following three networks : * * inflated google drive network * : we used the standard degree - driven approach for network topology generation to construct a larger google drive network based on the one in the _ privyseal dataset _ of section [ sec : privsealdataset ] @xcite . based on an input user degrees distribution from that dataset , we particularly used the _ configuration model _ as described by newman @xcite and implemented by the library networkx @xcite for inflating the graph . this model generates a random pseudograph ( a graph with parallel edges and self - loops ) by randomly assigning edges to match an input degree sequence . we removed the self - loops and parallel edges a posteriori from the generated graph . in the end , we had a collaboration graph with 18,000 users and 138,440 edges . this graph is , by construction , a connected graph , with an average node degree of 15 . * * paper collaboration network * : in an effort to have a realistic , large collaboration network without resorting to graph inflation , we relied on the microsoft academic graph , which consists of records of scientific papers along with the authors and their affiliations @xcite . we used a snapshot of 50,000 papers , and we constructed the collaboration graph based on it . we ended up with 41,000 collaborators and 199,980 edges . the graph itself is not a connected graph but is rather constructed of around 1700 connected components . the average node degree is 4 . our rationale is that this graph captures a realistic scenario of users collaborating on authoring documents , which is , in fact , an activity achieved via cloud services nowadays . hence , it is fit for showing the efficacy of our privacy indicators . * * team collaboration network * : we used the same academic graph in order to construct a network of teams . a team is defined as a frequently collaborating group of people . motivated by research around community detection @xcite , we use strongly connected components ( sccs ) in order to label teams in our graph . we ended up with 16,400 users split over 1700 teams . unlike the previous two networks where users themselves are the data subjects ( whose privacy is to be optimized ) , members of each team in this network consider their team as the data subject . * sharing and installation patterns : * in order to closely model the user characteristics in google drive , we assigned to each user in the collaboration networks a file sharing distribution and a number of apps corresponding to a user with a matching degree in the privyseal dataset . * apps : * as we wanted to perform the simulation with a much larger number of users than we had in the dataset described in section [ sec : privsealdataset ] , we also needed a larger collection of apps . given that google chrome store has only around 500 apps that are tagged by the `` works with google drive '' tag , we decided to also include all google chrome apps in the dataset ( i.e. , even those that do not have this tag ) . as far as the simulation is concerned , this step is justified since the only realistic information that we will rely on is the distribution of vendors per app . it is fair then to assume that this distribution does not differ significantly between the general category and the google drive category . hence , we augmented the privyseal dataset via apps from the google chrome store to arrive at 1000 apps . in addition to the app s installation count and vendor name , we also collected the set of `` _ _ related apps _ _ '' that the store displays for each app . this is because , in our simulation , we will assume that users have the choice to choose the app itself or one of its related apps . again , this is a fair assumption as these related apps are mostly the apps which deliver a close functionality to the app itself , and we will only rely on them to model the alternatives at each simulation step . * user decision models : * for the purpose of this simulation , we define 3 user decision models : * * fully aware model ( ): * the user always makes the decision that minimizes the privacy loss of the data subject , taking into account all previous installation decisions by her and by her collaborators . * * experimental history - based model ( ) * : the user takes decisions similar to what a random user of the _ hb _ experimental group does . in specific , we model those users as taking a history - based decision with probability @xmath39 and making a random app choice with probability @xmath40 . we set @xmath39 based on the number of users who mentioned the app existing access in _ writing _ as a reason for their choice in each module of section [ sec : study ] . based on module 1 s users responses , we set @xmath41 when the user encounters a vendor she previously authorized . based on module 2 , we set @xmath42 whenever the user is presented with one vendor previously authorized by a single collaborator . based on module 3 , we set @xmath43 for the cases where the user is presented with multiple vendors previously authorized by her collaborators . in all of these cases , the user will select the vendor with the minimal resulting _ aggregate _ @xmath18 with probability @xmath39 . * * experimental baseline model ( ): * the user takes decisions similar to what a random user of the _ bl _ experimental group does . as users in practice are rarely informed of what their friends have installed before , we do not integrate this knowledge into the model . hence , we only account for the case of module 1 , where the user s previous decisions are concerned . based on the fraction of users who mentioned the app s existing access as a motivation for their choice , we set the probability of taking history - based decision in this model as @xmath44 . in the special case of the team collaboration network , users who take history - based decisions account for their own decisions and the decisions of their team members only . we do not consider that users account for decisions taken by members of other teams . this is to demonstrate the potential of the privacy indicators under strict conditions . initialize @xmath18 value to 0 for each user select a random user @xmath45 based on user s app installation frequency[lst : line : userselect ] select a random new app @xmath35 based on app s installation count[lst : line : appselect ] @xmath46 @xmath47 set of vendors of apps in @xmath48 @xmath49 a random rational number in the range [ 0,1 ] select a random vendor @xmath50 install the app @xmath51 in @xmath48 from vendor @xmath52 install app @xmath35 compute @xmath53 for each vendor @xmath1 in @xmath54 at this time step select the vendor @xmath50 with highest @xmath53 at this time step install the app @xmath51 in @xmath48 from vendor @xmath52 install app @xmath35 install app @xmath35 update _ aggregate @xmath18 _ for @xmath0 update the average _ aggregate _ @xmath8 over all users we now move to the description of the simulation itself , which is detailed in algorithm [ alg : simulation ] . we had three simulation groups , named after the three decision models : group , group , and the group . the simulation was run until the average number of apps installed across by users reached 30 apps . on a high level , at each simulation step , the following actions are performed : * a user is selected from the collaboration network via a weighted random sampling based on the assigned app installation frequencies ( line [ lst : line : userselect ] ) . this accounts for the diversity of users installation frequencies . an app @xmath35 is selected from the simulation apps dataset via a weighted random sampling based on the actual app installations count in google chrome store ( line [ lst : line : appselect ] ) . that way , popular apps are installed more frequently ( as is the case in practice ) . * a user decision is simulated . the user is assumed to be choosing the app @xmath35 or one of its related apps . this choice is made depending on the user s decision model , as explained previously . * finally , the average _ aggregate _ @xmath8 is computed based on all users _ aggregate _ @xmath18 . [ fig : simulation_groups_inf ] [ fig : ratio_inf ] [ fig : events_inf ] + [ fig : simulation_groups_auth ] [ fig : ratio_auth ] [ fig : events_auth ] + [ fig : simulation_groups_team ] [ fig : ratio_team ] [ fig : events_team ] to demonstrate the simulation results , we show three types of figures per collaboration network . on a high level , in figures [ fig : simulation_groups_inf ] , [ fig : simulation_groups_auth ] , and [ fig : simulation_groups_team ] , we show how the privacy loss ( quantified using the average aggregate-@xmath8 ) in each group evolves as users install more apps . in figures [ fig : ratio_inf ] , [ fig : ratio_auth ] , and [ fig : ratio_team ] , we show ratios of the privacy loss in the two experimental groups and with respect to the baseline group . finally , figures [ fig : events_inf ] and [ fig : events_auth ] , and [ fig : events_team ] show the actual events contributing to the privacy loss growth , where we can specifically check the fraction of apps coming from new vendors , those coming from vendors previously authorized by the user , and those from vendors previously authorized by collaborators . based on these metrics we start by analyzing the results for the individuals networks , where we observe the following : * curtailed growth of privacy loss : * from figures [ fig : simulation_groups_inf ] and [ fig : simulation_groups_auth ] , we notice that the growth of the privacy loss is visibly curtailed in the cases of and groups compared to the baseline group . this significant divergence demonstrates the efficacy of our _ hb_privacy indicators . * impact of the network effect : * looking into the ratios in figures [ fig : ratio_inf ] and [ fig : ratio_auth ] , we see that the privacy loss in the group has dropped by 41% in the inflated network and by 28% in the authors - based network ( both with respect to the baseline ) . in the group , where users always optimize their privacy , the privacy loss has dropped by 70% in the inflated network and by 40% in the authors - based network . this higher impact in the case of the inflated network is due to the fact that it is a connected graph , unlike the authors - based network , which is composed of smaller connected components . nevertheless , we can state that , although our privacy indicators have a larger effect on highly connected networks , they are still significantly effective in less connected networks , like the authors - based dataset . * importance of accounting for collaborators decisions : * to dive further into events that lead to the observed privacy loss patterns , we look into figures [ fig : events_inf ] and [ fig : events_auth ] . first , we observe that users in the group are mainly installing new apps from vendors that had no previous access to their data . this is reflected in the almost linear increase of privacy loss in figures [ fig : simulation_groups_inf ] and [ fig : simulation_groups_auth ] . second , we observe that , in the case of the inflated network , users have been frequently installing apps from vendors with existing access through their collaborators . in fact , as apparent in figure [ fig : events_inf ] , this event outnumbers the event of installing from a new vendor . third , the number of installations from collaborators vendors is also significant in the case of the authors - based dataset . while it does not outnumber the installations from new vendors ( due to the low - graph connectivity ) , this is still enough to lead to 28% and 40% decrease in the privacy loss in the and the groups respectively . finally , we note that , although the users are more frequently encountering vendors authorized by their collaborators than by themselves , the latter event is still significantly impacting the results . this is because users still incur an incremental privacy loss with vendors authorized by their collaborators while this loss is zero with vendors they have previously authorized . accordingly , the obtained optimizations are a result of users accounting for their own and for others decisions . we now discuss the results for the case of the collaboration network where users work in teams and aim to protect the privacy of the team s data . we observe the following , based on figure [ fig : sim_team ] : * inherent usage of similar apps : * from figure [ fig : events_team ] , it is clear that the dominant event is that of users installing apps which have been authorized by other team members before . this is even in the case of the baseline group ( ) , which was not the case in the individuals networks . we justify that by the fact that we selected apps at each simulation step to match their realistic installation frequencies . in practice , apps installation counts follow a long - tail distribution , and users tend to mostly install a limited set of apps . that is why team members will naturally tend to install a set of similar apps . * curtailed growth of privacy loss : * still , we observe that the trend of slower growth of privacy loss also applies in the case of teams ( figure [ fig : simulation_groups_team ] ) . as we also observe in figure [ fig : ratio_team ] , the privacy loss has decreased by 23% for the group and by 45% for the group , both with respect to the baseline group . this implies that there is an ample room for privacy optimization in teams too . * effect due to internal collaborators : * we finally observe that the privacy loss decrease was achieved via decisions taken by each team s members independently , without relying on other teams decisions . this highlights the fact that _ hb_privacy indicators can still be effective even when users do not account for others decisions . obviously , taking the external members decisions into account can lead to further optimizations . + in sum , our simulations provide further evidence of the efficacy of using history - based privacy indicators in a large network of collaborators . it is worth noting too that , although users in our study were following the decision model , we believe that , in an actual deployment of such indicators , the model will move closer to the model . this is because users are more protective when their personal data is at risk than when they are put in a role playing scenario about fictitious data . moreover , users in our study were exposed to this indicator for the first time . when users are educated more about this feature , they might be more likely to take advantage of it . the problem of interdependent privacy has been tackled before in the context of social apps . the main approaches were high - level game - theoretic or economic modeling . in @xcite , the authors introduced the concept of interdependent privacy and modeled its impact via a game theoretic , ( 2-player , 1-app ) model . the work by pu and grossklags @xcite presented a more elaborate economic model that additionally accounts for the interplay among various social network parameters . they showed that app rankings do not accurately reflect the level of interdependent privacy harm the app can cause and that even rational users who consider their friends well - being might adopt apps with invasive privacy practices . evidently , these results do not apply in the cloud apps case , where _ all _ apps have the potential to inflict interdependent privacy harm . a later work by pu and grossklags @xcite used a conjoint study approach to quantify the monetary value which individuals associate with their friends personal data . they found that individuals place a significantly higher value on their own personal information than their friends personal information . this further supports our assumption of self - interested users in this work . the same authors also built on a user survey in @xcite to assess the factors affecting users own privacy concerns as well as friends privacy concerns in the context of social app adoption . in particular , they found evidence of negative association between past privacy invasion experiences and the trust in 3rd party apps handling of their own data . they also found partial support for a positive effect of privacy knowledge on concerns for users own privacy and their friends privacy . in this work , we are focused on quantifying the interdependence of privacy in the context of cloud apps before addressing it from a usable privacy perspective , thus bridging the gap between the theoretical studies and the end - user needs . our previous work @xcite was the first to study the privacy of 3rd party cloud apps and to expose that almost two thirds of those apps are over - privileged . in that work , we introduced a novel privacy indicator for deterring users from installing over - privileged apps by showing them far - reaching insights that apps can needlessly infer from their data ( e.g. , top topics , faces , or locations of interest ) . in the context of android apps , kelly et al . , showed that , by adding a set of privacy facts about an app , users will be more likely to choose apps with fewer permissions @xcite . harbach et al . , tackled the same problem but presented users with random examples from their data ( e.g. , pictures , contacts , etc . ) @xcite . almuhimedi et al . showed the effectiveness of privacy nudges , which regularly alert users about sensitive data collected by their apps , in encouraging users to review and adjust their permission @xcite . all these works , however , tackle the problem of over - privileged apps and try to lead the user into either avoiding them or adjusting their permissions whenever possible . our current work helps users improve their privacy by reducing the vendors with access to their data , even if the functionality delivered by the vendor abides by the least - privilege principle . hence , it complements these approaches and can be deployed alongside any of them . the findings in this work are the first to concretely delineate the various aspects of interdependent privacy in 3pc apps . one of the major outcomes is that a user s collaborators can be much more detrimental to her privacy than her own decisions . consequently , accounting for collaborators decisions should be a key component of future privacy indicators in 3rd party cloud apps . we have shown the impact of history - based insights as a privacy enhancing technology in this context , especially that , based on our user study , users are less likely to account for previous decisions on their own . our privacy indicators would optimally be implemented by the csps themselves as they control the authorization interface and the application stores . the indicators can also be realized by third party privacy providers with access to users data . our approach can also be easily mapped to other ecosystems . in the mobile apps scenario , it can enable users to reduce the number of vendors with access to her contacts . it can also be extended to the case where the goal is protection against 4th parties ( e.g. , ad providers and data brokers ) . there , the user can account for data previously held by a 4th party with which the app vendor cooperates . finally , due to their usability and effectiveness , we envision history - based insights as an important technique within the movement from static privacy indicators towards dynamic privacy assistants that lead users to data - driven privacy decisions . we would like to thank deniz taneli and nicolas hubacher for their help in exploratory work that led to this paper . we also thank rameez rahman for the helpful discussions and the anonymous reviewers for their valuable feedback . the research leading to these results has received funding from the eu in the context of the project _ cloudspaces _ : open service platform for the next generation of personal clouds ( fp7 - 317555 ) . in this section , we complement section [ sec : hbdesign ] by providing a proof the optimal user strategy for minimizing the privacy risk , given our assumptions . we follow the notation introduced in section [ sec : models ] . let us consider that each 3pc app vendor has a probability @xmath33 of exposing users data . as we do not assume that users are provided with a per - vendor risk estimation utility , we set this probability to be the same for all vendors . in general , at a time @xmath6 , a user @xmath0 would have exposed her data to a set @xmath9 of vendors , such that each vendor @xmath1 has access to a fraction @xmath55 of the files . without loss of generality , we will consider henceforth that the user has an all - files privacy goal ( cf . section [ sec : usermodel ] ) . however , the same reasoning applies in the case of a per - type privacy goal . in that case , we simply replace `` files '' by `` files of a specific type '' ( e.g. photos , documents ) . we will also be assuming that the users themselves are the data subjects ( i.e. , we consider individual - level subjects ) . for a vendor @xmath1 , we quantify the user s privacy risk magnitude as @xmath56 , i.e. , the fraction of user files possessed by the vendor multiplied by the probability that the vendor exposes the user s files . this vendor could have obtained access due to app installations by the user herself or by her collaborators . a user s privacy risk magnitude at time @xmath6 can thus be defined as the sum of the risk magnitude across vendors in @xmath9 : @xmath57 . when a user installs an app from a vendor @xmath52 at time @xmath58 , the vendor gets access to the whole set of user s files . hence , the risk magnitude is increased by @xmath59 . given that @xmath33 is constant , the risk magnitude can be minimized by choosing @xmath52 , such that @xmath60 ( which can also be written as @xmath61 ) . hence , the optimal , greedy strategy to minimize the risk is to select the vendor that already has the largest fraction of user files , thus minimizing @xmath59 . we call this strategy : `` history - based decisions '' .
there is considerable interest in hydrodynamic instabilities in dead zones of protoplanetary disks as a mechanism for driving angular momentum transport and as a source of particle - trapping vortices to mix chondrules and incubate planetesimal formation . we present simulations with a pseudo - spectral anelastic code and with the compressible code _ athena _ , showing that _ stably stratified _ flows in a shearing , rotating box are violently unstable and produce space - filling , sustained turbulence dominated by large vortices with rossby numbers of order . this _ zombie vortex instability _ ( zvi ) is observed in both codes and is triggered by kolmogorov turbulence with mach numbers less than . it is a common view that if a given constant density flow is stable , then stable vertical stratification should make the flow even more stable . yet , we show that sufficient vertical stratification can be unstable to zvi . zvi is robust and requires no special tuning of boundary conditions , or initial radial entropy or vortensity gradients ( though we have studied zvi only in the limit of infinite cooling time ) . the resolution of this paradox is that stable stratification allows for a new avenue to instability : baroclinic critical layers . zvi has not been seen in previous studies of flows in rotating , shearing boxes because those calculations frequently lacked vertical density stratification and/or sufficient numerical resolution . although we do not expect appreciable angular momentum transport from zvi in the small domains in this study , we hypothesize that zvi in larger domains with compressible equations may lead to angular transport via spiral density waves .
1410.8143
after 40 years of intense theoretical and computational research , the answer has been a qualified and unsatisfying `` maybe . '' it has long been known that molecular viscous diffusion is wholly inadequate to drive accretion because the transport timescale is several orders of magnitude longer than the age of the universe : @xmath3 gyr for conditions typical at @xmath4 au in a protoplanetary disk ( ppd ) with kinematic viscosity @xmath5 , thermal velocity @xmath6 km / s , molecular mean free path @xmath7 1 cm . @xcite proposed that keplerian differential rotation may be susceptible to hydrodynamic and/or magnetohydrodynamic instabilities that would drive turbulent diffusion , for which they introduced an `` eddy '' viscosity : @xmath8 ( the idea being that turbulent `` blobs '' would move at subsonic speeds over distances of order the scale height , and the parameter @xmath9 would subsume all our ignorance over the details of the transport efficiency ) . because the specific angular momentum increases with radius in keplerian differential rotation , it is often stated that keplerian shear is stable to purely hydrodynamic instabilities . however , this centrifugal stability criterion @xcite is strictly true only for axisymmetric perturbations in an unstratified , inviscid flow , and therefore is of limited relevance for _ stratified _ astrophysical disks . so , what is the nature of hydrodynamic instabilities that may lead to enhanced transport ? historically , by analogy with poiseuille ( pressure - driven ) and couette ( viscous - driven ) flows in both planar and cylindrical geometries , the go - to mechanism has been subcritical finite - amplitude instabilities that extract energy from the shear at very high reynolds numbers @xcite . this has been , and remains , highly controversial . @xcite and @xcite have argued , with theory and computational analyses , that the coriolis force stabilizes keplerian shear and shuts off any way ( in the absence of magnetic fields ) for perturbations to extract energy from the differential rotation in order to power sustained turbulence . others have argued that even the best to - date numerical simulations lack sufficient resolution to capture nonlinear triggering of instabilities at very high reynolds numbers @xcite . a related mechanism to generate shear turbulence starts with the transient amplification of a special class of initial conditions , which may trigger nonlinear feedback and continued growth if the amplitude of the transients peak above some critical threshold @xcite . however , none of these by - pass to turbulence " approaches regenerate the initial perturbations required to sustain turbulence after the initial burst of turbulence has decayed . with conflicting results in both theory and numerical work , one may hope that laboratory experiments may provide an independent test . unfortunately , two recent taylor - couette flow experiments ( water rotating between two concentric cylinders ) have yielded opposite results : experiments by @xcite claim a positive measurements of turbulence and outward transport of angular momentum ; whereas , a similar set of experiments by @xcite claim no such evidence of turbulence nor angular momentum transport . the measurements are difficult and sensitive to the ability of the experimentalists to control secondary flows from their apparatus ( e.g. , ekman pumping ) . direct numerical simulations of taylor - couette flows with laboratory boundary conditions seem to confirm that the axial boundaries are driving the instabilities and enhanced transport @xcite . @xcite and @xcite showed that magnetic fields could destabilize otherwise stable rotational flows . balbus and hawley were the first to apply this magnetorotational instability ( mri ) to keplerian shear flows and demonstrate how the resulting turbulence efficiently transported angular momentum outward @xcite . however , there exist relatively dense , cool and nearly neutral regions in ppds ( approx . 1 - 10 au ) that likely lack sufficient coupling between matter and magnetic fields @xcite , except perhaps in thin surface layers that have been ionized by cosmic rays or protostellar x - rays @xcite . the magnetically - decoupled regions are known as `` dead zones '' and there is continued research into whether waves and turbulence from the magnetically - coupled mri - active regions can propagate into the dead zones of ppds and drive motions that result in angular momentum transport or disrupt the settling of dust @xcite . in regions of a laminar near - keplerian disk that are strongly coupled to magnetic fields , there is no question that of all the known relevant linear instabilities , mri is the fastest to grow to large amplitude . however , the presence of dead zones in ppds still motivates researchers to investigate other mechanisms for instability and transport , especially because the dead zone seems to be coincident with planet - forming regions within disks @xcite . a large class of potential instabilities involve global properties of the disk , specifically radial gradients of thermodynamic quantities . extrema in the radial profile of potential vorticity ( a.k.a . vortensity ) is unstable to rossby waves which eventually roll - up into vortices @xcite . radial entropy gradients that are stable according to the solberg - hiland criterion can still be unstable to a `` subcritical baroclinic instability '' ( sbi ) that is most effective when the thermal cooling time is of order an orbital period @xcite . more recent work on the sbi by @xcite has shown that the instability is actually linear ( i.e. , does not require large amplitude perturbations ) and more akin to a convective overstability rather than a traditional baroclinic instability . still others have investigated vertical shear instabilities ( vsi ) related to the goldreich - shubert - fricke instability in rotating stars @xcite . while zvi needs vertical stable stratification ( e.g. , entropy increasing in direction opposite of gravity and sufficiently long thermal relaxation times ) the vsi needs vertically neutral stratification ( e.g. , no vertical entropy gradient , or sufficiently short thermal relaxation times ) . thus depending on the thermodynamic structure of the disk and its optical properties , one or the other might locally dominate . of course , by direct observation , we know that _ some _ mechanisms exist within a ppd that leads to the transport of angular momentum in the way that is needed to complete star formation and that allows dust to accumulate and agglomerate into kilometer - sized planetesimals . however , the elucidation of that mechanism or mechanisms can not be sloughed off as an unimportant `` detail . '' a more realistic treatment of turbulence is essential to quantify the spatial and temporal distribution of dust in a ppd and therefore affects our modeling of the infrared emission from dust ( relevant for alma & jwst observations ) , as well as an initial - mass function ( imf ) for planets . in ( * hereafter bm05 ) , we investigated the stability of three - dimensional vortices in the midplanes of protoplanetary disks . we created a model for a near - equilibrium vortex , which we initialized by hand in our spectral anelastic code that had specially tailored algorithms to handle shear , rotation and stratification @xcite . the vortex suffered a linear instability which resulted in its destruction near the midplane , but we unexpectedly discovered that the stratified regions away from the midplane naturally filled with robust vortices . at the time , we speculated that the initial midplane vortex excited internal gravity waves , which propagated away from the midplane and deposited their energy in stratified regions where the shear rolled vorticity perturbations into new vortices . however , our original explanation was not entirely complete nor satisfying . * hereafter mpjh13 ) correctly diagnosed the true mechanism for the formation of these vortices and in the process identified a new instability , which we now call the zombie vortex instability " or zvi . in order to get at the essential nature of the phenomenon , mpjh13 stripped out complicating features of a protoplanetary disk ( e.g. , spatially varying gravity and brunt - visl frequency ) and numerically investigated simple couette flow with constant gravity and constant brunt - visl frequency in the limit of the boussinesq approximation . in this far simpler system , mpjh13 found that a small perturbing vortex could trigger an instability in a rapidly - rotating , strongly stratified flow , yielding vigorous , space - filling vortices , vortex layers and turbulence . figure [ f : zombie_cartoon ] is a cartoon picture summary of the mechanism outlined in mpjh13 . each panel shows a small patch of a ppd in frame rotating with the gas ; upward is the forward azimuthal direction , rightward is increasing radius from the protostar . in the rotating frame , the background flow appears as an anticyclonic shear ( as indicated with the arrows ) . color illustrates vertical vorticity ( aligned with the rotation axis of the disk ) , with red indicating cyclonic ( out of the page ) vorticity and blue indicating anticyclonic ( into the page ) vorticity . the domain is initially filled with kolmogorov turbulence on small scales ( panel a ) . at low rossby number ( when the coriolis effect is dominant over advection ) , the turbulence can undergo a `` reverse cascade '' in which energy transfers to larger scales as small regions of vorticity merge with other regions of vorticity with the same sense of rotation . regions of vorticity that have the same sense of rotation as the background shear tend to roll - up into compact vortices , whereas vorticity patches with the opposite sense of rotation as the background shear tend to get stretched out into thin layers @xcite . after a period of reverse cascades and mergers , a region of the disk may only be populated with a few isolated anticyclones that move with the shear and interspersed with cyclonic vortex layers ( panel b ) . in our simplified cartoon , we consider a small region with only one remaining anticyclone . this vortex acts as a source of perturbations and waves . the crucial new insight in mpjh13 was the recognition of `` baroclinic critical layers '' as being sites that are receptive to perturbations . critical layers are special locations in a shear flow where the coefficients of the highest derivatives of the linearized equations vanish , indicating that the neutrally - stable eigenmodes are singular there @xcite . because our focus in this paper is to demonstrate the existence of zvi , we will defer the mathematics of critical layers to a future paper . for our purposes here , we note that there exist multiple critical layers on either side of the source vortex . let @xmath10 be local cartesianized radial , azimuthal , and vertical directions within a small box rotating with the gas . the cross - stream distance @xmath11 between the source and the @xmath12th critical layer is given by ( mpjh13 ) : @xmath13 where @xmath14 is the brunt - visl frequency , @xmath15 is the local shear rate , @xmath16 is a spatial wavenumber associated with a fourier component of the source vortex , @xmath17 is the domain size in the azimuthal direction , and @xmath12 is some non - zero integer . the critical layer closest to the source , without intersecting the source itself , has the largest integer @xmath18 . at first glance , it may appear that the closest critical layer depends on the arbitrary choice of the domain size in the azimuthal direction , but in numerical experiments , the physical location of the first excited critical layer is always just outside the perturber , and the value of the integer @xmath18 will scale with @xmath17 . critical layers for smaller integers @xmath19 are also excited . the critical later farthest from the source generally has @xmath20 and is at a distance @xmath21 . however , if a boundary of the domain in the cross - stream direction is too close to the source , or if there are physics that prohibit vortex coherence over distances as large @xmath22 , then the excited critical layer farthest from the perturbing source will have @xmath23 . in the calculations presented here the boundaries are chosen so that the farthest excited critical layer is 1 , and the equations of motion contain no physics that inhibit large vortices from forming , so the farthest distance is @xmath22 . however , as we discuss [ subsec : mach ] , the physics of the compressible gas in ppds will generally limit the size of vortices such that the farthest critical layers do not have @xmath20 . while the singular eigenmodes are neutrally stable ( do not grow nor decay ) to small perturbations , they are susceptible to being forced and excited by the nearby anticyclone . it is important to note that although the cartoon makes it appear that the excited critical layers are at the same height as the perturber , they are in fact located somewhat above and below the instability is inherently 3d . a dipolar vortex layer ( two juxtaposed oppositely - signed layers of vorticity ) develops at the location of each critical layers ( panel c ) . vortex layers that have the same sense of rotation as the background shear are linearly unstable ( panel d ) and roll - up into compact vortices @xcite , and so new anticyclones will eventually be spawned at the locations of the critical layers with @xmath12 vortices at the @xmath24 critical layer ( panel e ) . the newly spawned vortices from the different critical layers tend to merge into one larger vortex with that vortex located at the critical layer location farthest from the source ( generally with @xmath20 at a distance @xmath22 ) . the newly created large vortex then excites its neighboring critical layers ( panel f ) and the process continues . in numerical simulations , one can observe perturbations spawning new vortices farther and farther from the original perturbation vortex . we named the instability the zombie vortex instability not only because it occurs in the dead zones of ppd , but also because of the way one zombie vortex `` infects '' its neighboring critical layer , spawning new zombie vortices , which `` infect '' their neighboring critical layers , etc . in numerical simulations , one can see the entire domain rapidly fill with zombie vortices from one initial vortex . the instability is not an artifact of the numerical method as we have observed it with spectral codes and finite - volume codes ( e.g. , _ athena _ ) , with fully compressible , anelastic and boussinesq treatments of the continuity condition , with and without the shearing box , and with either hyperviscosity or real molecular viscosity . one may ask , if it is so robust , how was it missed in previous numerical calculations ? as will be elucidated in this series of papers , zvi requires vertical stratification , high resolution to resolve the narrow critical layers , a broad spectrum of perturbations ( i.e. , kolmogorov , but not gaussian - peaked ) , and enough simulation time to allow the critical layers to amplify perturbations . in this paper , we consider zvi in the astrophysical context , rather than the boussinesq approximation appropriate to laboratory couette flow as we did in mpjh13 . in [ sec : equations ] , we review the hydrodynamic equations of motion and outline the parameter space of interest for zvi . because the primary tuning parameter for zvi is the brunt - visl frequency @xmath25 , we shall replace linear gravity with a spatially constant value so that @xmath14 itself is also constant ( for vertically isothermal backgrounds ) . our previous numerical experiments ( including those in bm05 ) show that zvi is present even when @xmath25 is spatially varying . however , those experiments also convinced us that dealing with a constant @xmath14 makes it much easier to interpret the results and provide a pedagogical explanation for them . in [ sec : evidence ] , we shall study zvi and its development into turbulence . we shall also investigate triggers of zvi other than an initial vortex ( which was used as the initial condition in bm05 and mpjh13 ) , namely initial isotropic , homogeneous kolmogorov - like noise . we shall characterize properties of zombie turbulence : its magnitude , its space - filling nature , and its inhomogeneity and anisotropy due to persistent , but turbulent , large anticyclonic vortices and undulating cyclonic vortex layers . most of our numerical experiments presented here use the anelastic approximation ( which filters acoustic waves ) with our pseudo - spectral code that has specially tailored algorithms to handle shear , rotation and stratification @xcite . we have also done simulations with the fully compressible , godunov finite - volume code _ @xcite , and show that qualitatively and quantitatively we observe similar results . however , our timing tests show that _ athena _ is approximately 100 times more computationally expensive for these numerical experiments when using the same number of grid points as spectral modes in the pseudo - spectral code . in our discussion in [ sec : discussion ] , we explain why zvi was not observed in previous studies by others , we review zvi in the context of other hydrodynamic and magnetohydrodynamic instabilities in ppds , and we consider the role of zvi in angular momentum transport and planetesimal formation . in a future paper , we will confront zvi on a more mathematical and theoretical level and address thresholds required to trigger zvi . in particular , we will show why extraordinarily weak kolmogorov - like noise is so efficient in exciting zvi . consider a ppd in which the steady flow is only in the azimuthal direction with angular velocity @xmath26 where @xmath27 is the cylindrical radius from the protostar . a value of @xmath28 corresponds to keplerian rotation . now consider a three - dimensional box of gas centered at radius @xmath29 from the protostar that co - rotates with the gas with angular rate @xmath30 . the box is sufficiently small that we can ignore curvature of the disk and map the cylindrical coordinates onto local cartesian coordinates : @xmath31 @xcite . in what follows , symbols with `` hats '' will denote unit vectors : @xmath32 points in the local outward radial direction , @xmath33 points in the local azimuthal direction , and @xmath34 points in the local vertical direction . euler s equations ( continuity , momentum and energy with no viscous , radiative or diffusive terms ) for the gas in the rotating frame are : [ bhs35 ] @xmath35 where @xmath36 is the gas velocity with respect to the center of the box that co - rotates with the ppd . @xmath37 , @xmath38 , and @xmath39 are , respectively , the gas pressure , density and specific internal ( thermal ) energy . the vertical acceleration of gravity @xmath40 does not include any contribution from the self - gravity of the gas itself . the term @xmath41 is the coriolis force , and the term @xmath42 is the tidal acceleration that arises from the difference between the inward central force that causes circular motion @xmath43 and the outward centrifugal force @xmath44 due to being in a rotating frame . the term on the right - hand side of the thermal energy equation is the rate at which pressure forces do work due to volume changes of fluid parcels . to close the system of equations , we assume an ideal gas for the equation of state : @xmath45 where @xmath46 is the gas temperature , @xmath47 is the gas constant , and @xmath48 is the adiabatic index , or ratio of specific heats at constant pressure and constant volume . throughout this paper , we set ratio of specific heats @xmath49 . a time - independent equilibrium ( denoted with overbars ) to the above set of equations is : [ hydro ] @xmath50 because there is no radiative or diffusive effects in equations ( [ bhs35 ] ) , there is a degeneracy of allowable equilibrium thermodynamic profiles , so the equilibrium temperature @xmath51 corresponding to the equilibrium velocity @xmath52 is not unique . in fact , one thermodynamic quantity , @xmath53 , @xmath51 , @xmath54 , or @xmath55 can be arbitrarily specified , and the others will follow . in this work , we shall assume @xmath56 , so : @xmath57 , \label{hydrostatic1}\ ] ] where @xmath58 is the midplane density and @xmath59 is the midplane of the disk . the anelastic approximation has been extensively used in the study of deep , subsonic convection in planetary atmospheres @xcite and stars @xcite . we have previously used the anelastic approximation to study three - dimensional vortices in ppds @xcite and the kelvin - helmholtz instability of settled dust layers in ppds @xcite . the basic idea is that there may be large variations in the background pressure and density in hydrostatic equilibrium , but that at any height in the atmosphere , the fluctuations of the pressure and density are small compared to the background values at that height . mathematically , all flow variables are expanded in powers of the mach number ( presumed to be small ) , and only lowest order terms are retained in the equations of motion . in general , the anelastic approximation results in a valid description of the flow when : [ e : anelastic_validity ] @xmath60 where @xmath61 is the local sound speed . diagnostically , it is trivial to check every step during a numerical calculation whether the conditions ( [ e : anelastic_validity ] ) are satisfied . there are a number of variations of the anelastic approximation , all equivalent to the same order of the mach number , but varying in whether they retain or drop certain smaller terms or in exactly how they linearize the equation of state , often with an eye toward conserving some quantity such as energy ; see references above as well as @xcite . for the special case of uniform temperature equilibrium background , the most straight - forward application of the anelastic approximation yields energy - conserving equations , so we shall defer going into more detail until future papers in which we consider different background temperature profiles . the euler equations with the anelastic approximation for a vertically uniform background temperature are : [ e : anelastic_eqns ] @xmath62 & = & 0 , \label{anelastic1}\\ { \frac{\partial { \bf v}}{\partial t } } & = & -({\bf v}\cdot { \mbox{\boldmath $ \nabla$}}){\bf v}-2\omega_0 \ , \hat{\bf z } \times { \bf v } + 2q \omega_0 ^ 2 \ , x \ , \hat{\bf x}\nonumber\\ & { } & -{\mbox{\boldmath $ \nabla$}}\left[\frac{{p - \bar{p}(z)}}{{\bar{\rho}(z)}}\right ] + \left[{\frac{t - t_0}{t_0}}\right ] g(z ) \ , \hat{\bf z } , \label{e : anelastic2}\\ { \frac{\partial t}{\partial t } } & = & - ( { \bf v } \cdot { \mbox{\boldmath $ \nabla$ } } ) t - \left[\frac{n^2(z)}{g(z)}\right]tv_z , \label{e : anelastic3}\\ \left[{\frac{p - \bar{p}(z)}{\bar{p}(z)}}\right ] & = & \left[{\frac{\rho - \bar{\rho}(z)}{\bar{\rho}(z)}}\right ] + \left[{\frac{t - t_0}{t_0}}\right ] , \label{e : anelastic4}\end{aligned}\ ] ] where @xmath25 is the brunt - visl frequency , a measure of the convective stability of the atmosphere . in general , @xmath63 , where @xmath64 is the background entropy . here , we focus on vertical entropy gradients , so : @xmath65 one of the main dynamical consequences of this approximation is that the total density is replaced by the time - independent mean density in the mass continuity equation , which has the effect of filtering acoustic waves , but allowing slower wave phenomena such as internal gravity waves , rossby waves , etc . similarly to the fully compressible equations , the dissipationless anelastic equations contain a degeneracy , so any one thermodynamic quantity , @xmath53 , @xmath51 , @xmath54 , or @xmath55 can be arbitrarily specified . as before , we arbitrarily set the background equilibrium temperature to be constant , @xmath66 . we note that the isothermal equilibrium solutions to the fully compressible equations are the same isothermal equilibrium solutions to the anelastic equations . in a ppd , the vertical component of the protostellar gravity varies linearly with distance from the midplane , which yields gaussian profiles of pressure and density for isothermal vertical profiles , and a linear profile for the brunt - visl frequency ( see bm05 ) : [ linear_g ] @xmath67 where @xmath68 is the isothermal sound speed . for thin keplerian disks that have uniform temperature in the vertical direction . but of course , actual sound waves travel at the adiabatic sound speed even if the background is isothermal . ] as we shall demonstrate , the essential tuning parameter for the zvi is the brunt - visl frequency , so for this paper , rather than a vertical gravity that is linear in @xmath69 , we use a constant vertical gravity , which yields exponential profiles of pressure and density for isothermal vertical profiles , and a constant brunt - visl frequency @xmath70 : [ constant_g ] @xmath71(g_0/h_0)}.\end{aligned}\ ] ] in bm05 , we previously showed that zvi is present even when @xmath25 is spatially varying . in this paper , by focusing on constant @xmath70 , we will be able to probe what thresholds of stable stratification are required to trigger zvi . however , this restriction prevents us from exploring how gravity waves and turbulence generated by zvi will propagate into the nearly unstratified midplane ; we defer that important issue to future work . all of the simulations in this paper are periodic in the azimuthal or stream - wise direction @xmath72 , and use shearing box boundary conditions in the radial or cross - stream direction @xmath73 @xcite . in the pseudo - spectral calculations , we have performed computations in which the top and bottom boundary conditions are treated by ( _ a _ ) mapping the boundaries to @xmath74 ( as in bm05 ) , ( _ b _ ) setting the vertical velocity to zero ( i.e. , _ rigid _ boundary conditions ) , or ( _ c _ ) forcing all variables to be spatially periodic in @xmath75 . when choice ( _ c _ ) is used , it is necessary to include a small artificial damping layer adjacent to the top and bottom boundaries to smooth out the inherent non - periodicity of @xmath54 and to prevent spurious reflections of inertio - gravity waves and other unphysical effects . all three of these boundary conditions produce quantitatively similar solutions . however , option ( _ c _ ) leads to a triply - periodic and computationally fast code , and thus was used for exploring parameter space for almost all the pseudo - spectral simulations reported in this paper . the pseudo - spectral domain size was @xmath76 and was resolved with @xmath77 spectral modes . the algorithms used in the pseudo - spectral calculations have no inherent dissipations , so an explicit , weak hyperviscosity is required . _ athena _ is a godunov finite - volume code that can solve the fully compressible hydrodynamic equations in a shearing box @xcite . the _ athena _ simulations used a computational domain of size @xmath76 resolved with @xmath77 grid points . the vertical velocity was set to zero at the top and bottom boundaries . note that the _ athena _ code has no explicit dissipation , but its finite - volume algorithms are inherently dissipative . we found that lower resolution simulations with _ ( e.g. , @xmath78 grid points ) did not develop zvi . there are three ( reciprocal ) timescales of interest : the rotation rate @xmath79 , the shear rate @xmath80 , and the brunt - visl frequency @xmath70 . we consider only keplerian shear , @xmath28 ( with the exception of the calculations reported in fig . [ fig : balbus ] ) . we define the dimensionless ratio @xmath81 . for a ppd with linear vertical gravity and no vertical temperature gradient , @xmath82 . in bm05 , we observed that vortices naturally formed in stratified regions above and below the midplane : @xmath83 , corresponding to @xmath84 . for constant gravity @xmath85 , one can show from eq . ( [ constant_g ] ) that : @xmath86 with constant gravity , the only independent length scale in either the fully compressible or anelastic equations is the gas pressure scale height @xmath87 . therefore , the size of the computational domain @xmath88 introduces three more dimensionless parameters . in this study , we set @xmath89 . unless otherwise stated , in this section , all times are reported in units of the local orbital period @xmath90 or local `` years '' ; all lengths in units of @xmath91 ( see eq . [ delta ] ) ; all velocities in units of @xmath92 , and all energies per unit mass in terms of @xmath93 . table [ t : parameters ] summarizes the parameters for the figures in this section . the domain size and resolution for all calculations were @xmath94 with @xmath77 spectral modes , or @xmath77 grid points for _ athena _ calculations . all flows were for keplerian profiles @xmath28 ( with the exception of figs . [ fig : balbus ] which varied @xmath95 ) . the continuity equation was either fully compressible ( listed in the bottom row with a c ) , or was anelastic ( listed with an a ) . 4 , 5 , and 7 correspond to the same simulation ; figs . 3b , 6 , 10 , and 11b correspond to the same simulation ; and figs . 3a ( with the solid curve ) , 8 , 11a correspond to the same simulation . for reference , the mach number of the flows in this section are computed with the isothermal sound speed ( rather than the adiabatic sound speed ) : @xmath96 multiplied by the dimensionless velocity . .[t : parameters ] parameters values . the domain size and resolution for all calculations were @xmath94 with @xmath77 spectral modes or grid points . all flows were for keplerian profiles @xmath28 ( with the exception of fig . [ fig : balbus ] which varied @xmath95 ) . the continuity equation was either fully compressible ( c ) or anelastic ( a ) . 4 , 5 , and 7 are for the same simulation ; figs . 3b , 6 , 10 , and 11b are for the same simulation ; and figs . 3a ( with solid curve ) , 8 , and 11a are for the same simulation . [ cols="<,^,^,^,^,^,^",options="header " , ] fig . 2a is a copy of fig . 1 from @xcite . one of the most cited pieces of evidence that ppds are stable to purely - hydrodynamic instabilities is given by fig . 1 in @xcite , which we copy below as fig . [ fig : balbus]a . the figure shows the growth or decay of kinetic energy fluctuations as a function of time for various values of @xmath95 , where @xmath95 is defined in eq . ( [ q ] ) and where the kinetic energy fluctuation is defined as @xmath97 , where @xmath98 is the steady equilibrium flow in eq . ( [ hydrostatic2 ] ) . the initial - value calculations that produced fig . [ fig : balbus]a used the fully compressible equations ( [ bhs35])([bhs35d ] ) , and were initialized with the steady equilibrium velocity in eq . ( [ hydrostatic2 ] ) . however , this flow was computed with _ no _ vertical gravity and was therefore initialized with the equilibrium that had @xmath99 , @xmath100 , and @xmath101 . the initial flow was perturbed with small - amplitude noise ( with root - mean - square or rms velocity of @xmath102 ) that had a gaussian - like energy spectrum as a function of wavenumber . the equations were solved with periodic boundary conditions in @xmath72 and @xmath75 , and with shearing box boundaries in @xmath73 . because the energy fluctuations all increase in time for @xmath103 and decrease in time for @xmath104 , fig . [ fig : balbus]a is used to support the hypothesis that these flows of ideal gases are stable ( unstable ) to all purely - hydrodynamic perturbations when the angular momentum per unit mass of the flow , @xmath105 , increases ( decreases ) in the radially outward direction . this hypothesis is consistent with rayleigh s centrifugal stability theorem ; however , it must be noted that rayleigh s theorem was proved only for the case of constant density fluids and not for stratified ideal gases @xcite , and therefore , it may not be applicable to astrophysical flows in disks . in particular , initial conditions in which the density is constant prohibit baroclinic instabilities such as zvi ( mpjh13 ) . the curve labeled with `` shr '' in fig . [ fig : balbus]a corresponds to the case of pure cartesian shear with the coriolis and tidal acceleration terms dropped from eq . ( [ bhs35b ] ) , which would be appropriate for fully compressible flow in a channel with cross - stream shear , but no rotation . in fig . [ fig : balbus]b , we reproduce the results of fig . [ fig : balbus]a with our pseudo - spectral numerical code @xcite using the anelastic equations ( [ e : anelastic_eqns ] ) with no vertical gravity . the initial noise that we added to the equilibrium velocity in all of our numerical experiments was random , homogeneous , and isotropic with a kolmogorov spectrum so that the kinetic energy per unit mass was @xmath106 , where @xmath107 is the wavenumber . our primary reason for reproducing this figure from @xcite is to show that we get the same stability results for no vertical gravity with a pseudo - spectral code ( vs. finite - volume code ) which solve the anelastic equations ( vs. the fully compressible equations ) . how does vertical stratification affect stability ? the decaying dashed curve in fig . [ fig : growenergy]a is the time evolution of the kinetic energy fluctuations for a unstratified flow similar to the one as given by the curve in fig . [ fig : balbus]b with @xmath28 and @xmath108 , but plotted for a longer time and a smaller value of initial noise . the solid curve in fig . [ fig : growenergy]a shows the evolution of kinetic energy fluctuations when vertical gravity is included ; here , @xmath109 is constant , @xmath110 , and @xmath111 . the growth of the kinetic energy fluctuations shows that the flow is unstable with @xmath28 when vertical gravity is present . [ fig : growenergy]b shows the kinetic energy growth for the same flow as shown with the solid curve as in fig . [ fig : growenergy]a , but computed with the fully compressible equations with _ athena_. the rapid growth of the kinetic energy fluctuations shows that zvi is not an artifact of the anelastic approximation or spectral codes . the growth of kinetic energy fluctuations in figs . [ fig : growenergy ] is evidence of instability , but spatial plots of the relative vorticity , defined as @xmath112 , are more useful in illustrating zvi and its nonlinear evolution . the point - wise dimensionless rossby number , defined as the ratio of relative vorticity to the global keplerian vorticity @xmath113/(2 \omega_0)$ ] , is a better diagnostic of late - time zombie turbulence than , say , the mach number , because the amplitude of the mach number at late - times is sensitive to the values of @xmath114 and @xmath115 ( which are chosen arbitrarily in our simulations ) , whereas the rossby number is not see [ subsec : mach ] . the simulations in this section are initialized by random perturbations which are closer to the expected conditions in a ppd rather than by an _ ad hoc _ isolated initial vortex as was done in the simulations in bm05 and mpjh13 . [ fig : xz ] shows @xmath116 for an anelastic flow in the @xmath117 plane at four different times and at an arbitrary stream - wise location in @xmath72 . ( because the equations , boundary conditions and initial conditions are invariant under translation in @xmath72 , the flow in all @xmath117 planes is statistically the same for all time . ) [ fig : xy ] shows @xmath116 for the same flow in the @xmath118 plane at @xmath59 , which is midway between the upper and lower boundaries . the parameter values , initial conditions and boundary conditions of the flow in figs . [ fig : xz ] and [ fig : xy ] are similar to the flow shown by the solid curve in fig . [ fig : growenergy]a with the difference that @xmath119 , rather than 2 . the initial perturbing noise has a power - law spectrum with random velocity phases , so there are no inherent length scales or coherent features of any kind in the initial perturbation . for kolmogorov noise , the energy is dominated by fourier modes with the smallest spatial wave numbers , while the relative vorticity field is dominated by fourier modes with the largest spatial wavenumbers . thus , figs . [ fig : xz]a and [ fig : xy]a are dominated by the smallest length scales ( which are equal to the spatial resolution of the calculation , @xmath120 in each direction ) . much of the initial vorticity is quickly stretched and elongated in the stream - wise direction by the keplerian shear , as shown in figs . [ fig : xz]b and [ fig : xy]b . the stretching cascades the initial kinetic energy to spatial wavenumbers with higher fourier modes where it is quickly dissipated by hyperviscosity ( similar to fig . [ fig : growenergy]a at @xmath121 ) . in competition with the stretching , there is a weaker reverse cascade of energy from small to large length scales due to the fact the fluid is rapidly rotating . here , the reverse cascade manifests itself via vortex mergers . in a rapidly rotating flow without a background shear , like - signed vortices of both signs merge , but when a strong background shear is also present , vortices with the same sign as the background shear merge into larger vortices with shapes that are elongated in the stream - wise @xmath72 direction , while vortices with a sign opposite to the background shear flow are stretched into thin stream - wise - oriented filaments and layers @xcite . the shear in a ppd is anticyclonic , and therefore anticyclones merge , while cyclones are stretched . when @xmath122 and @xmath79 have the same [ opposite ] signs , or equivalently , when @xmath123 [ @xmath124 . ] the result of the competition between the forward energy cascade ( resulting in energy dissipation ) and the reverse energy cascade ( resulting in the mergers of anticyclones ) is visible in figs . [ fig : xz]c and [ fig : xy]c at time @xmath125 yrs ; the flow is dominated by a few scattered anticyclones . incidentally , this flow looks very much like the initial condition of a single isolated anticyclone that we studied in mpjh13 and illustrated in fig . [ f : zombie_cartoon ] . by @xmath126 yrs , zvi is well underway in figs . [ fig : xz ] and [ fig : xy ] . the scattered anticyclones which emerged from the reverse cascade have now excited neighboring critical layers to form dipolar vortex layers . the critical layers themselves are easy to identify in plots of @xmath127 ( not shown ) , but difficult to observe in plots of @xmath116 . however , the vortex layers that the critical layers create are easily identifiable in plots of @xmath116 because each dipolar sheet appears as a pair of opposite - signed `` stripes '' in the @xmath73-@xmath72 plane ( with the critical layer sandwiched between them fig . [ f : zombie_cartoon]c ) . in an initial - value simulation with an initial perturbation consisting of noise , it is sometimes difficult to see the dipolar vortex layers due to the fast instability of the anticyclonic sheet . however in carefully controlled simulations consisting of a single perturbing vortex , rather than noise , dipolar layers can be found easily , as shown in fig . 1a in mpjh13 . as discussed in @xmath128[sec : cartoon ] , the layers of cyclonic vorticity aligned in the stream - wise direction are linearly stable . in contrast , layers of anticyclonic vorticity aligned in the stream - wise direction are linearly unstable and roll - up into stable anticyclones . at late times ( figs . [ fig : xz]d and [ fig : xy]d ) , the flow has reached a statistically steady state of zombie turbulence . one can see by eye , especially in fig . [ fig : xy]d , that the flow has formed a pattern with cross - stream or @xmath73 wavenumber of @xmath0@xmath129 . the relative vorticity , although very turbulent , has developed spatial coherence and is neither homogeneous nor isotropic . the cyclonic vorticity has formed approximately 2-dimensional layers that are approximately aligned in the @xmath72-@xmath75 planes . between these planes are approximately ellipsoidal turbulent anticyclones . at intermediate times in our calculation , we would expect a dominant @xmath73-wavenumber of @xmath130 , and at times between that of figs . [ fig : xy]c and that of fig . [ fig : xy]d , the dominant @xmath73-wavenumber is indeed @xmath0@xmath131 . however , at later times , the vortices grow in size , run into each other in the @xmath73 direction , and become larger . due to mergers , the late - time @xmath73-diameters of the vortices and the average spacing in @xmath73 between the cyclonic sheers both become slightly larger than @xmath22 . therefore , the dominant wavenumber in @xmath73 becomes slightly less than @xmath132 . we have carried out initial - value calculations of zombie turbulence in anelastic flows and boussinesq flows ( mpjh13 ) for a wide variety of parameters , and the @xmath116 always look like the pattern in figs . [ fig : xz]d and [ fig : xy]d . our one fully compressible calculation ( fig . [ fig : athena ] ) also shows this same pattern at late times . in fig . [ fig : athena ] , @xmath133 and @xmath134 , so we would expect that @xmath135 , which argues that the @xmath73 wavenumber that initially dominates the flow is @xmath136 . this wavenumber agrees with this simulation at intermediate times . at late times , the wavenumber in fig . [ fig : athena ] has been reduced to @xmath0@xmath137 by vortex mergers and reverse energy cascades . note that the two flows shown in figs . [ fig : exam2 ] and [ fig : exam3 ] , also have @xmath133 and @xmath134 , and although we have not included figures of their @xmath116 , these flows also have dominant @xmath73-wavenumbers of @xmath0@xmath136 at intermediate times and @xmath0@xmath137 at late times . thus , the main signature of zombie turbulence that differentiates it from other types of turbulence in shear flows that are triggered by finite - amplitude instabilities is that the turbulence in zvi is never homogeneous or isotropic ; it is dominated by cyclonic layers and elongated anticyclones aligned in the stream - wise direction of the background shear , and the pattern of layers and anticyclones has a dominant @xmath73 wavenumber slightly less than @xmath132 . the aspect ratio @xmath138 ( defined as the stream - wise diameter of a vortex in the @xmath72 direction divided by its cross - stream diameter in @xmath73 ) of the zombie vortices is approximately the same as the laminar vortices studied by @xcite and investigated in bm05 in the ppd context , where @xmath138 is given implicitly by @xmath139 the moore - saffman relation was derived for a steady two - dimensional vortex with uniform relative vorticity embedded in flow with uniform shear . however , the turbulent zombie vortices have aspect ratios similar to that in eq . ( [ saffman ] ) because the relation is the quantification of the fact that a large relative vorticity tends to make a vortex `` round '' and a large background shear tends to elongate a vortex is its stream - wise direction . at late times the characteristic magnitude of @xmath140 of the anticyclones in zombie turbulence is always @xmath0 - 0.25 , so regardless of the parameters of the flow , eq . ( [ saffman ] ) shows that in a keplerian disk , the aspect ratios @xmath138 of zombie vortices are between 4 and 5 . figures [ fig : exam1 ] [ fig : exam4 ] show how the root - mean - square ( rms ) rossby numbers @xmath141 and rms mach number @xmath142 evolve in time for three anelastic flows and one fully compressible flow . the flows in figs . [ fig : exam2 ] and [ fig : exam3 ] have the same flow parameters , with the only difference being the rms value of their initial noise . all of the statistical properties , including @xmath141 and @xmath142 , of the late - time flows in fig . [ fig : exam2 ] and [ fig : exam3 ] are nearly the same , which suggests that they evolve to a common attracting solution that is independent of the initial conditions . for all four flows in figs . [ fig : exam1 ] [ fig : exam4 ] , the values of @xmath141 initially plummet due to small - scale hyperviscosity ( or numerical diffusion in _ athena _ ) , but then grow after zvi develops , and eventually reach a plateau with @xmath141 between @xmath143 and @xmath144 . the @xmath141 is the ratio of the strength of the vorticity in the turbulence to that of the unperturbed keplerian flow , so @xmath145 indicates large amplitude turbulence . we have no rigorous explanation for why the rossby numbers evolve to these values in _ all _ of our simulations that produce zombie turbulence , regardless of parameter values . however , this range of values of @xmath146 is consistent with the picture that we presented in mpjh13 and fig . [ f : zombie_cartoon ] of how zvi spreads throughout the computational domain after it starts at one or more discrete locations . we showed that after the first critical layer is excited , it creates a vortex dipolar layer that rolls up into the first zombie vortex . then , that first zombie vortex triggers an instability in an adjacent critical layer in an unperturbed region of the flow , which then launches a new vortex layer and zombie vortex that triggers the instability in its adjacent critical layer , and so on _ ad infinitum_. however , this scenario requires that the vortices that make up the zombie turbulence have sufficiently large amplitudes to trigger the finite - amplitude zvi in unperturbed regions of the flow . it is more common among astrophysicists to measure the amplitude of turbulence with the mach number , rather than the rossby number , so it is somewhat disconcerting that the late - time values of @xmath142 in our calculations depend on the values of @xmath147 , and that the rms mach numbers can be made big or small by adjusting the size of the computational box . it is also discouraging ( to those who might believe that zombie turbulence can transport angular momentum in ppds ) that @xmath142 is only @xmath0 @xmath148 in our calculations . we now show that the dependence of @xmath149 on the size of the computational box is easily explained , and that in a ppd we expect @xmath149 to be equal to @xmath146 at late times . note that the rms rossby number is approximately @xmath150 where @xmath151 is the characteristic length scale of the flow where the vorticity has its maximum value and @xmath152 is the characteristic velocity of a turbulent eddy of diameter @xmath153 . of the flow , where @xmath107 is the fourier wavenumber . the kinetic energy per unit mass of the total flow is @xmath154 , and the kinetic energy per unit mass of an eddy of diameter @xmath153 is defined as @xmath155 ^ 2/2 $ ] . the last equivalence is what is used to defined the eddy velocity @xmath152 . note that an eddy of diameter @xmath153 is made up of all velocity fourier modes with wavenumbers @xmath156 . kolmogorov turbulence has an energy spectrum @xmath106 , so in kolmogorov turbulence , @xmath152 is proportional to @xmath157 see @xcite . ] the rms mach number is approximately @xmath158 where @xmath159 is the characteristic length scale of the flow where the velocity has its maximum value . in a kolmogorov energy spectrum , @xmath160 because the velocity is dominated by the large length scales and the vorticity by the small length scales . however , zombie turbulence does not have a kolmogorov spectrum and is dominated by large vortices . in flows dominated by large vortices , @xmath159 and @xmath151 are both approximately the mean size of the vortices . therefore , in a flow dominated by zombie vortices , @xmath161 where @xmath162 is the characteristic diameter of a zombie vortex . in a flow with constant vertical gravity ( either an anelastic or a fully compressible flow with @xmath163 ) , eq . ( [ home ] ) determines @xmath61 , and eq . ( [ zdominated ] ) then shows that a flow with zombie vortices has @xmath164 in all of the calculations presented in this paper , @xmath165 , so using eq . ( [ delta ] ) for @xmath22 , eq . ( [ linear ] ) becomes @xmath166 where we used the fact that in our computations @xmath167 . equation ( [ almost ] ) shows that @xmath149 depends on the size of the computational box , is consistent with figs . [ fig : exam1 ] [ fig : exam3 ] at late times , and explains why our calculations produce zombie turbulence with @xmath168 . and [ fig : exam3 ] , the values of @xmath141 have plateaued at late times , but the values of @xmath142 are still slowly growing . this suggests that these flows have not quite yet reached a statistically steady state . the only way in which @xmath142 can grow while keeping @xmath141 fixed is if @xmath169 is still growing , and the latter is indicative that the reverse cascade of energy is still continuing at late times in the flows in figs . [ fig : exam2 ] and [ fig : exam3 ] . ] equations ( [ linear ] ) and ( [ almost ] ) do not apply to a ppd for two reasons . first , due to the fact that the vertical component of the protostellar gravity is linear in @xmath75 , rather than constant , the sound speed is given by eq . ( [ linear_g]b ) rather than eq . ( [ home ] ) . therefore , eq . ( [ linear ] is replaced by @xmath170 the second reason that eq . ( [ almost ] ) does not apply to a ppd is we expect @xmath171 for zombie vortices in a ppd , rather than @xmath165 . this expectation follows from the observation that in our numerical simulations , zombie vortices grow until they run into each other in the @xmath73 direction ( or they run into the @xmath73-boundary of the computational domain ) . because the spacing of the dominant critical layers in @xmath73 is @xmath22 , the spacing of the initial vortices is also @xmath22 and the vortices run into each other and stop growing when @xmath165 . ( we explicitly chose the parameters in our studies to have @xmath172 so that @xmath162 would be determined by @xmath22 and not by @xmath173 . ) however in a fully compressible flow , it is well known that vortex diameters are often constrained by the fact that the vortices must be subsonic , otherwise the vortices rapidly dissipate their energy via acoustic radiation . in particular , the change in the velocity from one side of the vortex to the opposite side must be less than the sound speed . therefore , in fully compressible flows , vortices stop growing when they become large enough to be supersonic , which can occur while the vortices are still too small to run into each other . by definition of the rossby number , the characteristic relative velocity of a zombie vortex is @xmath174 . however , the characteristic change in the velocity of the total flow across the @xmath73-diameter of the vortex is @xmath0 @xmath175 because the vortex is embedded in the keplerian disk with a shear of @xmath176 . for the velocity change across the diameter of the vortex to be less the sound speed , @xmath162 must be less than @xmath0 @xmath177 . in a ppd , @xmath178 , so we would expect that the diameters of zombie vortices have their growth limited by the constraint that the vortices remain subsonic , and not by the size @xmath22 . using @xmath171 in eq . ( [ ppdma ] ) makes the mach numbers of the zombie vortices in a ppd of order @xmath146 . note that in some fully compressible flows that it is possible that @xmath177 is greater than @xmath22 . in fact , in @xmath128[subsec : reverse ] we showed that for the fully compressible flow examined in this paper that @xmath179 . therefore for the fully compressible flow examined in this paper , we expect , and have observed numerically , that @xmath165 , that vortex diameters are not restricted by the subsonic constraint , that the zombie vortices remain subsonic , and that eq . ( [ almost ] ) is valid . figs . [ fig : xz]d , [ fig : xy]d , and [ fig : athena ] show that at late times zombie turbulence fills the computational domain . we can quantify this space - filling property by defining a spatial filling factor for the turbulent vorticity : @xmath180 is the volume fraction of the computational domain that has @xmath181 . fig . [ fig : fill ] shows that for the anelastic flow in fig . [ fig : exam2 ] , approximately @xmath182 of the flow is filled with vortices with rossby numbers with magnitudes greater than 0.3 ; @xmath183 with magnitudes greater than 0.2 ; and almost @xmath184 with magnitudes greater than 0.1 . these fill factors may still be growing in time . the fill factors in fig . [ fig : fill ] are representative of all of our anelastic calculations . for the flow in our fully compressible _ athena _ simulations , the fill factors are slightly larger with approximately @xmath185 of the flow is filled with vortices with rossby numbers with magnitudes greater than 0.3 ; @xmath186 with magnitudes greater than 0.2 ; and almost @xmath187 with magnitudes greater than 0.1 . we have shown that keplerian disks are unstable to the purely hydrodynamic zvi instability when a vertical component of gravity from the central star or protostar is included and the vertical thermodynamic structure is stably stratified . it may seem counterintuitive that introducing stable stratification makes a flow more unstable . yet we have demonstrated that vertical stratification introduces baroclinic critical layers into the flow . in this work , we have explored zvi in the limit of infinite cooling time . zvi has a fast growth rate , requiring only a few hundred local `` years '' to produce large - magnitude turbulence . even if the instability is initially spatially confined to a small region , the entire domain fills rapidly with turbulence with rms rossby numbers between 0.2 and 0.3 . we have argued that in a fully compressible ppd , we expect that the mach number of the space - filling zombie turbulence will be approximately equal to its rossby number . when turbulence created by zvi reaches a statistically - steady state , it is neither homogeneous nor isotropic , and the pattern of the turbulence is highly asymmetric with respect to cyclonic and anticyclonic vorticity . the flow is dominated by large , elongated , anticyclonic vortices and narrow cyclonic layers of vorticity that are aligned in the stream - wise direction ( i.e. , the azimuthal direction in a ppd ) . this pattern makes zombie turbulence easily identifiable and distinct from other forms of turbulence . in a ppd , the vortex size and the spacing between the layers would be expected to be of order the vertical pressure scale height . our calculations show that zombie turbulence is robust and does not decay . in mpjh13 , the permanence of the turbulence was shown to be due to the fact that it draws its energy from the energy stored in the keplerian shear . disks filled with statistically - steady zombie turbulence appear to be attracting solutions that are independent of the details of the initial conditions that trigger the turbulence ; properties , such as the energy of the non - keplerian velocity , the energy spectrum as a function of wavenumber , and the probability distributions of the mach and rossby numbers are independent of the triggering perturbations . the robustness of zvi prompts the question of how it was missed in previous studies . part of the answer is that zvi has four necessary ingredients : rotation , a component of gravity along the ( vertical ) rotation axis , horizontal shear , and stable vertical stratification . these ingredients are all present in a ppd , but many previous studies neglected stable vertical stratification . any stability study that does not include all four ingredients whether a numerical calculation or a laboratory experiment that uses a constant - density fluid can not produce zvi . * sufficient numerical resolution and/or control over numerical dissipation is required to resolve critical layers . [ fig : xz ] [ fig : athena ] show that these layers are very thin . simulations using the godunov finite - volume code _ athena _ show that zvi is present only when more than 128 grid points per vertical pressure scale height are used in the radial direction . thus the instability could not be observed in the simulations of vertically stratified disks by @xcite who used 30 radial points per pressure scale height or by @xcite who used 64 or fewer . * sufficient integration time is needed for vorticity at small length scales to reverse cascade to larger scales , for the cyclonic vorticity to homogenize , and for the anticyclones to merge into vortices that are sufficiently large and spatially scattered ( as in fig [ f : zombie_cartoon]b ) to excite critical layers . it then takes additional time for dipolar vortex layers to form and several more vortex turn - around times for them to roll up and create the first generation of zombie vortices . in our studies [ fig : growenergy ] ) , it takes at least 50 years for the zombie vortices to fill the domain , for energy to be drawn from the keplerian shear by the vortices , and for the non - keplerian energy component to increase substantially . if a calculation were terminated early ( e.g. , after 8 years , as in fig . [ fig : balbus ] ) , then zvi would not be observed . in bm05 , we hypothesized that vortices themselves could directly transport angular momentum radially outward in a ppd . however , we found that the rate of transport , as parameterized by @xmath188 , was a factor of @xmath0100 times smaller than the values needed for the observed rate of star formation from a protostar , where @xmath189 means the volume - averaged value of @xmath190 over the entire domain . the value of @xmath9 is of order : @xmath191 where @xmath192 is the correlation between @xmath193 and @xmath194 , @xmath195 is the time - averaged spatial fill factor of the flow with @xmath149 , and @xmath196 is the ratio of the mach - number - weighted average density of the gas to the density of the gas in the mid - plane of the ppd . in bm05 , the small values of @xmath9 were due to the small values of the correlation @xmath192 because of the inherent symmetry of a vortex . vortices in the disk are approximately mirror symmetric with respect to the @xmath73@xmath75 plane that passes through the vortex center ; thus , the values of @xmath197 on either side of the symmetry plane nearly cancel . we therefore believe that zombie vortices can not _ directly _ produce a sufficiently large @xmath9 for star formation . we now hypothesize that zombie turbulence in a fully compressible fluid _ indirectly _ may produce much larger values of @xmath9 . a fully compressible fluid has three orthogonal velocity components @xcite : two are rotational and divergence - free ( i.e. , the poloidal and toroidal components , which anelastic flows have ) , and one is an irrotational component with a non - zero divergence ( i.e. , the curl - free potential flow , which anelastic flows do not have ) . we suspect that the rotational divergence - free zombie turbulence will excite acoustic modes ( i.e. , equipartition of energy among the poloidal , toroidal and potential components ) . @xcite have shown that acoustic waves in a ppd transport angular momentum radially outward with values of @xmath9 of order @xmath198 . acoustic waves have correlations @xmath192 of order unity . for acoustic waves launched by zombie vortices , we estimate @xmath199 , @xmath200 , @xmath201 , and @xmath202 , which yields a value of @xmath203 , consistent with values inferred from observed rates of star formation . our estimate that the fill factor @xmath196 is small is based on the numerical simulations by bm05 and @xcite that found that vortices can exist above and below the disk midplane , but the midplane itself at @xmath204 is devoid of laminar vortices with @xmath205 . the calculation of @xmath9 in a fully compressible calculation of a ppd will be the subject of future papers . however , in this paper we have shown , regardless of the value of @xmath9 , that zvi destabilizes a ppd and creates space - filling turbulence with rms rossby numbers between 0.2 and 0.3 . other outstanding issues include : ( i ) varying different profiles of stable vertical stratification , ( ii ) determining how close to the nearly unstratified midplane can turbulence from zvi can penetrate , ( iii ) including explicit cooling terms , ( iv ) studying zvi in global simulations , ( v ) exploring how zvi interacts and competes with mri , ( vi ) investigating how zvi interacts and competes with streaming instabilities , and ( vii ) computing dust capture inside zombie vortices . psm is supported in part by nsf grants ast-0905801 and ast-1009907 , and by nasa patm grants nnx10ab93 g and nnx13ag56 g . part of the computational work used an allocation of computer resources from the extreme science and engineering discovery environment ( xsede ) , which was supported by national science foundation grant no . oci-1053575 , and part was supported by nasa - hec . jab is supported by nsf ast-1010052 . ph is supported by a ziff environmental fellowship from harvard university center for the environment . dl is supported by a hertz foundation fellowship , the national science foundation graduate research fellowship under grant no . dge 1106400 , a kavli institute for theoretical physics graduate student fellowship , and partially by the schneider chair in physics to eliot quataert . psm , jab , ph and dl would like to thank the kavli institute for theoretical physics ( kitp ) for hosting us while drafting this paper ( kitp is supported in part by the nsf grant phy-1125915 ) . the authors thank dejon phillips for graphic design services .
we derive the most general seiberg - witten maps for noncommutative gauge theories in second order of the noncommutative parameter . our results reveal the existence of more ambiguities than previously known . in particular , we demonstrate that some of these ambiguities enter observables like scattering cross sections and enlarge the parameter space of the noncommutative standard model beyond .
0707.3595
with the start of data taking at the large hadron collider ( lhc ) , particle physics will , for the first time , directly probe the tera scale , i.e. the scale of electroweak symmetry breaking according to the standard model ( sm ) , around @xmath2 . while the sm has been confirmed experimentally to be a very precise effective description of the physics below the tera scale , there are many serious contenders for the more fundamental theory beneath the sm . for quite some time , superstring theory has been a leading candidate for the fundamental theory unifying all known interactions . there are certain solutions of superstring theory with additional spatial dimensions , where the characteristic string scales are low enough to allow experimental tests at the lhc and the planned international @xmath3 linear collider ( ilc ) . one spectacular prediction @xcite of superstring theory is the emergence of a noncommutative ( nc ) structure of spacetime at a scale @xmath4 associated with nonvanishing commutators @xmath5 = { \mathrm{i}}\theta_{\mu\nu } = { \mathrm{i}}\frac{1}{\lambda_{\text{nc}}^2}c_{\mu\nu}\ ] ] of spacetime coordinates that correspond to oriented minimal resolvable areas of size @xmath6 . while a nonvanishing commutator like ( [ eq : nc - geometry ] ) had been proposed much earlier @xcite as a regulator of divergencies in quantum field theory ( qft ) and quantum gravity , the observation of @xcite caused intense renewed interest in qft on nc spacetimes ( ncqft ) . the commutator ( [ eq : nc - geometry ] ) can be conveniently realized on a _ commuting _ spacetime by replacing all products of functions by moyal - weyl @xmath7-products @xmath8 a prescription for constructing arbitrary gauge theories on a nc spacetime was presented in @xcite . these so - called seiberg - witten maps ( swm ) realize nc gauge transformations in the nc theory as ordinary commutative gauge transformations on an effective commutative gauge theory . by going to the enveloping algebra of the lie algebra of a given gauge group , this approach @xcite circumvents obstructions like charge quantization in @xmath9 gauge theories and the prohibition of @xmath10 gauge groups in the earlier attempts . in particular , this prescription allowed the construction of the so - called noncommutative standard model ( ncsm ) @xcite as an anomaly - free @xcite canonical nc extension of the sm without having to introduce additional particles gauge theory and subsequently break the extraneous symmetries , introducing additional particles that must be removed from the observable spectrum @xcite . ] . in the first order of an expansion in @xmath0 , one has only three new bounded parameters that depend on the choice of the representation of the enveloping algebra of the sm lie algebra and describe new couplings among gauge bosons @xcite . these couplings vanish in the minimal ncsm , where the enveloping algebra is realized by matrices acting in the vector space of the adjoint representation . furthermore , the bosonic sector of the minimal ncsm was shown to be renormalizable at one loop @xcite , where all counter terms can be expressed through the usual field strength and coupling constant renormalizations . also the nonminimal ncsm is renormalizable at one - loop , if a _ finite _ gauge invariant @xmath1-term is added to the action @xcite . finally , the fermionic sector can be shown to require a finite number of gauge invariant four - fermion operators as additional counter terms in one loop order @xcite . in euclidean nc space , the renormalizability of scalar and gauge models has been shown to all orders in @xmath0 @xcite . using the effective theory in the first order of the @xmath0-expansion , several phenomenological studies were performed for past , present , and future colliders @xcite . in a preceding paper @xcite , we have studied the associated production of photons and @xmath11-bosons at hadron colliders ( tevatron and lhc ) showing that at the lhc one can reach a noncommutativity scale @xmath4 slightly above @xmath2 @xcite . moreover , we have found that it is necessary to go beyond the first order in @xmath0 , because of significant contributions from partonic center of mass energies exceeding the noncommutativity scale that can actually be probed . while it is possible in simple cases to derive expressions for families of swm to all orders in @xmath0 @xcite , an explicit parameterization of the most general solution has not been given . therefore , we start in this paper by constructing the most general swm for the ncsm in the second order of the @xmath0-expansion . the importance of this systematic approach is stressed _ a posteriori _ by discovering ambiguities in the swm that have been missed in earlier @xmath12 constructions of swm @xcite . while these authors expected all ambiguities to cancel in observable quantities , we find that they do _ not_. in fact , using @xmath13 as an example , we will calculate the ambiguity in the corresponding scattering amplitude explicitly . the outline of this paper is as follows . in section [ sec : swm ] we derive the general swm up to second order in the noncommutativity @xmath0 . particular emphasis will be given to the ambiguities resulting from the homogeneous solutions of the gauge equivalence equations . furthermore , the lagrangian and the feynman rules of the neutral current sector of the ncsm are constructed in section [ sec : feynman ] . section [ sec : ambiguities ] presents our analysis of the impact of the swm ambiguities on physical observables . we will demonstrate by an explicit calculation of @xmath14 that not all ambiguities cancel in the @xmath12 contribution to the cross section . in section [ sec : summary ] we conclude with a brief summary . all expressions that are needed for the main results of this paper will be given in full , either in the main text or in the appendices . complete expressions in second order in @xmath0 that are too lengthy to be included in this paper can be found in the appendix of @xcite . the purpose of the swm is to realize noncommutative gauge transformations by representations of the enveloping algebra using nonlinear functions of ordinary , commutative fields that reside in representations of the given lie algebra . this requirement is expressed by a set of so - called gauge equivalence equations for noncommutative gauge fields @xmath15 , gauge parameters appear in the lagrangian in the guise of faddeev - popov ghosts , if the gauge fixing is performed before application of the swm . ] @xmath16 and matter fields @xmath17 as functions of the commutative gauge fields @xmath18 , gauge parameters @xmath19 and matter fields @xmath20 . general swm are defined as solutions of the gauge equivalence equations ) could be relaxed by demanding that the two sides of equation ( [ eq : swm - condition ] ) lie in the same gauge orbit , but are not identical @xcite . since by construction the corresponding ambiguities must cancel in the gauge invariant yang - mills action to all orders in @xmath0 , we can ignore them in the rest of this paper . ] : [ eq : swm - condition ] @xmath21 where @xmath18 and @xmath20 transform as usual : [ eq : gauge - transformations ] @xmath22 here , we have used the notation @xmath23 and @xmath24 , @xmath25 being the generators of the gauge group . in practice , the gauge equivalence equations ( [ eq : swm - condition ] ) can be solved order by order in an expansion in @xmath0 . the physical predictions of qft , in particular the on - shell @xmath26-matrix elements and scattering cross sections , do not depend on the choice of interpolating fields @xcite . in fact , any two theories which are related by non - singular local field redefinitions [ eq : reparametrization ] @xmath27 and the corresponding change in the lagrangian @xmath28 will predict identical scattering cross sections . this reparametrization invariance can be proven both in axiomatic qft @xcite and in perturbation theory for effective qft @xcite . it provides the basis for the application of the powerful and now ubiquitous methods of effective qft to elementary particle physics phenomenology @xcite . the fact that the reparametrization ( [ eq : reparametrization ] ) corresponds to a change of integration variables in the path integral , provides a particularly intuitive proof of the invariance . since the swm @xmath29 appear to be just a reparametrization like ( [ eq : reparametrization ] ) , one might expect that its application has no effect on the calculation of observables . for a noncommutative @xmath9 gauge theory with unit charge @xcite it can be shown that this is indeed the case . however , in theories with multiple @xmath9 charges and @xmath10 gauge groups , the enveloping algebra is strictly larger than the lie algebra and the swm ( [ eq : swm ] ) must therefore be singular . the ncsm is a prominent representative of such theories , whence we must expect nontrivial effects of the swm on observables , as can readily be seen from @xcite . it was noted from early on , that the solutions of ( [ eq : swm - condition ] ) are not unique @xcite . consequently , the construction of nc extensions of the sm via swm is a priori not unique as well . only those ambiguities that correspond to non - singular reparametrization like ( [ eq : reparametrization ] ) are guaranteed to cancel in observables . this leaves us with the crucial problem of genuine ambiguities in physical quantities . to first order in @xmath0 it has been shown by explicit calculations that all ambiguities cancel in on - shell scattering amplitudes . in turn , one can use these cancellations as a powerful consistency check for numerical calculations of cross sections @xcite . however , there is no general theorem or physical argument that implies the cancellation of all ambiguities in the swm to all orders in the @xmath0-expansion . below we will identify ambiguities in the second order swm that affect cross sections . in an examination of the effective action of @xmath1-ncqed @xcite showing that the swm in first order corresponds to a field redefinition , it was already conjectured that this is not the case in second order . our results provide a proof for this conjecture from a completely different angle . furthermore , some of the ambiguities in second order are discussed in @xcite , but no claim is made of completeness . in fact , we have found additional ambiguities . a recursive solution of the gauge equivalence equations ( [ eq : swm - condition ] ) to all orders in @xmath0 , including ambiguities , has been given in @xcite . unfortunately , the ambiguities are not given as an explicit function of an independent set of parameters and it is not straightforward to search for observable effects in the ncsm in this form . thus we do not rely on @xcite for a discussion of the general solution of ( [ eq : swm - condition ] ) and derive the higher order terms anew . still , the existence of ambiguities with observable effects does not render the ncsm proposed in @xcite unphysical and thus rather useless . it merely adds more free parameters , not unlike those originating from the freedom of choosing the representation of the enveloping algebra , already discussed in @xcite . the most serious aspect of such additional parameters is the question to which extent deviations from sm predictions could be hidden by particular choices of the ambiguous contributions , making the ncsm untestable by experiment . the non - linear gauge equivalence equations ( [ eq : swm - condition ] ) are most easily solved by going over to the equivalent set of equations for infinitesimal gauge transformations . in the commutative case , the latter are given by [ eq : delta ] @xmath30 \\ \label{eq : delta - psi } \delta_\alpha \psi & = { \mathrm{i}}\alpha\psi\,,\end{aligned}\ ] ] while the noncommutative infinitesimal gauge transformations read [ eq : hat - delta ] @xmath31 \\ \label{eq : hat - delta - psi } \hat\delta_\alpha \hat\psi(\psi , a,\theta ) & = { \mathrm{i}}\hat\lambda(\alpha , a,\theta)\ast\hat\psi(\psi , a,\theta)\,.\end{aligned}\ ] ] from ( [ eq : delta ] ) and ( [ eq : hat - delta ] ) , one readily obtains the infinitesimal versions of the gauge equivalence equations ( [ eq : swm - condition - a ] ) and ( [ eq : swm - condition - psi ] ) : [ eq : swm - condition - infinitesimal ] @xmath32 where the commutative gauge transformation @xmath33 acts on the arguments of the noncommutative gauge and matter fields via the chain rule . however , the equations ( [ eq : swm - condition - infinitesimal ] ) are not sufficient . the existence of a noncommutative gauge parameter @xmath16 in ( [ eq : swm - condition ] ) requires that the commutator of two infinitesimal gauge transformations closes to another gauge transformation just as in the commutative case : @xmath34 } \hat\psi\,,\ ] ] where @xmath35 $ ] denotes the bracket in the commutative lie algebra . applying ( [ eq : swm - condition - infinitesimal - psi ] ) twice and taking into account that the commutative gauge transformation of the noncommutative gauge parameter @xmath16 does not vanish because it depends on @xmath18 , one has @xmath36 where all unnecessary arguments have been omitted . substituting the above in ( [ eq : closure ] ) and factoring out @xmath37 , the additional consistency condition reads @xmath38 = \hat\lambda(-{\mathrm{i}}[\alpha,\beta],a,\theta)\,.\ ] ] the infinitesimal consistency equations ( [ eq : swm - condition - infinitesimal ] ) and ( [ eq : swm - condition - infinitesimal - lambda ] ) still contain all orders of @xmath0 and closed expressions for their solutions are not yet known . however , we can express the swm as formal power series in @xmath0 : @xmath39 expand the equations ( [ eq : swm - condition - infinitesimal ] ) and ( [ eq : swm - condition - infinitesimal - lambda ] ) and solve them order by order in @xmath0 , starting at @xmath40 with the commutative gauge theory : @xmath41 writing all terms involving the unknown function @xmath42 on the left - hand side , the expansion of ( [ eq : swm - condition - infinitesimal - lambda ] ) results in the inhomogeneous linear equation @xmath43 - { \mathrm{i}}\left[\alpha,\lambda^{(1)}(\beta , a,\theta)\right ] \\ - \lambda^{(1)}(-{\mathrm{i}}[\alpha,\beta],a,\theta ) = - \frac{\theta^{\mu\nu}}{2 } \left[\partial_\mu\alpha,\partial_\nu\beta\right]_+\,.\end{gathered}\ ] ] the general solution is given by @xmath44_+ + { \mathrm{i}}c_\lambda^{(1 ) } \theta^{\mu\nu } \left [ \partial_\mu \alpha , a_\nu \right]\ ] ] involving one free parameter @xmath45 . using this solution , we can proceed analogously with ( [ eq : swm - condition - infinitesimal ] ) and derive the linear equations for @xmath46 , @xmath47 \\ = \partial_\mu \lambda^{(1 ) } - { \mathrm{i}}\left[a_\mu,\lambda^{(1)}(\alpha , a,\theta)\right ] + \frac{\theta^{\rho\sigma}}{2 } \left[\partial_\rho a_\mu,\partial_\sigma\alpha\right]_+\,,\end{gathered}\ ] ] and @xmath48 , @xmath49 where the right - hand sides depend on the free parameter @xmath45 via @xmath50 . for each value of @xmath45 , we then find the general solutions @xmath51_+ - \frac{1}{4}\theta^{\mu\nu}[a_\mu,\partial_\nu a_\rho]_+ \\ + { \mathrm{i}}c_\lambda^{(1 ) } \,\theta^{\mu\nu } [ d_\rho^{\text{adj } } a_\mu , a_\nu ] - 2{\mathrm{i}}c_a^{(1)}\,\theta^{\mu\nu } d_\rho^{\text{adj } } f_{\mu\nu}\end{gathered}\ ] ] and @xmath52 \psi + \frac{c_\psi^{(1)}}{2}\theta^{\mu\nu}f_{\mu\nu}\psi\,,\ ] ] parametrized by the free parameters @xmath53 and @xmath54 , respectively . notice that all terms proportional to @xmath45 , @xmath53 , or @xmath54 are lie algebra valued . therefore , they correspond to field reparametrizations and will cancel in observables . in fact , only the anticommutators in the expression ( [ eq : swm - a-1 ] ) for @xmath55 require the enveloping algebra and carry the potential to affect observables . in second order in @xmath0 , we again start with the closure relation ( [ eq : swm - condition - infinitesimal - lambda ] ) for gauge transformations : @xmath56 - { \mathrm{i}}\left[\lambda^{(2)}(\alpha , a),\beta\right ] - \lambda^{(2)}(-i[\alpha,\beta],a ) \\ = - \frac{{\mathrm{i}}}{8}\theta^{\mu\nu}\theta^{\kappa\lambda } \left[\partial_\mu\partial_\kappa\alpha , \partial_\nu\partial_\lambda\beta\right ] + { \mathrm{i}}\left[\lambda^{(1)}(\alpha , a),\lambda^{(1)}(\beta , a)\right ] \\ - \frac{1}{2}\theta^{\mu\nu } \left(\left[\partial_\mu\lambda^{(1 ) } ( \alpha , a),\partial_\nu\beta\right]_+ - \left[\partial_\mu\lambda^{(1 ) } ( \beta , a),\partial_\nu\alpha\right]_+\right)\,,\end{gathered}\ ] ] where we have suppressed the dependence on @xmath0 in @xmath57 for brevity . again , the right - hand side depends on the free parameter @xmath45 via @xmath58 . for @xmath45 fixed , we find that the general hermitian solution of ( [ eq : swm - condition - infinitesimal - lambda-2 ] ) involves 15 new free parameters , @xmath59 , , @xmath60 . for the present discussion it is sufficient to describe the general characteristics of the solutions and to give explicit expressions only for a specific choice of the free parameters . in appendix [ app : swm - lambda ] we spell out the specific solution corresponding to @xmath61 in the case @xmath62 . the lengthy complete expression can be found in the appendix of @xcite . it can be shown that the solutions given in @xcite and @xcite are both contained in our 16-parameter family of solutions . proceeding with the expansion of the gauge equivalence equations ( [ eq : swm - condition - infinitesimal ] ) in powers of @xmath0 , one obtains the second order equations for gauge and matter fields : [ eq : ge2 ] @xmath63 \\ = \partial_\rho\lambda^2(\alpha , a ) - { \mathrm{i}}[a_\rho , \lambda^{(2)}(\alpha , a ) ] - { \mathrm{i}}[a^{(1)}_\rho(a ) , \lambda^{(1)}(\alpha , a ) ] \\ + \frac{1}{2}\theta^{\mu\nu } \big(\big[\partial_\mu a^{(1)}_\rho(a),\partial_\nu\alpha\big]_+ + \big[\partial_\mu a_\rho,\partial_\nu\lambda^{(1)}(\alpha , a)\big]_+\big ) + \frac{{\mathrm{i}}}{8}\theta^{\mu\nu}\theta^{\kappa\lambda } \partial_\mu\partial_\kappa a_\rho \partial_\nu\partial_\lambda\alpha\end{gathered}\ ] ] and @xmath64 where the right - hand sides now depend on @xmath45 , @xmath53 , @xmath54 , @xmath59 , , and @xmath60 . substituting ( [ eq : swm - lambda-1 ] ) and ( [ eq : swm - a-1 ] ) for @xmath58 and @xmath65 , respectively , and the general solution @xmath66 from @xcite in ( [ eq : ge2-a ] ) , one gets the general hermitian solution for @xmath67 which depends on 6 additional parameters @xmath68 . the latter define a convenient basis for the solutions of the homogeneous equation : [ eq : c2/a ] @xmath69_\pm \\ a^{(2)}_{\rho,2,\pm } & = c^{(2)}_{a,2,\pm } \frac{{\mathrm{i}}}{2 } \theta^{\mu\nu } \theta^{\kappa\lambda } \bigl [ f_{\mu\kappa } , d^{\text{adj}}_\rho f_{\nu\lambda } \bigr]_\pm\\ a^{(2)}_{\rho,3,\pm } & = c^{(2)}_{a,3,\pm } \frac{{\mathrm{i}}}{2 } \theta^{\mu\nu } \theta^{\kappa\lambda } \bigl [ f_{\kappa\lambda } , d^{\text{adj}}_\rho f_{\mu\nu } \bigr]_\pm\,.\end{aligned}\ ] ] the three commutator terms @xmath70 are contained in the lie algebra and correspond to field redefinitions that must cancel in observables . in contrast , the three anticommutator terms @xmath71 need not be part of the lie algebra and can , in principle , give non - vanishing contributions to scattering amplitudes as we will demonstrate in section [ sec : example ] . an explicit expression for a specific solution of @xmath72 is given in appendix [ app : swm - a ] . the solution presented in @xcite can be shown to be contained in our set of general solutions @xcite . in contrast , the solution presented in @xcite is not . in fact , one can verify that the solution as written in @xcite does not satisfy the gauge equivalence equation ( [ eq : ge2-a ] ) and must therefore be considered incorrect . in addition , the solutions @xmath73 of the homogeneous equation are missing . as we will see later , @xmath74 plays a particularly important rle . while the size of the expressions makes it hard to pinpoint the actual error in @xcite , we want to mention that our results have been obtained and verified by long but straightforward algebraic manipulations using ` form ` @xcite . plugging ( [ eq : swm - lambda-1 ] ) for @xmath58 , ( [ eq : swm - psi-1 ] ) for @xmath75 , and the general solution for @xmath66 given in @xcite into ( [ eq : ge2-psi ] ) , we obtain a three - parameter family of matter field swm . a suitable basis for the solutions to the homogeneous equations is given by @xmath76 this result confirms the corresponding result in @xcite . in appendix [ app : swm - psi ] we present a specific solution for @xmath77 . for the general solution we again refer to @xcite . following the prescription of @xcite for constructing a ncqft , we replace all field products in the yang - mills lagrangian @xmath78 by @xmath7-products , and obtain an action for fields in the enveloping algebra that is invariant under nc gauge transformations and follows from the lagrangian @xmath79 with @xmath80 $ ] and @xmath81 $ ] , respectively . in a second step , we apply the swm ( [ eq : swm ] ) to obtain an action for fields residing in the lie algebra that is invariant under commutative gauge transformations and results from @xmath82 from ( [ eq : ncym ] ) , one can derive the feynman rules for the nc extension of the original commutative gauge theory . since the terms in ( [ eq : ym ] ) and ( [ eq : ncym ] ) that are quadratic in the fields are identical , we can perform gauge - fixing and introduce faddeev - popov ghosts directly to ( [ eq : ncym ] ) in terms of the commuting gauge fields . hence , the ghost interactions are not modified by nc contributions . for the purposes of the present paper , we are mainly interested in the cubic and quartic couplings in the neutral current sector of the ncsm that contribute to boson pair production in fermion - antifermion annihilation , @xmath83 , at tree level up to second order in @xmath0 . with _ all _ momenta incoming , including outgoing fermions , we define the vertex factors as follows : [ eq : vertices ] @xmath84 } \cdot v_{\mu_1\mu_2\mu_3}(k_1,k_2,k_3)\,.\end{aligned}\ ] ] where @xmath85}$ ] denotes the three - gauge boson coupling that depends on the choice for the representation @xmath86 of the enveloping algebra . in a @xmath9 gauge theory we can choose an arbitrary hermitian matrix @xmath87 as a generator , normalized to @xmath88 . consequently the squares of the eigenvalues of @xmath87 are bounded , @xmath89 , and we find @xmath90 } \le g$ ] , with @xmath85}=0 $ ] for @xmath91 and @xmath85}=g$ ] for @xmath92 . only in the latter case , the anticommutator remains in the lie algebra representation . since the chiral structure of the fermionic currents remains unaffected by the swm , we have written the following vertex factors for pure vector currents . the necessary substitutions @xmath93 depending on the fermion flavor and the type of vector boson coupled can be copied directly from the sm lagrangian . using the notations @xmath94 and @xmath95 , the vertex factors ( [ eq : vertices ] ) up to second order in @xmath0 are given by @xmath96\ ] ] @xmath97\end{gathered}\ ] ] @xmath98\end{gathered}\ ] ] @xmath99 \\ + \text{terms vanishing by equations of motion}\end{gathered}\ ] ] @xmath100 + \left ( k_1 \theta k_2 \right ) \left [ k_{3,\mu_1 } g_{\mu_2\mu_3 } - g_{\mu_1\mu_3 } k_{3,\mu_2 } \right ] \\ + \bigl [ ( k_1\theta)_{\mu_1 } \left [ k_{2,\mu_3 } k_{3,\mu_2 } - ( k_2 k_3 ) g_{\mu_2\mu_3 } \right ] - ( \mu_1 \leftrightarrow \mu_2 ) - ( \mu_1 \leftrightarrow \mu_3 ) \bigr ] \\ + \text{cyclical permutations of}\ ; \bigl\ { ( \mu_1,k_1 ) , ( \mu_2,k_2 ) , ( \mu_3,k_3 ) \ , \bigr\}\end{gathered}\ ] ] @xmath101 \\ + \text{cyclical permutations of}\ ; \bigl\ { ( \mu_1,k_1 ) , ( \mu_2,k_2 ) , ( \mu_3,k_3 ) \ , \bigr\}\,.\end{gathered}\ ] ] note that the triple gauge boson vertex ( [ eq : ggg2 ] ) at @xmath12 is generated by the swm alone . there are no contributions from the moyal - weyl @xmath7-product . in this paper , we will apply the ncsm feynman rules to the process @xmath102 at tree level . in the above , we have therefore given only the on - shell expression for the @xmath103 contact term ( [ eq : fbarvvf - on - shell ] ) , dropping terms which vanish by equation of motion . the complete expressions can be found in @xcite . as noted in section [ sec : swm1 ] , all swm ambiguities to first order in @xmath0 , i.e. all terms in the swm proportional to @xmath45 , @xmath53 , and @xmath54 correspond to lie algebra valued field redefinitions and must cancel in observables such as on - shell scattering amplitudes to this order . this is explicitly checked for @xmath83 in ( [ eq : amplitude(1 ) ] ) . note , however , that the above @xmath1 parameters reappear in the swm in higher order . we have already pointed out , that beyond @xmath1 there are no such general arguments for or against the cancellation of the swm ambiguities that do not correspond to non - singular field redefinitions in observables . therefore , we can approach this question presently only by studying specific examples . for that we choose the ncqed process @xmath13 as a prototype process in the neutral current sector of the ncsm . the additional @xmath11-boson couplings , their chiral structure and the @xmath11-mass in the ncsm will not add to our conclusions about the swm ambiguities . the feynman diagrams contributing to @xmath13 in ncqed are depicted in figs . [ fig : diagrams0]-[fig : diagrams2 ] . the feynman rules to @xmath1 and @xmath12 are given in section [ sec : feynman ] . + feynman diagrams contributing to @xmath13 in qed.,title="fig : " ] + feynman diagrams contributing to @xmath13 in @xmath1 in ncqed . the open squares denote @xmath1-vertices.,title="fig : " ] + feynman diagrams contributing to @xmath13 in @xmath12 in ncqed . the filled squares denote @xmath12-vertices.,title="fig : " ] actually , it is possible to demonstrate the dependence of the scattering amplitude on the free @xmath12 parameter @xmath104 and the interplay of reparametrization invariance and the enveloping algebra even without performing a complete calculation . the term in ( [ eq : c2/a/1 ] ) relevant for the tree level diagrams of fig . [ fig : diagrams2 ] contains two gauge fields . suppressing terms with more than two gauge fields , one has @xmath105 this term contributes both to the contact and the three - gauge boson vertex . explicitly , from the feynman rule ( [ eq : fbarvvf - on - shell ] ) for the contact term one finds the following contribution to the scattering amplitude : @xmath106 + \ldots\,.\end{gathered}\ ] ] since @xmath104 is absent from the on - shell @xmath107-vertex , this contribution can not be cancelled by any term coming from the @xmath108- or @xmath109-channel diagrams . therefore , a cancellation must involve @xmath110-channel diagrams , which however are proportional to the representation dependent coupling @xmath85}$ ] as can be seen from ( [ eq : vvv ] ) . consequently , the cancellation can at most take place for a particular value of @xmath85}$ ] , namely for @xmath85}=g$ ] , when the noncommutative gauge fields do not leave the lie algebra and the swm are just field reparametrizations . as a side remark , the non - vanishing , a priori undetermined contribution ( [ eq : contact2r ] ) to the tree level amplitude of @xmath111 results from one of the swm ambiguities missed in @xcite . we will now corroborate these preliminary remarks by a complete tree level calculation of @xmath13 up to second order in @xmath0 . it is useful to split the full scattering amplitude in the following pieces : @xmath112 } a_s^{(1 ) } + g^2 a^{(2 ) } + gg_{[\rho ] } a_s^{(2)}\,,\ ] ] which are self - explaining . it should be noted that the @xmath110-channel contributions @xmath113 must be separately gauge invariant , because their normalization depends on the choice of the representation of the enveloping algebra . for completeness , we restate the familiar qed amplitude from the diagrams in fig . [ fig : diagrams0 ] : @xmath114 with @xmath115 and @xmath116 , as well as the @xmath1-amplitude in ncqed @xcite : @xmath117 with @xmath118 u(p_1)\,,\ ] ] which collects all contributions from the @xmath108- and @xmath109-channel diagrams as well as the contact diagram in fig . [ fig : diagrams1 ] containing no pole . as expected , all contributions from ambiguous terms in the swm cancel in @xmath119 after application of the equations of motion . note that the @xmath120- and @xmath121-pole terms in ( [ eq : amplitude(1 ) ] ) result solely from the expansion of the moyal phases to order @xmath0 . this can be easily seen by combining the phase factors which multiply the vertices in a given diagram . for momenta @xmath122 , @xmath123 and @xmath124 flowing into a vertex momentum conservation implies the identities @xmath125 in the @xmath108-channel one thus obtains the total phase factor @xmath126 from which the corresponding phase factor in the @xmath109-channel follows by exchanging @xmath127 : @xmath128 turning to the @xmath110-channel amplitude @xmath129 , one finds that the @xmath130-pole contributions cancel in all contributions from the swm yielding @xmath131 with the @xmath130-pole coming from the moyal - weyl @xmath7-product alone : @xmath132 u(p_1 ) \,,\ ] ] and @xmath133 given in ( [ eq : a - nopole ] ) . this ensures that all effects from the swm cancel for @xmath85}=g$ ] as they should , since in this case the swm are non - singular field reparametrizations . to second order in @xmath0 , we obtain from the diagrams in fig . [ fig : diagrams2 ] @xmath134 with @xmath135 u(p_1)\,.\end{gathered}\ ] ] note that the @xmath120- and @xmath121-pole terms in ( [ eq : amplitude(2 ) ] ) follow again from the expansion of the combined moyal phases and contain no contribution from the swm . however , in contrast to @xmath133 , the second order amplitude @xmath136 does depend on ambiguous terms in the swm . in the case at hand , this is signalled by the appearance of the free parameters @xmath53 and @xmath104 . the exact cancellation for the choice @xmath137 appears to be accidental . as in first order in @xmath0 , we find that all @xmath130-poles cancel in the @xmath110-channel terms coming from the swm . since the moyal - weyl @xmath7-product only contributes to the three - photon couplings in odd orders of @xmath0 there is no @xmath130-pole term at all in @xmath12 . the result is exactly the negative of [ eq : a2-nopole ] : @xmath138 leading again to the cancellation of all contributions from the swm in the case @xmath85}=g$ ] , including the ambiguous terms involving the free parameter @xmath139 . however , in general , the cross section for @xmath13 is affected by this swm ambiguity . we have investigated the noncommutative extension of the standard model up to second order in @xmath0 . as our main result , we find that the general solution for the corresponding seiberg - witten maps contains more ambiguous terms than those reported previously , and that the swm ambiguities do not necessarily cancel in observables . furthermore , studying the scattering amplitude for @xmath13 as an explicit example we have shown that the ambiguities remaining in the scattering amplitude can be traced to the necessary extension of the lie algebra of the gauge group to its enveloping algebra , which elevates the swm from a unobservable field reparametrization to a source of new effective interactions . our results imply that the parameter space of the ncsm @xcite in @xmath12 is larger than assumed so far . there is every reason to expect that higher orders in @xmath0 will introduce even more ambiguities and evidence for this has been found in ncqed @xcite . as an outlook , phenomenological consequences of the ncsm at @xmath12 with the focus on collider searches will be presented in an upcoming paper @xcite . this research is supported by deutsche forschungsgemeinschaft , grant ru 311/1 - 1 and research training group 1147 _ theoretical astrophysics and particle physics _ , and by bundesministerium fr bildung und forschung germany , grant 05h4wwa/2 . a.a . gratefully acknowledges support from evangelisches studienwerk e.v . thanks the members of slac theory group for their kind hospitality . for each value of @xmath45 , we find a family of hermitian solutions to ( [ eq : swm - condition - infinitesimal - lambda-2 ] ) depending on the free parameters @xmath59 , , @xmath60 . the specific solution corresponding to @xmath61 for the case @xmath62 is given by @xmath140 for the general expression we refer to the appendix of @xcite . a representative of the six - parameter family of second order swm @xmath141 corresponding to the parameter choice @xmath142 is given by @xmath143 + [ \partial_\mu a_\kappa , \partial_\rho \partial_\lambda a_\nu ] \big)\\ \nonumber + \frac{1}{16}\theta^{\mu\nu}\theta^{\kappa\lambda } & \big ( + 2a_\mu a_\kappa \partial_\nu \partial_\lambda a_\rho - a_\mu a_\kappa \partial_\nu \partial_\rho a_\lambda + a_\mu a_\kappa \partial_\rho \partial_\lambda a_\nu\\ \nonumber & + 4a_\mu \partial_\nu a_\kappa \partial_\lambda a_\rho - 4a_\mu \partial_\nu a_\kappa \partial_\rho a_\lambda - 2a_\mu \partial_\kappa a_\nu \partial_\lambda a_\rho\\ \nonumber & - 2a_\nu \partial_\lambda a_\rho \partial_\mu a_\kappa + 3a_\nu \partial_\rho a_\lambda \partial_\mu a_\kappa + 4a_\kappa \partial_\nu \partial_\lambda a_\rho a_\mu\\ \nonumber & + 2a_\lambda \partial_\nu a_\rho \partial_\mu a_\kappa - a_\lambda \partial_\rho a_\nu \partial_\mu a_\kappa - a_\rho \partial_\mu a_\kappa \partial_\lambda a_\nu\\ \nonumber & - 4\partial_\mu a_\kappa a_\nu \partial_\lambda a_\rho + \partial_\mu a_\kappa a_\nu \partial_\rho a_\lambda + \partial_\mu a_\kappa a_\lambda \partial_\rho a_\nu\\ \nonumber & - \partial_\mu a_\kappa \partial_\lambda a_\nu a_\rho + 2\partial_\nu a_\kappa \partial_\lambda a_\rho a_\mu - 3\partial_\nu a_\kappa \partial_\rho a_\lambda a_\mu\\ \nonumber & + 2\partial_\nu a_\rho \partial_\mu a_\kappa a_\lambda - 2\partial_\kappa a_\nu \partial_\lambda a_\rho a_\mu + \partial_\kappa a_\nu \partial_\rho a_\lambda a_\mu\\ \nonumber & + 2\partial_\lambda a_\nu a_\rho \partial_\mu a_\kappa + 2\partial_\lambda a_\rho a_\mu \partial_\nu a_\kappa - 4\partial_\lambda a_\rho \partial_\mu a_\kappa a_\nu\\ \nonumber & - \partial_\rho a_\lambda a_\mu \partial_\nu a_\kappa - \partial_\rho a_\lambda a_\mu \partial_\kappa a_\nu + 4\partial_\rho a_\lambda \partial_\mu a_\kappa a_\nu\\ \nonumber & + 2\partial_\nu \partial_\lambda a_\rho a_\mu a_\kappa + \partial_\nu \partial_\rho a_\lambda a_\mu a_\kappa - \partial_\rho \partial_\lambda a_\nu a_\mu a_\kappa \big)\\ \nonumber + \frac{{\mathrm{i}}}{32 } \theta^{\mu\nu}\theta^{\kappa\lambda } & \big ( - 4a_\mu a_\nu a_\kappa \partial_\lambda a_\rho + 3a_\mu a_\nu a_\kappa \partial_\rho a_\lambda - 2\partial_\rho a_\lambda a_\mu a_\kappa a_\nu \\ \nonumber & + 4a_\mu a_\kappa a_\nu \partial_\lambda a_\rho - 2a_\mu a_\kappa a_\nu \partial_\rho a_\lambda - 4a_\mu a_\kappa a_\lambda \partial_\nu a_\rho \\ \nonumber & - 2a_\mu a_\kappa \partial_\nu a_\lambda a_\rho + 2a_\mu a_\kappa \partial_\lambda a_\nu a_\rho - 8a_\mu \partial_\nu a_\kappa a_\lambda a_\rho \\ \nonumber & - 4a_\nu a_\kappa \partial_\lambda a_\rho a_\mu + 4a_\nu a_\kappa \partial_\rho a_\lambda a_\mu + 8a_\nu a_\lambda a_\rho \partial_\mu a_\kappa \\ \nonumber & - 4a_\nu \partial_\lambda a_\rho a_\mu a_\kappa - 2a_\nu \partial_\rho a_\lambda a_\mu a_\kappa - 4a_\kappa a_\nu \partial_\lambda a_\rho a_\mu \\ \nonumber & - 2a_\kappa a_\nu \partial_\rho a_\lambda a_\mu - 4a_\kappa a_\lambda \partial_\nu a_\rho a_\mu + a_\kappa a_\lambda \partial_\rho a_\nu a_\mu \\ \nonumber & - 8a_\kappa \partial_\lambda a_\nu a_\rho a_\mu - 4a_\kappa \partial_\lambda a_\rho a_\mu a_\nu + a_\kappa \partial_\rho a_\lambda a_\mu a_\nu \\ \nonumber & - 4a_\lambda a_\nu a_\rho \partial_\mu a_\kappa - 4a_\lambda a_\rho a_\mu \partial_\nu a_\kappa + 4a_\lambda a_\rho a_\mu \partial_\kappa a_\nu \\ \nonumber & + 8a_\lambda a_\rho \partial_\mu a_\kappa a_\nu - 4a_\lambda \partial_\nu a_\rho a_\mu a_\kappa + 4a_\lambda \partial_\rho a_\nu a_\mu a_\kappa \\ \nonumber & - 2a_\rho a_\mu a_\kappa \partial_\nu a_\lambda - 2a_\rho a_\mu a_\kappa \partial_\lambda a_\nu + 8a_\rho a_\mu \partial_\kappa a_\nu a_\lambda \\ \nonumber & - 2a_\rho \partial_\mu a_\kappa a_\nu a_\lambda + 2a_\rho \partial_\mu a_\kappa a_\lambda a_\nu + 2\partial_\mu a_\kappa a_\nu a_\lambda a_\rho \\ \nonumber & + 2\partial_\mu a_\kappa a_\lambda a_\nu a_\rho - 4\partial_\nu a_\kappa a_\lambda a_\rho a_\mu + 4\partial_\nu a_\lambda a_\rho a_\mu a_\kappa \\ \nonumber & - 4\partial_\nu a_\rho a_\mu a_\kappa a_\lambda + 4\partial_\kappa a_\nu a_\lambda a_\rho a_\mu - 8\partial_\lambda a_\nu a_\rho a_\mu a_\kappa \\ \nonumber & - 4\partial_\lambda a_\rho a_\mu a_\nu a_\kappa + 4\partial_\lambda a_\rho a_\mu a_\kappa a_\nu + 3\partial_\rho a_\nu a_\mu a_\kappa a_\lambda \big)\\ \nonumber + \frac{1}{32}\theta^{\mu\nu}\theta^{\kappa\lambda } & \big ( - 3a_\mu a_\nu a_\kappa a_\lambda a_\rho + 2a_\mu a_\kappa a_\nu a_\lambda a_\rho - 4a_\nu a_\kappa a_\lambda a_\rho a_\mu\\ \nonumber & + 4a_\nu a_\lambda a_\rho a_\mu a_\kappa - 4a_\nu a_\rho a_\mu a_\kappa a_\lambda + 4a_\kappa a_\nu a_\lambda a_\rho a_\mu\\ \nonumber & - 4a_\kappa a_\lambda a_\nu a_\rho a_\mu - 2a_\kappa a_\lambda a_\rho a_\mu a_\nu - 8a_\lambda a_\nu a_\rho a_\mu a_\kappa\\ \nonumber & - 4a_\lambda a_\rho a_\mu a_\nu a_\kappa + 4a_\lambda a_\rho a_\mu a_\kappa a_\nu\\ \nonumber & - 3a_\rho a_\mu a_\nu a_\kappa a_\lambda + 2a_\rho a_\mu a_\kappa a_\nu a_\lambda\big)\,.\end{aligned}\ ] ] here , we have arranged the terms according to the power of @xmath18 , for the benefit of deriving feynman rules . the general solution can be found in @xcite . as pointed out in section [ sec : swm2-a ] , this solution is related to the one presented in @xcite , but incompatible with the one in @xcite . choosing @xmath144 , we obtain the following representative of the three - parameter family of matter field swm : @xmath145 again , the terms are ordered according to the power in the gauge field . for the general solution one may consult @xcite . b. jurco , l. mller , s. schraml , p. schupp and j. wess , eur . j. c * 21 * , 383 ( 2001 ) [ arxiv : hep - th/0104153 ] . x. calmet , b. juro , p. schupp , j. wess and m. wohlgenannt , eur . j. * c23 * , 363 ( 2002 ) [ arxiv : hep - ph/0111115 ] ; b. melic , k. passek - kumericki , j. trampetic , p. schupp and m. wohlgenannt , eur . j. c * 42 * , 483 ( 2005 ) [ arxiv : hep - ph/0502249 ] ; b. melic , k. passek - kumericki , j. trampetic , p. schupp and m. wohlgenannt , eur . j. c * 42 * , 499 ( 2005 ) [ arxiv : hep - ph/0503064 ] . c. p. martin , nucl . phys . * b652 * , 72 ( 2003 ) [ arxiv : hep - th/0211164 ] ; f. brandt , c. p. martin and f. r. ruiz , jhep * 0307 * , 068 ( 2003 ) [ arxiv : hep - th/0307292 ] . m. chaichian , p. presnajder , m. m. sheikh - jabbari and a. tureanu , eur j. c * 29 * , 413 ( 2003 ) [ arxiv : hep - th/0107055 ] . m. buric , d. latas and v. radovanovic , jhep * 0602 * , 046 ( 2006 ) [ arxiv : hep - th/0510133 ] ; m. buric , v. radovanovic and j. trampetic , jhep * 0703 * , 030 ( 2007 ) [ arxiv : hep - th/0609073 ] ; d. latas , v. radovanovic and j. trampetic , [ arxiv : hep - th/0703018 ] . r. wulkenhaar , jhep * 0203 * , 024 ( 2002 ) [ arxiv : hep - th/0112248 ] . h. grosse and r. wulkenhaar , lett . * 71 * , 13 ( 2005 ) ; j. nonlin . phys . * 11 * suppl . , 9 ( 2004 ) ; commun . phys . * 256 * , 305 ( 2005 ) [ arxiv : hep - th/0401128 ] ; v. rivasseau , f. vignes - tourneret and r. wulkenhaar , commun . phys . * 262 * , 565 ( 2006 ) [ arxiv : hep - th/0501036 ] ; h. grosse and m. wohlgenannt , [ arxiv:0706.2167 ] . w. behr , n. g. deshpande , g. duplancic , p. schupp , j. trampetic and j. wess , eur . phys . j. c * 29 * , 441 ( 2003 ) [ arxiv : hep - ph/0202121 ] ; g. duplancic , p. schupp and j. trampetic , eur . j. c * 32 * , 141 ( 2003 ) [ arxiv : hep - ph/0309138 ] ; p. schupp , j. trampetic , j. wess and g. raffelt , eur . j. c * 36 * , 405 ( 2004 ) [ arxiv : hep - ph/0212292 ] ; p. minkowski , p. schupp and j. trampetic , eur . j. c * 37 * , 123 ( 2004 ) [ arxiv : hep - th/0302175 ] ; b. melic , k. passek - kumericki and j. trampetic , phys * d72 * , 054004 ( 2005 ) [ arxiv : hep - ph/0503133 ] ; b. melic , k. passek - kumericki and j. trampetic , phys . rev . * d72 * , 057502 ( 2005 ) [ arxiv : hep - ph/0507231 ] ; m. buric , d. latas , v. radovanovic and j. trampetic , phys . rev . * d75 * , 097701 ( 2007 ) . t. ohl and j. reuter , phys . rev . * d70 * , 076007 ( 2004 ) [ arxiv : hep - ph/0406098 ] ; t. ohl and j. reuter , [ arxiv : hep - ph/0407337 ] . a. alboteanu , t. ohl and r. rckl , phys . rev . * d74 * , 096004 ( 2006 ) [ arxiv : hep - ph/0608155 ] ; pos * hep2005 * ( 2006 ) , 322 [ arxiv : hep - ph/0511188 ] . g. barnich , f. brandt and m. grigoriev , nucl . phys . * b677 * , 503 ( 2004 ) [ arxiv : hep - th/0308092 ] . j. rauh , diploma thesis , university of wrzburg , 2006 ; j. zeiner , phd thesis , university of wrzburg , 2007 . t. asakawa and i. kishimoto , jhep * 9911 * , 024 ( 1999 ) [ arxiv : hep - th/9909139 ] . r. haag , phys . rev . * 112 * , 669 ( 1958 ) ; h. j. borchers , nuovo cim . * 25 * , 270 ( 1960 ) ; d. ruelle , helv . acta * 35 * , 34 ( 1962 ) . s. r. coleman , j. wess and b. zumino , phys . rev . * 177 * , 2239 ( 1969 ) ; c. g. callan , s. r. coleman , j. wess and b. zumino , phys . rev . * 177 * , 2247 ( 1969 ) . s. weinberg , physica a * 96 * , 327 ( 1979 ) . j. m. grimstrup and r. wulkenhaar , eur . j. c * 26 * , 139 ( 2002 ) [ arxiv : hep - th/0205153 ] . j. a. m. vermaseren , _ symbolic manipulation with ` form ` version 2 : tutorial and reference manual _ , can ( amsterdam , 1992 ) .
we show that the lagrangian for interacting nonrelativistic particles can be coupled to an external gauge field and metric tensor in a way that exhibits a nonrelativistic version of general coordinate invariance . we explore the consequences of this invariance on the example of the degenerate fermi gas at infinite scattering length , where conformal invariance also plays an important role . we find the most general effective lagrangian consistent with both general coordinate and conformal invariance to leading and next - to - leading orders in the momentum expansion . at the leading order the lagrangian contains one phenomenological constant and reproduces the results of the thomas - fermi theory and superfluid hydrodynamics . at the next - to - leading order there are two additional constants . we express various physical quantities through these constants . int - pub 05 - 23
cond-mat0509786
symmetry is one of the most important ideas in modern physics . frequently , the identification of the symmetries of a physical system is the key to understanding it . in this paper we identify a new type of symmetries of nonrelativistic physics , which are manifest when the system is put in curved space , yet less evident in flat space . the uncovering of these symmetries comes from our attempt to understand the properties of the so - called `` unitary '' fermi gas , which is a gas of fermions which interact with each other through a zero - range two - body potential tuned to infinite scattering length @xmath0 . this system has been studied theoretically for some time @xcite , but only recently did it become realized in experiments with trapped atoms using a feshbach resonance @xcite . what is special about this system the absence of any intrinsic parameter , except for the number density . unfortunately , this same feature means that there is no small coupling constant on which to build a reliable perturbation theory . it is therefore imperative that we find and exploit all the symmetries of this system . despite the strong coupling , we are now certain that the ground state of the unitary fermi gas is a superfluid . we also have , as our starting point , at least two theoretical frameworks which allow one to solve certain problems involving the unitary fermi gas in a reliable fashion . these are the thomas - fermi theory and superfluid hydrodynamics . the two theories are very similar : both make no attempt to solve the problem at the microscopic level ; they are both valid in the long - wavelength regime , when the characteristic length scale of the problem is much larger than all microscopic scales ( including the healing length ) ; and they need as an input at most one parameter ( which is denoted as @xmath1 in eq . ( [ xi - def ] ) and characterizes the zero - temperature equation of state . ) the total energy and the oscillation frequencies of unitary fermi gas in a trap @xcite can be computed using these approaches . the input parameter @xmath1 is not fixed within these theories , but can either be measured experimentally or computed by monte - carlo simulations . however , until now it is not possible to systematically include corrections to either the thomas - fermi theory or superfluid hydrodynamics . at weak coupling there is no fundamental obstacle to compute these corrections , but our tools do not work at strong coupling . these corrections can be considerable if we want to study the physics at length scales not much larger than the healing length . the main , utilitarian motivation of this paper is to develop a systematic way to go beyond the leading order in the long - wavelength expansion . this is done using an effective field theory ( eft ) @xcite . the central idea of the eft philosophy is a power expansion over momentum ( or inverse wavelength ) , instead of the usual perturbative expansion over powers of the coupling constant . this makes eft particularly suitable for problems involving strongly coupled systems in the long - wavelength limit . naturally we find that at leading order the eft reproduces the thomas - fermi theory and superfluid hydrodynamics . however , in principle , the eft can be extended to arbitrary order in the expansion over momentum at the cost of introducing an increasing number of free parameters . the form of the effective lagrangian is constrained only by the symmetries of the system under study . somewhat unexpectedly , putting all possible constraints on the next - to - leading order ( nlo ) lagrangian turns out to be a very nontrivial problem in the case of the unitary fermi gas . in this paper , this task is accomplished by temporarily putting the system in curved space and exploiting the properties of general coordinate invariance and conformal invariance . working in curved space necessitates the use of certain mathematical tools normally employed only in general relativity , but what we find at the end is quite nontrivial and worth the effort . in addition to one free parameter at the leading order , one needs just two parameters to characterize the nlo lagrangian . these parameters can not be computed within the eft , but since the number of physical quantities that can be computed is greater than three , the eft has real predictive power . we start the paper with a summary of some results that can be obtained using the eft ( sec . [ sec : summary ] ) . then in sec . [ sec : nrgci ] we show that the lagrangian for interacting nonrelativistic particles can be coupled to background gauge fields and spatial metric in a way that respects gauge invariance and the general coordinate invariance . the general coordinate invariance is a gauge version of the translation and galilean transformations . in sec . [ sec : conformal ] we study the conformal invariance of unitary fermi gas . we then proceed , in sec . [ sec : leff ] and [ sec : nlo ] , to find the most general effective lagrangian , to leading and next - to - leading orders , that is consistent with these symmetries . some applications of the eft are considered in sec . [ sec : applications ] . we conclude with sec . [ sec : conclusion ] . here we give a brief summary of the result of this paper for the unitary fermi gas . we do it here to give the reader a sense of the power and the limitations of the effective field theory derived in the paper . we work in the unit system where @xmath2 . we denote the number density as @xmath3 and the fermion mass as @xmath4 , from which we define the fermi momentum @xmath5 , the fermi energy @xmath6 , and the fermi velocity @xmath7 . we introduce the dimensionless parameter @xmath1 , defined as the ratio of the energy density of the unitary fermi gas to the energy density of a free fermi gas at the same density , @xmath8 first we shall show that the dynamics of the gas is described , to leading order , by the lagrangian , @xmath9^{5/2},\ ] ] where @xmath10 is the phase of the condensate , @xmath11 is the chemical potential , @xmath12 is the trapping potential , and @xmath13 is a dimensionless parameter . the leading - order lagrangian captures the same information as the thomas - fermi theory and superfluid hydrodynamics . the coefficient @xmath13 is directly related to the parameter @xmath1 , defined in eq . ( [ xi - def ] ) , by @xmath14 and has to be determined microscopically . to the next - to - leading order in the low - momentum expansion , the effective lagrangian has the form @xmath15 \sqrt x\,.\ ] ] where we have introduced the shorthand notation @xmath16 and @xmath17 and @xmath18 are two additional dimensionless parameters that appear at the nlo level . many quantities can be computed using the effective lagrangian . let us consider infinite system first ( @xmath19 ) . the dispersion relation for low - energy phonons is @xmath20 + o(q^5\ln q ) , \qquad c_s = \sqrt{\frac\xi3}\ , { v_{\mathrm{f}}}\,.\ ] ] the static density and transverse response functions are [ sum : response ] @xmath21 + o(q^4\ln q),\\ \chi^t(q ) & = -9c_2 \sqrt{\frac\xi2}\ , { v_{\mathrm{f}}}q^2 + o(q^4\ln q).\end{aligned}\ ] ] the predictive power of the eft is illustrated by a relation between the nonlinear @xmath22 term in the phonon dispersion ( [ sum : dispersion ] ) and the static response functions , @xmath23 in addition , we also find the ground - state energy of a system of @xmath24 particles in a harmonic trap to be @xmath25 where @xmath26 , @xmath27 are the trap frequencies , @xmath28 . let us first consider a system of noninteracting particles ( fermions or bosons ) . instead of letting particles move freely in flat space , we will put the system in an external gauge field @xmath29 and a curved three - dimensional manifold with the spatial line element @xmath30 as indicated in eq . ( [ line - element ] ) , the metric on the manifold is , in general , time - dependent , so the manifold can change shape with time . the spatial indices @xmath31 will be raised and lowered by the metric @xmath32 and its inverse @xmath33 . the determinant of @xmath32 is denoted as @xmath34 . the action of the system is @xmath35.\ ] ] here @xmath36 . note that @xmath37 , @xmath38 , and @xmath32 appear as external parameters of the lagrangian ; they have no dynamics of their own . the most common situation is when the system is put in a trapping potential @xmath39 . this corresponds to @xmath40 , @xmath41 , @xmath42 . another case , not considered in this paper , is that of rotating systems where @xmath38 are not zero . although space is flat in all practical applications , keeping @xmath32 arbitrary in intermediate steps allow one to derive nontrivial results , as we shall see . we have written the time - derivative term in eq . ( [ free - l ] ) in the symmetric form @xmath43 . in flat space this is equivalent to the asymmetric form @xmath44 ; the two differ from each other by a full time derivative . this continues to be true if the metric is time - independent . for a general time - dependent metric , however , the two forms are not equivalent . as at the end we shall put the metric to flat , the choice between the two forms is a matter of convenience . we will soon find that the symmetric form in ( [ free - l ] ) is much more advantageous . the action ( [ free - l ] ) is invariant under gauge transformations , @xmath45 where @xmath46 . it is also invariant under a change of coordinate on the 3d manifold , @xmath47 the infinitesimal version of ( [ gauge - tr ] ) and ( [ 3d - gci ] ) is @xmath48 we note that while the gauge parameter @xmath49 in eq . ( [ static - gci ] ) can be a function of both time and space , so far we have considered the gauge parameter @xmath50 ( which is the coordinate shift , @xmath51 ) to be time - independent . in fact , if one attempts to make @xmath50 time - dependent , one finds that the action is not invariant , but acquires a change proportional to the time derivative of @xmath50 , @xmath52 however , the invariance of the action can be restored by a small modification of the transformation laws ( [ static - gci ] ) . the modified transformations read [ nonrel - gci ] @xmath53 it is straightforward to verify that terms proportional to @xmath54 now cancel in @xmath55 . ( had we written the time - derivative term in the asymmetric form @xmath44 this would not be true . ) we shall call ( [ nonrel - gci ] ) the _ nonrelativistic general coordinate transformations _ , or simply general coordinate transformations . the action is said to be general coordinate invariant . let us discuss the relationship between the transformations ( [ nonrel - gci ] ) with the more familiar galilean transformations . in contrast to ( [ nonrel - gci ] ) , the galilean transformations are defined in flat space , @xmath42 . they can be thought of as a special class of ( [ nonrel - gci ] ) with a particular choice of @xmath49 and @xmath50 : @xmath56 this leaves the flat metric unchanged while transforming other fields , @xmath57 note that in eqs . ( [ galilean ] ) terms proportional to @xmath58 and higher orders of @xmath59 have been dropped . thus any theory with general coordinate symmetry is automatically galilean invariant when restricted to flat space.the transformations ( [ nonrel - gci ] ) can be thought of as a gauge version of the translation and galilean transformations . a similar `` gauging '' of nonrelativistic spatial translations has been considered previously @xcite . a subclass of ( [ nonrel - gci ] ) which does not change the flat metric has been considered in ref . @xcite . so far we have considered the noninteracting theory . it is easy to introduce interaction into the theory in a way that respects general coordinate invariance by introducing auxiliary fields . for example , the theory with the following lagrangian , @xmath60 is consistent with general coordinate invariance , as one can make @xmath61 transform as @xmath62 in flat space ( [ int - l ] ) describes a system of particles interacting though a yukawa two - body potential , @xmath63 we thus have an example of an interacting theory where coupling to an external gauge field and a metric can be introduced in a way consistent with general coordinate invariance . we have not demonstrated that general coordinate invariance can be achieved for _ any _ two - body potential . however , our foremost interest is in systems with universal behavior , where only the scattering length @xmath64 is important , not the details of the two - body potential . for this purpose , the lagrangian ( [ int - l ] ) is sufficient . in particular , infinite scattering length is achieved by fine - tuning @xmath65 and @xmath66 so that @xmath67 when the first bound state appears at threshold . to achieve universality we also need to keep @xmath66 smaller than any physical length scale . the universality of the fermionic system with infinite scattering length means that one can choose @xmath65 and @xmath66 arbitrarily ( keeping @xmath66 small ) , and the physics does not depends on our choice if eq . ( [ fine - tuning ] ) is satisfied . one can even make @xmath65 and @xmath66 time - dependent and nothing will depend on their values , if the fine - tuning condition is kept at all time . this remark will be important in our subsequent discussion . we have found the transformations ( [ nonrel - gci ] ) by trial and error . an alternative way to arrive to these transformations is by taking the nonrelativistic limit of the relativistic general coordinate invariance . the second way is instructive and will be useful for us when we try to write down an effective lagrangian consistent with general coordinate invariance . we will describe it in the next section . consider a relativistic field theory of one free complex scalar field @xmath68 in a external four - dimensional metric @xmath69 . we use the @xmath70 metric signature with @xmath71 and @xmath72 . we keep track of the speed of light @xmath73 in order to send @xmath74 in the nonrelativistic limit . the action , @xmath75 is invariant under the infinitesimal general coordinate transformation [ relat - gci ] @xmath76 in the nonrelativistic limit we take @xmath74 . we assume that in this limit @xmath68 has a fast phase rotation in time solely due to its rest mass , which can be factored out to yield the nonrelativistic field @xmath77 , @xmath78 the metric is taken to have the @xmath79 component close to @xmath80 and small @xmath81 components , [ gag ] @xmath82{0mm}{6 mm } \\ -\dfrac{a_i}{mc } & & g_{ij } \end{array}\right),\ ] ] where @xmath37 , @xmath38 , and @xmath32 are kept fixed in the limit @xmath74 . the inverse metric is @xmath83{0mm}{6mm}\\ -\dfrac{a^i}{mc } + o(c^{-3 } ) & & g^{ij } + o(c^{-2 } ) \end{array}\right),\ ] ] where @xmath84 . substituting eqs . ( [ psipsi ] ) and ( [ ginv ] ) into the action ( [ relat - s ] ) , one reproduces ( [ free - l ] ) after discarding terms containing negative powers of @xmath73 . note that the @xmath85 and @xmath86 components of the 4d metric tensor now appear as the gauge field in the nonrelativistic action , in a manner reminiscent of the kaluza - klein mechanism @xcite . taking the coordinate shift @xmath87 of the form @xmath88 where @xmath49 is fixed in the limit @xmath74 , eqs . ( [ relat - gci ] ) reduce to eqs . ( [ nonrel - gci ] ) . thus we can understand ( [ nonrel - gci ] ) as a nonrelativistic limit of the relativistic general coordinate invariance . we now show that the general coordinate invariance implies conservation of particle number and momentum . we also will find that in theories with general coordinate invariance the number current coincides ( up to a factor of @xmath4 ) with the momentum density . the connected part of the generating functional @xmath89 is formally defined as a path integral @xmath90 } = \int\!d\psi^{\dagger}d\psi\ , e^{is[\psi,\psi^{\dagger},a_0,a_i , g_{ij } ] } \,.\ ] ] the correlation functions of operators coupled to the sources can be found by differentiating @xmath89 with respect to its arguments . general coordinate invariance implies , to linear order in @xmath49 and @xmath50 , @xmath91 = w[a_0,a_i , g_{ij}],\ ] ] where @xmath92 , @xmath93 , and @xmath94 are defined as in eqs . ( [ nonrel - gci ] ) . expanding the left - hand side of eq . ( [ wdelta ] ) to linear order in @xmath49 and @xmath50 , one finds [ conservation ] @xmath95 where @xmath96{0mm}{6mm}}\\ j^i = - \dfrac{\delta w}{\delta a_i}\ , , \qquad & t^i_k = 2g_{kj}\dfrac{\delta w}{\delta g_{ij}}-\delta^i_k w + a_k\dfrac{\delta w}{\delta a_i}\ , . \end{array}\ ] ] eqs . ( [ conservation ] ) can be interpreted as conservation of particle number and momentum . here @xmath3 is the number density , @xmath97 is the number flux , @xmath98 is the momentum density and @xmath99 is the stress tensor . we also notice that in flat space and in the absence of the vector potential the momentum density is proportional to the number current , @xmath100 this is the condition previously used by greiter , wilczek , and witten @xcite to find the effective field theory of superfluids . any theory which is general coordinate invariant automatically satisfies ( [ gww - cond ] ) . the lagrangian of noninteracting particles ( [ free - l ] ) exhibits scale invariance , which is an overall rescaling of time and fields , @xmath101 it is useful to introduce the notion of _ scaling dimension_. an operator @xmath102 is said to have scaling dimension @xmath103 ( in shorthand notation , @xmath104=\delta_o$ ] ) if under scale transformation in transforms as @xmath105 note that the scaling dimension of a product is the sum of scaling dimensions : @xmath106=[o_1]+[o_2]$ ] . also , one can assign scaling dimensions to space and time derivatives , @xmath107=0 $ ] and @xmath108=1 $ ] , so @xmath109=[{\partial}_i{\partial}_j o]=[o]$ ] , while @xmath110=[o]+1 $ ] . in this language , @xmath111 = -1 , \quad [ x^i]= 0,\quad [ \psi ] = \frac34\ , , \quad [ a_0 ] = 1,\quad [ a_i ] = 0,\quad [ g_{ij } ] = -1.\ ] ] and @xmath112=\frac52 $ ] , so @xmath113 is invariant . and @xmath38 in eq . ( [ scaling ] ) may seem somewhat unnatural , but that is because we have assigned @xmath32 a scaling dimension of @xmath80 . ] in fact , the action is invariant under a larger set of transformations , which involves arbitrary reparametrization of time @xmath114 , simultaneous with rescaling of @xmath37 and @xmath32 , @xmath115{0mm}{6mm}}\qquad & a_0(t ) \to a'_0(t ' ) = \left(\dfrac{dt}{dt'}\right)a_0(t ) , \\ g_{ij}(t ) \to g'_{ij}(t ' ) = \left(\dfrac{dt'}{dt } \right ) g_{ij}(t ) , & a_i(t ) \to a'_i(t ' ) = a_i(t).\\ \end{array}\ ] ] the infinitesimal version of ( [ conf - large ] ) is @xmath116 we shall call this transformation the _ conformal transformation_. ) and the general coordinate transformation ( [ nonrel - gci ] ) with @xmath117 , @xmath118 and @xmath119 , where @xmath120 is a constant . this combination leaves the trivial background @xmath121 , @xmath42 unchanged . ] the scale transformation ( [ scaling ] ) is a particular case of conformal transformations with @xmath122 being a linear function of time : @xmath123 . in general , if a field @xmath102 transforms under conformal transformation as @xmath124 then @xmath102 is said to have conformal dimension @xmath103 , @xmath104_{{\mathrm{c}}}=\delta_o$ ] . note that we still have @xmath125_{{\mathrm{c}}}=[o_1]_{{\mathrm{c}}}+[o_2]_{{\mathrm{c}}}$ ] , and @xmath109_{{\mathrm{c}}}=[o]_{{\mathrm{c}}}$ ] . however , we can not assign a conformal dimension to the time derivative @xmath126 . in fact , if @xmath102 transforms as in eq . ( [ conf - o ] ) , then in general @xmath127 does not have a well - defined conformal dimension at all , @xmath128 the extra term proportional to @xmath129 prevents us from assigning the conformal dimension @xmath130 to @xmath127 , unless @xmath131 . ( this term is absent for scale transformations where @xmath122 is linear in @xmath132 . ) therefore while an operator with a conformal dimension has a scaling dimension , the reverse is not true in general . note that the lagrangian ( [ free - l ] ) contains a term with a time derivative but still have a well - defined conformal dimension of @xmath133 , since the @xmath129 terms cancel out between the variations of @xmath134 and @xmath135 . again we see that the symmetric from of the time - derivative term is preferable . the conformal symmetry is generally broken when one turns on an interaction . we now show that conformal invariance becomes exact again in the limit of infinite scattering length . for that we can use the model with yukawa interaction ( [ int - l ] ) . if we take @xmath61 to transform the same way as @xmath77 , @xmath136 then after substitution , we see that two terms in the action , @xmath137 and @xmath138 , change under the conformal transformation . the effect is that the constants @xmath65 and @xmath66 are replaced by new time - dependent constants @xmath139 however , we recall that the condition of infinite scattering length corresponds to @xmath140 equal the critical value ( [ fine - tuning ] ) . we can see that this condition is not violated under the conformal transformation . thus we can make use of the freedom in choosing @xmath65 and @xmath66 to transform it back to the original constant @xmath65 and @xmath66 , without changing the generating functional @xmath89 . therefore , the generating functional of a system tuned to infinite scattering length is conformally invariant . this property of unitary fermi gas will be used in sec . [ sec : nlo ] . finally , we note that in contrast to the general coordinate transformations , the scale and conformal transformations described above are not descended from the relativistic counterparts . it is now well established that fermi gas remain superfluid in the whole range of the parameter @xmath141 from the bec to the bcs regimes . superfluids provide the simplest nontrivial application of general coordinate invariance . at long distance the only physical degree of freedom is the phase of the condensate . we shall therefore construct an effective field theory of the this phase . for a system consisting of bosons , the field is defined as the phase of the microscopic field , @xmath142 for fermions , we define @xmath143 as half of the phase of the cooper pair , @xmath144 frequently , one describes a system at finite chemical potential @xmath11 by including the term @xmath145 into the lagrangian . this approach is not suitable for our purposes since this term breaks general coordinate , scale and conformal symmetries . instead we shall write the lagrangian without the @xmath145 term . the chemical potential enters the problem through the choice of the ground state . in particular , the superfluid ground state of a system at chemical potential @xmath11 corresponds to @xmath146 . we will frequently expand @xmath147 where fluctuations of @xmath10 around zero correspond to phonon excitations . the breaking of symmetries is now _ spontaneous _ , i.e. , it comes from the choice of the ground state ( [ thetavarphi ] ) . let us first discuss the theory in free flat space ( @xmath121 , @xmath42 ) . the effective field theory is formulated as a lagrangian @xmath148 . this lagrangian has a shift symmetry @xmath149 , due to the global symmetry with respect to phase rotation of @xmath77 . due to this symmetry , @xmath143 does not appear without derivative . so when we expand the lagrangian in power series over derivatives and over @xmath143 , we encounter in each term at least as many derivatives as @xmath143 . consider a few representative term in this expansion : @xmath150 the most naive power counting scheme is to consider @xmath151 and each derivative as @xmath152 . in this scheme , the first term in ( [ l - symb - exp ] ) is @xmath153 , while the last two terms are @xmath154 . however , the power counting scheme that is most useful is to count @xmath155 and @xmath156 as @xmath157 , and additional time and space derivatives as @xmath152 . this is possible since , as we have seen , @xmath143 always appears with at least one derivative . in this second power counting scheme , the first two terms in ( [ l - symb - exp ] ) are of order 1 while the last term is @xmath153 . a generic term in the which contains @xmath158 $ ] time derivatives , @xmath159 $ ] spatial derivatives , and the @xmath160 $ ] power of the field will be counted as @xmath161 } { \partial}_i^{n[{\partial}_i ] } \theta^{n[\theta ] } \sim p^{n[{\partial}_t]+n[{\partial}_i]-n[\theta ] } .\ ] ] in the second power counting scheme , the leading order effective lagrangian sums up all terms where the power of @xmath143 is equal to the total number of space and time derivatives . thus it is a function of first derivatives of @xmath143 , @xmath162 which may contain , upon series expansion , terms with any power of @xmath163 . so far we have not discussed the background fields @xmath37 , @xmath38 , and @xmath32 . since @xmath37 and @xmath38 appears with in covariant derivatives , it is natural to assume that all fields scale as @xmath157 and their first derivatives as @xmath152 . in fact , in a trap where one has to treat the trapping potential @xmath37 as a quantity of order 1 , the second power counting scheme with @xmath164 is the only consistent one . the effective theory should inherit the general coordinate invariance of the microscopic theory . this means that the effective action is invariant under the general coordinate transformations . since @xmath165 we need the lagrangian density to transform as @xmath166 so that the action is invariant @xmath167 equation ( [ deltalscalar ] ) means that @xmath168 has to transform as a scalar under general coordinate transformations . we shall assume that the microscopic theory is invariant under time reversal , which means that the effective theory is invariant under @xmath169 this time reversal invariance has an important consequences for the momentum expansion of the eft . this invariance implies that in each term , the difference between of the number of time derivatives and the number of @xmath143 is even , @xmath170 - n[\theta ] = \textrm{even } .\ ] ] due to rotational symmetry and parity ( which forbids @xmath171 ) , the number of spatial derivatives have to be even , since the spatial indices have to contract pairwise . therefore we have @xmath170 + n[{\partial}_i ] - n[\theta ] = \textrm{even},\ ] ] which means that the terms in the effective lagrangian are proportional to even powers of @xmath172 . therefore the next - to - leading order terms in the lagrangian are @xmath153 , the next - to - next - to - leading terms are @xmath154 , etc . let us now turn on @xmath173 and the curved metric . the leading - order effective lagrangian , as we have discussed , should be a function of a finite number of variables which are @xmath174 : @xmath175 gauge invariance and invariance with respect to three - dimensional general coordinate transformations ( with time - independent @xmath50 ) limit the lagrangian to be a function of two variables , @xmath176 where @xmath177 now we require the invariance of the effective theory with respect to general coordinate transformations @xmath178 under ( [ nonrel - gci - theta ] ) we have @xmath179 this means that only one combination of these two variables transforms as a scalar , which is @xmath180 and the leading - order effective lagrangian has to be a function of this combination , @xmath181 if we now set the metric to flat and external fields to zero , then we find the most general form of the lagrangian for superfluids , @xmath182 which was previously found by greiter , wilczek , and witten @xcite using a different line of arguments . we now show that the function @xmath183 in the leading - order lagrangian is identical to the function that defines , at zero temperature , the pressure as a function of the chemical potential , @xmath184 . for that purpose , we notice that the particle number density can be found by differentiating the lagrangian with respect to @xmath37 , and is equal to @xmath185 the ground state at a chemical potential @xmath11 corresponds to @xmath146 , which means @xmath186 . this implies @xmath183 is a function satisfying @xmath187 where @xmath188 is the equilibrium number density at chemical potential @xmath11 . that means @xmath183 is the thermodynamic pressure , up to a constant . the constant is not important in any situation and will be set to zero . it is also possible to obtain the lagrangian ( [ gww - l ] ) by taking the nonrelativistic limit of the effective lagrangian of relativistic superfluids . the field appearing in the effective lagrangian is the phase @xmath189 of the relativistic field @xmath68 ( see eq . ( [ relat - s ] ) , @xmath190 from eq . ( [ psipsi ] ) , the relation between @xmath189 and the nonrelativistic phase @xmath143 is @xmath191 to leading order , the effective lagrangian is a function of @xmath192 . due to lorentz invariance , it has to have the form @xmath193 from eqs . ( [ phiphi ] ) and ( [ ginv ] ) one finds @xmath194 + o(c^{-2 } ) = \frac{m^2c^2}2 + mx + o(c^{-2}).\ ] ] so , in the nonrelativistic limit the lagrangian is a function of @xmath195 , as found in eq . ( [ lx ] ) . it is shown in ref . @xcite that the function @xmath196 coincides with pressure at chemical potential equal to @xmath197 . this is consistent with our previous result , since the difference between the relativistic and nonrelativistic normalizations of the chemical potential is the rest energy @xmath198 . for completeness , we include here a discussion of the relation of the lagrangian ( [ gww - l ] ) and its derivation with other , more frequently used , approaches . we also derive some well - known results using the effective lagrangian . galilean invariance is sufficient to obtain the leading order lagrangian . we recall that the galilean transformations ( [ alphaxigal ] ) are particular cases of general coordinate transformations . the galilean transformations leave flat metric unchanged . under these transformations , @xmath199 from which we see that the only combination which is rotationally and galilean invariant is @xmath200 . a natural question arises : does galilean invariance in general guarantee general coordinate invariance ? the answer is no . we shall see in sec . [ sec : nlo ] that at the next - to - leading order , galilean invariance is not sufficient to constrain the form of the effective lagrangian . from eqs . ( [ gww - l ] ) and ( [ thetavarphi ] ) we can derive the field equation for @xmath10 , @xmath201 in eq . ( [ continuity ] ) @xmath3 is the equilibrium number density at chemical potential equal to @xmath202 . this is the continuity equation of superfluid hydrodynamics , if we identify the superfluid velocity with the gradient of @xmath143 , @xmath203 in superfluid hydrodynamics one regards @xmath3 and @xmath10 as independent variables . the second equation of superfluid hydrodynamics describes time evolution of @xmath10 comes from the definition of @xmath3 , @xmath204 here @xmath205 is the chemical potential at which the number density is equal to @xmath3 . we can determine the phonon spectrum by expanding the lagrangian ( [ gww - l ] ) to quadratic order in @xmath10 ( recall that @xmath206 ) . the lagrangian is then @xmath207 the first term is constant , and the second term is a full derivative and does not contribute to the field equation . the two last terms determine the phonon speed to be @xmath208 in performing the last transformation we have use the thermodynamic relation @xmath209 and the definition of mass density @xmath210 . the speed of sound is given by the same formula as in normal fluids ( however here the sound is not collisional . ) expanding the lagrangian further , one can find the interaction vertices . the interaction between phonons , at leading order in momentum , is determined solely from the equation of state . this nontrivial result is actually not new . its relativistic equivalence was derived in ref . @xcite . in the nonrelativistic setting , it was known in the hamiltonian formalism , which we discuss next . from the effective lagrangian ( [ gww - l ] ) one can derive the equivalent hamiltonian . the variable conjugate to the phase @xmath143 is the number density , @xmath211 the equal - time commutation relation between @xmath3 and @xmath143 is @xmath212 = -i \delta({\mathbf{x}}-{\mathbf{y}}).\ ] ] the hamiltonian is then @xmath213 note that the function @xmath214 is the legendre transform of @xmath215 . since @xmath215 is the pressure as function of chemical potential , its legendre transform is energy as function of number density @xmath216 . therefore , @xmath217 the hamiltonian is the sum of the kinetic energy @xmath218 and the internal energy not associated with the macroscopic flow , @xmath216 . this hamiltonian is well known . it was used , e.g. , to find the interaction between phonons @xcite . the lagrangian @xmath219 is sometimes called popov s hydrodynamic lagrangian @xcite . the equation of state of a unitary fermi gas is known up to a constant . due to the absence of any internal scale , by dimensional analysis the pressure has to be of the form @xmath220 where @xmath13 is a dimensionless variable . the leading - order effective action is then , @xmath221 the fractional power in eq . ( [ s - conf ] ) is not problematic , since we always expand around the ground state @xmath146 . as expected , the lagrangian ( [ s - conf ] ) is invariant under infinitesimal conformal transformations @xmath222 taking the legendre transform of @xmath223 , we find the energy density as a function of the number density . this is proportional to the energy density of a free fermi gas at the same density , with the coefficient of proportionality @xmath1 , @xmath224 the parameter @xmath1 is estimated to be around @xmath225 from quantum monte carlo simulations @xcite ( but see also ref . we can express @xmath13 via @xmath1 , @xmath226 at large , but finite , scattering length @xmath64 the effective lagrangian contains an extra term proportional to @xmath227 , @xmath228 where @xmath229 is some constant . the term proportional to @xmath229 violates scale and conformal symmetries explicitly . the leading - order effective lagrangian allows one to compute all quantities of interest , e.g. , scatterings amplitudes or correlation functions , to leading order in momentum . the corrections to these leading - order results come from two sources . first , the lagrangian contains terms subleading in derivatives ; these terms contribute , at tree level , subleading contributions to the scattering amplitudes and correlation function . as explained above , these corrections start at @xmath153 order . the second source of corrections are loop diagrams , constructed using the leading - order lagrangian . we shall demonstrate here that the loop contributions start at order @xmath154 , i.e. , are comparable to those coming from the next - to - next - to - leading order terms in the effective lagrangian . therefore , if one is interested in the leading @xmath157 and next - to - leading @xmath153 results , one can work at tree level . to estimate the size of loop corrections , it is useful to work in the unit system where the sound speed is @xmath230 , so energy and momentum has the same dimension . the dimension of the lagrangian is 4 ( momentum@xmath231 ) . we also rescale @xmath10 so that the quadratic terms in the lagrangian is normalized canonically . the dimension of @xmath10 is then 1 . the interaction terms will have coefficients which can be estimated from dimensional analysis , @xmath232 here @xmath11 is the chemical potential and all dimensionless coefficients are , in general , of order one . as an example of a loop diagram , consider a one - loop correction to the phonon dispersion relation ( fig . [ fig : selfe ] ) . the graph is divergent and has to be regularized . after dimensional regularization it becomes a finite function of the external momentum @xmath172 , with a possible logarithmic singularity . however we know that the graph contains two three - phonon vertices and hence is proportional to @xmath233 . the self - energy has dimension 2 . therefore we conclude that @xmath234 which is suppressed by @xmath235 compared to the tree - level result @xmath153 . consider another example , which is a one - loop correction to phonon - phonon scattering amplitude ( fig . [ fig : scatt ] ) . the scattering amplitude is @xmath236 at tree level . the one - loop graph contains two four - phonon vertices and hence is proportional to @xmath237 . by dimension , it has to behave as @xmath238 as a function of the external momenta @xmath172 . the result is by a factor of @xmath235 smaller than the tree level . generalizing to more complicated diagrams and processes , we see that we have to compute loops only when we want to know the answer to order @xmath239 compared to the leading - order result . if we want to know the answer only to the leading and next - to - leading order , tree - level calculation using the lagrangian with subleading terms should suffice . to find all terms allowed to the next - to - leading order , it is most convenient to work in the relativistic theory and then take the nonrelativistic limit . in the relativistic theory , the next - to - leading terms contain two derivatives more than the field @xmath189 . they have the form @xmath240 , @xmath241 , where @xmath242 is an arbitrary function of x , and @xmath243 are lorentz scalars which are of order @xmath244 : @xmath245 here @xmath246 is the covariant derivative , @xmath247 , @xmath248 is the ricci tensor , and @xmath249 ( see ref . @xcite and appendix [ sec : ricci ] ) . the terms proportional to the ricci tensor are @xmath153 since @xmath248 contains second derivatives of @xmath250 . we now use the field equation to exclude redundant terms from ( [ list ] ) . to leading order , where the lagrangian is @xmath251 , the field equation can be written as @xmath252 ; after adding @xmath153 terms ( [ list ] ) to the lagrangian , there will be higher corrections to this equation : @xmath253 by using the field equation one can express @xmath254 and @xmath255 through @xmath256 : @xmath257 so they are not independent . in addition , one term from ( [ list ] ) can be written as a linear combination of other terms , up to a full derivative , @xmath258.\end{gathered}\ ] ] therefore there are only four independent terms remaining from the original seven , and @xmath259 where @xmath260 , @xmath261 , @xmath262 , and @xmath263 are arbitrary functions . now all we have to do is to take the nonrelativistic limit of the terms appearing in ( [ lnlo - rel ] ) . consider the @xmath260 term . notice that @xmath264 therefore @xmath265 for the @xmath261 term , we first recall that @xmath266 therefore @xmath267 the coefficient in front of @xmath262 can be expanded as @xmath268 where all three terms survive in the limit @xmath74 since @xmath269 , @xmath270 , and @xmath271 [ see appendix [ sec : ricci ] where the components of the ricci tensor are computed for the nonrelativistic metric ( [ gag ] ) ] . the coefficient in front of @xmath263 is the ricci scalar , which reduces to the ricci scalar of the 3d space in the nonrelativistic limit after neglecting terms of order @xmath272 and higher : @xmath273 thus , we arrive to the most general nlo lagrangian consistent with general coordinate invariance , @xmath274 where @xmath275 ^ 2 f_2(x),\\ \begin{split } { \mathcal{l}}_3 & = \biggl\{-\frac{m^2}4 ( 2 g^{ij}\ddot g_{ij } + \dot g^{ij}\dot g_{ij } ) -m \nabla_i e^i + \frac14 f^{ij}f_{ij}\\ & \qquad + \big[m{\partial}_t{\partial}_i\ln g -\nabla_j(mg^{jk}\dot g_{ki } + g^{jk}f_{ki } ) \big ] d^i\theta + r_{ij } d^i\theta d^j\theta \biggr\ } f_3(x ) , \end{split}\\ { \mathcal{l}}_4 & = r_{\rm 3d } f_4(x),\end{aligned}\ ] ] where @xmath276 , @xmath277 , and raising / lowering indices and covariant derivatives are done using the 3d metric @xmath32 . now having made use of the general coordinate invariance , we dispense with the curved space and set the metric to flat , @xmath42 . the nlo lagrangian becomes much simpler , @xmath278 the @xmath279 term is not important in flat space . if one retains only @xmath37 and sets @xmath41 ( which would preclude one from calculating the transverse response function , see sec . [ sec : applications ] ) , then the lagrangian is even simpler @xmath280 we have mentioned above that , at the leading order , general coordinate invariance is not more powerful than galilean invariance : both yield the greiter - wilczek - witten lagrangian . however , at the next to leading order , general coordinate invariance is more restrictive than galilean invariance . in other worlds , one can write down galilean invariant terms which are not general coordinate invariant . to see that , we write down the infinitesimal galilean transforms ( recall that we are in flat space ) , @xmath281 all three terms in eq . ( [ lnlo - flat ] ) are galilean invariant , as they should be since general coordinate invariance implies galilean invariance . however , one can directly check that each of the following three expressions is galilean invariant , @xmath282 but , in fact , the third expression does not enter the lagrangian ( [ lnlo - flat ] ) at all , and the first two enter the lagrangian only through one particular linear combination . therefore if one relies only on galilean invariance one would conclude incorrectly that the nlo lagrangian involves five arbitrary functions , instead of three . when one only keeps the trapping potential @xmath37 , setting @xmath41 , general coordinate invariance allows one to exclude a term of the form @xmath283 from appearing in the lagrangian ( [ lnlo - trap ] ) . unitary fermi gas does not possess an intrinsic scale , therefore the functions @xmath284 , @xmath285 above should be powers of @xmath195 . requiring the invariance of the action under scale transformations @xmath286 , @xmath287 , one finds that @xmath288 the powers of @xmath195 can be found from counting the scaling dimension and requiring that @xmath112=\frac52 $ ] ( note that @xmath289=1 $ ] ) ; the powers of @xmath4 are written to make @xmath290 dimensionless . therefore , instead of three arbitrary functions of @xmath195 , the nlo lagrangian for unitary fermi gas involves three dimensionless phenomenological _ constants _ @xmath17 , @xmath18 , and @xmath291 ( four in curved space , with @xmath292 ) . remarkably , conformal invariance further reduces the number of independent constants to two ( three in curved space ) . performing the infinitesimal transformation ( [ conf - theta ] ) , we find that @xmath293 and @xmath294 are separately invariant . indeed , their expressions contain only spatial derivatives of elementary fields , and thus have conformal dimension equal to the scaling dimension . however , @xmath295 and @xmath296 contain time derivatives of the metric tensor and hence do not have well - defined conformal dimension . thus @xmath297 and @xmath298 change under conformal transformations , @xmath299 x^{1/2 } , \\ \delta s_3 & = -\frac{c_3}2 m^{3/2 } \!\int\!dt\,d{\mathbf{x}}\,\sqrt g\ , \dddot\beta x^{1/2}. \label{deltas3}\end{aligned}\ ] ] a closer investigation reveals that the two contributions cancel each other for a particular ratio of @xmath18 and @xmath291 . to see that , we perform integration by part in eq . ( [ deltas3 ] ) , and make use of the leading - order field equation , @xmath300 to obtain @xmath301.\ ] ] the contribution coming from the last term in the brackets can be further integrated by parts , @xmath302 one thus finds @xmath303 x^{1/2}.\ ] ] therefore , conformal symmetry can be satisfied if one requires @xmath304 ( in @xmath305 spatial dimensions we would have @xmath306 . ) thus , we find that the nlo effective lagrangian of unitary fermi gas in flat space and in the presence of background gauge fields involves only two independent additional constants , @xmath307 x^{1/2},\end{gathered}\ ] ] where @xmath308 . if we retain only the trapping potential @xmath37 and set @xmath41 , then the effective lagrangian becomes @xmath309 x^{1/2}.\ ] ] the two parameters @xmath17 and @xmath18 are universal constants and need to be determined from microscopic computations . in sec . [ sec : applications ] we shall see how these coefficients appear in physical quantities , which suggests methods to extract them . we will also show that @xmath18 is positive . in this section , we compute some physical quantities using the nlo effective lagrangian . our aim is not to go through an exhaustive list of applications , but just to demonstrate the predictive powers of the eft . further calculations are deferred to future work . in this section we use the unit system @xmath310 . consider an infinite system with no trapping potential @xmath121 . expanding the effective lagrangian to quadratic order in @xmath10 , we find @xmath311.\ ] ] the phonon dispersion relation is found by solving the equation @xmath312 the solution is @xmath313 using @xmath314 and eq . ( [ c0xi ] ) , we can write the result in the form @xmath315.\ ] ] we reproduce the linear part of the phonon dispersion with the velocity @xmath316 , but we also find the nonlinear @xmath22 term in the dispersion . this correction depends on the combination @xmath317 . in particular , if @xmath318 the dispersion curves up and a low - energy phonon can decay into two phonons . in the opposite case such a process is kinematically forbidden . the eft allows one to compute any correlation functions in the long - wavelength , low - frequency regime . here we concentrate on the static response functions , as these functions are most readily computable using monte - carlo techniques . the static density response function @xmath319 is defined through the response of the medium to a small potential @xmath37 @xcite . if one introduces the time - ordered density - density correlation function , @xmath320 then @xmath321 it is related to the dynamic structure function @xmath322 , @xmath323 through a dispersion relation @xmath324 at zero momentum the static response function is fixed by the compressibility sum rule , @xmath325 we shall now compute the leading correction at small @xmath326 . from the general formalism , the density correlation function can be obtained by differentiating the generating functional @xmath89 , @xmath327 } { \delta a_0(t,{\mathbf{x}})\delta a_0(0,\mathbf{0})}\bigg|_{a_0=a_i=0}.\ ] ] in eft @xmath89 can be found from diagrammatic techniques . since to the nlo we have only tree graphs , @xmath89 coincides with the classical action at the saddle point , @xmath328 = s[\varphi_{\rm cl } , a_0 , a_i],\ ] ] where @xmath329 is the solution to the field equation in the background @xmath37 , @xmath38 . to compute the static density response function one can keep @xmath41 and turn on only a static @xmath37 . the leading - order field equation ( [ loeq ] ) has a trivial solution @xmath330 on this background . if one is interested in the nlo action one can just insert this solution into the nlo lagrangian . the lagrangian density for this solution is @xmath331 = c_0 ( \mu - a_0)^{5/2 } + \frac{c_1}{\sqrt{\mu - a_0 } } ( { \partial}_i a_0)^2 - 9c_2 ( \nabla^2 a_0)\sqrt{\mu - a_0}\,,\ ] ] from which we find @xmath332.\ ] ] at @xmath333 the compressibility sum rule is reproduced . note that the @xmath334 correction is proportional to a different linear combination of @xmath17 and @xmath18 compared to the phonon dispersion relation . the transverse response function measures the `` paramagnetic '' current @xmath335 induced by a transverse static background field @xmath336 : @xmath337 the `` paramagnetic '' current @xmath338 in this formula is defined without the `` diagmagnetic part '' @xmath339 , @xmath340 therefore if we define the transverse current - current correlator , @xmath341 then the transverse response function is related to @xmath342 in the following way : @xmath343 at zero temperature , the static transverse response function approaches zero in the low momentum limit @xmath344 a fact that corresponds to the vanishing of the normal density . there is a dispersion relation relating @xmath345 with the dynamic transverse structure function @xmath346 : @xmath347 where @xmath348 is positive , so @xmath349 . the transverse response function can be computed from a formula analogous to ( [ gnnw ] ) , where @xmath37 is replaced by the @xmath350 . again in the presence of transverse static @xmath38 the field equation yields @xmath351 , and the lagrangian density is simply @xmath352= c_0 \left(\mu-\frac{\mathbf{a}_\perp^2}2\right)^{5/2 } + c_2 \sqrt\mu\left ( -\frac94 f_{ij } f_{ij } -9 { \partial}_i f_{ij } a_j\right ) + o(a_\perp^4).\ ] ] we find @xmath353 but since @xmath349 we conclude that @xmath354 . consider unitary fermi gas in an asymmetric harmonic trap with the potential @xmath355 to leading order , the ground state at chemical potential @xmath11 is @xmath146 , and the lagrangian density , evaluated for this configuration and integrated over space , is the free energy @xmath356 in eq . ( [ ftrap ] ) the integral is taken over the volume occupied by the fermion cloud , which is determined by @xmath357 . taking the integrals one obtains @xmath358 where @xmath28 . the particle number as a function of the chemical potential is found by differentiating @xmath359 , @xmath360 , and the energy as a function of particle number is the legendre transform of @xmath361 : @xmath362 note that @xmath17 and @xmath18 appear in the same combination as they appear in the static density response function ( [ chic1c2 ] ) . we were somewhat fortunate that all integrals in eq . ( [ ftrap ] ) converge . we would not be so lucky beyond the nlo . indeed , consider as an example a @xmath363 term in the effective lagrangian , @xmath364 this term respects both general coordinate and conformal invariance . the contribution to the free energy in a trap would be @xmath365 ^ 2}{(\mu - a_0)^{7/2}}\,.\ ] ] for simplicity let us assuming a spherical trap , @xmath366 . denote the distance to the boundary of the fermion cloud as @xmath367 , @xmath368 then the integral in ( [ nnloint ] ) , in the region of small @xmath367 , looks like @xmath369 clearly , the integral diverges near @xmath370 . this divergence can be traced to the breakdown of the derivative expansion near the boundary of the cloud . indeed , near the boundary the effective chemical potential @xmath371 changes substantially over distances of order @xmath367 . when @xmath367 is comparable to the fermi wavelength , i.e. , @xmath372 the expansion parameter in the eft becomes of order one . in other words , there is a boundary layer with the thickness determined by eq . ( [ skin ] ) where the eft theory break down . one can estimate ( but not calculate ) the contribution of this region to the free energy by placing ( [ skin ] ) as a cutoff in ( [ intdelta ] ) , and find @xmath373 the eft , in principle , can not give an answer with an error smaller than @xmath374 . this uncertainty translates into an uncertainty of order @xmath375 in eq . ( [ en ] ) . note that this is smaller than the nlo correction by only a factor of @xmath376 . in this paper we have discussed general coordinate invariance and conformal invariance in the nonrelativistic many - body theory . we also extract physical consequences of these symmetries for the unitary fermi gas . we find that to the leading and next - to - leading orders in the momentum expansion we need three constants @xmath13 ( or @xmath1 ) , @xmath17 and @xmath18 to completely characterize the dynamics of the system and its response to external probes . while we have some numerical estimates for @xmath13 from monte carlo measurements of thermodynamics , we still lack estimates for @xmath17 and @xmath18 . we propose to extract @xmath17 and @xmath18 by finding the density and transverse response functions in the low - momentum region using monte carlo techniques and matching the results with eqs . ( [ sum : response ] ) . the effective lagrangian which we have derived can be used to compute many physical quantities . one interesting application is the computation of the kinetic coefficients of a unitary fermi gas at low temperatures . the knowledge of these coefficients is important for computing the damping rate of oscillations of the trapped gas at low temperatures @xcite . taking a broader perspective , we expect the general coordinate invariance to have a wide range of applications , since it is a common property of all nonrelativistic systems . in the paper we have considered only systems with one species of particles , but it is straightforward to generalize the general coordinate transformations ( [ nonrel - gci ] ) for systems consisting of many species of particles . for example , in the case of a @xmath231he@xmath377he mixture one should introduce two gauge fields , one ( say , @xmath173 ) coupled to the @xmath231he current and another ( @xmath378 ) to the @xmath377he current . the general coordinate transformations contain two copies of eqs . ( [ nonrel - gci - a0 ] ) and ( [ nonrel - gci - ai ] ) ; in particular we have @xmath379 where @xmath380 and @xmath381 are the masses of the @xmath231he and @xmath377he atoms , respectively . on the other hand , conformal invariance , at least in three spatial dimensions , seems to be specific to the unitary fermi gas . it would be interesting to explore conformal invariance in lower - dimensional systems , for example for the anyon gas in two dimensions , which has been suggested to be a superfluid @xcite and respects scale invariance @xcite . * acknowledgments * we thank eric braaten , hans hammer , and misha stephanov for discussions , and roman jackiw and peter stichel for comments . this work is supported , in part , by doe grant number de - fg02 - 00er41132 . d.t.s . is supported , in part , by the alfred p. sloan foundation . our convention for the christoffel symbols and the ricci tensor is @xmath382 the covariant derivatives of scalars and vectors are given by @xmath383 to derive eq . ( [ q5 ] ) the following equation is used : @xmath384 for the metric of the form ( [ gag ] ) , one finds @xmath385 where @xmath386 and @xmath387 are the electric and magnetic fields constructed from the field potential @xmath388 : @xmath389 the completely spatial components of @xmath390 coincides with the christoffel symbols of the three - dimensional space with metric @xmath32 . for the ricci tensor , one has @xmath391.\end{aligned}\ ] ] again , the spatial components @xmath392 coincides with the ricci tensor of the three - dimensional space with metric @xmath32 . in particular , in the limit @xmath74 , the 4d ricci scalar @xmath393 coincides with the 3d counterpart . m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j. hecker denschlag , and r. grimm , `` crossover from a molecular bose - einstein condensate to a degenerate fermi gas , '' phys . lett . * 92 * , 120401 ( 2004 ) [ arxiv : cond - mat/0401109 ] . m. w. zwierlein , c. a. stan , c. h. schunck , s. m. f. raupach , a. j. kerman , and w. ketterle , `` condensation of pairs of fermionic atoms near a feshbach resonance , '' phys . 92 * , 120403 ( 2004 ) . t. bourdel , l. khaykovich , j. cubizolles , j. zhang , f. chevy , m. teichmann , l. tarruell , s. j. j. m. f. kokkelmans , and c. salomon , `` experimental study of the bec - bcs crossover region in lithium 6 , '' phys . lett . * 93 * , 050401 ( 2004 ) [ arxiv : cond - mat/0403091 ] . see , e.g. , s .weinberg , _ the quantum theory of fields . ii : modern applications _ ( cambridge university press , cambridge , 1996 ) , chapter 19 . j. lukierski , p. c. stichel , and w. j. zakrzewski , `` from gauging nonrelativistic translations to n - body dynamics , '' ann . phys . ( n.y . ) * 288 * , 164 ( 2001 ) [ arxiv : hep - th/0007102 ] , and references therein . r. jackiw and s. y. pi , `` finite and infinite symmetries in ( 2 + 1)-dimensional field theory , '' arxiv : hep - th/9206092 , nucl . suppl . * 33c * ( 1993 ) 104 . modern kaluza - klein theories , _ edited by t. appelquist , a. chodos , and p. g. o. freund ( addison - wesley , menlo park , 1987 ) . m. greiter , f. wilczek , and e. witten , `` hydrodynamic relations in superconductivity , '' mod . b * 3 * , 903 ( 1989 ) . c. r. hagen , `` scale and conformal transformations in galilean - covariant field theory , '' phys . d * 5 * , 377 ( 1972 ) . t. mehen , i. w. stewart , and m. b. wise , `` conformal invariance for non - relativistic field theory , '' phys . b * 474 * , 145 ( 2000 ) [ arxiv : hep - th/9910025 ] . d. t. son , `` low - energy quantum effective action for relativistic superfluids , '' arxiv : hep - ph/0204199 . e. m. lifshitz and l. p. pitaevskii , _ statistical physics part 2 _ ( pergamon press , 1980 ) . g. e. astrakharchik , j. boronat , j. casulleras , and s. giorgini , `` equation of state of a fermi gas in the bec - bcs crossover : a quantum monte - carlo study , '' phys . lett . * 93 * , 200404 ( 2004 ) [ arxiv : cond - mat/0406113 ] . d. lee and t. schfer , `` cold dilute neutron matter on the lattice ii : results in the unitary limit , '' arxiv : nucl - th/0509018 . l. d. landau and e. m. lifshitz , _ the classical theory of fields _ ( pergamon press , 1975 ) . d. pines and p. nozires , _ the theory of quantum liquids . volume i : normal fermi liquids _ ( benjamin , new york , 1966 ) ; _ the theory of quantum liquids . volume ii : superfluid bose liquids _ ( reading , massachusetts , 1989 ) . m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j. hecker denschlag , and r. grimm , `` collective excitations of a degenerate gas at the bec - bcs crossover , '' phys . rev . lett . * 92 * , 203201 ( 2004 ) [ arxiv : cond - mat/0403716 ] . see , e.g. , f. wilczek , _ fractional statistics and anyon superconductivity _ ( world scientific , singapore , 1990 ) . o. bergman and g. lozano , `` aharonov - bohm scattering , contact interactions and scale invariance , '' ann . * 229 * , 416 ( 1994 ) [ arxiv : hep - th/9302116 ] .
we investigate the fermions of the standard model without a higgs scalar . instead , we consider a non - local four - quark interaction in the tensor channel which is characterized by a single dimensionless coupling . quantization leads to a consistent perturbative expansion for small . the running of is asymptotically free and therefore induces a non - perturbative scale , in analogy to the strong interactions . we argue that spontaneous electroweak symmetry breaking is triggered at a scale where grows large and find the top quark mass of the order of . we also present a first estimate of the effective yukawa coupling of a composite higgs scalar to the top quark , as well as the associated mass ratio between the top quark and the w boson . _ institut fr theoretische physik , philosophenweg 16 , 69120 heidelberg , germany _ _ e - mail : [email protected] , [email protected]_
0812.4223
the large hadron collider ( lhc ) will soon test the mechanism of spontaneous electroweak symmetry breaking . it is widely agreed that this phenomenon is associated to the expectation value of a scalar field which transforms as a doublet with respect to the weak su(2)-symmetry . the origin and status of this order parameter wait , however , for experimental clarification . in particular , an effective description in terms of a scalar field does not tell us if this scalar is fundamental " in the sense that it constitutes a dynamical degree of freedom in a microscopic theory which is formulated at momentum scales much larger than the fermi scale . alternatively , no fundamental higgs scalar may be present , and the order parameter rather involves an effective composite field " . in this note we investigate the second possibility and therefore consider the fermions of the standard model without a fundamental higgs scalar . as long as we include only the gauge interactions of the standard model , the microscopic or classical action of such a theory does not involve any mass scale . an effective mass scale @xmath2 will only be generated by the running of the strong gauge coupling , which is asymptotically free @xcite . confinement removes then the gluons and quarks from the massless spectrum . in addition , the spontaneous chiral symmetry breaking by quark - antiquark condensates would also imply electroweak symmerty breaking by a composite order parameter , in this case the chiral condensate . in this setting all particle masses would be of the order @xmath2 or zero , such that this scenario can not explain why the top quark mass or the @xmath3,@xmath4-boson masses are much larger than 1gev . furthermore , all leptons would remain massless . one concludes that any realistic model needs further interactions beyond the standard model gauge interactions . since the strong and electroweak gauge couplings remain actually quite small at the fermi scale of electroweak symmetry breaking , they are expected to produce only some quantitative corrections to the dominant mechanism of electroweak symmetry breaking . we will therefore neglect the gauge couplings in this paper . our knowledge about the effective interactions between the quarks and leptons at some microscopic or ultraviolet " scale @xmath5 is very limited . furthermore , it is not known which scale @xmath5 has to be taken . typically , one may associate @xmath5 with the scale where further unification takes place , as a grand unified scale or the planck scale for the unification with gravity . this would suggest a very high scale , @xmath6 gev and we will have this scenario in mind for our discussion . however , much smaller values of @xmath5 are also possible . in practice , we will only assume here that @xmath5 is sufficiently above the fermi scale ( say @xmath7 tev ) , such that an effective description involving only the fermions of the standard model becomes possible in the momentum range @xmath8 . we will formulate our model in terms of effective fermion interactions at the scale @xmath5 and restrict the discussion to a four - fermion interaction involving only the right - handed top quark and the left - handed top and bottom quarks . this is motivated by the observation that only the top quark has a mass comparable to the fermi scale . interactions with the other quarks and leptons are assumed to be much smaller than the top quark interactions - typically their relative suppression is reflected in the much smaller masses of the other fermions . for the discussion in this paper we omit all light " fermions and the gauge bosons . since we do not know the effective degrees of freedom at the scale @xmath5 , the effective interaction is not necessarily local . non - localities involving inverse powers of the exchanged momenta are typically generated by the propagators of exchanged massless particles . with the inclusion of possibly non - local interactions the limitation to an effective four - fermion interaction poses no severe restriction . many models with additional degrees of freedom can be effectively described in this way . local four - fermion interactions have already been investigated earlier , for example in the models of top quark condensation " @xcite . by simple dimensional analysis a local interaction involves a coupling @xmath9(mass)@xmath10 . such models therefore exhibit an explicit mass scale in the microscopic action . it is indeed possible to obtain spontaneous electroweak symmetry breaking in this way - the prototype is the nambu - jona - lasinio model @xcite . without a tuning of parameters the top quark mass @xmath11 turns out , however , to be of the same order as @xmath5 , in contradiction to the assumed separation of scales . by a tuning of parameters it is possible to obtain @xmath12 , but the issue is now similar to the gauge hierarchy problem " in presence of a fundamental scalar field . in order to obtain @xmath12 the microscopic effective action must be close to an ultraviolet fixed point . the necessity of tuning arises from a relevant parameter " in the vicinity of the fixed point ( in the sense of statistical physics for critical phenomena ) which has a dimension not much smaller than one . a rather extensive search for possible ultraviolet fixed points for pointlike four - fermion interactions has been performed in @xcite . many fixed points have been found , but all show a relevant direction with substantial dimension , and therefore the need for a parameter tuning for @xmath12 . we will therefore concentrate in this paper on non - local effective interactions . for such interactions the coupling @xmath13 is replaced by @xmath14 , with @xmath15 the square of some appropriate exchanged momentum and @xmath0 a dimensionless coupling . interactions of this type do not involve a mass scale on the level of the microscopic action - the classical action exhibits dilatation symmetry . still , the quantum fluctuations typically induce an anomaly for the scale symmetry , associated to the running of the dimensionless coupling @xmath0 . this may be responsible for the generation of the fermi scale , in analogy to the confinement scale " @xmath2 for qcd . since the running of dimensionless couplings is only logarithmic , this offers a chance for a large natural hierarchy @xmath16 , without tuning of parameters . we will present a model of this type . let us first discuss the possible tensor structures for a non - local four - fermion interaction @xmath17 . the basic building block is a color singlet fermion bilinear @xmath18 , where the color indices are contracted . the tensor structure with respect to the lorentz symmetry is determined by @xmath19 such that @xmath18 is a scalar or pseudoscalar , a vector or pseudovector , or a second rank antisymmetric tensor . interactions in the vector or pseudovector channels involve bilinears with two left - handed or two right - handed fermions , @xmath20 or @xmath21 . they conserve chiral flavour symmetries which act separately on @xmath22 and @xmath23 and therefore forbid mass terms for the fermions @xmath24 . interactions capable of producing masses for the top quark and w / z - bosons of comparable magnitude must therefore involve the ( pseudo)scalar or tensor channels . a non - local scalar interaction @xmath25 , with @xmath15 the squared momentum in the scalar exchange channel , has very similar properties as a model with a fundamental higgs scalar which is massless at the scale @xmath5 . we therefore expect the usual necessity of parameter tuning if we want to achieve a small ratio @xmath26 . a local coupling @xmath27 is allowed by the symmetries and will be generated by quantum fluctuations . the interesting remaining candidate is a tensor interaction , with @xmath28 $ ] . for chiral tensors no local interaction in this channel is consistent with the su(3)@xmath29su(2)@xmath29u(1 ) symmetries of the standard model as well as lorentz symmetry . we will therefore investigate a model with a non - local interaction of this type . we define the microscopic or classical action for a lorentz invariant theory of massless interacting fermions by @xmath30 , defined in momentum space as @xmath31 and @xmath32 \ ; \left [ \bar{t}(p')\sigma_-^l t(p'+q)\right ] \\ \nonumber & + & \left [ \bar{t}(q+p)\sigma_+^k b(p)\right ] \ ; \left [ \bar{b}(p')\sigma_-^l t(p'+q)\right ] \bigg\}. \end{aligned}\ ] ] here @xmath33 and @xmath34 are dirac spinor fields describing the top and bottom quark , respectively . ( the theory can be easily extended to all three generations of quarks and also to leptons . ) contracted indices are summed . the @xmath35 matrix @xmath36 involves the spacelike indices @xmath37,@xmath38 and is defined by @xmath39 it has the properties @xmath40 the non - local character of the interaction arises from the factor @xmath41 . in a standard spinor basis in which @xmath42 , @xmath43 , the @xmath44 matrices @xmath45 are defined in terms of the pauli matrices @xmath46 , @xmath47 the fermion bilinears @xmath48 and @xmath49 therefore mix left- and right - handed spinors and violate the chiral symmetry that would protect the top quark from acquiring a mass . the matrices @xmath50 correspond to appropriate projections of the commutator @xmath51 $ ] on left ( right ) handed spinors , such that eq . ( [ a2 ] ) indeed describes a tensor exchange interaction ( cf . the appendix a ) . the lorentz - invariance of the interaction can be checked explicitly . furthermore , the action ( [ a1 ] ) , ( [ a2 ] ) has a global su(2)@xmath29u(1 ) symmetry - the remnant of the electroweak gauge symmetry in the limit of neglected gauge couplings . this symmetry forbids mass terms for the quarks , such that a mass term can be generated only by spontaneous symmetry breaking . similarly , the interaction is invariant under the color symmetry su(3 ) , with implicitly summed color indices in the bilinears . we note that only the left - handed bottom quarks are involved in the interaction and we will therefore omit the right - handed bottom quark together with the other light quarks and the leptons . similar to the photon - exchange description of the non - local coulomb interaction we may obtain @xmath52 from the exchange of chiral tensor fields @xcite , according to the feynman diagrams in fig . 1 . the lorentz symmetry of @xmath52 becomes more apparent in this formulation . a summary of the properties of the associated tensor fields and a proof of equivalence of a local theory with massless chiral tensor fields with our non - local fermion interaction ( [ a2 ] ) is provided in appendix a. in this paper we will not use tensor fields and concentrate on the purely fermionic action ( [ a1 ] ) , ( [ a2 ] ) . in particular , this avoids the delicate issues of consistency of the quantization of chiral antisymmetric tensors @xcite . one can write a consistent functional integral and therefore a consistent quantum field theory based on the action ( [ a1 ] ) , ( [ a2 ] ) . ( anomaly cancellation involves the omitted light quarks and leptons . ) ( 400,120)(0,0 ) ( 50,20)(70,60 ) ( 70,60)(50,100 ) ( 150,20)(130,60 ) ( 130,60)(150,100 ) ( 70,60)(130,60)3 ( 55,100)[tl]@xmath53 ( 55,20)[bl]@xmath54 ( 145,100)[tr]@xmath54 ( 145,20)[br]@xmath53 ( 100,65)[b]@xmath55 ( 250,20)(270,60 ) ( 270,60)(250,100 ) ( 350,20)(330,60 ) ( 330,60)(350,100 ) ( 270,60)(330,60)3 ( 255,100)[tl]@xmath56 ( 255,20)[bl]@xmath54 ( 345,100)[tr]@xmath54 ( 345,20)[br]@xmath56 ( 300,65)[b]@xmath57 this paper is organized as follows . in sect . 2 we compute the running of the dimensionless coupling @xmath0 and demonstrate that it is asymptotically free . in consequence , a non - perturbative chiral " scale @xmath1 is generated where the running coupling @xmath0 grows large . 3 evaluates the effective interactions in the scalar and vector channels which are induced in one loop - order . we find that the induced interaction in the scalar channel @xmath58 also gets large once the chiral scale is approached . as a consequence of this large interaction spontaneous electroweak symmetry breaking can be generated similar to the njl - model @xcite . we compute the resulting top quark mass @xmath11 in sect . 4 , using a schwinger - dyson @xcite or gap equation . we find @xmath59 . in sect . 5 we introduce a composite higgs field by partial bosonization " of the interaction in the scalar channel . we compute the running of its yukawa coupling @xmath60 to the top quark . the latter is directly related to the mass ratio @xmath61 , with @xmath62 and @xmath63 the mass and gauge coupling of the @xmath3 boson . in sect . 6 we present our conclusions . the dimensionless coupling constant @xmath0 is the only free parameter in our model . we will show in this section that the corresponding renormalized running coupling is asymptotically free in the ultraviolet . on the other side , there is a characteristic infrared scale @xmath1 where the coupling @xmath0 grows large . this is in complete analogy to qcd , where @xmath2 corresponds to the scale where the strong gauge coupling grows large . by dimensional transmutation " we may therefore trade @xmath0 for the mass scale @xmath1 . besides this mass scale the model has no free dimensionless parameter . in particular , we will argue in sect . 4 that our model leads to spontaneous breaking of the su(2)@xmath29u(1 ) symmetry . the fermi scale @xmath64 characterizing the electroweak symmetry breaking must be proportional to @xmath1 , @xmath65 , with a dimensionless proportionality coefficient @xmath66 that is , in principle , calculable without involving a further free parameter . if the model leads to a composite higgs scalar , its mass in units of @xmath64 , @xmath67 , as well as its effective yukawa coupling to the top quark , @xmath68 , are calculable quantities in our model . in order to investigate the running of @xmath0 , we wish to compute the effective action @xmath69 corresponding to the classical action ( [ a1]),([a2 ] ) , to one - loop order . the one - loop correction reads @xmath70 where the field dependent inverse propagator @xmath71 is defined as the second functional derivative of the action with respect to the quark fields , @xmath72 here @xmath73 are the quark fields , with color indices @xmath74 , flavor indices @xmath75 ( taking values @xmath76 ) , spinor indices @xmath77 and momenta @xmath78 . the trace tr involves an integral over momentum and summation over all kinds of indices . we write @xmath79 , where @xmath80 is the free " part of the propagator , derived from @xmath81 , @xmath82\ ; \delta_{cc'}\delta(p - p'),\ ] ] and @xmath83 . we treat @xmath84 as a perturbative correction due to @xmath52 . then @xmath85 reads , up to an infinite constant " , @xmath86 where the dots stand for neglected six - quark and higher interactions . we display the explicit expressions for @xmath87 , as well as the formal expressions for the first two terms on the r.h.s . ( [ c4 ] ) , in appendix b. ( 400,120)(0,0 ) ( 40,10)(40,35 ) ( 40,35)(40,85 ) ( 40,85)(40,110 ) ( 40,60)(25,270,90)3 ( 35,20)[r]@xmath54 ( 35,60)[r]@xmath53 ( 35,100)[r]@xmath54 ( 70,60)[l]@xmath55 ( 140,10)(140,35 ) ( 140,35)(140,85 ) ( 140,85)(140,110 ) ( 140,60)(25,270,90)3 ( 135,20)[r]@xmath54 ( 135,60)[r]@xmath56 ( 135,100)[r]@xmath54 ( 170,60)[l]@xmath57 ( 240,10)(240,35 ) ( 240,35)(240,85 ) ( 240,85)(240,110 ) ( 240,60)(25,90,270)3 ( 235,20)[r]@xmath53 ( 235,60)[r]@xmath54 ( 235,100)[r]@xmath53 ( 270,60)[l]@xmath55 ( 340,10)(340,35 ) ( 340,35)(340,85 ) ( 340,85)(340,110 ) ( 340,60)(25,90,270)3 ( 335,20)[r]@xmath56 ( 335,60)[r]@xmath54 ( 335,100)[r]@xmath56 ( 370,60)[l]@xmath57 ( 320,130)(0,0 ) ( 20,20)(20,60 ) ( 20,60)(20,100 ) ( 120,20)(120,60 ) ( 120,60)(120,100 ) ( 20,60)(50,60)3 ( 90,60)(120,60)3 ( 70,60)(20,0,180 ) ( 70,60)(20,180,0 ) ( 15,40)[tr]@xmath53 ( 15,80)[br]@xmath54 ( 125,40)[tl]@xmath54 ( 125,80)[bl]@xmath53 ( 35,63)[b]@xmath55 ( 105,63)[b]@xmath55 ( 70,83)[b]@xmath54 ( 70,37)[t]@xmath53 ( 190,20)(190,60 ) ( 190,60)(190,100 ) ( 290,20)(290,60 ) ( 290,60)(290,100 ) ( 190,60)(220,60)3 ( 260,60)(290,60)3 ( 240,60)(20,0,180 ) ( 240,60)(20,180,0 ) ( 185,40)[tr]@xmath56 ( 185,80)[br]@xmath54 ( 295,40)[tl]@xmath54 ( 295,80)[bl]@xmath56 ( 205,63)[b]@xmath57 ( 275,63)[b]@xmath57 ( 240,83)[b]@xmath54 ( 240,37)[t]@xmath56 the nonlocal factors @xmath88 appearing in the interaction ( [ a2 ] ) are attached in different ways to the fermion lines . we represent them as dashed lines in the corresponding feynman diagrams in figs . 2 - 4 . our notation recalls the one - to - one correspondence with the exchange of the associated chiral tensor fields . the diagrams in fig . 2 correspond to the first term @xmath89 in eq . ( [ c4 ] ) , while figs . 3,4 reflect the terms @xmath90 . + our one - loop calculation results in an effective action @xmath91 with a kinetic term @xmath92 and three different types of quartic interactions @xmath93 . ( note the relative minus sign between @xmath69 and the classical action @xmath94 which is chosen in order to make analytic continuation to the euclidean effective action straightforward by replacing @xmath95 . ) the tensorial part " , @xmath96 , corresponds to the exchange of a tensor field and has the form of the classical interaction @xmath52 . the other two parts , @xmath97 and @xmath98 , correspond to the exchange of vectors and scalars , respectively , and will be further discussed in sect . the momentum intergals in the loop expansion ( [ c4 ] ) are logarithmically divergent , both in the ultraviolet and the infrared . we regularize our model in the ultraviolet by a suitable momentum cutoff @xmath5 . in order to investigate the flow of effective couplings we also introduce an effective infrared cutoff @xmath37 . the effective action @xmath99 depends then on the scale @xmath37 , resulting in an effective coupling @xmath100 . instead of the infrared scale induced by non - vanishing external momenta for the vertices ( as most common for perturbative renormalization ) we introduce the cutoff by modifying the quark propagators + @xmath101 . this is a procedure known from functional renormalization . indeed , @xmath99 may be considered as the average action " or flowing action " @xcite . the precise implementation of the infrared cutoff is not important and does not influence the one - loop beta function for the running coupling @xmath100 that we will derive next . the fermion anomalous dimensions arise from eq . ( [ c9 ] ) or fig . 2 . our computation in the purely fermionic model yields the same result as in ref . @xcite , where it was computed in the equivalent model with chiral tensors , namely @xmath102 only the terms visualized diagramatically in fig . 3 , which are @xmath103 in eq . ( [ c10 ] ) , generate the same tensor structure as the classical interaction ( [ a2 ] ) . they provide the one - loop correction to the inverse chiron propagator , i.e. to the matrix @xmath104 , and one obtains @xmath105 again our fermionic computation reproduces the computation in ref . @xcite with chiral tensors . the correction results in an anomalous dimension for the chiron , @xmath106 there are no further one loop corrections in the tensor exchange channel . among the quartic corrections shown in fig . 4 , the first four terms in eq . ( [ c10 ] ) contribute to an interaction channel which is equivalent to the exchange of a vector particle . these diagrams will be evaluated in the next section . similarly , the fifth and sixth term in eq . ( [ c10 ] ) contribute to an interaction with a tensor structure different from the classical action ( [ a2 ] ) . it can be interpreted as an effective scalar exchange and will also be discussed in the next section . the running of the renormalized coupling @xmath107 ( to one - loop order ) is therefore given by the anomalous dimensions of the fermions and the correction to @xmath104 , @xmath108 this implies that the interaction is asymptotically free . the solution to the renormalization group equation ( [ c13 ] ) is @xmath109 where the chiral scale " @xmath1 is the scale at which @xmath110 becomes much larger than one . this is completely equivalent to the result of ref . the asymptotic freedom of the chiral fermion - tensor interaction is simply taken over to the non - local four - fermion interaction . ( 450,240)(0,0 ) ( 30,130)(30,155 ) ( 30,155)(30,205 ) ( 30,205)(30,230 ) ( 110,130)(110,155 ) ( 110,155)(110,205 ) ( 110,205)(110,230 ) ( 30,205)(70,180)3 ( 110,205)(70,180)3 ( 70,180)(110,155)3 ( 70,180)(30,155)3 ( 25,140)[r]@xmath53 ( 25,180)[r]@xmath54 ( 25,220)[r]@xmath53 ( 115,140)[l]@xmath53 ( 115,180)[l]@xmath54 ( 115,220)[l]@xmath53 ( 40,202)[bl]@xmath55 ( 100,202)[br]@xmath55 ( 170,130)(170,155 ) ( 170,155)(170,205 ) ( 170,205)(170,230 ) ( 250,130)(250,155 ) ( 250,155)(250,205 ) ( 250,205)(250,230 ) ( 170,205)(210,180)3 ( 250,205)(210,180)3 ( 250,155)(210,180)3 ( 170,155)(210,180)3 ( 165,140)[r]@xmath54 ( 165,180)[r]@xmath53 ( 165,220)[r]@xmath54 ( 255,140)[l]@xmath54 ( 255,180)[l]@xmath53 ( 255,220)[l]@xmath54 ( 180,162)[tl]@xmath55 ( 240,162)[tr]@xmath55 ( 310,130)(310,155 ) ( 310,155)(310,205 ) ( 310,205)(310,230 ) ( 390,130)(390,155 ) ( 390,155)(390,205 ) ( 390,205)(390,230 ) ( 310,205)(350,180)3 ( 390,205)(350,180)3 ( 350,180)(390,155)3 ( 350,180)(310,155)3 ( 305,140)[r]@xmath53 ( 305,180)[r]@xmath54 ( 305,220)[r]@xmath56 ( 395,140)[l]@xmath56 ( 395,180)[l]@xmath54 ( 395,220)[l]@xmath53 ( 320,202)[bl]@xmath57 ( 380,202)[br]@xmath55 ( 30,10)(30,35 ) ( 30,35)(30,85 ) ( 30,85)(30,110 ) ( 110,10)(110,35 ) ( 110,35)(110,85 ) ( 110,85)(110,110 ) ( 30,85)(70,60)3 ( 110,85)(70,60)3 ( 70,60)(110,35)3 ( 70,60)(30,35)3 ( 25,20)[r]@xmath56 ( 25,60)[r]@xmath54 ( 25,100)[r]@xmath56 ( 115,20)[l]@xmath56 ( 115,60)[l]@xmath54 ( 115,100)[l]@xmath56 ( 40,82)[bl]@xmath57 ( 100,82)[br]@xmath57 ( 170,10)(170,35 ) ( 170,35)(170,85 ) ( 170,85)(170,110 ) ( 250,10)(250,35 ) ( 250,35)(250,85 ) ( 250,85)(250,110 ) ( 250,85)(170,85)3 ( 170,35)(250,35)3 ( 165,20)[r]@xmath54 ( 165,60)[r]@xmath53 ( 165,100)[r]@xmath54 ( 255,20)[l]@xmath53 ( 255,60)[l]@xmath54 ( 255,100)[l]@xmath53 ( 210,88)[b]@xmath55 ( 210,38)[b]@xmath55 ( 310,10)(310,35 ) ( 310,35)(310,85 ) ( 310,85)(310,110 ) ( 390,10)(390,35 ) ( 390,35)(390,85 ) ( 390,85)(390,110 ) ( 390,85)(310,85)3 ( 310,35)(390,35)3 ( 305,20)[r]@xmath54 ( 305,60)[r]@xmath56 ( 305,100)[r]@xmath54 ( 395,20)[l]@xmath56 ( 395,60)[l]@xmath54 ( 395,100)[l]@xmath56 ( 350,88)[b]@xmath57 ( 350,38)[b]@xmath57 in this section we evaluate the diagrams in fig . 4 , representing the terms @xmath111 in eq . ( [ c10 ] ) . we begin with the fifth diagram with all external momenta set to zero . ( the contribution of the sixth diagram is equivalent , with @xmath53 substituted by @xmath56 . ) the contribution to @xmath69 is @xmath112 \left[\bar{t}\sigma^n_- \frac{\slash\hspace{-0.2 cm } q}{q^2 } \sigma^k_+ t \right].\ ] ] this may be rewritten in terms of weyl spinors @xmath113 \left [ t^\dagger_l \tau^n \frac{\slash\hspace{-0.2 cm } \bar{q}}{q^2 } \tau^k t_l \right],\ ] ] where now @xmath114 with the identity @xmath115_{\alpha\beta } \left [ \tau^n \slash\hspace{-0.2 cm } \bar{q } \tau^k \right]_{\lambda\eta } = 5 q^4 \left [ \slash\hspace{-0.2 cm } q \right]_{\alpha\beta } \left [ \slash\hspace{-0.2 cm } \bar{q } \right]_{\lambda\eta } + 4 q^6 \delta_{\alpha\eta}\delta_{\beta\lambda}\ ] ] this simplifies to @xmath116 [ t^\dagger_l \slash\hspace{-0.2 cm } \bar{q } t_l ] + 4 q^2 [ t^{c \dagger}_r t^{c'}_l][t^{c ' \dagger}_l t^c_r ] \right).\ ] ] the relative minus sign is due to an exchange of grassmann variables . note the different color structure of the second term . the momentum integral can be evaluated by analytic continuation to euclidean space , @xmath117 . introducing the infrared cutoff in the quark propagators ( i.e. substituting @xmath118 ) , one obtains @xmath119 [ t^\dagger_l \bar{\tau}^\nu t_l ] - 16 [ t^{c \dagger}_r t^{c'}_l][t^{c ' \dagger}_l t^c_r ] \right).\ ] ] using the identities @xmath120 and @xmath121 ( where @xmath122 are the eight gell - mann matrices ) this can be further simplified to @xmath123[\bar{t}_l t_r ] + \frac{3 f^4}{4 \pi^2 k^2}[\bar{t}^c_r t^z_{cd } t^d_l ] [ \bar{t}^{c'}_l t^z_{c'd ' } t^{d'}_r].\ ] ] the total contribution to @xmath69 from the fifth and sixth diagram in fig . 4 reads @xmath124 the first term in eq . ( [ d8 ] ) is equivalent to the four - fermion interaction generated at tree level by a yukawa interaction with a scalar field @xmath125 , which has a mass @xmath37 and a yukawa coupling @xmath126 given by @xmath127 this scalar transforms as a singlet under color and a doublet with respect to the electroweak interactions . it therefore has the same quantum numbers as a ( composite ) higgs doublet . the second term in eq . ( [ d8 ] ) corresponds to the exchange of a second scalar which is an octet with respect to color . finally we evaluate the first four diagrams of fig . 4 , which generate an interaction equivalent to the exchange of a vector particle . the expression for the first diagram is @xmath128 \left[\bar{t}\sigma^n_- \frac{\slash\hspace{-0.2 cm } q}{q^2 } \sigma^k_+ t \right].\ ] ] this can again be rewritten in terms of weyl spinors @xmath129 \left [ t^\dagger_l \tau^n \frac{\slash\hspace{-0.2 cm } \bar{q}}{q^2 } \tau^k t_l \right].\ ] ] with the identities @xmath130_{\alpha\beta } \left [ \tau^n \slash\hspace{-0.2 cm } \bar{q } \tau^k \right]_{\lambda\eta } = 5 q^4 \left [ \slash\hspace{-0.2 cm } \bar{q } \right]_{\alpha\beta } \left [ \slash\hspace{-0.2 cm } \bar{q } \right]_{\lambda\eta } + 4 q^6 ( \delta_{\alpha\beta}\delta_{\eta\lambda}-\delta_{\alpha\eta}\delta_{\beta\lambda})\ ] ] and @xmath131 the expression ( [ e2 ] ) simplifies to @xmath132 [ t^\dagger_l \slash\hspace{-0.2 cm } \bar{q } t_l ] -2 q^2 [ t^\dagger_l \bar{\tau}^\mu t_l][t^\dagger_l \bar{\tau}_\mu t_l ] \right).\ ] ] with an infrared cutoff @xmath37 in the quark propagator the momentum integral yields @xmath133 [ \bar{t}_l \gamma_\mu t_l].\ ] ] this is equivalent to the four - fermion interaction generated at tree level by the exchange of a vector field with mass @xmath37 and coupling @xmath134 given by @xmath135 the evaluation of the diagrams 3 and 4 is equivalent , with two or four external @xmath53 fields substituted by @xmath56 . the expression for the second diagram , in terms of weyl spinors , is @xmath136 \left [ t^\dagger_r \tau^k \frac{\slash\hspace{-0.2 cm } q}{q^2 } \tau^n t_r \right].\ ] ] identities similar to eqs . ( [ e4 ] ) and ( [ e7 ] ) result in @xmath137 [ \bar{t}_r \gamma_\mu t_r].\ ] ] the vector exchange " interaction generated at one loop level can be summarized as @xmath138 [ \bar{t}_l \gamma_\mu t_l ] + 2 [ \bar{b}_l \gamma^\mu t_l ] [ \bar{t}_l \gamma_\mu b_l ] \qquad \\ \nonumber & + [ \bar{b}_l \gamma^\mu b_l ] [ \bar{b}_l \gamma_\mu b_l ] + 2 [ \bar{t}_r \gamma^\mu t_r ] [ \bar{t}_r \gamma_\mu t_r]\bigg).\end{aligned}\ ] ] the presence of the effective yukawa interaction in eq . ( [ d8 ] ) indicates the possibility of spontaneous symmetry breaking and an analogue of the higgs mechanism . if we neglect for @xmath37 close to @xmath1 all interactions except for the scalar singlet exchange channel in the first term of eq . ( [ d8 ] ) , we may characterize the effective action by a scalar four - fermion coupling " @xmath139 , @xmath140 = 2 \lambda ( \bar{\psi}_l \psi_r)(\bar{\psi}_r \psi_l)\ ] ] with @xmath141 in the limit where the momentum dependence of the scalar four - fermion interactions can be neglected this can be interpreted as an effective njl model . here the role of the uv cutoff in the effective njl model is played by the scale @xmath37 , since the computation of @xmath142 has involved only fluctuations with momenta @xmath143 ( due to the infrared cutoff ) such that the remaining fluctuations with @xmath144 still have to be included . it is well known that for @xmath145 exceeding a critical value the njl model leads to spontaneous symmetry breaking . for a rough estimate of the top quark mass @xmath11 induced by spontaneous electroweak symmetry breaking we consider the schwinger dyson equation @xmath146 setting @xmath147 and performing the momentum integral yields the gap equation for the top quark mass @xmath148 @xmath149 we have indicated the scale @xmath37 for @xmath148 in order to recall that the solution of the gap equation will depend on the choice of the scale @xmath37 which we use as an uv cutoff for the effective njl model of course , for an exact treatment the physical top quark mass should no longer depend on @xmath37 . let us next discuss the solution of eq . ( [ g2 ] ) . obviously , @xmath150 is always one possible solution . the expression in brackets is always @xmath151 . if we write @xmath152 we see that non - zero solutions for @xmath148 ( indicating spontaneous symmetry breaking ) occur for @xmath153 once the condition ( [ g4 ] ) is obeyed , one finds indeed a non - zero @xmath148 obeying @xmath154 since @xmath155 grows arbitrarily large as @xmath37 approaches @xmath1 and @xmath100 diverges , the condition ( [ g4 ] ) is always fulfilled for @xmath37 sufficiently close to @xmath1 . for a qualitative investigation we first use the one loop result @xmath156 and replace in eq . ( [ c14 ] ) @xmath157 with @xmath158 of order 1 . this is motivated by the effective infrared cutoff @xmath159 which stops or slows down the running of @xmath107 . in fig . 5 we show @xmath160 as a function of @xmath161 for @xmath162 . after the onset of a nonzero @xmath148 for @xmath163 we find first a very rapid increase until @xmath148 settles at @xmath164 for @xmath165 . it is obvious that @xmath1 sets the scale for the top quark mass . this coincides with the result of a two - loop schwinger - dyson equation in a formulation with chiral tensor fields in @xcite . indeed , since the generation of @xmath166 is a one - loop effect , and the gap equation ( [ g1 ] ) involves a further loop , the generation of the top quark mass consists of two nested one - loop integrals , which are equivalent to a two - loop integral . in the njl approximation . ] the solution of eq . ( [ xx ] ) for @xmath165 , @xmath167 corresponds to an asymptotic value which is obtained from the condition @xmath168 . as an alternative to the ad hoc insertion of the quark mass cutoff in eq . ( [ c14 ] ) we may take into account the additional infrared cutoff due to @xmath148 by replacing in the quark propagator @xmath169 . as a consequence , the anomalous dimensions involve a threshold function @xmath170 , @xmath171 given by @xmath172 the one loop expression for @xmath173 ( [ g1b ] ) can be replaced by a flow equation @xmath174 ( for @xmath150 , @xmath175 and constant @xmath0 this reproduces eq . ( [ g1b ] ) . ) nonzero @xmath148 results in a threshold function for the running of @xmath166 , @xmath176 @xmath177 we show the running of @xmath107 and @xmath166 in fig . 6 , for different values of @xmath178 . again we stop the flow at some value of @xmath179 and solve the schwinger - dyson equation with this uv cutoff . the value of @xmath37 for which the sde yields the given @xmath178 is indicated in fig . 6 by a dot . the dots in fig . 5 show the corresponding @xmath37 dependence of @xmath160 . ( red curves ) and @xmath166 ( blue curves ) . solid ( dashed , dotted ) lines correspond to @xmath180 ( 0.6 , 0.9 ) @xmath1.,title="fig : " ] ( red curves ) and @xmath166 ( blue curves ) . solid ( dashed , dotted ) lines correspond to @xmath180 ( 0.6 , 0.9 ) @xmath1.,title="fig : " ] we are aware that our treatment of the infrared cutoff is only qualitative . while the one - loop form of the flow equation can be maintained if we interpret the flow as an approximation to the exact functional renormalization group equation for the average action @xcite , the approximation of the exact inverse quark and chiron propagators by @xmath181 and @xmath182 with momentum independent @xmath4-factors becomes questionable in the presence of large anomalous dimensions . an elegant method for the description of composite particles in the context of functional renormalization uses partial bosonization @xcite . it is based on the observation that an interaction of the type ( [ g1a ] ) can be described by the exchange of a composite scalar field . indeed , one may add to the flowing action @xmath99 a scalar piece , with @xmath183 a complex doublet scalar field @xmath184 integrating out " the scalar field by solving its field equation as a functional of @xmath185 and reinserting the solution into eq . ( [ s1 ] ) yields eq . ( [ g1a ] ) , with @xmath173 dependent on the squared exchanged momentum in the scalar channel @xmath186 as long as the momentum dependence of the effective scalar exchange vertex is not resolved ( as in our computation where the vertex is only evaluated for @xmath187 ) , one may take arbitrary @xmath188 . we will therefore replace @xmath142 in eq . ( [ g1a ] ) by eq . ( [ s1 ] ) , and replace the flow of @xmath173 in eq . ( [ x1 ] ) by @xmath189 at this stage our reformulation precisely reproduces the results in sect . the rule for the replacement of the flow of @xmath190 by a running of the renormalized yukawa coupling and dimensionless mass term @xmath191 is to adjust the flow of @xmath192 and @xmath193 such that the flow of @xmath190 is reproduced , with @xmath194 for the approximately scale invariant situation for small coupling one expects that the relative split into @xmath15-dependent and @xmath15-independent parts does not depend much on @xmath37 . we therefore make the approximation that the flow of @xmath193 receives no contribution from bosonization , such that @xmath195 or @xmath196 the effective initial value @xmath197 can be computed by evaluating the momentum dependence of the four - fermion interaction in the scalar channel . at present , it remains a free parameter . in a more complete calculation one should also evaluate the diagrams in fig . 4 for non - zero external momenta and choose the flow of @xmath192 and @xmath193 such that the flow of the vertex @xmath190 is well approximated by eq . ( [ s3a ] ) . we expect that the resulting flow for @xmath193 will be attracted to an approximate fixed point . we note that for a fixed ( point ) value of @xmath193 the mass term @xmath198 decreases roughly @xmath199 . ( 320,120)(0,0 ) ( 20,40)(45,40 ) ( 45,40)(95,40 ) ( 95,40)(120,40 ) ( 45,40)2 ( 95,40)2 ( 70,40)(26,180,0 ) ( 70,40)(24,180,0 ) ( 20,37)[tl]@xmath53 ( 45,37)[t]@xmath126 ( 70,37)[t]@xmath54 ( 95,37)[t]@xmath126 ( 120,37)[tr]@xmath53 ( 70,70)[b]@xmath64 ( 300,10)(300,35 ) ( 300,35)(300,85 ) ( 300,85)(300,110 ) ( 300,35)(250,35)3 ( 250,85)(300,85)3 ( 250,60)(25,90,270 ) ( 250,60)(25,270,90 ) ( 304,22)[tl]@xmath53 ( 304,60)[l]@xmath54 ( 304,98)[bl]@xmath53 ( 275,32)[t]@xmath55 ( 275,88)[b]@xmath55 ( 279,60)[l]@xmath53 ( 222,60)[r]@xmath54 ( 10,115)[tl](a ) ( 190,115)[tl](b ) ( 160,60)[t]@xmath200 the formulation in terms of a composite scalar field allows an inclusion of the scalar field fluctuations as well . this adds new diagrams , which can be viewed as a partial resummation of the box diagrams in fig . 4 . in particular , the quark wave function renormalizations @xmath201 get additional contributions from the scalar exchange diagrams in fig . 7 ( with scalars represented as double lines ) , as well known from computations in the standard model . in consequence , the anomalous dimensions acquire a scalar contribution with opposite sign to the tensor contribution . @xmath202 @xmath203 with @xmath204 and @xmath205 } { 1+\tilde{m}_t^2+\tilde{m}_\varphi^2 } .\ ] ] for @xmath206 the scalar correction to @xmath207 becomes approximately @xmath208 . as long as @xmath107 remains small , the effective two - loop contribution corresponding to scalar exchange ( cf . fig . 7b ) remains subleading . however , for large @xmath107 the scalar contribution to @xmath207 may become important and modify the anomalous dimensions towards positive values . this could contribute to a final stop of the increase of @xmath107 which obeys now @xmath209 with @xmath207 given by eq . ( [ s4 ] ) . ( 200,100)(0,0 ) ( 75,50)2 ( 125,50)2 ( 100,50)(25,180,0 ) ( 100,50)(25,0,180 ) ( 35,51)(75,51 ) ( 35,49)(75,49 ) ( 125,51)(165,51 ) ( 125,49)(165,49 ) ( 145,55)[b]@xmath64 ( 55,55)[b]@xmath64 ( 72,47)[tr]@xmath126 ( 128,47)[tl]@xmath126 ( 100,22)[t]@xmath54 ( 100,78)[b]@xmath210 we next turn to the scalar contributions to the running of @xmath198 and @xmath188 . the fermion loop correction to the inverse scalar propagator , as depicted in fig . 8 , results in the standard result for the anomalous dimension of the scalar field , @xmath211 due to the yukawa coupling @xmath212 a positive non - vanishing @xmath188 is generated , even if it is absent at some microscopic scale . this produces a pole in the scalar propagator for @xmath213 , such that the scalar behaves as a dynamical particle . the contribution of the quark loop shown in fig . 8 to the flow of the scalar mass term is @xmath214 . within functional renormalization it has been investigated in @xcite and one finds with our cutoff in the fermion propagator @xmath215 @xmath216 the momentum integral for the contribution of fig . 8 to the flow of @xmath198 has been cut at some effective scale @xmath217 . here @xmath217 is a characteristic scale below which the description of the flow in terms of scalar fluctuations becomes a reasonable approximation . this should be somewhat above @xmath1 , but the precise value remains somewhat arbitrary without additional computations . ( in any case the use of an improved infrared cutoff within functional renormalization would remove this spurious dependence on a scale . ) for sufficiently large @xmath192 the flow ( [ s8 ] ) drives @xmath193 to negative values , indicating the onset of spontaneous symmetry breaking . this is the same physics that is responsible for the nontrivial solutions of the schwinger - dyson equation for @xmath11 in the preceding section . indeed , for a qualitative picture we can take for @xmath218 a constant @xmath212 and @xmath219 , and neglect @xmath220 as well as the contribution from bosonization . this replaces eq . ( [ s8 ] ) by @xmath221 for the solution one finds the critical value @xmath222 for which @xmath198 reaches zero for @xmath165 . inserting @xmath223 and using @xmath224 this indeed corresponds to @xmath225 . the vacuum expectation value @xmath226 , with @xmath227 the location of the minimum of the scalar effective potential , differs from zero if @xmath198 gets negative . for the computation of its value , which should be @xmath228175 gev in a realistic model , one further needs to compute the quartic scalar self interaction @xmath229 . we can adjust the value of @xmath1 ( or the ultraviolet value @xmath230 ) such that the fermi scale @xmath231 has the correct value . a particularly interesting quantity is the renormalized yukawa coupling @xmath60 ( [ s5 ] ) . its value for @xmath232 determines the top quark mass in terms of the fermi scale @xmath233 in other words , the knowledge of @xmath234 is equivalent to a determination of the mass ratio @xmath235 , where we use @xmath236 , with @xmath63 the known weak gauge coupling ( @xmath237 ) . the observational value is @xmath238 . while the scale @xmath231 is set by dimensional transmutation and therefore a free parameter , a computation of @xmath239 is equivalent to a parameter - free pre "- diction for the mass ratio @xmath235 . in our approximation we can infer the flow equation for @xmath192 from eq . ( [ s3 ] ) , @xmath240 -\frac{f^4}{\pi^2 } \tilde{m}_\varphi^2 \tilde{s}(\tilde{m}_t^2 ) .\end{aligned}\ ] ] ( red curves ) , @xmath60 ( blue curves in the left picture ) and @xmath193 ( picture on the right ) . at large values of @xmath37 we started with @xmath241 ( solid lines ) and @xmath242 ( dashed lines).,title="fig : " ] ( red curves ) , @xmath60 ( blue curves in the left picture ) and @xmath193 ( picture on the right ) . at large values of @xmath37 we started with @xmath241 ( solid lines ) and @xmath242 ( dashed lines).,title="fig : " ] we have solved the flow equations numerically until the onset of spontaneous symmetry breaking at @xmath243 where @xmath244 . for @xmath245 one has @xmath206 such that many threshold functions equal one . we display the running of @xmath60 and @xmath0 in fig . 9 . for @xmath246 one has to continue the flow in the regime with spontaneous symmetry breaking and non - zero @xmath247 , as well known from many studies of functional flow equations @xcite . we will not do so here . in the present approximation to the flow equations we observe an increase of @xmath0 and @xmath60 to very large values as @xmath248 approaches zero . we do not expect our approximations to remain valid for large couplings , even though the one - loop form of the functional flow equations is exact @xcite . one of the main shortcomings is the inaccurate truncation of the general form of momentum dependence for the fermion propagators and non - local interactions in a region where the anomalous dimensions @xmath249 are of the order one . for example , the quartic interactions for the bare quarks in the tensor channel replaces in our approximation @xmath250 in eq . ( [ a2 ] ) . we may consider an effective momentum dependence of @xmath251 given by the anomalous dimension @xmath252 evaluated for @xmath253 , i.e. @xmath254 - this would indeed be a valid approximation for small @xmath255 and @xmath256 . however , for @xmath257 the momentum dependence of @xmath251 would effectively cancel the nonlocality @xmath258 of the interaction . for a quasi - local interaction in the tensor channel we expect strong modifications of the flow of @xmath0 - for example , it could reach a fixed point @xmath259 , replacing in the interaction @xmath260 . ( in the language of chiral tensor fields this would correspond to the generation of a non - local mass for the chirons " @xcite . ) for a realistic model of electroweak interactions the increase of the yukawa coupling should stop or be substantially slowed down in the vicinity of its final value at @xmath232 , say for @xmath261 . for the corresponding region in the scale @xmath37 , @xmath262 we find values @xmath263 . ( we use @xmath264 for the quantitative estimates . ) in view of our discussion , it seems not unreasonable that the non - perturbative infrared physics stops the further increase of @xmath0 and @xmath60 in this range . if this happens , the ratio @xmath235 may come out in a reasonable range . a typical value of the scale for non - perturbative physics where the increase of @xmath0 and @xmath60 stops , may be @xmath265 . at this scale the quantity @xmath166 ( cf . ( [ s8a ] ) with @xmath266 has reached a value @xmath267 , not too far from the critical value in the schwinger - dyson approach . it is well conceavable that the top quark mass is substantially below the scale @xmath268 , such that the region of strong interactions may correspond to the multi - tev - scales and not disturb too much the lep precision tests of the electroweak theory . our computation of the flow of @xmath0 and the yukawa coupling @xmath60 has further substantial quantitative uncertainties . for example , we have neglected effects from the interactions in sect . 3 that correspond to the exchange of color - octet scalars or vectors . this may be motivated for the region of @xmath37 where @xmath248 is already small , since a resonance type behavior and spontaneous symmetry breaking is only expected in the scalar singlet sector . on the other hand , these contributions may still play a role in the interesting region where the increase of @xmath0 may stop . we conclude that models for quarks and leptons with a non - local four - fermion interaction in the tensor channel appear to be promising candidates for an understanding of the electroweak symmetry breaking . no fundamental higgs scalar is needed . the theory is asymptotically free and generates an exponentially small chiral scale " @xmath1 where the dimensionless coupling @xmath0 grows large . furthermore , a strong interaction in the scalar channel is generated at scales where @xmath0 is large . a solution of schwinger - dyson equations suggests that this interaction triggers the spontaneous breaking of the electroweak symmetry at a scale determined by @xmath1 . this would solve the gauge hierarchy problem . we have also introduced a composite higgs scalar and investigated the flow of its mass . we find indeed spontaneous symmetry breaking with a fermi scale somewhat below the chiral scale . our first attempt of an estimate of the yukawa coupling of the top quark is encouraging , yielding a reasonable range for the ratio @xmath235 . the model has also other interesting features . it was shown @xcite that the flavor and cp - violation is completely described by the ckm matrix @xcite . masses of the light quarks and leptons arise from appropriate four - fermion couplings @xcite . first phenomenological constraints from lep precision tests and the anomalous magnetic moment of the muon have been computed @xcite . at the present stage the understanding of the strong interactions around the scale @xmath1 , which would be a few tev in a realistic model , is still in its infancy . for this reason our estimate of @xmath235 is only very crude . nevertheless , no free parameter enters in the determination of this ratio in our model . if the strong interactions can be understood quantitatively , our model leads to a unique pre "- diction of @xmath235 . it would also predict the masses and interactions of the composite scalar fields . * appendix a : non - local four - fermion interaction and chiral tensor fields * + in this appendix we show the equivalence of the microscopic action with a theory of chiral tensor fields @xcite . this is similar to the equivalence of quantum electrodynamics to a theory with a non - local four - fermion interaction , generated by integrating out the photon . we may start from the non - local four - fermion interaction and obtain the chiral tensor theory by a hubbard - stratonovich transformation . we proceed here in the opposite way , starting with a chiral tensor theory and integrating out the chiral tensors . starting point is the theory of chiral antisymmetric tensor fields investigated in ref . we consider a complex antisymmetric tensor field @xmath269 which is a doublet of weak isospin and carries hypercharge @xmath270 . the field can be decomposed into two parts which correspond to irreducible representations of the lorentz group . the two parts are the chiral " components of @xmath271 : @xmath272 these components can be written as @xmath44 matrices acting in the space of dirac spinors via @xmath273 where @xmath274 $ ] . the matrix @xmath275 ( @xmath276 ) acts only on left - handed ( right - handed ) fermions , i.e. @xmath277 one may introduce an interaction between the fermions and the chiral tensors : @xmath278 here the chiral couplings @xmath279 are @xmath35 matrices in generation space , @xmath280 are the left - handed quark doublets , @xmath281 ( @xmath282 ) are the right - handed up - type ( down - type ) quarks , @xmath283 are the left - handed lepton doublets , @xmath284 are the right - handed electron - type leptons , and we defined @xmath285 where the transposition @xmath286 and the pauli matrix @xmath287 act in weak isospin space , i.e. on the two components of the weak doublet @xmath288 . the fields @xmath289 can be represented by three - vectors @xmath290 , @xmath291 rewriting the kinetic term @xmath292 in terms of the @xmath293-fields and in fourier space gives @xmath294.\ ] ] the propagator @xmath295 has the properties @xmath296 in the following we ignore for simplicity all couplings except @xmath297 , i.e. we assume @xmath298 in this case , only the @xmath299 fields couple to top quarks @xmath33 and left - handed bottom quarks @xmath34 . all other fields are free if we ignore gauge couplings . we denote the two weak isospin components of @xmath299 as @xmath57 and @xmath55 , since they correspond to electric charge @xmath300 and @xmath301 after electroweak symmetry breaking . the action for the chiral interactions is then @xmath302,\ ] ] in a spinor basis in which @xmath303 the matrices @xmath45 are defined in terms of the pauli matrices @xmath46 as @xmath304 in the action @xmath305 , a summation over quark color is understood . the classical field equations are obtained by varying @xmath306 with respect to the @xmath299 fields . they determine these fields as functionals of the quark fields . in momentum space , these relations are @xmath307 we then insert these relations into @xmath308 and obtain the action @xmath52 of the non - local four - fermion interaction given in eq . ( [ a2 ] ) . in a functional integral formulation this procedure is equivalent to performing the gaussian integral over the @xmath299 fields . + * appendix b : one loop expressions for the effective action * + in this appendix we evaluate the interaction contribution @xmath87 to the inverse propagator @xmath71 in eq . ( [ c2 ] ) for the different quark types separately , ( @xmath309 ) . one obtains @xmath310 \left ( -\frac{p_{kl}^*(q)}{q^4}\right)\\ \nonumber & + \delta_{cc'}\bar{t}^{c''}_\gamma(q+p')t^{c''}_\eta(q+p ) \left [ \sigma^k_{+ \alpha\beta } \sigma^l_{- \gamma\eta } + \sigma^l_{- \alpha\beta } \sigma^k_{+ \gamma\eta}\right ] \frac{p_{kl}^*(p'-p)}{(p'-p)^4}\\ \nonumber & + \bar{b}^{c'}_\gamma(q+p')b^c_\eta(q+p ) \sigma^k_{+ \alpha\eta } \sigma^l_{- \gamma\beta } \left ( -\frac{p_{kl}^*(q)}{q^4}\right)\bigg\},\end{aligned}\ ] ] @xmath311 @xmath312 \sigma^k_{+\alpha\beta } \frac{p_{kl}^*(p'-p)}{(p'-p)^4},\ ] ] @xmath313 \sigma^k_{-\alpha\beta } \frac{p_{kl}^*(p'-p)}{(p'-p)^4},\ ] ] @xmath314 \frac{p_{kl}^*(q)}{q^4},\ ] ] @xmath315 \frac{p_{kl}^*(q)}{q^4},\ ] ] @xmath316 @xmath317 @xmath318 inserting this into the expression ( [ c4 ] ) , we find an anomalous dimension term for @xmath33 and for the left - handed @xmath34 @xmath319 the term @xmath90 in eq . ( [ c4 ] ) produces an effective four fermion interaction @xmath320 with @xmath321 and @xmath322 color indices are suppressed , since all pairs of fermions in large brackets are color singlets . gross and f. wilczek , phys . * 30 * , 1343 ( 1973 ) ; + h.d . politzer , phys . 30 * , 1346 ( 1973 ) v.a . miransky , m. tanabashi and k. yamawaki , phys . * b 221 * , 177 ( 1989 ) ; mod . phys . lett . * a 4 * , 1043 ( 1989 ) ; w. bardeen , c. hill and m. lindner , phys . * d 41 * , 1647 ( 1990 ) y. nambu and g. jona - lasinio , phys . * 122 * , 345 ( 1961 ) h. gies , j. jaeckel and c. wetterich , phys . rev . * d 69 * , 105008 ( 2004 ) c. wetterich , phys . * d 74 * , 095009 ( 2006 ) ; mod . lett . * a 23 * , 677 ( 2008 ) c. wetterich , int . j. mod . phys . * a 23 * , 1545 ( 2008 ) f.j . dyson , phys . rev . * 75 * , 1736 ( 1949 ) ; + j.s . schwinger , proc . sci . * 37 * 452 ( 1951 ) c. wetterich , phys . * b 301 * , 90 ( 1993 ) h. gies and c. wetterich , phys . rev . * d 65 * , 065001 ( 2002 ) c. wetterich , z. phys . * c 48 * , 693 ( 1990 ) j. berges , n. tetradis and c. wetterich , phys . rept . * 363 * , 223 ( 2002 ) m. kobayashi and t. maskawa , prog . . phys . * 49 * , 652 ( 1973 ) ; + n. cabibbo , phys . * 10 * , 531 ( 1963 ) c. wetterich , arxiv:0709.1102 [ hep - ph ] j .- m . schwindt and c. wetterich , int . j. mod . phys . * a 23 * , 4345 ( 2008 ) b. de wit and j.w . van holten , nucl . * b 155 * , 530 ( 1979 ) ; e. bergshoeff , m. de roo and b. de wit , nucl . * b 182 * , 173 ( 1981 ) ; i.t . todorov , m.c . mintchev and v.b . petkova , _ conformal invariance in quantum field theory _ , p.89 ( ets , pisa , 1978 ) ; m.v . chizhov , mod . lett . * a 8 * , 2753 ( 1993 ) ; l.v . avdeev and m.v . chizhov , phys . * b 321 * , 212 ( 1994 ) n. kemmer , proc . r. soc . * a 166 * , 127 ( 1938 ) ; v.i . ogievetsky and i.v . polubarinov , yad . fiz . * 4 * , 216 ( 1966 ) ; a. salam and e. sezgin ( eds . ) , _ supergravities in diverse dimensions _ ( north holland and world scientific , 1989 ) ; m. kalb and p. ramond , phys . rev . * d 9 * , 2273 ( 1974 ) ; e. cremmer and j. scherk , nucl . * b 72 * , 117 ( 1974 ) ; s.a . frolov and a.a . slavnov , phys . * b218 * , 461 ( 1989 )
we study the relationship between unit - distance representations and lovsz theta number of graphs , originally established by lovsz . we derive and prove min - max theorems . this framework allows us to derive a weighted version of the hypersphere number of a graph and a related min - max theorem . then , we connect to sandwich theorems via graph homomorphisms . we present and study a generalization of the hypersphere number of a graph and the related optimization problems . the generalized problem involves finding the smallest ellipsoid of a given shape which contains a unit - distance representation of the graph . we prove that arbitrary positive semidefinite forms describing the ellipsoids yield np - hard problems .
1010.6036
geometric representation of graphs is a beautiful area where combinatorial optimization , graph theory and semidefinite optimization meet and connect with many other research areas . in this paper , we start by studying geometric representations of graphs where each node is mapped to a point on a hypersphere so that each edge has unit length and the radius of the hypersphere is minimum . lovsz @xcite proved that this graph invariant is related to the lovsz theta number of the complement of the graph via a simple but nonlinear equation . we show that this tight relationship leads to min - max theorems and to a `` dictionary '' to translate existing results about the theta function and its variants to the hypersphere representation setting and vice versa . based on our approach , we derive a weighted version of the hypersphere number of a graph and deduce related min - max theorems . our viewpoint allows us to make new connections , strengthen some facts and correct some inaccuracies in the literature . after observing that the hypersphere number of a graph is equal to the radius of the smallest euclidean ball containing a unit - distance representation of the graph , we propose generalizations of the underlying optimization problems . given a graph , the generalized optimization problem seeks the smallest ellipsoid of given shape which contains a unit - distance representation of the graph . we finally show that at this end of the new spectrum of unit - distance representations , arbitrary positive semidefinite forms describing the shapes of the ellipsoids yield np - hard geometric representation problems . we denote the set of symmetric @xmath0 matrices by @xmath1 , the set of symmetric @xmath0 positive semidefinite matrices by @xmath2 , and the set of symmetric @xmath0 positive definite matrices by @xmath3 . for a finite set @xmath4 , the set of symmetric @xmath5 matrices is denoted by @xmath6 , and the symbols @xmath7 and @xmath8 are defined analogously . for @xmath9 , we write @xmath10 meaning @xmath11 . define an inner product on @xmath1 by @xmath12 , where @xmath13 is the trace of @xmath14 . the linear map @xmath15 extracts the diagonal of a matrix ; its adjoint is denoted by @xmath16 . the vector of all ones is denoted by @xmath17 . we abbreviate @xmath18 \colonequals { \ { { 1,\dotsc , n } \}}$ ] . the notation @xmath19 for a norm is the euclidean norm unless otherwise specified . for a finite set @xmath4 , the set of orthogonal @xmath20 matrices is denoted by @xmath21 . the set of nonnegative reals is denoted by @xmath22 . the set of positive reals is denoted by @xmath23 . define the notations @xmath24 and @xmath25 analogously for integer numbers . for any function @xmath26 on graphs , we denote by @xmath27 the function defined by @xmath28 for every graph @xmath29 , where @xmath30 denotes the complement of @xmath29 . for a graph @xmath29 , we denote the clique number of @xmath29 by @xmath31 and the chromatic number of @xmath29 by @xmath32 . the complete graph on @xmath18 $ ] is denoted by @xmath33 . let @xmath29 be a graph . its vertex set is @xmath34 and its edge set is @xmath35 . for @xmath36 , the subgraph of @xmath29 induced by @xmath37 , denoted by @xmath38 $ ] , is the subgraph of @xmath29 on @xmath37 whose edges are the edges of @xmath29 that have both ends in @xmath37 . for @xmath39 , the neighbourhood of @xmath40 , denoted by @xmath41 , is the set of nodes of @xmath29 adjacent to @xmath40 . a block of @xmath29 is an inclusionwise maximal induced subgraph of @xmath29 with no cut - nodes , where a cut - node of a graph @xmath42 is a node @xmath43 such that @xmath44 $ ] has more connected components than @xmath42 . for a graph @xmath45 , the _ laplacian of @xmath29 _ is the linear extension @xmath46 of the map @xmath47 for every @xmath48 , where @xmath49 denotes the @xmath40th unit vector . laplacians arise naturally in spectral graph theory and spectral geometry ; see @xcite . let @xmath45 be a graph . a _ unit - distance representation of @xmath29 _ is a function @xmath50 for some @xmath51 such that @xmath52 whenever @xmath53 . a _ hypersphere representation of @xmath29 _ is a unit - distance representation of @xmath29 that is contained in a hypersphere centered at the origin , and the _ hypersphere number of @xmath29 , _ denoted by @xmath54 , is the square of the smallest radius of a hypersphere that contains a unit - distance representation of @xmath29 . the _ theta number of @xmath29 _ is defined by @xmath55 this parameter was introduced by lovsz in the seminal paper @xcite ; see also @xcite for further properties and alternative definitions . lovsz @xcite noted the following formula relating @xmath56 and @xmath57 : [ thm : t - theta ] for every graph @xmath29 , we have @xmath58 we will show how the relation can be used to better understand some of the properties of the theta number and the hypersphere number . this will allow us to obtain simpler proofs of some facts about the theta number and new results about hypersphere representations . we include a proof of theorem [ thm : t - theta ] for the sake of completeness . we may formulate @xmath54 as the sdp @xmath59 here , @xmath60 is the adjoint of the laplacian @xmath61 of @xmath29 . the dual of is @xmath62 both and have slater points , so sdp strong duality holds for this dual pair of sdps , i.e. , their optimal values coincide and both optima are attained . in particular , @xmath54 is equal to . if we write an optimal solution @xmath63 of as @xmath64 , then @xmath65 is a hypersphere representation of @xmath29 with squared radius @xmath54 . we can rewrite the dual as @xmath66 by taking @xmath67 . the objective value of a feasible solution @xmath68 is @xmath69 . thus , @xmath70 , where @xmath71 it is easy to check that @xmath72 . let @xmath45 be a graph . an _ orthonormal representation of @xmath29 _ is a function from @xmath4 to the unit hypersphere in @xmath73 for some @xmath51 that maps non - adjacent nodes to orthogonal vectors . it is well - known that , if @xmath50 is a hypersphere representation of @xmath29 with squared radius @xmath74 , then the map @xmath75 } \in { \mathbb{r}}\oplus { \mathbb{r}}^d\ ] ] is an orthonormal representation of @xmath30 . define @xmath76 as the set of all @xmath77 such that @xmath78 for every orthonormal representation @xmath79 of @xmath29 and unit vector @xmath80 . then @xmath81 . the transformation allows us to interpret theorem [ thm : t - theta ] as strong duality for a nonlinear min - max relation : [ prop : t - minmax ] let @xmath29 be a graph . for every hypersphere representation of @xmath29 with squared radius @xmath56 and every nonzero @xmath82 , we have @xmath83 with equality if and only if @xmath84 and @xmath85 . set @xmath86 . let @xmath50 be a hypersphere representation of @xmath29 with squared radius @xmath56 . we may assume that @xmath87 . let @xmath88 . define an orthonormal representation @xmath89 of @xmath30 from @xmath90 as in . . then @xmath92 . the equality case now follows from theorem [ thm : t - theta ] . proposition [ prop : t - minmax ] shows that @xmath93 and elements from @xmath94 are natural dual objects for @xmath54 and hypersphere representations of @xmath29 . in fact , using a well - known description of the elements of @xmath94 , we recover from proposition [ prop : t - minmax ] the following sdp - free purely geometric min - max relation : [ prop : t - minmax - sdpfree ] let @xmath45 be a graph . for every hypersphere representation of @xmath29 with squared radius @xmath56 , every orthonormal representation @xmath95 of @xmath29 , and every unit vector @xmath96 such that @xmath97 , we have @xmath98}^{-1 } \geq 1,\ ] ] with equality if and only if @xmath84 and @xmath99 . the transformation may be reversed as follows . suppose that @xmath100 is an orthonormal representation of @xmath30 such that , for some positive @xmath101 and some @xmath102 , we have @xmath103 } \qquad \forall i \in v.\ ] ] then @xmath104 is a hypersphere representation of @xmath29 with squared radius @xmath105 . we can use and to obtain an identity involving these objects . [ prop : th - gallai ] let @xmath45 be a graph . then @xmath106 where @xmath90 ranges over all orthonormal representations of @xmath30 and @xmath107 over unit vectors of the appropriate dimension . we first prove `` @xmath108 '' in . let @xmath109 be an orthonormal representation of @xmath30 and let @xmath80 be a unit vector . we will show that @xmath110 it is well - known that there exists an orthonormal representation @xmath89 of @xmath30 and a unit vector @xmath111 such that @xmath112 for all @xmath113 . if @xmath114 , then @xmath115 shows that @xmath116 , so assume that @xmath117 . we may assume that @xmath118 and @xmath119 for every @xmath120 . now use with @xmath121 to get a hypersphere representation @xmath104 of @xmath29 from @xmath89 with squared radius @xmath122 . this proves . next we prove `` @xmath123 '' in . let @xmath124 be a hypersphere representation of @xmath29 with squared radius @xmath54 . build an orthonormal representation @xmath89 of @xmath30 as in and pick @xmath125 . then @xmath126 for every @xmath120 . ( the reciprocal of the second term of the sum on the lhs of was used as the original definition of @xmath93 by lovsz @xcite . ) note that does not provide a good characterization of either @xmath54 or the maximization problem on the lhs of . in this sense , proposition [ prop : th - gallai ] is akin to gallai s identities for graphs ( * ? ? ? * lemmas 1.0.1 and 1.0.2 ) . for a graph @xmath29 , let @xmath127 be the square of the smallest radius of an euclidean ball that contains a unit - distance representation of @xmath29 . this parameter is also mentioned by lovsz ( * ? ? ? * proposition 4.1 ) . to formulate @xmath127 as an sdp , replace the constraint @xmath128 in by @xmath129 . the resulting sdp and its dual have slater points , so sdp strong duality holds , i.e. , both optima are attained and the optimal values coincide . evidently , @xmath130 for every graph @xmath29 . in fact , equality holds : [ thm : tb - th ] for every graph @xmath29 , we have @xmath131 . if we mimic the proof of theorem [ thm : t - theta ] for @xmath127 , we find that @xmath132 where @xmath133 is defined by adding the constraint @xmath134 to the sdp . thus , by and , theorem [ thm : tb - th ] is equivalent to the fact that @xmath135 for every graph @xmath29 . this follows from next result ( * ? ? ? * proposition 9 ) ( this was pointed out to the first author by fernando mrio de oliveira filho ) : [ prop : gijswijt ] let @xmath136 be such that @xmath137 whenever @xmath138 and @xmath139 . if @xmath140 is an optimal solution for the optimization problem @xmath141 , then @xmath142 for some positive @xmath143 . since @xmath144 is a relaxation of @xmath133 , we have @xmath145 . to prove the reverse inequality , let @xmath140 be an optimal solution for . by proposition [ prop : gijswijt ] , we have @xmath146 for some @xmath147 . hence @xmath140 is feasible for the sdp that defines @xmath133 , whence @xmath148 . the formula relating @xmath54 and @xmath93 allows us to infer some basic facts about the theta number from a geometrically simpler viewpoint . [ thm : sandwich ] for any graph @xmath29 , we have @xmath149 . by theorem [ thm : t - theta ] and the fact that @xmath150 for every @xmath151 , the sandwich theorem is equivalent to the inequalities @xmath152 for every graph @xmath29 . the first inequality is obvious : if @xmath42 is a subgraph of @xmath29 , then @xmath153 . the second one is also obvious : if @xmath154 { \to}{\mathbb{r}}^d$ ] is a hypersphere representation of @xmath155 and @xmath156 $ ] is a colouring of @xmath29 , then @xmath157 is a hypersphere representation of @xmath29 . this hints at a strong connection with graph homomorphisms , which we will look at more closely in section [ sec : hom ] . lovsz @xcite mentions that a graph @xmath29 is bipartite if and only if @xmath158 . the less obvious of the implications may be easily proved by showing that @xmath159 for every odd cycle @xmath160 . however , we find that the following proof using hypersphere representations gives a more enlightening geometric interpretation . by theorem [ thm : t - theta ] , we must show that @xmath161 if and only if @xmath29 is bipartite . if @xmath29 is bipartite , then @xmath29 has a hypersphere representation with radius @xmath162 even in @xmath163 . suppose @xmath29 has a hypersphere representation with radius @xmath164 . the only pairs of points at distance @xmath165 in a hypersphere of radius @xmath162 are the pairs of antipodal points , so @xmath29 is bipartite . given graphs @xmath45 and @xmath166 with @xmath167 , the _ direct sum of @xmath29 and @xmath42 _ is the graph @xmath168 . it is proved in @xcite that @xmath169 . by theorem [ thm : t - theta ] , this is equivalent to the geometrically obvious equation @xmath170 . in particular , @xmath171 . more generally , @xmath172 . this follows from the next result , where we denote @xmath173 and @xmath174 for graphs @xmath175 and @xmath176 . [ prop : clique - separation ] let @xmath45 be a graph , and suppose @xmath177 for graphs @xmath178 and @xmath179 , with @xmath180 a complete graph . then @xmath181 by theorem [ thm : t - theta ] , it suffices to prove that @xmath182 . clearly ` @xmath123 ' holds in the desired equation . assume @xmath183 . since the feasible region of is convex and contains @xmath184 , there are hypersphere representations @xmath104 and @xmath185 of @xmath178 and @xmath179 , respectively , both with squared radius @xmath186 . we may assume that the images of @xmath104 and @xmath185 live in the same euclidean space . since @xmath180 is a complete graph , there is an orthogonal matrix @xmath187 such that @xmath188 for every @xmath189 . if we glue the hypersphere representation @xmath190 of @xmath179 with @xmath104 , we get a hypersphere representation of @xmath29 with squared radius @xmath186 . this behavior of @xmath56 and @xmath191 with respect to clique sums is shared by many other graph parameters , e.g. , @xmath192 , @xmath193 , the hadwiger number ( the size of the largest clique minor ) , and the graph invariant @xmath194 introduced in @xcite . proposition [ prop : clique - separation ] and theorem [ thm : tb - th ] imply the following purely geometric result : let @xmath45 be a graph , and suppose @xmath177 for graphs @xmath178 and @xmath179 , with @xmath180 a complete graph . for @xmath195 , let @xmath196 be a unit - distance representation of @xmath197 contained in an euclidean ball of radius @xmath198 . then there is a unit - distance representation of @xmath29 contained in an euclidean ball of radius @xmath199 . the proof contains an algorithm to build the desired unit - distance representation of @xmath29 . however , whereas one would expect such an algorithm to provide a geometric construction from @xmath200 and @xmath201 , the one presented essentially needs to solve an sdp , and it may ignore @xmath200 and @xmath201 altogether . using basic properties of laplacians , we can prove the following behaviour of @xmath56 and @xmath191 with respect to edge contraction : [ prop : t - contraction ] let @xmath45 be a graph and let @xmath202 . if @xmath203 is an optimal solution for , then @xmath204 . if @xmath205 is an optimal solution for applied to @xmath93 , then @xmath206 . see appendix [ sec : delayed ] . finally , using basic properties about the intersection of two hyperspheres , we can prove a property of @xmath191 that is shared by the parameters @xmath192 , @xmath193 , and the fractional chromatic number @xmath207 . the proof is based on ( * ? ? ? * lemma 4.3 ) . [ prop : nbhood ] let @xmath29 be a graph and @xmath39 with @xmath208 . then @xmath209 ) \leq 1 - 1/[4t(g ) ] \quad \text{and } \quad \overline{\vartheta}(g ) \geq \overline{\vartheta}(g[n(i)])+1.\ ] ] see appendix [ sec : delayed ] . for @xmath210 , define @xmath211 by replacing the objective function in by @xmath212 , where @xmath213 for every @xmath120 . it is natural to define a weighted hypersphere number @xmath214 so that it satisfies a natural generalization of , namely , @xmath215 whenever @xmath216 . by using the proof of theorem [ thm : t - theta ] as a guide , we arrive at the definition : t(g , w ) = & t & + & ( x ) = |e+ ( t-)w , & + & _ g^*(x ) = |e+ ( t-)_g^*(w ) , & + & x _ + ^ , t . & this sdp and its dual have slater points , so sdp strong duality holds . even though we can not offer a nice direct interpretation for this definition of @xmath214 , by construction , we can generalize proposition [ prop : t - minmax ] : [ thm : thw - minmax ] let @xmath29 be a graph and @xmath217 . then , for every feasible solution @xmath218 of and every nonzero @xmath88 , we have @xmath219 with equality if and only if @xmath218 is optimal for and @xmath220 . set @xmath86 . we may assume that @xmath87 . write @xmath221 for some @xmath222 \times v$ ] matrix @xmath223 , and define @xmath109 by @xmath224 . the map @xmath225 } \in { \mathbb{r}}\oplus { \mathbb{r}}^d$ ] is an orthonormal representation of @xmath30 . put @xmath226 . then @xmath227 . the equality case now follows by construction . if @xmath228 , it can be shown that @xmath229 , where @xmath230 is obtained from @xmath29 by replacing each node @xmath40 by a clique @xmath197 on @xmath231 nodes ; if @xmath232 , then every node in @xmath197 is adjacent in @xmath230 to every node in @xmath233 . in fact , every feasible solution @xmath234 of encodes a hypersphere representation of @xmath230 with squared radius @xmath235 . indeed , write @xmath236 for some @xmath222 \times v$ ] matrix @xmath223 , and define @xmath237 . for @xmath120 , let @xmath238 be a hypersphere representation of @xmath197 with squared radius @xmath239 . define @xmath240 as follows . for @xmath241 , set @xmath242 to be the vector whose block in @xmath73 is @xmath243 and whose block in @xmath244 is @xmath245 ; all other blocks of @xmath242 are zero . then @xmath104 is a hypersphere representation of @xmath230 with squared radius @xmath235 . let @xmath29 and @xmath42 be graphs . a _ homomorphism from @xmath29 to @xmath42 _ is a function @xmath246 such that @xmath247 whenever @xmath232 . if there is a homomorphism from @xmath29 to @xmath42 , we write @xmath248 . note that @xmath249 whenever @xmath248 . indeed , if @xmath26 is a homomorphism from @xmath29 to @xmath42 and @xmath185 is a hypersphere representation of @xmath42 , then @xmath250 is a hypersphere representation of @xmath29 . this combines with the graph - theoretic observation that @xmath251 to yield @xmath252 , which by theorem [ thm : t - theta ] is equivalent to the sandwich theorem [ thm : sandwich ] . motivated by this , we call a real - valued graph invariant @xmath26 _ hom - monotone _ if @xmath253 whenever @xmath248 and the following `` nondegeneracy '' condition holds : there is a non - decreasing function @xmath254 such that @xmath255 for every integer @xmath151 . using these properties for an arbitrary graph @xmath29 and the fact that @xmath256 , we get @xmath257 , and thus @xmath258 ( see @xcite for a similar use of these ideas . ) we point out that hom - monotonicity can not recover strong sandwich theorems which state that @xmath259 since this inequality fails to hold for the hom - monotone invariant @xmath193 . the function @xmath260 is non - decreasing on @xmath261 , so @xmath56 is hom - monotone , and we recover from the sandwich theorem [ thm : sandwich ] . the reason why @xmath56 satisfies the first condition of hom - monotonicity roughly comes from the fact that the constraints for the sdp of @xmath56 are `` uniform '' for the edges , i.e. , all edges are treated in the same way . we are thus led to define other sdps of the same type . one such example is the parameter @xmath262 . however , as we have seen in theorem [ thm : tb - th ] , this parameter is equal to @xmath56 . now define @xmath263 clearly , @xmath264 for every graph @xmath29 , and it is easy to see that equality holds if @xmath29 is node - transitive . in particular , @xmath265 for every @xmath266 . thus , the function @xmath267 proves that @xmath268 is hom - monotone . using and @xmath264 , we obtain @xmath269 for every graph @xmath29 . if we mimic the proof of theorem [ thm : t - theta ] for @xmath270 , we find that @xmath271 , where @xmath272 is defined by adding the constraint @xmath273 to , i.e. , @xmath274 is the graph parameter introduced in @xcite and @xcite . let @xmath275 be the minimum @xmath276 such that there is a unit - distance representation of @xmath29 in @xmath73 ; consider @xmath277 . as before , @xmath278 implies @xmath279 . since @xmath280 , the function @xmath281 shows that @xmath282 is hom - monotone , so @xmath283 . however , we will see later that computing @xmath275 is np - hard . ( a similar parameter was introduced in @xcite . ) define @xmath284 similarly as @xmath275 but for hypersphere representations of @xmath29 with squared radius @xmath164 and @xmath285 for orthonormal representations of @xmath30 . such parameters are also hom - monotone . clearly @xmath286 for every graph @xmath29 , but strict inequality occurs for the mosers spindle ( see figure [ fig : moser ] and the proof of theorem [ thm : dim - hard ] ) . since shows that @xmath287 and @xcite shows that @xmath288 , these parameters are related by @xmath289 . in particular , by , we find that @xmath290 . also @xmath291 , where @xmath292 is the maximum degree of @xmath29 . in fact , by brooks theorem , @xmath293 when @xmath29 is connected but not complete nor an odd cycle . the following relaxation of graph colouring was introduced in @xcite . let @xmath45 be a graph . for a real number @xmath294 , a _ vector @xmath295-colouring of @xmath29 _ is a function @xmath90 from @xmath4 to the unit hypersphere in @xmath73 for some @xmath51 such that @xmath296 whenever @xmath48 ; we consider the fraction to be @xmath297 if @xmath298 , so the only graphs that have a vector @xmath165-colouring are the graphs with no edges . a vector @xmath295-colouring @xmath90 of @xmath29 is _ strict _ if @xmath299 for every @xmath53 , and a strict vector @xmath295-colouring @xmath90 of @xmath29 is _ strong _ if @xmath300 whenever @xmath301 . the _ vector chromatic number of @xmath29 _ is the smallest @xmath302 for which there exists a vector @xmath295-colouring of @xmath29 , and the strict vector chromatic number and strong vector chromatic number are defined analogously . it is easy to show ( see , e.g. , @xcite ) that the vector chromatic number of @xmath29 is @xmath303 , the strict vector chromatic number of @xmath29 is @xmath93 , and the strong vector chromatic number of @xmath29 is @xmath304 , known as szegedy s number @xcite , where @xmath305 is defined by replacing the constraints @xmath306 for every @xmath53 in by @xmath307 for every @xmath53 . here , we note that a scaling map yields a correspondence between these variations of vector colourings and unit - distance representations , provided that the graph @xmath29 has at least one edge . let @xmath90 be a strict vector @xmath295-colouring of @xmath29 . then the map @xmath308 , where @xmath309 , is a hypersphere representation of @xmath29 with squared radius @xmath56 . conversely , if @xmath89 is a hypersphere representation of @xmath29 with squared radius @xmath87 , then the map @xmath310 , is a strict vector @xmath295-colouring of @xmath29 , where @xmath311 . this correspondence shows that @xmath312 , where @xmath313 denotes the strict vector chromatic number of @xmath29 . the same scaling maps as above yield correspondences between vector @xmath295-colourings and the geometric representations arising from the graph invariant @xmath268 , and also between strong vector @xmath295-colourings and geometric representations arising from the graph invariant t^+(g ) & t & + & ( x ) = t |e , & + & x_ii-2x_ij+x_jj = 1 , & \ { i , j } e(g ) , + & x_ii-2x_ij+x_jj 1 , & \ { i , j } ( g ) , + & x _ + ^ , & note however , that the parameter @xmath314 does not fit into the framework of hom - monotone graph invariants since the sdp has non - edge constraints . we point out here that , while these equivalences between variants of vector chromatic number and variants of theta number are easy to prove , they are not as widely known as they should be . for instance , in @xcite it is shown that the vector chromatic number @xmath315 of @xmath29 satisfies @xmath316 where @xmath317 and @xmath318 denote the largest and smallest eigenvalue , respectively , and @xmath319 denotes the set of all weighted adjacency matrices of @xmath45 , i.e. , all symmetric @xmath20 matrices @xmath320 such that @xmath321 . however , since @xmath322 , it is possible to adapt the proof of the hoffman bounds for @xmath144 ( see , e.g. , ( * ? ? ? * corollary 33 ) ) to show that actually holds with equality . also , in ( * ? ? ? * remark 3.1 ) it is reported that a certain graph @xmath29 has vector chromatic number strictly smaller than its strict vector chromatic number , and that it was unknown whether some such graph existed . however , this statement about the vector chromatic numbers is equivalent to @xmath323 , and the existence of graphs satisfying this strict inequality was already known as far back as 1979 ( see @xcite ) . we also mention that one of the characterizations of @xmath272 in @xcite and @xcite is inaccurate . define an _ obtuse representation of a graph @xmath45 _ to be a map @xmath324 for some @xmath51 such that a. @xmath325 for every @xmath120 , and b. @xmath326 for every @xmath327 . in ( * ? ? ? * theorem 1 ) and @xcite it is claimed that @xmath328 where @xmath90 ranges over obtuse representations of @xmath29 and @xmath107 ranges over unit vectors of appropriate dimension . let @xmath29 be a @xmath329-partite graph with color classes @xmath330 such that @xmath331 . thus , @xmath332 . let @xmath333 for every @xmath334 and @xmath335 $ ] , and @xmath336 for every @xmath337 and @xmath335 $ ] . set @xmath338 . by , we get @xmath339 , a contradiction . now we show how to fix the formula . given an obtuse representation @xmath79 of a graph @xmath340 , we say that a vector @xmath80 is _ consistent with @xmath90 _ if @xmath341 for every @xmath120 . the next result is a gallai - type identity involving @xmath270 , parallel to proposition [ prop : th - gallai ] for @xmath54 . [ prop : t-gallai ] let @xmath45 be a graph . then @xmath342 where @xmath90 ranges over all obtuse representations of @xmath30 and @xmath107 over unit vectors consistent with @xmath90 . this proof is analogous to the proof of proposition [ prop : th - gallai ] , with the following slight adjustments . in the notation of the proof of , the vector @xmath111 may be chosen to be consistent with the obtuse representation @xmath89 , so we do not need to replace any of the @xmath343 s by their opposites . let @xmath45 be a graph . then @xmath272 is given by , where @xmath90 ranges over obtuse representations of @xmath29 and @xmath107 ranges over unit vectors consistent with @xmath90 . this follows from proposition [ prop : t-gallai ] together with the formula @xmath271 . the graph parameter @xmath262 encodes the problem of finding the smallest euclidean ball that contains a unit - distance representation of a given graph . in this section , we study graph parameters that encode the problem of finding the smallest ellipsoid of a given shape that contains a unit - distance representation of a given graph . let @xmath45 be a graph . in section [ sec : balls ] , we defined @xmath127 as the minimum infinity - norm of the vector @xmath344 over all unit - distance representations @xmath104 of @xmath29 , where we are using the notation @xmath345 . it is natural to replace the vector @xmath344 in the objective function of the previous optimization problem with the vector @xmath346 for some fixed @xmath347 . the resulting optimization problem corresponds to finding the minimum squared radius @xmath56 such that the ellipsoid @xmath348 contains a unit - distance representation of @xmath29 . we are thus led to define , for every graph @xmath45 , every @xmath349 for some @xmath51 , and every @xmath350 $ ] , the number @xmath351 as the infimum of @xmath352 as @xmath104 ranges over all unit - distance representations of @xmath29 in @xmath73 , or equivalently , @xmath353 } } \big\}}.\ ] ] note that we allow @xmath320 to be singular . since the feasible region in is invariant under right - multiplication by matrices in @xmath354 , we have @xmath355 for every @xmath356 . in particular , @xmath357 is a spectral function . let us derive some basic properties of the optimal solutions of @xmath351 . first , we prove that if @xmath351 is finite then the corresponding optimal geometric representation exists . the first observation towards this goal is that , if @xmath29 is connected , then the maximum distance between any pair of points in every unit - distance representation is at most @xmath358 . [ thm : e - attains ] let @xmath45 be a graph . let @xmath359 for some @xmath51 and let @xmath350 $ ] . if @xmath360 , then there exists @xmath361}$ ] such that @xmath362 and @xmath363 . we may assume that @xmath29 is connected . ( if not , it suffices to focus on the component @xmath42 of @xmath29 with @xmath364 . ) we may further assume @xmath365 where @xmath366 , where @xmath367 denotes the vector of eigenvalues of @xmath320 , with multiplicities , arranged in a nonincreasing order . so , there exists a largest @xmath368 $ ] so that @xmath369 . let @xmath370 . throughout this proof , let @xmath371 denote the projection onto the first @xmath295 components , i.e. , @xmath372 , and let @xmath373 denote the projection onto the last @xmath374 components . note that @xmath375 and @xmath376 . let @xmath377 such that @xmath378 . fix @xmath113 arbitrarily . we claim that the following constraints may be added to the rhs of without changing its optimal value : @xmath379 let us see why this proves the theorem . let @xmath380}$ ] be feasible for and satisfy and . let @xmath381 be arbitrary . since the columns of @xmath382 form a unit - distance representation of @xmath29 , the distance in @xmath29 between @xmath40 and @xmath383 is an upper bound for @xmath384 . hence , @xmath385 . thus , the new feasible region is compact and we will be done . first , we prove that the constraints may be added to without changing the optimal value . suppose @xmath386}$ ] violates for some @xmath120 . then @xmath387 , so @xmath388 may be discarded from the feasible set of . next , we add the constraint . let @xmath389}$ ] be feasible for and satisfy . define @xmath390}$ ] by setting @xmath391 for every @xmath120 and @xmath392 for every @xmath120 . hence , @xmath393 is feasible for and satisfies and . moreover , @xmath394 . this completes the proof . a geometrically pleasing , intuitive conjecture is that a suitably defined notion of a `` centre '' of an optimal representation of every graph must coincide with the centre of the ellipsoid . the next result takes a step along this direction by refining the previous theorem . [ thm : e - attains - and - conv ] let @xmath45 be a graph . let @xmath359 for some @xmath51 and let @xmath350 $ ] . if @xmath360 , then there is a unit - distance representation @xmath50 of @xmath29 such that @xmath395 and @xmath396 . we use the same assumptions and notation defined in the first paragraph of the proof of theorem [ thm : e - attains ] . let @xmath50 be a feasible solution for @xmath351 . let @xmath397 be the set of all unit - distance representations of @xmath29 of the form @xmath398 for some vector @xmath399 such that @xmath400 . note that if @xmath401 , then @xmath397 is a singleton . clearly , every element of @xmath397 has the same objective value as @xmath104 . we will show that if there does not exist some element @xmath402 such that @xmath403 , then @xmath404 . then this theorem will follow from theorem [ thm : e - attains ] . so , assume that @xmath405 . since @xmath406 is a polyhedron and @xmath407 , there exists @xmath408 and @xmath409 such that @xmath410 for every @xmath411 and @xmath120 . note that @xmath412 since for each @xmath413 the linear function @xmath414 of @xmath56 is bounded below by @xmath415 . thus , @xmath416 let @xmath417 such that @xmath418 and let @xmath419 , where @xmath420 will be chosen later . define @xmath421 by @xmath422 . then @xmath423 . hence @xmath424 if and only if @xmath425 . thus , we will be done if we can find @xmath420 such that @xmath426 . since @xmath427 , such @xmath428 exists . this shows that , for some choice of @xmath420 , we have @xmath424 for every @xmath120 , whence @xmath429 . the next result shows that it is not very interesting to use arbitrarily large prescribed embedding dimension @xmath111 : [ thm : e - lower - dim ] let @xmath45 be a graph . let @xmath359 for some @xmath51 and let @xmath350 $ ] . if @xmath430 $ ] is such that @xmath431 has an optimal solution @xmath50 with @xmath432 , then @xmath433 where @xmath434 . in particular , @xmath435 if @xmath436 . we may assume that @xmath365 where @xmath437 ( here , @xmath438 denotes the vector of eigenvalues of @xmath320 , with multiplicities , arranged in a nondecreasing order ) . note that @xmath439 . the proof of ` @xmath108 ' in is immediate by appending extra zero coordinates to an optimal solution of @xmath440 . to prove ` @xmath123 , ' let @xmath50 be an optimal solution for @xmath431 such that @xmath441 . then , there exists @xmath356 such that , for each @xmath120 , the final @xmath374 coordinates of @xmath442 are zero . let @xmath443 be obtained from @xmath442 by dropping the final @xmath374 ( zero ) coordinates . if @xmath444 is the principal submatrix of @xmath445 indexed by @xmath446 $ ] , then @xmath447 . hence , @xmath448 . by interlacing of eigenvalues , @xmath449 . hence , @xmath450 . it follows from theorem [ thm : e - attains - and - conv ] that @xmath435 if @xmath436 . it is clear that @xmath451 if and only if @xmath452 . so deciding whether @xmath453 for any fixed @xmath295 reduces to computing @xmath351 for any @xmath454 $ ] where @xmath320 is a matrix of nullity @xmath295 . it is easy to see that the former decision problem is np - hard ( see ( * ? ? ? * theorem 4 ) ) . we give below a shorter proof . [ thm : dim - hard ] the problem of deciding whether @xmath455 for a given input graph @xmath29 is np - hard . let @xmath295 be a fixed positive integer . saxe @xcite showed that the following problem is np - hard : given an input graph @xmath456 and @xmath457 , decide whether there exists @xmath458 such that @xmath459 for every @xmath53 . saxe showed that the problem remains np - hard even if we require @xmath460 . we will show a polynomial - time reduction from the above problem with @xmath461 and @xmath460 to the problem of deciding whether @xmath455 . it suffices to show how we can replace any edge of the input graph @xmath29 which is required to be embedded as a line segment of length 2 by some gadget graph @xmath42 so that every unit - distance representation of @xmath42 in @xmath462 maps two specified nodes of @xmath42 to points at distance @xmath463 . consider the graph @xmath464 known as the mosers spindle shown in figure [ fig : moser ] . the subgraph of @xmath464 induced by @xmath465 has exactly two unit - distance representations in @xmath462 modulo rigid motions : one of them as displayed in figure [ fig : moser ] , and the other one maps nodes @xmath466 and @xmath467 to the same point . we claim that , in any unit - distance representation @xmath104 of @xmath464 in @xmath462 , the nodes @xmath466 and @xmath467 are not mapped to the same point . suppose otherwise . since the points @xmath468 are at distance @xmath165 from @xmath469 and from each other , @xmath104 shows that @xmath470 , whereas clearly @xmath471 . . ] let @xmath42 be the gadget shown in figure [ fig : gadget ] , which consists of two copies of @xmath464 sharing a triangle ( some edges of @xmath464 are drawn in dots for the sake of ease of visualization ) . then , every unit - distance representation of @xmath42 in @xmath462 maps the nodes @xmath40 and @xmath383 to points at distance @xmath463 . thus , if we replace the corresponding edges @xmath472 of the input graph @xmath29 by @xmath42 , we obtain a graph @xmath473 such that @xmath474 if and only if @xmath29 can be embedded in @xmath462 with the prescribed edge lengths . it follows from theorem [ thm : dim - hard ] that , for any fixed @xmath454 $ ] , the problem of computing @xmath351 for an input graph @xmath29 and @xmath475 is np - hard . hence the graph parameter @xmath476 is in a sense on the borderline of tractability . for every matrix @xmath478 , if we set @xmath479 , then there exists an orthogonal @xmath480 matrix @xmath187 such that @xmath481 . hence , if @xmath482 , then _ p(g;a ) = & ( x^1/2q^taqx^1/2)_p & + & _ g^*(x ) = |e & + & x _ + ^ , q ^. & when @xmath483 , the objective function in is @xmath484 so we can write @xmath485 where @xmath486 is defined for any @xmath487 as the sdp @xmath488 [ prop : tw - finite ] let @xmath45 be a connected graph and let @xmath487 . then @xmath486 is finite if and only if @xmath489 or @xmath490 . moreover , whenever @xmath486 is finite , both and its dual sdp have optimal solutions and their optimal values coincide . the parameter @xmath486 thus underlies the parameters @xmath491 as well as the hypersphere number @xmath54 , since shows that @xmath492 if @xmath393 is feasible in for @xmath493 , then @xmath393 is completely determined by its diagonal entries . using this fact , it is easy to prove that the feasible region of for @xmath494 is @xmath495 using a second - order cone programming formulation , we can show that @xmath496 let us use and to compute @xmath497 . let @xmath498 be nonzero . since @xmath499 for some @xmath500 , it follows from and that @xmath501 the supremum may be replaced by @xmath502 , which is easily seen to be @xmath503 . this implies with theorem [ thm : e - lower - dim ] that @xmath504 for the other extreme @xmath505 , the first property of hom - monotonicity holds . more precisely , let @xmath506 be a nondecreasing sequence of positive reals . define @xmath507 for every @xmath508 . then , @xmath509 we do not know whether the invariant @xmath510 satisfies the second property of hom - monotonicity . in fact , we do not know an analytical formula to compute @xmath511 in terms of @xmath320 . however , we have such a formula for an infinite family of complete graphs , as we now describe . let @xmath42 be an @xmath0 hadamard matrix , i.e. , @xmath42 is @xmath512-valued and @xmath513 . we may assume that @xmath42 has the form @xmath514 then @xmath515 , so @xmath516 , i.e. , the map @xmath517 is a unit - distance representation of @xmath33 . this map is called a _ hadamard representation of @xmath33 _ . let @xmath518 such that there exists an @xmath0 hadamard matrix . then , for any @xmath350 $ ] and diagonal @xmath519 , every hadamard representation of @xmath33 is an optimal solution for @xmath520 . the objective value of the hadamard representation @xmath521 of @xmath33 in the optimization problem @xmath520 is @xmath522 } { \|{\bar{e}}\|_{p}}$ ] . thus , @xmath521 is optimal for @xmath523 by . from the inequality @xmath524 we get @xmath525 , which shows that @xmath521 is optimal for @xmath505 . therefore , @xmath521 is optimal for every @xmath350 $ ] . it is natural to lift a hadamard representation @xmath526 of @xmath33 to obtain a frugal feasible solution for @xmath527 . the image of @xmath526 is an @xmath528-dimensional simplex @xmath529 . if @xmath185 is a vertex of an @xmath266-dimensional simplex whose opposite facet is @xmath529 , then the line segment @xmath530 joining @xmath185 to the barycenter of @xmath529 is the shortest line segment joining @xmath185 to @xmath529 . it makes sense to align @xmath530 with the most expensive axis , i.e. , the one corresponding to @xmath503 . suppose @xmath365 and @xmath531 . we thus obtain a unit - distance representation @xmath104 of @xmath532 in @xmath533 of the form @xmath534 \\ 0 \oplus { \big [ { \alpha + { \big ( { \frac { n+1 } { 2n } } \big)}^{\scriptscriptstyle 1/2 } } \big ] } , & \text{if } i = n+1 . \end{cases}\ ] ] by optimizing the shift parameter @xmath415 , we obtain the following upper bound : let @xmath518 such that there exists an @xmath0 hadamard matrix . if @xmath498 , then @xmath535 equality holds for @xmath536 if @xmath537 . the proof of equality for @xmath536 involves the obvious parametrization of @xmath538 and basic trigonometry . 10 y. bilu . tales of hoffman : three extensions of hoffman s bound on the graph chromatic number . , 96(4):608613 , 2006 . p. j. cameron , a. montanaro , m. w. newman , s. severini , and a. winter . on the quantum chromatic number of a graph . , 14(1):research paper 81 , 15 pp . ( electronic ) , 2007 . f. r. k. chung . , volume 92 of _ cbms regional conference series in mathematics_. published for the conference board of the mathematical sciences , washington , dc , 1997 . p. erds , f. harary , and w. t. tutte . on the dimension of a graph . , 12:118122 , 1965 . a. galtman . spectral characterizations of the lovsz number and the delsarte number of a graph . , 12(2):131143 , 2000 . d. gijswijt . . phd thesis , university of amsterdam , 2005 . m. x. goemans . semidefinite programming in combinatorial optimization . , 79(1 - 3 , ser . b):143161 , 1997 . lectures on mathematical programming ( ismp97 ) ( lausanne , 1997 ) . m. grtschel , l. lovsz , and a. schrijver . , volume 2 of _ algorithms and combinatorics_. springer - verlag , berlin , second edition , 1993 . h. van der holst , m. laurent , and a. schrijver . on a minor - monotone graph invariant . , 65(2):291304 , 1995 . b. horvat , j. kratochvl , and t. pisanski . on the computational complexity of degenerate unit distance representations of graphs . in _ combinatorial algorithms _ , volume 6460 of _ lecture notes in comput . _ , pages 274285 . springer , heidelberg , 2011 . d. karger , r. motwani , and m. sudan . approximate graph coloring by semidefinite programming . , 45(2):246265 , 1998 . d. e. knuth . the sandwich theorem . , 1:article 1 , approx . ( electronic ) , 1994 . m. laurent and f. rendl . semidefinite programming and integer programming . in _ handbook on discrete optimization _ , pages 393514 . elsevier b. v. , amsterdam , 2005 . l. lovsz . on the shannon capacity of a graph . , 25(1):17 , 1979 . l. lovsz . semidefinite programs and combinatorial optimization . in _ recent advances in algorithms and combinatorics _ , pages 137194 . springer , new york , 2003 . l. lovsz and m. d. plummer . , volume 121 of _ north - holland mathematics studies_. north - holland publishing co. , amsterdam , 1986 . annals of discrete mathematics , 29 . r. j. mceliece , e. r. rodemich , and h. c. rumsey , jr . the lovsz bound and some generalizations . , 3(3):134152 , 1978 . p. meurdesoif . strengthening the lovsz @xmath539 bound for graph coloring . , 102(3 , ser . a):577588 , 2005 . j. b. saxe . two papers on graph embedding problems . technical report cmu - cs-80 - 102 , department of computer science , carnegie - mellon university , 1980 . a. schrijver . a comparison of the delsarte and lovsz bounds . , 25(4):425429 , 1979 . a. soifer . . springer , new york , 2009 . mathematics of coloring and the colorful life of its creators , with forewords by branko grnbaum , peter d. johnson , jr . and cecil rousseau . m. szegedy . a note on the theta number of lovsz and the generalized delsarte bound . in _ proceedings of the 35th annual ieee symposium on foundations of computer science _ , 1994 let @xmath203 be an optimal solution for . we will construct a feasible solution for applied to @xmath540 with objective value @xmath541 . assume @xmath542 and @xmath543 , so we are denoting the contracted node of @xmath540 by @xmath466 . let @xmath223 be the @xmath544 matrix defined by @xmath545 . then @xmath546 , where @xmath547 is obtained from @xmath548 as follows . in taking the contraction @xmath540 from @xmath29 , immediately after we identify the ends of @xmath549 , but before we remove resulting parallel edges , there are at most two edges between each pair of nodes of @xmath540 , as we assume that @xmath29 is simple . if there is exactly one edge between nodes @xmath40 and @xmath383 , we just set @xmath550 . if there are two edges joining nodes @xmath40 and @xmath383 , say @xmath26 and @xmath551 , we put @xmath552 . similarly , if we define @xmath553 by putting @xmath554 for @xmath555 and @xmath556 , then @xmath557 . since @xmath558 , we see that @xmath559 is a feasible solution for applied to @xmath540 , and its objective value is @xmath560 . to prove the inequality involving @xmath93 , use together with its proof to see that @xmath205 corresponds to an optimal solution @xmath203 for with @xmath561 , so @xmath562 . by theorem [ thm : t - theta ] , it suffices to show @xmath563 ) \leq 1 - 1/{[{4t(g)}]}$ ] . let @xmath79 be a hypersphere representation of @xmath29 with squared radius @xmath564 . we may assume that @xmath565 . for every @xmath566 , we have @xmath567_1 $ ] . hence , @xmath568_1 = ( 2t-1)/(2t^{1/2 } ) \equalscolon \beta$ ] for every @xmath566 . define the following hypersphere representation of @xmath569 $ ] : for each @xmath570 , let @xmath571 be obtained from @xmath572 by dropping the first coordinate . the squared radius of the resulting hypersphere representation is @xmath573 . let @xmath574 . then @xmath575 if and only if @xmath576 for some @xmath577 ; for the ` only if ' part , use @xmath578 . let @xmath577 . the smallest eigenvalue of @xmath579 is @xmath580 . thus , @xmath581 if and only if @xmath582 . assume first that @xmath583 for some @xmath584 . by proposition [ prop : tw - finite ] , finiteness of @xmath585 implies @xmath586 or @xmath587 . assume the former . by , @xmath588 where @xmath589 . the second - order cone program on the rhs of has @xmath590 as a slater point , and its dual is @xmath591 . since @xmath592 is optimal for the dual , it follows that @xmath593 now we drop the diagonal assumption , so let @xmath594 such that @xmath489 or @xmath490 . for @xmath595 , we can write @xmath596 , so @xmath597 , i.e. , @xmath598 by . hence , follows from . if @xmath599 for some @xmath120 , then @xmath600 and we are done by induction on @xmath266 . so we may assume that @xmath601 . define @xmath602 by @xmath603 and let @xmath604 . note that @xmath605 and @xmath606 . for every @xmath139 with @xmath607 , the matrix @xmath608 is feasible in the optimization problem with objective value @xmath609 . since @xmath610 is an optimal solution , we see that @xmath611 attains the maximum of @xmath609 over all @xmath139 with @xmath607 . since @xmath612 , then @xmath611 attains the maximum of @xmath609 also over all @xmath613 with @xmath614 . thus , for @xmath615 , we have @xmath616 , so @xmath617 . let @xmath388 be a feasible solution for . let @xmath621 . then @xmath622 for some @xmath623 . the objective value of @xmath624 in the rhs of is @xmath625 , which is the objective value of @xmath388 in . suppose that @xmath489 . let @xmath630 and assume @xmath631 $ ] . let @xmath500 such that @xmath632 . then @xmath633 for some @xmath634 , since @xmath29 . let @xmath635 , @xmath636 and @xmath637 such that @xmath638 note that @xmath639 . thus , for every @xmath640 , we have @xmath641 if and only if @xmath642 . since @xmath643 , we know that for @xmath194 negative and with sufficiently large magnitude , we have @xmath641 , and hence @xmath644 . assume @xmath647 . since has @xmath649 as a slater point , the dual @xmath650 of has an optimal solution @xmath651 . assume @xmath631 $ ] and let @xmath500 such that @xmath652 . then @xmath653 and @xmath654 imply that @xmath655 for every @xmath656 $ ] . thus , @xmath657 , which together with @xmath646 implies @xmath490 . since @xmath658 , it is easy to check that the constraint @xmath659 may be added to without changing the optimal value . the dual of this augmented sdp is @xmath660 . by proposition [ prop : tw - slater ] , this dual has a slater point @xmath661 with @xmath662 , so has an optimal solution . since has a slater point and is bounded below , its dual has an optimal solution and the optimal values coincide . if @xmath489 , then and its dual have slater points by proposition [ prop : tw - slater ] . if @xmath663 , then @xmath664 with @xmath665 shows that @xmath666 . assume now that @xmath667 . if @xmath668 , then @xmath666 . otherwise , apply proposition [ prop : tw - ones - in - nullw ] .
the disordered potential landscape in an ingaas / inalas two - dimensional electron gas patterned into narrow wires is investigated by means of scanning gate microscopy . it is found that scanning a negatively charged tip above particular sites of the wires produces conductance oscillations that are periodic in the tip voltage . these oscillations take the shape of concentric circles whose number and diameter increase for more negative tip voltages until full depletion occurs in the probed region . these observations can not be explained by charging events in material traps , but are consistent with coulomb blockade in quantum dots forming when the potential fluctuations are raised locally at the fermi level by the gating action of the tip . this interpretation is supported by simple electrostatic simulations in the case of a disorder potential induced by ionized dopants . this work represents a local investigation of the mechanisms responsible for the disorder - induced metal - to - insulator transition observed in macroscopic two - dimensional electron systems at low enough density .
1502.07857
two - dimensional electron gases @xcite ( 2degs ) buried inside semiconductor heterostructures @xcite show ballistic transport over micrometers at low temperature . their very long electron mean free path results from the combination of a high growth quality by molecular beam epitaxy and a remote doping technique that drastically reduces scattering by impurities . @xcite in such heterostructures , conduction electrons are confined at the interface between two different band gap materials and spatially separated from the dopants , which are placed a few tens of nanometers above the heterojunction . however , the random distribution of the ionized dopants produces long - range potential fluctuations in the 2deg that strongly affect electron transport at low temperature . @xcite below a critical electron density , this disorder potential breaks the 2deg into several electron puddles , @xcite and conduction is described by a percolation process in a two - dimensional network with thermally activated hopping . @xcite this 2d metal - to - insulator transition ( mit ) has been extensively studied by transport experiments in macroscopic samples using large planar gates to control the overall electron density . @xcite investigations of the mit in small samples revealed that long - range and short - range disorder potentials produce different behaviors . in particular , insulating samples with short gate length show a metallic behavior at very low temperature @xcite that may result from resonant tunneling between conducting domains . @xcite in samples with even shorter gate length , strong conductance fluctuations are observed versus gate voltage @xcite due to sample specific disorder configurations . @xcite these potential fluctuations also explain the tremendous difficulty to fabricate ballistic one - dimensional wires . @xcite the presence of potential barriers along the wire results in the formation of localized states with coulomb blockade , especially in long wires of several microns in length , @xcite but also in submicrometer - length wires . @xcite in quantum point contacts ( qpcs ) , the presence of potential fluctuations in the constriction @xcite is often invoked to explain resonances in the quantized conductance plateaus . @xcite alternatively , qpcs could be used to probe locally the disorder potential , since only a small region between the split - gates dominates the transport . @xcite finally , in mesoscopic devices of intermediate dimension at very low temperature , quantum interferences of electron waves spreading coherently in the disordered potential landscape give rise to universal conductance fluctuations @xcite ( ucfs ) . imaging the disorder potential of a surface 2deg can be achieved by scanning tunneling microscopy , @xcite but the case of 2degs buried tens of nanometers below the surface requires specific local probe techniques . most of the studies have been done in the quantum hall regime at high magnetic field , using techniques based on subsurface charge accumulation , @xcite single electron transistors , @xcite and scanning gate microscopy ( sgm ) . @xcite surprisingly , very few studies have been done at zero magnetic field . scanning capacitance microscopy is the only technique that succeeded in imaging directly the disorder potential at zero field and revealed fluctuations on a length scale much larger than expected from the distance between dopants and the 2deg . @xcite however , sgm can also provide indirect information on the disorder potential inside or close to a nanoscale device by imaging , for example , the complex branched electron flow spreading in a 2deg out of a qpc , @xcite the ucf pattern in a small constriction etched in a 2deg , @xcite the irregular fringe pattern in a quantum ring , @xcite or the presence of charge traps in the 2deg heterostructure . @xcite in this paper , we use sgm @xcite to probe the disorder potential in a low - density ingaas / inalas 2deg , patterned by etching into a network of wires . we show that transport through the wires is dominated by a few spots where the electrostatic potential forms a valley surrounded by two hills . when the tip is placed above these spots with a negative voltage , the conductance decreases strongly and can even drop to zero . in addition , the conductance does not decrease smoothly when the tip approaches these spots , but shows several oscillations , which are clearly revealed by sensitive transconductance measurements . these oscillations are interpreted as a signature of localized states with coulomb blockade in quantum dots that form in the 2deg local potential valley . by lowering locally the electron density close to zero under the tip , we indeed expect the disordered potential landscape to form a series of potential barriers delimiting quantum dots with localized states , as depicted schematically in fig . [ fig1 ] . in the case of macroscopic 2d gates , it would give rise to the formation of isolated 2deg islands and to a percolation - driven mit . our experiment therefore represents a local investigation of the disorder - induced mit . similar conductance oscillations have been observed in sgm images of various systems , but were explained by the presence quantum dots only for inas nanowires , @xcite carbon nanotubes , @xcite and graphene . @xcite so far , none of the sgm studies has found conductance oscillations due to the charging of quantum dots within a 2deg but rather due to charging of traps or impurities in the heterostructure surrounding the conducting channel . @xcite here , we show that the features observed in our experiment are not consistent with charging events in traps , but should instead be explained by the formation of quantum dots in the disordered potential landscape . we substantiate our finding by approximate electrostatic calculations of the disorder potential within the wire and the induced tip potential revealing nearly quantitative agreement with the experimental data . the paper is organized as follows . section [ sec : exp1 ] gives technical information about the experiment . sections [ sec : exp2 ] and [ sec : exp3 ] present the sgm images and their analysis . section [ sec : sim1 ] presents simulations of the disordered potential landscape induced by ionized dopants . section [ sec : sim2 ] demonstrates that the sgm tip can reveal the presence of quantum dots and supports our analysis of the experimental data . supplementary information , measurements , and analysis are given in appendix sections . the sample is based on a pseudomorphic in@xmath0ga@xmath1as / inalas heterostructure grown by molecular beam epitaxy on a semi - insulating inp substrate @xcite with the following layer sequence : 100 nm lattice - matched inalas buffer layer , 50 nm alassb barrier , 400 nm inalas layer , 15 nm in@xmath0ga@xmath1as channel , 20 nm inalas spacer , @xmath2-doping si plane ( @xmath3@xmath4 ) , 15 nm inalas barrier , 7 nm doped ingaas cap layer . the 2deg is formed 42 nm below the sample surface with a carrier density @xmath5@xmath4 and a mobility @xmath6@xmath7v@xmath8s@xmath8 as measured by magneto - transport at 4.2k in a hall bar patterned on the same sample . the investigated nanostructure is a @xmath9m@xmath10 network made of three 180 nm wide parallel wires , linked together by two 210 nm wide wires , connected to the source and drain reservoirs by 370 nm wide openings ( see fig . [ fig2](a ) ) . this complex sample geometry will be simply considered here as a set of three independent wires measured in parallel . the pattern is written by electron beam lithography and transferred into a mesa by wet etching of 65 nm deep trenches . sgm measurements are performed in a homemade atomic force microscope @xcite ( afm ) cooled at 4.2k by exchange gas in a liquid helium cryostat . a commercial silicon tip coated with a ptir conducting layer is glued at the extremity of a tuning fork , which is used as a force sensor in the afm imaging mode . experiments start by recording a topographic image at 4.2k to locate the device . for sgm measurements , the tip is lifted by 100 nm ( for all the data presented here ) and scanned in a plane at a constant height above the sample . usually , a negative voltage relative to the 2deg is applied to the tip , and the device conductance and/or transconductance are recorded as a function of the tip position during scanning . as a result of the capacitive coupling between the tip and the 2deg , the electron density under the tip is reduced and the electrostatic potential is raised towards the fermi level : the tip acts as a local movable gate . the device conductance is measured with a lock - in using a small ac source - drain excitation at 68hz , while a dc voltage is applied to the tip . the transconductance is measured with a small dc source - drain bias , while a 40mv ac excitation at 939hz is applied to the tip in addition to the main dc voltage . the two signals can be recorded simultaneously with a dual reference lock - in amplifier . the unperturbed device resistance being around 10k@xmath11 , voltage or current bias can be used for the source - drain polarization , corresponding to the measurement of a conductance @xmath12 or a resistance @xmath13 , respectively . both configurations have been used depending on the highest resistance recorded in the sgm map , but all data are plotted here in terms of conductance and transconductance , using the conversion @xmath14 and @xmath15 . note that @xmath16 instead of @xmath17 will be plotted for the transconductance signal in order to keep all quantities ( @xmath12 and @xmath16 ) in units of @xmath18 . measured with an ac current bias @xmath19na . the dc tip voltage is indicated on each image . ( c ) sgm conductance profiles along the red lines drawn in ( a ) for tip voltages from @xmath203.6v ( bottom curve ) to + 4.2v ( top curve ) with 0.6v steps . left and right graphs correspond to bottom and top red lines , respectively . ( d ) average conductance @xmath21 and standard deviation @xmath22 calculated from the sgm images in ( b ) versus tip voltage . ( e ) difference @xmath23 between two consecutive sgm images as explained in the text.,width=302 ] the conductance images shown in fig . [ fig2](b ) are obtained by scanning the tip above the entire device for decreasing negative tip voltages . they show a complex pattern of conductance drops covering the device area between the two openings . the device geometry can hardly be recognized because the tip - induced potential has a broad lateral extension and influences electron transport in the device even if the tip is not directly above the wires . the largest changes are observed along the central path , which probably carries the largest current , and in particular at its ends which are critical nodes for transmission . at some locations , the conductance drops by a factor of 4 at @xmath24v and can even drop to zero at larger negative tip voltage as shown later . the narrow width of the arms and the low electron density make the device very sensitive to potential changes induced by the tip . sgm profiles recorded along two selected lines are plotted in fig . [ fig2](c ) for different tip voltages . it is found that the profiles recorded at @xmath25v ( black curves ) show no conductance change , i.e. , the tip does not produce any potential perturbation . this particular value , the so - called flat band voltage , corresponds to the work function difference between the ptir coating of the tip and the ingaas cap layer of the heterostructure taking into account a surface fermi level pinning at mid - gap ( see appendix a ) . similar values were found for ptir tips and gaas surfaces in other sgm experiments @xcite since ingaas and gaas have similar work functions . the average conductance calculated over the full scanning area is shown in fig . [ fig2](d ) . it varies roughly linearly with the tip voltage as in the case of a macroscopic field effect transistor , except for the lower slope observed at positive voltages that we attribute to a larger screening in case of charge accumulation . the standard deviation of the conductance maps , also shown in fig . [ fig2](d ) , is found to drop very close to zero at the flat band voltage @xmath25v , showing that the tip is free from charged dust particles that would have disturbed its local gate action . @xcite the linear increase of the standard deviation on both sides of @xmath26 is consistent with an in - average linear gate effect of the tip ( linear response ) , since the conductance does not drop to zero in this tip voltage range . @xcite careful examination of the conductance images in fig . [ fig2](b ) reveals that they contain several spots , growing in size and amplitude for decreasing tip voltages . the edge of these spots can be made more visible in fig . [ fig2](e ) by plotting the difference @xmath27 between conductance maps recorded at two consecutive voltages separated by 0.6v . many overlapping circles appear in these images , four in the left branch , two in the right one , and even more in the central one , which are difficult to distinguish . their diameter increases for more negative tip voltages , but their center remains at a fixed position , always located inside the wires , never in the etched regions . these images show that the device is very sensitive to a local potential change at these particular locations . similar isolated features were observed previously in sgm images by other groups and were interpreted as the presence of charged traps in the semiconductor heterostructure , possibly in the doping plane , but not in the 2deg itself . @xcite in this interpretation , the trapped charges create potential perturbations in the potential landscape of the 2deg , and changing the number of these charges modifies the device conductance . in this case , approaching the tip with a negative voltage removes electrons from the traps and restores a larger device conductance . in our case , the exact opposite behavior is observed , since approaching the tip with a negative voltage strongly decreases the conductance . the phenomenon observed in our experiment is therefore not related to traps in the heterostructure , and we propose instead that the tip affects directly the 2deg potential within the following mechanism . when the tip scans above a high hill of the potential landscape , the gate effect of the tip is stronger , and it produces a spot of low conductance in the sgm image . low density regions have indeed a weak screening capability and can be easily depleted by the repulsive potential of the tip . in addition , our particular device geometry composed of narrow wires makes the conductance very sensitive to a local depletion of the 2deg . according to this proposal , which will be sustained later in the paper , sgm images reveal the spatial inhomogeneity of the 2deg and show that it is characterized by a discrete distribution of small regions where the electron density is much lower than the average value . some of the features in fig . [ fig2](e ) consist of two concentric circles that may arise either from two very close spots , or from a single spot with two successive changes . to distinguish these two possibilities , we need higher resolution images . for this purpose , we now present transconductance measurements , which are more sensitive than a simple difference between two conductance maps . the transconductance signal @xmath17 is measured with a small additional ac voltage on the tip . a series of transconductance images is shown in fig . [ fig3](b ) for different tip voltages from @xmath26 down to @xmath203.6v . note that these images have been recorded during the same cool down as in fig . [ fig2 ] , but a small electrostatic discharge may have occurred during the change of the measurement configuration resulting in a slightly different potential landscape . measured with a dc current bias @xmath19na and an ac tip voltage modulation @xmath28mv . the dc tip voltage is indicated below each image . a logarithmic color scale is used and only positive values of the transconductance are plotted . ( c ) sgm profiles extracted along the red line drawn in ( a ) . the successive profiles recorded from @xmath203.6v to + 0.6v are shifted upwards by @xmath29 for the sake of clarity ( the dotted lines indicate the zeros ) . ] at @xmath200.6v tip voltage , the sgm image shows several spots and circles which correspond to those visible in fig . [ fig2](e ) . as clearly seen in the image at @xmath201.2v , all these features are located along the device wires whose topography is shown in fig . [ fig3](a ) . for more negative tip voltages , the spots evolve into narrow circles with increasing diameters . in the central wire , the presence of many overlapping circles makes the pattern rather complex to analyze . in the lateral wires however , only a limited number of spots dominate the conductance ( one spot in the left wire , two spots in the right wire ) . several concentric circles are visible around each spot , with at least two circles for each , and up to four circles for the left spot . these circles look very much like the coulomb blockade oscillations observed previously by sgm in different kinds of quantum dots made by lithography , @xcite or present accidentally in nanowires , @xcite carbon nanotubes , @xcite and graphene . @xcite in our experiment , each spot showing concentric circles can therefore be interpreted by the presence of a quantum dot with coulomb blockade . when the tip is approached towards the dot , the electrostatic potential is raised , which results in the discharging of electrons outside the dot , one - by - one , with a conductance maximum each time a charge state crosses the fermi level . because of the large electron density in the unperturbed 2deg , these dots do not pre - exist in absence of the tip , but appear when the electron density is lowered under the tip , such that the potential fluctuations of the 2deg are brought around the fermi level . fig . [ fig1 ] illustrates this effect by showing a localized state under the tip with discrete energy levels close to the fermi level . according to this interpretation , each set of concentric circles observed in the sgm images reveals the presence of a quantum dot formed in the 2deg potential fluctuations . note that the dots are not created by the tip , as done in the past with a scanning tunneling microscope on a clean inas surface using a positive tip voltage . @xcite here , the dots result from a local lowering of the density , such as to induce locally an equivalent to the disorder - induced metal - to - insulator transition , well - known in macroscopic 2degs at low enough electron density . @xcite these sgm images are reproducible within the same cool - down in absence of external perturbation , but change if light is shined on the sample or if an electrostatic discharge occurs in the setup . an example of images obtained after a small electrostatic perturbation is given in appendix b. this behavior gives information about the origin of the electrostatic disorder , which is not structural , but results from a particular distribution of charges , located either in the doping plane , or at the surface , and which are frozen in a given configuration at low temperature . a striking feature in the transconductance images is the appearance of a disk with constant signal inside the innermost circle , which grows in size for more negative tip voltage without any new circle appearing inside . this phenomenon is also visible in the sgm profiles recorded along a single line and plotted in fig . [ fig3](c ) : the transconductance oscillations are progressively shifted away from the center and a region with flat signal develops in the middle . simultaneous conductance and transconductance measurements on a single constriction ( see appendix c ) have shown that the absence of feature inside the innermost circle corresponds to zero current in the wire . this effect can be understood by the quantum dot being completely emptied and/or the barriers becoming too high to give significant tunneling . since no current can flow around the dot ( the wire is too narrow ) , the conductance of the wire vanishes . the total conductance of the device is however not zero because some current still flows in the two other wires of the network , and circles from quantum dots in these other wires are therefore visible inside the depleted areas ( see fig . [ fig3](b ) below @xmath202.4v in the left wire ) . measured with a dc current bias @xmath30na and an ac tip voltage modulation @xmath28mv . the dc tip voltage is indicated on each image . ( c ) sgm profiles extracted along the white line in ( b ) . the successive profiles recorded from @xmath203.4v to @xmath201.3v are shifted upwards by @xmath31 . ] the concentric circles and their evolution with tip voltage are now analyzed in more details thanks to the high - resolution sgm images of the left wire plotted in fig . [ fig4](b ) for a slightly different disorder potential . four dots in series with a similar response can be identified in this wire . each dot shows a set of concentric circles , whose diameter increases for more negative tip voltages , and new circles emerge progressively from the center . below a given tip voltage , a region of constant signal appears in the middle , corresponding to zero current in the wire ( in parallel with the two other wires ) . this uniform region grows in size and merges progressively with the uniform regions of the nearby dots . careful examination of these areas without signal shows that they are not exactly centered on the dots , and sometimes , cover partly the innermost circles . this indicates that a uniform region may not correspond to an empty dot with the last electron being removed , but rather to the appearance of a thick barrier around the dot that suppresses the current . the sgm profiles in fig . [ fig4](c ) show the conductance oscillations for a single dot . about six oscillations are visible on both sides of the flat region where the wire is blocked . for a correct interpretation of the data , it is important to note that the transconductance signal corresponds to the derivative of the conductance curve with respect to gate voltage . each coulomb blockade conductance peak therefore appears as a transconductance oscillation with a negative peak immediately followed by a positive peak when the tip approaches the dot . each oscillation is progressively shifted outwards the center when the tip voltage is decreased . the shift versus tip voltage is linear below @xmath202.2v , which indicates an unscreened tip - induced potential . a detailed discussion of the potential induced by the tip in the wire in presence of screening effects is given in appendix d. in our experiment , the successive coulomb resonances are not separated by coulomb - blocked regions , indicating that the charging energy is smaller than the temperature or the intrinsic resonance width , probably broadened by the poor confinement of the disorder potential . a tip - dot capacitance @xmath32f can be deduced from the tip voltage change @xmath33v required to shift the coulomb oscillations by one period . note that the transconductance is measured with a 40mv tip voltage modulation smaller than @xmath34 in order to fully resolve the coulomb oscillations . the determination of the charging energy , however , requires the knowledge of the total dot capacitance , usually measured by source - drain bias spectroscopy of the dot . the presence of several dots in series between source and drain in this device prevents the investigation of an individual dot . in this section , we develop a simple model to show that isolated dots hosting a few electrons can appear under the sgm tip due to the disordered potential landscape inside the narrow wire . potential fluctuations in the 2deg have several origins , including alloy disorder , fluctuations of the barrier thickness , random distribution of ionized dopants , inhomogeneous density of surface charges . for simplicity , we consider only the distribution of ionized dopants as source of potential fluctuations , since it is an intrinsic source that can not be suppressed . following ref . @xcite , we calculate the potential induced by positively charged ions distributed randomly in a plane located at a distance @xmath35 nm from the 2deg , with a mean density @xmath36m@xmath37 . we use a boundary condition with a uniform potential on the surface located at a distance @xmath38 nm from the 2deg . @xcite we assume the fermi level to be pinned at mid - gap by the surface states of the ingaas cap layer , such that the conduction band edge of ingaas is at an energy @xmath39mev above the fermi level . for simplicity , the dielectric constant is taken uniform over the heterostructure , using the value @xmath40 of the inalas barrier . in the case of a uniform dopant distribution with a continuous density @xmath41 , the positively charged dopants induce an attractive potential energy @xmath42 for electrons located below the doping plane with respect to the fixed surface potential . therefore , electrons accumulate at the ingaas / inalas interface and form a 2deg with a uniform density @xmath43 , which in turn induces a repulsive potential energy @xmath44 in the 2deg plane . the electron density @xmath45 depends self - consistently on the total potential energy @xmath46 calculated with respect to the fermi level ( @xmath47 is the electron effective mass and @xmath48 is the heaviside step function ) . the condition for a non - zero electron density in the 2deg is having @xmath49 negative , such that the conduction band edge is below the fermi level ( which is set to zero as the energy reference ) . when the density is non - zero , the self - consistent potential energy writes : @xmath50 where the coefficient before the parenthesis represents the screening by the 2deg . for the chosen heterostructure parameters , this coefficient equals 0.09 and the attractive potential energy of the dopants is @xmath51mev . in order to reproduce the measured electron density @xmath52m@xmath37 , we have to set the surface potential to @xmath53mev , which is close to the value expected from a fermi level pinning at mid - gap on the surface . . the regions in gray are depleted . ( c ) energy potential profile along the central line at @xmath54 . the fermi energy is about 8mev with peak - to - peak fluctuations as large as 6mev . ] in the case of a random distribution of ionized dopants , the attractive potential energy @xmath55 is non - uniform and writes : @xmath56\end{gathered}\ ] ] where the fixed surface potential is equivalent to the presence of an image charge with opposite sign . in this case , the electron density @xmath57 , the repulsive self - energy @xmath58 , and the total potential energy @xmath59 are also non - uniform , and their exact determination would require self - consistent quantum calculations in the 2deg plane . here , we keep the calculation classical , and make the approximation of a local response using the same relations as for the uniform case . this is a first - order approximation to give an estimate of the potential fluctuations . [ fig5](a ) shows a random distribution of ionized dopants in a 200 nm wide and infinitely long wire , while fig . [ fig5](b ) shows the resulting screened potential energy in the 2deg . the finite width of the wire results in a larger attractive potential in the central region and 20 nm wide depleted regions on each side ( in gray ) . the fermi energy in this wire geometry ( about 8mev in the center ) is significantly lower than the value for the infinite plane ( 20mev ) . the most striking property is the presence of strong potential fluctuations along the wire ( fig . [ fig5](c ) ) with peak - to - peak variations ( about 6mev ) of the same order as the fermi energy ( about 8mev ) . these fluctuations are proportional to the square root of the mean dopant density @xcite and their typical length scale ( about 50 nm ) is governed by the distance between the doping plane and the 2deg . @xcite this length scale indeed corresponds to the extension of the potential induced by each dopant , which is much larger than the mean dopant spacing ( 7 nm ) . nm ( indicated by the vertical bar ) and @xmath60 nm . the dopant distribution is the same as in fig . the spatial extension of the tip - induced potential is given in the text and its amplitude is increased from 6 to 12mev as indicated in each panel . the tip creates a low density region and a small quantum dot is formed for potentials between 8 and 10mev with a small number of electrons ( indicated by black dots ) . ] in sgm experiments , for large enough negative tip voltage , the tip - induced potential brings locally the total electron potential above the fermi level and builds a barrier that blocks electron transport along the narrow wire . in presence of potential fluctuations with a local minimum under the tip , a pocket of electrons can survive in this barrier , forming a small dot between two barriers as drawn schematically in fig . [ fig1](c ) . when these confining barriers are rather symmetric , a resonant tunneling process through the dot can restore a high electron transmission for discrete energy levels . if the resistance of the barriers is larger than @xmath61 , coulomb blockade will also occur with charge quantization in the dot and a finite energy spacing between successive charge states . in this case , if the temperature is lower than the charging energy @xmath62 , discrete conductance peaks should appear as a function of the gate voltage on the tip . fig . [ fig6 ] shows an example of the formation of such a quantum dot when the tip is placed right above a local potential minimum . panels ( a ) to ( d ) show the evolution of the potential landscape in the wire when the potential energy under the tip is raised from 6 to 12mev . the shape of the tip - induced potential is chosen of the form @xmath63 with @xmath64 nm as in the experiment , corresponding to an unscreened potential ( see appendix d ) . by increasing locally the potential energy , an island of electrons forms , then shrinks , and finally disappears . to quantify the number of electrons in this island , the total charge is calculated by integration of the electron density , and each additional charge @xmath65 along the @xmath66 axis is marked by a black dot . for example , three electrons are present in the dot for the 8mev tip - induced potential . these simulations show that isolated islands with a few electrons can indeed form under the tip in presence of potential fluctuations along the wire . this result supports our interpretation of the experimental data in terms of coulomb blockade in quantum dots formed in the 2deg disorder potential . sgm has been used to investigate locally the disorder - induced potential fluctuations in a 2deg patterned into narrow wires . the sgm images reveal that a few discrete spots dominate the total resistance , corresponding to hills in the potential landscape . in addition , several concentric circles appear in the transconductance images , which are very similar to those observed previously for real quantum dots in the coulomb blockade regime . these features indicate the presence of localized states in the 2deg , confined between two hills of the disorder potential , when the tip lowers locally the electron density . additional characterizations of these dots should be done in the future , in particular source - drain bias spectroscopy of a single dot in a short constriction , to measure their charging energy and level spacing . this work has been supported by the french agence nationale de la recherche ( micatec project ) , by the grenoble nanosciences foundation ( scanning - gate nanoelectronics project ) , by frs - fnrs projects j.0067.13 and u.n025.14 , by frfc grant no . 2.4503.12 , and by the arc project `` stresstronics '' . b.h . and f.m . are research associate and postdoctoral fellow of the frs - fnrs , respectively . the work function of a semiconductor with fermi level pinning at mid - gap is given by @xmath67 where @xmath68 is the electron affinity and @xmath69 the band gap . according to this formula , the in@xmath70ga@xmath71as cap layer has a work function @xmath72ev ( similar to the value @xmath73ev for gaas ) . the afm tip ( pointprobeplus from nanosensors ) coated with a layer of pt@xmath74ir@xmath75 alloy has a work function @xmath76ev , as measured by kelvin probe force microscopy @xcite ( note that @xmath77ev and @xmath78ev ) . the tip voltage @xmath26 that compensates for the work function difference between the tip and the surface ( also called flat band potential ) is therefore equal to @xmath79v for an ingaas surface ( @xmath80v for a gaas surface ) . this value is consistent with the value @xmath80v extracted from fig . [ fig2](d ) . for a slightly different disorder potential . ( a ) topography recorded just before the sgm measurements . ( b ) sgm images of the transconductance @xmath17 measured with a dc current bias @xmath30na and an ac tip voltage modulation @xmath28mv . the dc tip voltage is indicated on each image . ( c ) sgm profiles extracted along the red line drawn in ( a ) . the successive profiles recorded from @xmath205v to @xmath203v are shifted upwards by @xmath81 ( the dotted lines indicate the zeros ) . ] [ fig7 ] shows a set of sgm images recorded during the same cool - down as for fig . [ fig3 ] but after a refilling of the cryostat with liquid helium . several dots can be recognized in the two sets of images , but some are new and others have disappeared . a small electrostatic discharge may have occurred in the cryostat during this operation , explaining a change of the charge distribution in the heterostructure , resulting in a slightly different potential landscape in the 2deg . sgm profiles across a single dot in the upper wire are plotted in fig . [ fig7](c ) . the conductance oscillations are rather large ( amplitude up to @xmath81 ) because this dot blocks the transport through two of the three wires of the device . when the tip voltage is lowered , the coulomb peaks move away from the center and become sharper . this entails the faster potential change experienced by the dot when the tip is scanned with a larger negative voltage . at @xmath205v tip voltage , the transconductance becomes flat and almost zero in the center because the local potential under the tip is so high that the current in the wire is completely blocked and the transconductance signal is suppressed . measured with a dc voltage bias @xmath82mv and an ac tip voltage modulation @xmath28mv . the dc tip voltage is indicated on each image . ( c ) sgm images of the conductance @xmath12 measured with an ac voltage bias @xmath83v simultaneously to the transconductance . ( d , e ) sgm transconductance and conductance profiles extracted along the white line in ( b , c ) for tip voltages from @xmath202v to @xmath209v ( from top to bottom ) . in ( d ) , the successive profiles are shifted by @xmath31 . ] a different configuration of the disorder potential was obtained by shining light on the sample at low temperature and waiting for charge noise relaxation . in this configuration , the upper device constriction shown in fig . [ fig8](a ) exhibits the strongest sgm response and dominates the device resistance . this situation corresponds to the presence of several negative charges in this region , which are frozen at the surface or in the doping plane , and raise locally the 2deg potential . simultaneous conductance and transconductance sgm measurements have been carried out in this region using a dual reference lock - in and plotted in figs . [ fig8](b , c ) . up to six dots can be identified in these images , arranged in parallel with respect to the current flow and controlling the amount of current flowing between the reservoir and the device . for tip voltages below @xmath206v , a region with zero current and zero transconductance appears in the middle of the image , with a contour delimited by portions of different circles . this region corresponds to the overlap of the blocked regions created by the different dots . individually , the dots can not block the current because they are arranged in parallel rather than in series and several parallel paths are available for the current [ fig8](d , e ) show that weak conductance modulations correspond to sharp transconductance oscillations : when coulomb blockade effects are weak , transconductance measurements strongly improve their detection in sgm images . on curves recorded with tip voltages lower than @xmath207v , the region with a flat transconductance signal corresponds exactly to the region where the conductance is zero . this result shows that a flat transconductance signal usually indicates a vanishing current in the probed region . a direct measure of the energy change induced by the tip in the dots would require source - drain bias spectroscopy of an individual dot , @xcite but this study can not be done here because of the multichannel character of the branched device . alternatively , we investigate the potential induced by the tip in the 2deg by measuring continuously the size of the concentric circles versus tip voltage while scanning a single line . @xcite fig . [ fig9](a ) shows the evolution of the transconductance signal along a vertical line in the middle of fig . [ fig8](a ) while sweeping the tip voltage . following a given transconductance peak in this voltage - position diagram gives a trace @xmath84 that corresponds to an iso - potential line for the dot , i.e. a line where the potential induced in the dot is constant . @xcite from a theoretical point of view , the tip - induced potential along the @xmath85 axis can be written : @xmath86 where @xmath87 is the tip - dot capacitance , @xmath88 is the total dot capacitance , and @xmath26 is the flat band voltage . the quantity @xmath89 represents the position - dependent lever - arm parameter between the tip voltage and the potential induced in the 2deg . this parameter can be determined from an iso - potential line @xmath84 , and then used to get the spatial dependence of the tip - induced potential at fixed tip voltage : @xmath90 in fig . [ fig9](a ) , it is not clear if the different traces correspond to different dots or to the successive charge states of the same dot , and the precise extraction of an iso - potential line is difficult . in the following , we adopt an alternative approach and compare the experimental figure with one resulting from the modeling of the potential induced by the tip in the dot with and without screening . axis at @xmath91 m in fig . [ fig8](a ) for several tip voltages from 0 to @xmath2010v with 10mv steps . the transconductance @xmath17 is measured with a dc voltage bias @xmath82mv and an ac tip voltage modulation @xmath28mv ( note that the weak tilted parallel lines covering the full plot are artifacts from a parasitic interference signal ) . ( b ) simulation of the transconductance signal versus tip position and voltage ( same axes as in ( a ) ) , using the modeled conductance curve shown in ( c ) and the tip - dot couplings shown in ( d ) without screening ( left ) and with screening ( right ) . ( c ) model for the quantum dot conductance versus tip voltage , when the tip is exactly above the dot . this curve simulates the gate effect around threshold and includes coulomb blockade oscillations . ( d ) model of the tip - dot coupling ( or potential induced in the 2deg plane ) versus tip position , according to eq . [ eqn1 ] without screening ( red line ) and eq . [ eqn2 ] with screening ( green line ) . the minimum tip - dot separation @xmath92 is 200 nm and 500 nm for the red and green lines , respectively . ] for this purpose , we approximate the tip as a point charge @xmath93 for which an analytic solution is possible . this charge is placed in vacuum above the surface of a semiconductor with dielectric constant @xmath94 . the potential created inside a semiconductor , at coordinates @xmath95 relative to the charge , writes : @xmath96 this expression is plotted as a red curve in fig . [ fig9](d ) for fixed values of @xmath66 and @xmath97 . if screening from the surrounding 2deg can be neglected , this relation gives the potential variations of a dot inside the semiconductor , when a charge @xmath98 is scanned in an horizontal plane above the surface with coordinates @xmath95 relative to the dot ( @xmath97 is the sum of the charge height above the surface and the dot depth below the surface ) . to check if this unscreened @xmath99 dependence is consistent with the data in fig . [ fig9](a ) , we simulate the transconductance signal in the presence of the tip . for this purpose , we model the dot conductance as shown in fig . [ fig9](c ) , with a global drop due to the local depletion and weak coulomb oscillations due to the disorder potential . this phenomenological model reproduces the typical behavior of the conductance curve versus gate voltage for a quantum dot in a disordered wire . @xcite the tip plays here the role of the gate . the left panel of fig . [ fig9](b ) shows the expected transconductance signal versus tip position and voltage . this plot reproduces qualitatively the experimental traces in fig . [ fig9](a ) if the minimum tip - dot distance is adjusted to @xmath100 nm . since the dot is in the 2deg plane located at 42 nm below the surface and the tip is at 100 nm above the surface , i.e. @xmath101 nm , the horizontal distance between the scanning line and the dot is found to be @xmath102 nm . the linear asymptotic behavior of the experimental traces at large distance is well reproduced by this model without screening . note that the successive charge states of the dot have different asymptotic slopes in the model , whereas the experimental traces are parallel to each other and may therefore correspond to different dots . the above expression without screening predicts a very large tip - induced potential , which is not realistic . in reality , this potential is partially screened by the 2deg , which is grounded to zero volt at the ohmic contacts . unfortunately , no analytical expression exists for the real case of a 2deg embedded inside a semiconductor host and perturbed by a charge above the surface . in the following , we treat the closest situation as possible , which has an analytical solution , i.e. a 2deg at the surface of the semiconductor . we therefore neglect the dielectric constant of the semiconductor barrier above the 2deg and keep it only below the 2deg . in the regime of linear response ( no depleted region in the 2deg ) and in the thomas - fermi approximation ( short fermi wavelength ) , the potential in the 2deg at a radial distance @xmath103 from a point charge @xmath98 placed in vacuum at distance @xmath97 above the 2deg , can be calculated with the formula : @xcite @xmath104 where @xmath105 is the screening wave vector , @xmath106 is the zeroth - order bessel function , and @xmath94 is the dielectric constant of the semiconductor located below the 2deg . this formula is almost equivalent to the expression : @xmath107 where the integral @xmath108 is a function of the dimensionless parameter @xmath109 . using @xmath110 for ingaas gives @xmath111nm@xmath8 , then @xmath112 and @xmath101 nm give @xmath113 . in this situation , @xmath114 can be approximated by @xmath115 and the potential becomes : @xmath116 this expression is plotted as a green curve in fig . [ fig9](d ) and is independent of the semiconductor dielectric constant @xmath94 because of the large screening by the 2deg located at the surface . this expression predicts a faster @xmath117 potential decay at large distance than eq . [ eqn1 ] without screening . the expected transconductance traces for this screened potential are shown in the right panel of fig . [ fig9](b ) . their nonlinear asymptotic behavior at large distance differ from the linear behavior of the experimental traces in fig . [ fig9](a ) , which are better reproduced by the model without screening . this result might be explained by the very low electron density close to the depletion threshold where these traces have been measured . axis in the middle of the device ( @xmath118 m in fig . [ fig2](a ) ) for several tip voltages from @xmath203.6 to 0v with 30mv steps . these data are recorded for a large device resistance ( 40k@xmath11 ) after an electrostatic discharge . the transconductance @xmath17 is measured with an ac tip voltage modulation @xmath28mv . ( b ) simulation of the transconductance signal versus tip position and voltage ( same axes as in ( a ) ) , without screening ( left ) and with screening ( right ) , like in fig . [ fig9](b ) . two conductance curves slightly different from that in fig . [ fig9](c ) are used to model two different dots contributing in parallel to the total conductance . the tip - induced potentials with and without screening are the same as in fig . [ fig9](d ) . ] according to eq . [ eqn2 ] , screening by the 2deg gives a reduction of the tip - induced potential by a factor @xmath119 , which gives a more realistic estimate of the potential in the 2deg , as explained in the following . the charge @xmath98 that dresses the sgm tip can be estimated using the sphere - plane capacitance model . the conical part of the tip above the apex also contributes to the tip - induced potential , but is not considered here . the tip is modeled by a metallic sphere of radius @xmath120 biased at a voltage @xmath121 relative to the grounded 2deg at a gap distance @xmath97 . its capacitance can be written @xcite @xmath122 , where the function @xmath123 for @xmath124 , @xmath125 , @xmath126 , @xmath127 . since @xmath128 for a sharp tip with small curvature radius , we can reasonably assume @xmath129 . in this case , the charge is given by @xmath130 . in this model , the screened potential in the 2deg under the tip writes : @xmath131 this potential is of the order of 3mv for a 3v tip voltage and a 30 nm tip radius ( tip with metallic coating ) . since the depletion of the 2deg is obtained experimentally for a tip voltage of a few volts in the regions where dots are observed , we can estimate the fermi energy to be about 3mev in these regions . this small energy is consistent with our simulation of the disorder potential in the wire ( see fig . [ fig5](c ) ) where a fermi energy as small as 5mev is obtained on the highest potentiall hill . the existence of such high potential hills explains the strong response in the sgm images and the formation of quantum dots . note that this model assumes an infinite 2deg , whereas the device is etched into wires , which reduces the amount of screening as compared to the model . for some disorder configuration , the different dots can be sufficiently far from each other to make the analysis of the dot characteristics easier . [ fig10](a ) corresponds to such a case , with only three dots , one in each of the three parallel wires ( the scanning line is in the @xmath66 direction along the symmetry axis of the device ) . this plot was recorded in very different conditions than for previous data , with a very large device resistance . about five traces are visible for the central dot and for the right dot . two traces close to zero tip voltage come from a third dot in the left wire ( this dot is rapidly depleted for negative tip voltages ) . in fig . [ fig10](b ) , we compare these experimental traces with theoretical ones calculated using the above model . the figure shows the iso - potential traces for successive charge states of two different dots contributing in parallel to the conductance . the left panel corresponds to the unscreened tip - induced potential with @xmath99 decay ( eq . [ eqn1 ] ) and the right panel to the screened potential with @xmath117 decay ( eq . [ eqn2 ] ) . the traces in the left panel reproduce better the shape of the experimental ones at large distance , as if the potential would not be screened . however , with eq . [ eqn1 ] , the tip voltage has to be artificially reduced by a factor 100 to give a realistic potential change in the 2deg on the order of the fermi energy , whereas eq . [ eqn2 ] gives reasonable values . this analysis shows that screening effects are rather difficult to understand quantitatively in highly nonuniform systems like nanoscale devices , and would require 3d self - consistent calculations @xcite to be correctly taken into account . in addition , part of the discrepancy may result from the point charge model used for the tip , and numerical calculations would be necessary to treat correctly the actual shape of the tip . 99 t. ando , a. b. fowler , and f. stern , rev . phys . * 54 * , 437 ( 1982 ) . j. j. harris , j. a. pals , and r. woltjer , rep . prog . phys . * 52 * , 1217 ( 1989 ) . h. l. strmer , surface sci . * 132 * , 519 ( 1983 ) . j. m. rorison , m. j. kane , d. c. herbert , m. s. skolnick , l. l. taylor , and s. j. bass , semicond . technol . * 3 * , 12 ( 1988 ) . s. ilani , a. yacoby , d. mahalu , and h. shtrikman , science * 292 * , 1354 ( 2001 ) . j. shi and x. c. xie , phys . lett . * 88 * , 086401 ( 2002 ) . s. das sarma , m. p. lilly , e. h. hwang , l. n. pfeiffer , k.w . west , and j. l. reno , phys . 94 * , 136401 ( 2005 ) . m. baenninger , a. ghosh , m. pepper , h. e. beere , i. farrer , and d. a. ritchie , phys . rev 100 * , 016805 ( 2008 ) . d. neilson and a. r. hamilton , phys . b * 82 * , 035310 ( 2010 ) . s. washburn , a. b. fowler , h. schmid , and d. kern , phys . b * 38 * , 1554 ( 1988 ) . j. h. davies and j. a. nixon , phys . rev . b * 39 * , 3423 ( 1989 ) . j. a. nixon and j. h. davies , phys . b * 41 * , 7929 ( 1990 ) . m. j. laughton , j. r. barker , j. a. nixon , and j. h. davies , phys . b * 44 * , 1150 ( 1991 ) . j. h. f. scott - thomas , s. b. field , m. a. kastner , h. i. smith , and d. a. antoniadis , phys . lett . * 62 * , 583 ( 1989 ) . u. meirav , m. a. kastner , m. heiblum , and s. j. wind , phys . b * 40 * , 5871 ( 1989 ) . s. b. field , m. a. kastner , u. meirav , j. h. f. scott - thomas , d. a. antoniadis , h. i. smith , and s. j. wind , phys . b * 42 * , 3523 ( 1990 ) . a. a. m. staring , h. van houten , c. w. j. beenakker , and c. t. foxon , phys . rev . b * 45 * , 9222 ( 1992 ) . j. t. nicholls , j. e. f. frost , m. pepper , d. a. ritchie , m. p. grimshaw , and g. a. c. jones , phys . b * 48 * , 8866 ( 1993 ) . w. poirier , d. mailly , and m. sanquer , phys . b * 59 * , 10856 ( 1999 ) . j. a. nixon , j. h. davies , and h. u. baranger , phys . rev . b * 43 * , 12638 ( 1991 ) . j. faist , p. guret , and h. rothuizen , phys . b * 42 * , 3217 ( 1990 ) . i. a. larkin and e. v. sukhorukov , phys . rev . b * 49 * , 5498 ( 1994 ) . p. a. lee and a. d. stone , phys . * 55 * , 1622 ( 1985 ) . h. van houten , b. j. van wees , m. g. j. heijman , and j. p. andr , appl . phys . lett . * 49 * , 1781 ( 1986 ) . m. morgenstern , j. klijn , chr . meyer , m. getzlaff , r. adelung , r. a. rmer , k. rossnagel , l. kipp , m. skibowski , and r. wiesendanger , phys . * 89 * , 136806 ( 2002 ) . j. wiebe , chr . meyer , j. klijn , m. morgenstern , and r. wiesendanger , phys . b * 68 * , 041402 ( 2003 ) . s. h. tessmer , p. i. glicofridis , r. c. ashoori , l. s. levitov , and m. r. melloch , nature * 392 * , 51 ( 1998 ) . g. finkelstein , p. i. glicofridis , r. c. ashoori , and m. shayegan , science * 289 * , 90 ( 2000 ) . g. a. steele , r. c. ashoori , l. n. pfeiffer , and k. w. west , phys . lett . * 95 * , 136804 ( 2005 ) . m. j. yoo , t. a. fulton , h. f. hess , r. l. willett , l. n. dunkleberger , r. j. chichester , l. n. pfeiffer , and k. w. west , science * 276 * , 579 ( 1997 ) . n. b. zhitenev , t. a. fulton , a. yacoby , h. f. hess , l. n. pfeiffer , and k. w. west , nature * 404 * , 473 ( 2000 ) . s. ilani , j. martin , e. teitelbaum , j. h. smet , d. mahalu , v. umansky , and a. yacoby , nature * 427 * , 328 ( 2004 ) . m. t. woodside , c. vale , p. l. mceuen , c. kadow , k. d. maranowski , and a. c. gossard , phys . rev . b * 64 * , 041310(r ) ( 2001 ) . a. baumgartner , t. ihn , k. ensslin , k. maranowski , and a. c. gossard , phys . b * 76 * , 085316 ( 2007 ) . n. paradiso , s. heun , s. roddaro , d. venturelli , f. taddei , v. giovannetti , r. fazio , g. biasiol , l. sorba , and f. beltram , phys . b * 83 * , 155305 ( 2011 ) . f. martins , s. faniel , b. rosenow , m. g. pala , h. sellier , s. huant , l. desplanque , x. wallart , v. bayot , and b. hackens , new j. phys . * 15 * , 013049 ( 2013 ) . s. chakraborty , i. j. maasilta , s. h. tessmer , and m. r. melloch , phys . b * 69 * , 073308 ( 2004 ) . m. a. topinka , b. j. leroy , r. m. westervelt , s. e. j. shaw , r. fleischmann , e. j. heller , k. d. maranowski , and a. c. gossard , nature * 410 * , 183 ( 2001 ) . m. p. jura , m. a. topinka , l. urban , a. yazdani , h. shtrikman , l. n. pfeiffer , k. w. west , and d. goldhaber - gordon , nat . phys . * 3 * , 841 ( 2007 ) . a. a. kozikov , c. rssler , t. ihn , k. ensslin , c. reichl , and w. wegscheider , new . j. phys . * 15 * , 013056 ( 2013 ) . b. brun , f. martins , s. faniel , b. hackens , g. bachelier , a. cavanna , c. ulysse , a. ouerghi , u. gennser , d. mailly , s. huant , v. bayot , m. sanquer , and h. sellier , nat . commun . * 5 * , 4290 ( 2014 ) . n. aoki , c. r. da cunha , r. akis , d. k. ferry , and y. ochiai , appl . phys . lett . * 87 * , 223501 ( 2005 ) . c. r. da cunha , n. aoki , t. morimoto , y. ochiai , r. akis , and d. k. ferry , appl . lett . * 89 * , 242109 ( 2006 ) . m. g. pala , b. hackens , f. martins , h. sellier , v. bayot , s. huant , and t. ouisse , phys . b * 77 * , 125310 ( 2008 ) . m. g. pala , s. baltazar , f. martins , b. hackens , h. sellier , t. ouisse , v. bayot , and s. huant , nanotechnol . * 20 * , 264021 ( 2009 ) . t. chwiej and b. szafran , phys . b * 87 * , 085302 ( 2013 ) . a. pioda , s. kicin , d. brunner , t. ihn , m. sigrist , k. ensslin , m. reinwald , and w. wegscheider , phys . b * 75 * , 045433 ( 2007 ) . a. e. gildemeister , t. ihn , r. schleser , k. ensslin , d. c. driscoll , and a. c. gossard , j. appl . phys . * 102 * , 083703 ( 2007 ) . h. sellier , b. hackens , m. g. pala , f. martins , s. baltazar , x. wallart , l. desplanque , v. bayot , and s. huant , semicond . * 26 * , 064008 ( 2011 ) . d. k. ferry , a. m. burke , r. akis , r. brunner , t. e. day , r. meisels , f. kuchar , j. p. bird , and b. r. bennett , semicond . * 26 * , 043001 ( 2011 ) . a. c. bleszynski , f. a. zwanenburg , r. m. westervelt , a. l. roest , e. p. a. m. bakkers , and l. p. kouwenhoven , nano lett . * 7 * , 2559 ( 2007 ) . m. t. woodside and p. l. mceuen , science * 10 * , 1098 ( 2002 ) . m. r. connolly , k. l. chiu , a. lombardo , a. fasoli , a. c. ferrari , d. anderson , g. a. c. jones , and c. g. smith , phys . b * 83 * , 115441 ( 2011 ) . n. pascher , d. bischoff , t. ihn , and k. ensslin , appl . * 101 * , 063101 ( 2012 ) . a. g. f. garcia , m. knig , d. goldhaber - gordon , and k. todd , phys . b * 87 * , 085446 ( 2013 ) . r. crook , c. g. smith , w. r. tribe , s. j. oshea , m. y. simmons , and d. a. ritchie , phys . b * 66 * , 121301(r ) ( 2002 ) . n. aoki , a. burke , c. r. da cunha , r. akis , d. k. ferry , and y. ochiai , j. phys . : conf . ser . * 38 * , 79 ( 2006 ) . x. wallart , b. pinsard , and f. mollot , j. appl . phys . * 97 * , 053706 ( 2005 ) . f. martins , b. hackens , m. g. pala , t. ouisse , h. sellier , x. wallart , s. bollaert , a. cappy , j. chevrier , v. bayot , and s. huant , phys . lett . * 99 * , 136807 ( 2007 ) . t. vancura , s. kicin , t. ihn , k. ensslin , m. bichler , and w. wegscheider , appl . . lett . * 83 * , 2602 ( 2003 ) . a. pioda , s. kicin , t. ihn , m. sigrist , a. fuhrer , k. ensslin , a. weichselbaum , s. e. ulloa , m. reinwald , and w. wegscheider , phys . lett . * 93 * , 216801 ( 2004 ) . a. e. gildemeister , t. ihn , m. sigrist , k. ensslin , d. c. driscoll , and a. c. gossard , phys . b * 75 * , 195338 ( 2007 ) . r. crook , c. g. smith , m. y. simmons , and d. a. ritchie , j. phys . : condens . matter * 13 * , l249 ( 2001 ) . s. kicin , a. pioda , t. ihn , m. sigrist , a. fuhrer , k. ensslin , m. reinwald , and w. wegscheider , new j. phys . * 7 * , 185 ( 2005 ) . p. fallahi , a. c. bleszynski , r. m. westervelt , j. huang , j. d. walls , e. j. heller , m. hanson , and a. c. gossard , nano lett . * 5 * , 223 ( 2005 ) . r. dombrowski , chr . steinebach , chr . wittneven , m. morgenstern , and r. wiesendanger , phys . b * 59 * , 8043 ( 1999 ) . this `` pinned surface '' boundary condition provides simple expressions for the potentials , but corresponds in principle to a room temperature situation where the surface states are conducting . t. heim , k. lmimouni , and d. vuillaume , nano lett . * 4 * , 2145 ( 2004 ) . c. kim , b. lee , h. j. yang , h. m. lee , j. g. lee , and h. shin , j. korean phys . soc . * 47 * , s417 ( 2005 ) . e. j. spadafora , r. demadrille , b. ratier , and b. grvin , nano lett . * 10 * , 3337 ( 2010 ) . f. stern , phys . lett . * 18 * , 546 ( 1967 ) . a. d. karsono and d. r. tilley , j. phys . c. : solid st . * 10 * , 2123 ( 1977 ) . m. r. krcmar and w. m. saslow , phys . b * 66 * , 235310 ( 2002 ) . y. oyama , y. majima , and m. iwamotoa , j. appl . phys . * 86 * , 7087 ( 1999 ) . b. szafran , phys . b * 84 * , 075336 ( 2011 ) .
in this paper we investigate a class of swing options with firm constraints in view of the modeling of supply agreements . we show , for a fully general payoff process , that the premium , solution to a stochastic control problem , is concave and piecewise affine as a function of the global constraints of the contract . the existence of bang - bang optimal controls is established for a set of constraints which generates by affinity the whole premium function . when the payoff process is driven by an underlying markov process , we propose a quantization based recursive backward procedure to price these contracts . a priori error bounds are established , uniformly with respect to the global constraints . _ key words : swing option , stochastic control , optimal quantization , energy . _
0705.0466
the deregulation of energy markets has given rise to various families of contracts . many of them appear as some derivative products whose underlying is some tradable futures ( day - ahead , etc ) on gas or electricity ( see @xcite for an introduction ) . the class of swing options has been paid a special attention in the literature , because it includes many of these derivative products . a common feature to all these options is that they introduce some risk sharing between a producer and a trader , of gas or electricity for example . from a probabilistic viewpoint , they appear as some stochastic control problems modeling multiple optimal stopping problems ( the control variable is the purchased quantity of energy ) ; see @xmath0 @xcite in a continuous time setting . gas storage contracts ( see @xcite , @xcite ) or electricity supply agreements ( see @xcite , @xcite ) are examples of such swing options . indeed , energy supply contracts are one simple and important example of such swing options that will be deeply investigated in this paper ( see below , see also @xcite for an introduction ) . it is worth mentioning that this kind of contracts are slightly different from multiple exercises american options as considered in @xcite for example . in our setting the volumetric constraints play a key role and thus , the flexibility is not restricted to time decisions , but also has to take into account volumes management . designing efficient numerical procedures for the pricing of swing option contracts remains a very challenging question as can be expected from a possibly multi - dimensional stochastic control problem subject to various constraints ( due to the physical properties of the assets like in storage contracts ) . most recent approaches developed in mathematical finance , especially for the pricing of american options , have been adapted and transposed to the swing framework : tree ( or forest " ) algorithms in the pioneering work @xcite , least squares regression mc methods ( see @xcite ) , pde s numerical methods ( finite elements , see @xcite ) . the aim of this paper is to deeply investigate an old question , namely to elucidate the structure of the optimal control in supply contracts ( with firm constraints ) and how it impacts the numerical methods of pricing . we will provide in a quite general ( and abstract ) setting some natural " ( and simple ) conditions involving the local and global purchased volume constraints to ensure the existence of _ bang - bang optimal strategy _ ( such controls usually do not exist ) . it is possible to design _ a priori _ the contract so that their parameters satisfy these conditions . to our knowledge very few theoretical results have been established so far on this problem ( see however @xcite in a markovian framework for contracts with penalized constraints and @xcite , also in a markovian framework ) . this first result of the paper not only enlightens the understanding of the management of a swing contract : it also has some deep repercussions on the numerical methods to price it . as a matter of fact , taking advantage of the existence of a bang - bang optimal strategy , we propose and analyze in details ( when the underlying asset has a markovian dynamics ) a quantized dynamic programming procedure to price any swing options whose volume constraints satisfy the bang - bang " assumption . furthermore some _ a priori _ error bounds are established . this procedure turns out to be dramatically efficient , as emphasized in the companion paper @xcite where the method is extensively tested with assets having multi - factor gaussian underlying dynamics and compared to the least squares regression method . [ [ the - abstract - swing - contract - with - firm - constraints ] ] the abstract swing contract with firm constraints + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the holder of a supply contract has the right to purchase periodically ( daily , monthly , etc ) an amount of energy at a given unitary price . this amount of energy is subject to some lower and upper local " constraints . the total amount of energy purchased at the end of the contract is also subject to a global " constraint . given dynamics on the energy price process , the problem is to evaluate the price of such a contract , at time @xmath1 when it is emitted and during its whole life up to its maturity . to be precise , the owner of the contract is allowed to purchase at times @xmath2 , @xmath3 a quantity @xmath4 of energy at a unitary _ strike price _ @xmath5 . at every date @xmath2 , the purchased quantity @xmath4 is subject to the firm _ local " constraint _ , @xmath6 whereas the global purchased quantity @xmath7 is subject to the ( firm ) global constraint @xmath8\qquad ( 0<q_{\min } \le q_{\max}<+\infty).\ ] ] the strike price process @xmath9 can be either deterministic ( even constant ) or stochastic , @xmath0 indexed on past values of other commodities ( oil , etc ) . usually , on energy markets the price is known through future contracts @xmath10 where @xmath11 denotes the price at time @xmath12 of the forward contract delivered at maturity @xmath13 . the available data at time @xmath14 are @xmath15 ( in real markets this is of course not a continuum ) . the underlying asset price process , temporarily denoted @xmath16 , is often the so - called day - ahead " contract @xmath17 which is a tradable instrument or the spot price @xmath18 which is not . all the decisions about the contract need to be adapted to the filtration of @xmath19 @xmath20 @xmath21 , @xmath3 ( with @xmath22 ) . this means that the price of such a contract is given at any time @xmath23 , by @xmath24,j = k,\ldots , n-1 , \sum_{j = k}^{n-1}q_j\!\in [ q^k_{\min } , q^k_{\max}]\right\}\end{aligned}\ ] ] where @xmath25 , @xmath26 denote the _ residual global constraints _ and @xmath27 denotes the ( deterministic ) interest rate . this pricing problem clearly appears as a stochastic control problem . in the pioneering work by @xcite , this type of contract was computed by using some forests of ( multinomial ) trees . a natural variant , at least for numerical purpose , is to consider a penalized version of this stochastic control . thus , in @xcite , a penalization @xmath28 with @xmath29 is added ( @xmath30 is negative outside @xmath31 $ ] and zero inside ) . as concerns more sophisticated contracts ( like storages ) , the holder of the contract receive a quantity @xmath32 when deciding @xmath33 . when dealing with gas this is due to the storing constraints since injecting or withdrawing gas from its storing units induce fixed costs ( and physical constraints ( pressure , etc ) ) . as concerns the underlying asset dynamics , it is commonly shared in finance to assume that the traded asset has a markovian dynamics ( or is a component of a markov process like with stochastic volatility models ) . the dynamics of physical assets for many reasons ( some of them simply coming from history ) are often modeled using some more deeply non - markovian models like long memory processes , etc . all these specific features of energy derivatives suggest to tackle the above pricing problem in a rather general framework , trying to avoid as long as possible to call upon markov properties . this is what we do in the first part of the paper where the general setting of a swing option defined by an abstract sequence of @xmath34-adapted payoffs is deeply investigated as a function of its global constraints @xmath35 ( when the local constraints are normalized @xmath20 @xmath33 is @xmath36$]-valued for every @xmath37 ) . we show that this premium is a concave , piecewise affine , function of the global constraints , affine on triangles of the @xmath38 , @xmath39 , @xmath40 and @xmath41 , @xmath42 , @xmath43 . we also show that for integral valued global constraints , the optimal controls are always bang - bang @xmath20 the _ a priori _ @xmath36$]-valued optimal purchased quantities @xmath44 are in fact always equal to @xmath14 or @xmath45 . such a result can be extended in some way to any couple of global constraints when all the payoffs are nonnegative . then , when there is an underlying markov structure process " , we propose an optimal quantization based on numerical approach to price efficiently swing options . this markov structure process " can be the underlying traded asset itself or a higher dimensional hidden markov process : such a framework comes out in case of multi - factor processes having some long - memory properties . optimal quantization was first introduced as a numerical method to solve nonlinear problem arising in mathematical finance in a series of papers @xcite devoted to the pricing and hedging of american style multi - asset options . it has also been applied to stochastic control problem , namely portfolio optimization in @xcite . the purely numerical aspects induced by optimal quantization , with a special emphasis on the gaussian distribution , have been investigated in @xcite . see @xcite for a survey on numerical application of optimal quantization to finance . for other applications ( to automatic classification , clustering , etc ) , see also @xcite . in this paper , we propose a quantized backward dynamic programming to approximate the premium of a swing contract . we analyze the rate of convergence of this algorithm and provide some _ a priori _ error bounds in terms of quantization errors . we illustrate the method by computing the whole graph of the premium viewed as a function of the global constraints , combining the affine property of the premium and the quantized algorithm in toy model " : the future prices of gas are modeled by a two factor gaussian model . an extensive study of the pricing method by optimal quantization is carried out from both a theoretical and numerical point of view in @xcite . the paper is organized as follows . in the section below we detail the decomposition of swing options into a swap contract and a normalized swing option . in section [ abstractswing ] , we precisely describe our abstract setting for normalized swing options with firm constraints and the variable of interest ( global constraints , local constraints , etc ) . in section [ progdynabst ] , we establish the dynamic programming formula satisfied in full generality by the premium as a function of the global constraints ( this unifies the similar results obtained in markov settings , see @xcite , @xcite , etc ) and we show this is a concave function with respect to the global constraints . then , in section [ piecaffine ] , we prove in our abstract framework that the premium function is piecewise affine and that the optimal purchased quantities satisfy a @xmath14-@xmath45 " or bang - bang principle ( theorem [ main ] ) . a special attention is paid to the @xmath46-period model which provides an intuitive interpretation of the results . in section [ swingquant ] , after some short background on quantization and its optimization , we propose a quantization based backward dynamic programming formula as a numerical method to solve the swing pricing problem . then we provide some error bounds for the procedure depending on the quantization error induced by the quantization of the markov structure process . @xmath47 the lipschitz coefficient of a function @xmath48 is defined by @xmath49_{\rm lip}:=\sup_{x\neq y}\frac{|f(x)-f(y)|}{|x - y|}\le + \infty$ ] . the coefficient @xmath50_{\rm lip}$ ] is finite if and only if @xmath51 lipschitz continuous . @xmath47 the canonical euclidean norm on @xmath52 will be denoted @xmath53 . as a first step we need to normalize this contract to reduce some useless technical aspects . in practice this normalization , in fact decomposition , corresponds to the splitting of the contract into a swap and a normalized swing . the decomposition can be derived from the fact that an @xmath54-measurable random variable @xmath55 is @xmath56$]-valued if and only if there exists a @xmath36$]-valued @xmath54-measurable random variable @xmath57 such that @xmath58 . then , for every @xmath59 , @xmath60,n}_k\left(\widetilde q^k_{\min},\widetilde q^k_{\max}\right)}_{\mbox{\em normalized contract}}\end{aligned}\ ] ] where @xmath61 and @xmath62,n}_k\left(\widetilde q^k_{\min},\widetilde q^k_{\max}\right)$ ] is a _ normalized swing contract _ in which the local constraints are @xmath36$]-valued . one considers a sequence @xmath63 of integrable random variables defined on a probability space @xmath64 . let @xmath65 denote its natural filtration . for convenience we introduce a more general discrete time filtration @xmath66 to which @xmath67 is adapted @xmath20 satisfying @xmath68 , @xmath69 @xmath20 such that the sequence @xmath63 is @xmath70-adapted . we aim to solve the following abstract stochastic control problem ( with _ maturity _ @xmath71 ) @xmath72,0\le k\le \ldots , n-1 , \sum_{k=0}^{n-1}q_k\!\in [ q_{\min } , q_{\max}]\right\}\ ] ] where @xmath73 and @xmath74 are two non - negative @xmath75-measurable random variables satisfying @xmath76 ( the inequality @xmath77 induces no loss of generality : one can always replace @xmath74 by @xmath78 in ( [ swing ] ) since in any case @xmath79 ) . note that no assumption is made on the dynamics of the state process @xmath67 . we need to introduce the following notations and terminology : throughout the paper , @xmath80 will always be taken with respect to the probability @xmath81 so @xmath81 will be dropped from now on . a couple @xmath82 of non - negative @xmath83-measurable random variables satisfying @xmath84 is called a couple of _ global constraints _ at time @xmath85 . an @xmath86-adapted sequence @xmath87 of @xmath36$]-valued r.v . is called a _ locally admissible control_. for any locally admissible control , one defines the cumulative purchase process by @xmath88 if @xmath89 $ ] , @xmath55 is called an _ @xmath90-admissible control_. for every @xmath91 and every couple of global constraints @xmath82 at time @xmath85 , set @xmath92 so that @xmath93 is the value function of the stochastic control problem ( [ swing ] ) when the global constraints @xmath73 and @xmath74 ( at time @xmath14 ) satisfy ( [ globconstr ] ) . note that the standard convention @xmath94 yields @xmath95 . to alleviate notations , the payoff process @xmath96 the filtration @xmath70 will be often dropped in @xmath97 . to be precise we will answer the following questions : @xmath47 existence of an optimal control @xmath98 . @xmath47 regularity of the value function @xmath99 . @xmath47 existence of a _ bang - bang _ optimal control @xmath100 for certain values of the global constraints @xmath101 ( namely when @xmath101 has integral components ) ? by bang - bang we mean that @xmath102-@xmath103 all the _ local _ constraints on the @xmath104 are saturated @xmath20 for every @xmath59 , @xmath105 or there exists at most one instant @xmath106 such that @xmath107 and one global constraint is saturated . note that if @xmath108 , then a bang - bang @xmath101-admissible control necessary satisfies @xmath102-@xmath103 @xmath109 , @xmath110 . the existence of bang - bang optimal controls combined with the piecewise affinity of @xmath111 will be the key in the design of a numerical . @xmath47 when there is an underlying structure markov process ( @xmath112 , we will show that the optimal control turns out to be a function of @xmath113 at every time @xmath85 as well . as a first step , we need to establish the following easy properties of @xmath114 as a function of the global constraints @xmath115 . this is obvious from ( [ ( 2.3 ) ] ) . owing to * p1 * one may assume without loss of generality that @xmath116 . let @xmath55 and @xmath117 be two admissible controls with respect to @xmath101 and @xmath118 respectively . set @xmath119 , @xmath120 . the control @xmath121 is admissible with respect to @xmath118 since @xmath122 . furthermore , @xmath123 hence , @xmath124 the equality follows by symmetry.@xmath125 one may assume by * p1 * that @xmath116 . the convexity of admissible global constraints is obvious . as concerns the concavity of the value function , note that if @xmath55 and @xmath57 are locally admissible controls then @xmath126 is still locally admissible . if @xmath55 and @xmath57 satisfy the global constraints induced by @xmath101 and @xmath127 respectively , then @xmath128 always satisfies that induced by @xmath129 . consequently , using that @xmath130 is @xmath75-measurable , @xmath131,\,\bar q'_{n}\!\in [ q'_{\min},q'_{\max } ] \right\}\\ \!&\!=\!&\ ! \lambda\ , \supess\left\{\e(\sum_{k=0}^{n-1 } q_kv_k\,|\,{\cal f}_0 ) , \ , q \hbox { locally admissible } , \ , \bar q_{n}\!\in [ q_{\min},q_{\max } ] \right\}\\ & & + ( 1-\lambda)\,\supess\left\{\e(\sum_{k=0}^{n-1}q'_kv_k\,|\,{\cal f}_0 ) , \ , q ' \hbox { locally admissible } , \,\bar q'_{n}\!\in [ q'_{\min},q'_{\max } ] \right\}\\ \!&\!=\!&\ ! \lambda p^n_0(q)+ ( 1-\lambda)p^n_0(q').\end{aligned}\ ] ] the monotony property is as follows : let @xmath132 and let @xmath133 be a fixed @xmath127-admissible control . then , for every @xmath101-admissible control @xmath55 , set @xmath134 then @xmath57 is clearly @xmath127-admissible and @xmath135 owing to * p1 * , one may assume @xmath116 without loss of generality . the result is straightforward once noticed that for every @xmath101-admissible control @xmath136 classical consequence of * p3 * : for every @xmath137 set @xmath138 then , set for every @xmath139 @xmath140 one shows using the concavity and monotony properties established in * p3 * that the above limit does exist and that , @xmath102-@xmath103 @xmath141 is continuous on @xmath142 and that @xmath143 @xmath81-@xmath103 . let @xmath144 be a couple of global constraints ( at time @xmath14 ) . @xmath145 this follows from the simple remark that one can define from any @xmath146-admissible control @xmath55 a new @xmath146-admissible control @xmath147 by @xmath148 and @xmath149 the main consequence of * p5 * is that at every time @xmath85 , one may assume without loss of generality that the couple of admissible global constraints @xmath115 is deterministic since , for any possibly random admissible constraint @xmath101 ( at time @xmath85 ) and every @xmath150 , @xmath151 . as a consequence , for notational convenience , we will still denote @xmath114 instead of @xmath152 so that for any admissible global constraints @xmath101 at time @xmath85 @xmath153 ( backward dynamic programming principle ) set @xmath95 . @xmath154 local dynamic programming formula . for every @xmath155 and every couple @xmath115 of deterministic admissible global constraints at time @xmath85 @xmath156 where @xmath157 and @xmath158 $ ] . @xmath159 global dynamic programming formula . for every couple @xmath115 of admissible global constraints at time @xmath14 , the price of the contract at time @xmath59 is given by @xmath160 where @xmath161 furthermore , @xmath162 * remark . * the definition ( [ optiqk ] ) may be ambiguous when argmax is not reduced to a single point . then , one considers @xmath163argmax to define @xmath164 . @xmath154 it is clear that owing to * p1 * the case @xmath116 is the only one to be proved . as a first step , we prove that @xmath165 : let @xmath87 be a @xmath101-admissible control . then , @xmath166 is @xmath36$]-valued ( as well as the @xmath167 s ) and @xmath168 so that @xmath166 is @xmath169-valued . furthermore , @xmath170 note that @xmath171 is an admissible couple of ( @xmath172-measurable ) constraints at time @xmath45 . consequently , @xmath173 where the last inequality follows from the definition of @xmath174 , @xmath175 , ( [ ineqreccontr ] ) and the monotony of conditional expectation . then , @xmath176 one concludes that @xmath177 : we proceed as usual by proving a bifurcation property for the controls . let @xmath166 and @xmath178 be two @xmath169-valued @xmath172-measurable random variables . set @xmath179 and @xmath180 then , one checks that @xmath181 consequently there exists a sequence @xmath182 of @xmath183$]-valued random variables such that @xmath184 one may assume by applying the above bifurcation property that the above supremum holds as a nondecreasing limit as @xmath185 . now , for every fixed @xmath186 , there exists a sequence of @xmath36^{n-1}$]-valued random vectors @xmath187 such that @xmath188 $ ] and @xmath189 where we used that the admissible sequences @xmath190 for the problem starting at @xmath45 clearly satisfy the bifurcation principle due to the homogeneity of conditional expectation @xmath191 with respect to @xmath75 -measurable r.v .. consequently , @xmath192 where we used the conditional beppo levi theorem . note that for every @xmath193 , @xmath194 is an admissible control with respect to @xmath101 since @xmath195 . hence @xmath196 to pass from ( [ pgmdyn2 ] ) to ( [ pgmdyn ] ) is standard using * p5*. let @xmath197 . @xmath198 conversely , setting @xmath199 yields the reverse inequality . @xmath159 this item follows from * p5*. @xmath200 we recall that , for every integer @xmath186 , the triangular set of admissible values for a couple of ( deterministic ) global constraint ( at time @xmath14 ) , as introduced in * p5 * , is defined as : @xmath201 then we will define a triangular tiling of @xmath142 as follows : for every couple of integers @xmath202 , @xmath203 , @xmath204\!\times\![j , j+1],\ ; v\ge u+j - i\}\quad \mbox{and}\quad t^-_{i\,j}:= \{(u , v)\!\in [ i , i+1]\!\times\![j , j+1],\ ; v\le u+j - i\}.\ ] ] one checks that @xmath205 [ main ] the multi - period swing option premium with deterministic global constraints @xmath206 as defined by ( [ ( 2.3 ) ] ) is always obtained as the result of an optimal strategy . @xmath154 the value function ( premium ) : the mapping @xmath207 is a concave , continuous , piecewise affine process , affine on every triangle @xmath208 of the tiling of @xmath142 . furthermore , @xmath209 if @xmath210 @xmath103 for every @xmath211 , then , for every @xmath212 , @xmath213 @xmath103 . ( in particular @xmath214 is @xmath103 non - decreasing ) . @xmath159 the optimal control : if the global constraint @xmath215 , then there always exists a bang - bang optimal control @xmath98 with @xmath216 is @xmath217-valued for every @xmath69 . . if @xmath210 @xmath103 for every @xmath211 , then there always exists a bang - bang optimal control which satisfies @xmath218 . otherwise the optimal control is not bang - bang as emphasized by the case @xmath219 ( see proposition [ n=2 ] below ) we will first inspect the case of a two period swing contract . it will illustrate in a simpler setting the approach developed in the general case . furthermore , we will obtain a slightly more precise result about the optimal controls . we assume @xmath219 throughout this section . the first result is the following let @xmath220 denote an admissible global constraint and @xmath221 $ ] . there is an optimal control @xmath222 given by @xmath223 so that @xmath224 * proof . * let @xmath225 be an admissible control : @xmath226 $ ] and @xmath33 are @xmath36$]-valued @xmath227-measurable , @xmath228 . consequently @xmath166 is @xmath229-valued and @xmath230 is @xmath231$]-valued . hence @xmath232 on the other hand @xmath233 the mapping @xmath234 ( called the _ objective variable _ from now on ) is piecewise affine on @xmath229 with @xmath172-measurable coefficients so the above definition of @xmath235 defines an @xmath172-measurable @xmath229-valued random variable . now , combining the above inequalities yields @xmath236 in the proposition below we investigate in full details the case @xmath219 . [ n=2 ] let @xmath219 . the two period swing option premium with admissible global constraints @xmath237 as defined by ( [ swing ] ) is always obtained as the result of an optimal strategy . @xmath154 the optimal control : if the global constraints @xmath238 only take integral values ( in @xmath239 ) then there always exists a @xmath217-valued bang - bang optimal control . when @xmath101 simply satisfies @xmath240 , there always exists a bang - bang optimal control . if @xmath241 @xmath103 , then any optimal control @xmath242 is @xmath103 bang - bang and satisfies @xmath243 on @xmath244 . otherwise the optimal control is generally not bang - bang . @xmath159 the value function ( premium ) : the mapping @xmath245 is affine on the four triangles @xmath208 that tile @xmath246 . furthermore , when @xmath247 and @xmath248 are @xmath103 non negative , @xmath249 the objective variable being piecewise affine on @xmath229 , @xmath235 is equal either to one of its monotony breaks or to the endpoints of @xmath229 . consequently , a careful inspection of all possible situations for the global constraints yields the complete set of explicit optimal rules for the optimal exercise of the swing option involving the values @xmath247 and @xmath250 ( expected gain or loss at time @xmath14 ) at time @xmath14 and @xmath248 at time @xmath45 . @xmath251 $ ] and the objective variable reads @xmath252 with one monotony break at @xmath73 . one checks that @xmath253 note that @xmath254 so that the above optimal control is not bang - bang on this event except if @xmath255 or @xmath256 . @xmath257 $ ] and the objective variable reads @xmath258 with monotony breaks at @xmath73 , @xmath259 . one checks that @xmath260 note that @xmath261 so that the control is not bang - bang on this event , except if @xmath262 or @xmath263 , since the local control @xmath235 and the global constraint are not saturated . likewise @xmath264 and the optimal control is not bang - bang on this event , except when @xmath255 or @xmath263 . note that _ both events correspond to prediction errors _ : @xmath248 has not the predicted sign . moreover , these events are @xmath103 empty when @xmath265 @xmath103 , @xmath266 . on all other events the optimal control is bang - bang . then the monotony breaks of the objective process ( with the same expression as in the former case ) still are @xmath73 , @xmath259 . a careful inspection of the four possible cases leads to @xmath267 note that on the event @xmath268 the optimal control is not bang - bang , except if @xmath269 or @xmath256 ( both @xmath235 and @xmath270 are @xmath271-valued ) or @xmath272 ( @xmath273 , @xmath274 ) ; and on the event @xmath275 the optimal control is not bang - bang either ( except if @xmath255 or @xmath256 or @xmath272 ) by similar arguments . note that these events _ do not correspond to an error of prediction_. on all other events the optimal control is bang - bang . the objective variable is defined on @xmath276 $ ] by @xmath277 with only one breakpoint at @xmath259 . one checks that @xmath278 once again on the event @xmath268 the optimal control is not bang - bang , except if @xmath269 or @xmath256 or @xmath272 . finally , note that when @xmath241 , the events on which the optimal controls are not bang - bang are empty . @xmath125 @xmath154 we proceed by induction on @xmath71 . for @xmath279 the result is trivial since @xmath280 and @xmath281 . ( when @xmath219 this follows from proposition [ n=2 ] . ) now , we pass from @xmath71 to @xmath282 . note that combining the backward programming principle and @xmath283 yields @xmath284 we inspect successively all the triangles of the tiling of @xmath285 as follows : the upper and lower triangles which lie strictly inside the tiling , then the triangles which lie on the boundary of the tiling . then , @xmath286 and @xmath287 $ ] . one checks that @xmath288 if @xmath289 $ ] , @xmath290 if @xmath291 $ ] and @xmath292 if @xmath293 $ ] ( see figure [ figure2a ] ) . these three triangles @xmath294 , @xmath295 and @xmath296 are included in @xmath142 . it follows from the induction assumption that @xmath297 is @xmath103 affine on them . hence there exists three triplets of @xmath298-measurable random variables @xmath299 , @xmath300 , such that , for every @xmath301 , @xmath302 where @xmath303 $ ] , @xmath304 $ ] and @xmath305 $ ] . note these random coefficients satisfy some compatibility constraints to ensure concavity ( and continuity ) . consequently @xmath306 where @xmath307 , etc . a piecewise affine function reaches its maximum on a compact interval either at its endpoint or at its monotony breakpoints @xmath308 , @xmath309 , @xmath310 , @xmath311 . hence , @xmath312 it is clear that the right hand side of the previous equality stands as the maximum of four affine functions of @xmath101 . one derives that @xmath313 is a convex function on @xmath314 as the maximum of affine functions . hence it is affine since we know that it is also concave . ( 60,60 ) ( 0,0)(0,1)60 ( 0,0)(1,0)60 ( 10,10)(0,1)40 ( 10,10)(1,0)40 ( 10,10)(1,1)40 ( 10,30)(1,0)40 ( 30,10)(0,1)40 ( 10,50)(1,0)40 ( 50,10)(0,1)40 ( 30,10)(1,1)20 ( 10,30)(1,1)20 ( 20,30)(-1,-1)5 ( 30,40)(-1,-1)10 ( 35,45)(-1,-1)5 ( 0,-5)@xmath14 ( -5,30)@xmath315 ( 30,-5)@xmath316 ( 30,36)@xmath317 ( 15,36)@xmath318 ( 8,20)@xmath319 this case can be treated likewise . in that case @xmath320 $ ] , @xmath321 . if @xmath322 , @xmath323 , @xmath324 $ ] , @xmath325 , @xmath326 $ ] , @xmath327 , @xmath328 $ ] ( see figure [ figure2b ] ) . the induction assumption implies that @xmath329 is piecewise affine with monotony breaks at @xmath73 and @xmath330 . if @xmath331 , @xmath332 crosses the upper ( horizontal ) edge of @xmath333 at @xmath334 and the left ( vertical ) edge of @xmath142 at @xmath335 . hence @xmath329 is again piecewise affine with monotony breaks at @xmath73 and @xmath330 . in both cases one concludes as above . ( 60,60 ) ( 0,0)(0,1)60 ( 0,0)(1,0)50 ( 0,30)(1,0)20 ( 0,50)(1,0)40 ( 0,10)(1,1)40 ( 0,30)(1,1)20 ( 20,30)(0,1)20 ( 0,30)(0,-1)10 ( 0,40)(0,-1)10 ( 5,45)(-1,-1)5 ( 0,-5)@xmath14 ( -5,30)@xmath315 ( 8,26)@xmath336 ( 8,46)@xmath337 on proceeds like with @xmath338 except that @xmath339 $ ] which yields only one monotony break at @xmath73 . assume first @xmath340 . @xmath341 $ ] and @xmath342 if @xmath343 $ ] . otherwise @xmath344 . it follows ( see figure [ figure3 ] ) that @xmath345 if @xmath346 $ ] and @xmath347 if @xmath348 $ ] . both @xmath349 and @xmath350 are included in @xmath142 . hence @xmath351 is affine on both triangles , one derives that @xmath352 } } ( a^1(q_{\min}-x)+b^1n+c^1)+\mbox{\bf 1}_{_{x\in[q_{\max}-n,1]}}(a^2(q_{\min}-x)+b^2(q_{\max}-x)+c^2).\end{aligned}\ ] ] where @xmath353 are @xmath298-measurable r.v .. then , one concludes like in the first case . if @xmath354 , one proceeds likewise except that the two visited " triangles of @xmath142 are @xmath355 . ( 60,60 ) ( 0,0)(0,1)40 ( 0,0)(1,0)60 ( 10,10)(0,1)20 ( 10,10)(1,0)20 ( 10,10)(1,1)20 ( 10,30)(1,0)40 ( 30,10)(1,1)20 ( 30,10)(0,1)20 ( 20,30)(-1,-1)5 ( 30,30)(-1,0)10 ( 40,30)(-1,0)10 ( 0,-5)@xmath14 ( -5,30)@xmath71 ( 30,-5)@xmath316 ( 30,20)@xmath356 ( 10,20)@xmath357 @xmath341 $ ] and @xmath358 if @xmath359 , @xmath360 $ ] , @xmath361 $ ] if @xmath362 . in both cases the only visited " triangle is @xmath363 and one concludes as usual . @xmath364 $ ] and @xmath342 if @xmath365 $ ] , @xmath366 otherwise . hence @xmath332 takes its values in @xmath367 on which @xmath351 is affine . the conclusion follows . the inspection of all these cases completes the proof of the induction . the values of @xmath368 when @xmath369 are obvious consequences of the degeneracy of the global constraints . @xmath159 we deal successively with the two announced settings . global constraints in @xmath370 : let @xmath371 . we rely on the characterization ( [ optiqk ] ) of @xmath235 @xmath372 we know from item @xmath154 that @xmath373 is affine on every tile @xmath374 of @xmath142 . if @xmath375 then one checks that @xmath376 , @xmath377 , is always affine with @xmath169 having @xmath14 and/or @xmath45 as endpoints . to be precise if @xmath378 , @xmath379 , @xmath380 , @xmath381 $ ] , if @xmath382 , @xmath383 , @xmath384 , @xmath381 $ ] , if @xmath385 , @xmath386 , @xmath387 , if @xmath388 , @xmath389 , @xmath390 , @xmath381 $ ] , if @xmath391 , @xmath392 , @xmath381 $ ] , if @xmath393 , @xmath394 , @xmath395 . as a consequence , affinity being stable by composition , @xmath396 is affine on @xmath397 , \{0\ } , \{1\}\}$ ] . in turn , @xmath398 is affine and reaches its maximum at some endpoint of @xmath169 @xmath20 @xmath399 or at @xmath400 . then , inspecting the above cases shows that @xmath401 . using ( [ pgmdynglob ] ) and ( [ optiqk ] ) , one shows by induction on @xmath85 that @xmath216 is always @xmath217-valued . non negative @xmath402 : _ step 1 : global constraint saturated . _ let @xmath403 . let @xmath98 be an optimal @xmath101-admissible control . we introduce the @xmath404-stopping time @xmath405 with the convention @xmath406 . on @xmath407 , @xmath408 for every @xmath69 . in particular the global constraint is saturated at time @xmath409 , @xmath20 @xmath410 set @xmath411 one check that @xmath121 is a @xmath101-admissible control : this follows from the fact that @xmath412 is an @xmath404-stopping time ( note that @xmath413 is @xmath36$]-valued on @xmath414 since then @xmath415 and @xmath416 ) . likewise one shows that @xmath417 for every @xmath69 . furthermore , note that if @xmath74 is an integer and @xmath242 is @xmath217-valued then @xmath121 is still @xmath217-valued . the @xmath418 being non negative @xmath419 hence @xmath121 is still an optimal control . furthermore @xmath420 on @xmath421 so that the stopping time @xmath422 satisfies by construction @xmath423 note that if the control @xmath242 is bang - bang , iterating the above construction at most @xmath71 times yields an optimal control @xmath424 for which @xmath425 @xmath103 . such a control @xmath424 saturates the global constraint . as a consequence , this shows that @xmath426 @xmath103 so that @xmath427 is @xmath103 non - decreasing and concave . since there is an optimal control @xmath242 which saturates the global constraint , one may assume without loss of generality that @xmath428 . we proceed again by induction on @xmath71 based on the dynamic programming formula ( [ pgmdyn3 ] ) . when @xmath279 the result is obvious ( and true when @xmath219 as well ) . assume now the announced result is true for @xmath371 . let @xmath429 and @xmath430 $ ] . then , @xmath431 $ ] and @xmath432 , @xmath433 . hence @xmath434 , @xmath435 $ ] and @xmath436 , @xmath437 $ ] . now @xmath438 is @xmath103 concave , non - decreasing , affine on @xmath439 $ ] and on @xmath440 $ ] and non - decreasing . consequently , there exists @xmath441 , @xmath442 , @xmath298-measurable random variables satisfying @xmath443,\\ p_0(\chi^n((q_{\max},q_{\max}),x),\f_{1 + . } ) & = & b^2(q_{\max}-x)+c^2,\ ; x\!\in [ q_{\max}-j , q_{\max}\!\wedge\!1].\end{aligned}\ ] ] with @xmath444 and @xmath445 @xmath103 . set temporarily @xmath446 hence , @xmath447 set @xmath448 and @xmath449 and note that @xmath450 and @xmath451 @xmath103 . elementary computations show that : @xmath452 on @xmath453 and @xmath454 on @xmath455 , @xmath456 on @xmath457 and @xmath458 on @xmath459 . consequently @xmath235 can be chosen @xmath460-valued on @xmath461 and equal to @xmath330 on @xmath462 . on @xmath463 , one has @xmath464 . then , the dynamic programming formula shows that the other components @xmath465 of the optimal control on @xmath466 can be obtained as the optimal control of the pricing problem @xmath467 . one derives from the induction assumption at time @xmath71 that @xmath465 can be chosen bang - bang and @xmath468-admissible which implies that @xmath55 is @xmath101-admissible and bang - bang since @xmath469 ( on @xmath470 ) . a similar proof holds on @xmath471 . on @xmath472 , one has @xmath473 . then , the dynamic programming formula shows that the other components @xmath474 of the optimal control on @xmath475 can be obtained as the optimal control of the pricing problem @xmath476 . as @xmath477 there exists a @xmath478-admissible bang - bang optimal control @xmath100 ( with respect to @xmath479 on @xmath472 . then @xmath480 is @xmath217-valued for every @xmath110 ( in fact identically @xmath14 if @xmath481 ) . at this stage one can recursively modify @xmath482 using the procedure described in step 1 to saturate the upper global constraint . finally one may assume that @xmath483 which in turn implies that @xmath484 is a bang - bang @xmath485-admissible optimal control.@xmath200 when a global constraint @xmath101 belongs to the interior of a triangle @xmath486 , _ one only needs to compute the value of @xmath487 at the vertices of this triangle to derive the value of the premium at every @xmath488_. when @xmath101 is itself an integral valued couple , at most six further points allow to compute the premium in a neighborhood of @xmath101 . we will use this result extensively when designing our quantization based numerical procedure in section [ swingquant ] . an additional result . proposition [ n=2 ] shows that it is hopeless to produce in full generality some bang - bang optimal control when @xmath489 . this comes from the fact that at integral valued global constraints the bang - bang optimal control may saturate none of the global constraints ( indeed , so is the case at @xmath490 when @xmath219 ) . however , using the same approach as that developed in that in the proof of theorem [ main ] , one can show the following result , whose details of proof are left to the reader . assume the assumptions of theorem [ main ] hold . if a couple of admissible constraint @xmath491 satisfies @xmath492 then there exist a _ quasi - bang - bang _ control in the following sense : @xmath81-@xmath103 , @xmath216 is @xmath217-valued except for at most one local constraint @xmath493 . by markov setting we simply mean that the payoffs @xmath418 are function of an @xmath52-valued underlying @xmath404-markov structure process @xmath494 @xmath20 @xmath495 the markovian dynamics of @xmath496 reads on borel functions @xmath497 @xmath498 where @xmath499 is sequence of borel probability transitions on @xmath500 . then the backward dynamic programming principle ( [ pgmdyn ] ) can be rewritten as follows @xmath501 with @xmath502 and for every @xmath69 and every @xmath503 , @xmath504 where @xmath157 and @xmath158 $ ] . * pointwise estimation of @xmath505 . * as established in theorem [ main ] , one only needs to compute the value function @xmath368 at global constraints @xmath506 @xmath20 with integral components . moreover , for these constraints , the local optimal control @xmath216 is always bang - bang @xmath20 @xmath507 . let @xmath508 . for every @xmath69 , one defines the set of attainable residual global constraints at time @xmath85 , namely @xmath509 ( thus @xmath510 ) . note that the running parameter @xmath511 represents the possible values of the cumulative purchase process @xmath512 . one checks that for every @xmath513 , @xmath514 since @xmath515 consequently the backward dynamic programming formula having @xmath516 as a result reads : @xmath517 at this stage no numerical computation is possible yet since no space discretization has been achieved . this is the aim of the next section where we will approximate the above dynamic programming principle by ( optimal ) quantization of the state process @xmath496 . in this section , we propose a quantization based approach to compute the premium of the swing contracts with firm constraints . quantization has been originally introduced and developed in the early 1950 in signal processing ( see @xcite ) . the starting idea is simply to replace every random vector @xmath518 by a random vector @xmath519 taking finitely many values in a _ grid _ ( or _ codebook _ ) @xmath520 ( of size @xmath521 ) . the grid @xmath522 is also called an @xmath521-quantizer of @xmath496 . when the borel function @xmath523 satisfies @xmath524 @xmath525 is called a _ voronoi quantization _ of @xmath496 ( and @xmath523 as well ) . one easily checks that @xmath523 is necessarily a nearest neighbor projection on @xmath522 @xmath20 satisfies @xmath526 the so - called _ voronoi cells _ @xmath527 , make up a _ voronoi tessellation _ or partition of @xmath52 ( induced by @xmath522 . note that when the distribution @xmath528 of @xmath496 weights no hyperplanes the boundary of the voronoi tessellation of @xmath522 are @xmath528-negligible so that the @xmath528-weights of the voronoi cells entirely characterize the distribution of @xmath529 . when @xmath530 , the @xmath531-mean error induced by replacing @xmath496 by @xmath525 , namely @xmath532 is called the _ @xmath531-mean quantization error _ induced by @xmath533 and its @xmath534 power is known as the @xmath531-distortion . we will see in the next section that the codebook @xmath522 can be optimized so as to minimize the @xmath531-quantization error with respect to the ( distribution of ) @xmath496 . our aim in this section is to design an algorithm based on the quantization of the markov chain @xmath535 at every time @xmath85 to approximate the premium of the swing contract with firm constraints and to provide some _ a priori _ error estimates in terms of quantization errors . as a first step we consider at every time @xmath85 a grid @xmath536 ( of size @xmath537 ) . then , we design the quantized tree algorithm to price swing contracts by simply mimicking the original dynamic programming formula ( [ pgmdyn ] ) . this means in particular that we force in some way the markov property on @xmath538 by considering the quantized transition operator @xmath539 so that @xmath540 @xmath541 let @xmath542 be a couple of ( deterministic ) global constraints ( at time @xmath14 ) . the quantized dynamic programming principle is defined by @xmath543 one easily shows by induction that , for every @xmath544 ( residual global constraint at time @xmath85 ) , @xmath545 @xmath546 let us briefly discuss the complexity of this quantized backward dynamic procedure . let @xmath550 . at every nod " @xmath551 the computation of @xmath552 requires @xmath553 products ( up to a constant ) , so that for a given residual global constraint the complexity at time @xmath85 in the dynamic programming is proportional to @xmath554 . on the other hand , one checks that @xmath555 consequently , the complexity of the computation of @xmath556 is proportional to @xmath557 a simple upper - bound is provided by @xmath558 and a uniform one by @xmath559 note that this last upper bound corresponds to the complexity of the quantized version of the algorithm based on some _ penalized global volume constraints _ ( see the companion paper @xcite ) . [ quantrate ] assume that the markov process @xmath560 is lipschitz feller in the following sense : for every bounded lipschitz continuous function @xmath561 and every @xmath59 , @xmath562 is a lipschitz function satisfying @xmath563_{\rm lip}\le[\theta_k ] _ { \rm lip } [ g]_{\rm lip}$ ] . assume that every function @xmath564 is lipschitz continuous with lipschitz coefficient @xmath565_{\rm lip}$ ] . let @xmath530 such that @xmath566 . then , there exists a real constant @xmath567 such that @xmath568 for every @xmath155 , the function @xmath574 is lipschitz on @xmath52 , uniformly with respect to @xmath575 and its lipschitz coefficient @xmath576_{{\rm lip},y}:=\sup_{q\in t^+(n - k)}[p^n_k(q,.)]_{\rm lip}$ ] satisfies for every @xmath577 , @xmath578_{{\rm lip},y}\le [ v_{n-1}]_{\rm lip},\qquad [ p^n_k]_{{\rm lip},y}\le [ v_k]_{\rm lip}+ [ \theta_k]_{\rm lip}[p^n_{k+1}]_{{\rm lip},y}.\ ] ] temporarily set for convenience , @xmath593 . one derives that for every @xmath110 , @xmath594_{\rm lip}+2[\theta_k]_{\rm lip}[p^n_{k+1}]_{{\rm lip},y } ) \|y_k-\widehat y_k\|_{_p } + \delta^{n , p}_{k+1}.\ ] ] furthermore , @xmath595_{\rm lip}\|y_{n-1}-\widehat y_{n-1}\|_{_p}$ ] . the result follows by induction.@xmath125 in this section , we provide a few basic elements about optimal quantization in order to give some error bounds for the premium of the swing option . we refer to @xcite for more details about theoretical aspects and to @xcite for the algorithmic aspects numerical applications . let @xmath596 . let @xmath597 be an @xmath52-valued random vector and let @xmath598 be a given grid size . the best @xmath531-approximation of @xmath496 by a random vector taking its values in a given grid @xmath522 of size ( at most ) @xmath521 is given by a voronoi quantizer @xmath599 which induces an @xmath600- mean quantization error @xmath601 it has been shown independently by several authors ( in various finite and infinite dimensional frameworks ) that when the grids @xmath522 runs over all the subsets of @xmath52 of size at most @xmath521 , that @xmath602 reaches a minimum denoted @xmath603 ( see @xmath0 @xcite or @xcite ) @xmath20 the minimization problem @xmath604 has at least a solution temporarily denoted @xmath605 . several algorithms have been designed to compute some optimal or close to optimality quantizers , especially in the quadratic case @xmath606 . they all rely on the stationarity property satisfied by optimal quantizers . in the quadratic case , a grid @xmath522 is stationary @xmath607 this follows from some differentiability property of the @xmath531-distortion . for a formula in the general case we refer to @xcite . in @xmath45-dimension , a regular newton - raphson zero search procedure turns out to be quite efficient . in higher dimension ( at least when @xmath608 or @xmath609 ) only stochastic procedures can be implemented like the @xmath610 ( a stochastic gradient descent , see @xcite or the lloyd i procedure ( a randomized fixed point procedure , see @xcite ) . for more details and result we refer to @xcite . as a result of these methods , some optimized grids of the ( centered ) normal distribution @xmath611 are available on line at the url @xmath612 for dimensions @xmath613 and sizes from @xmath614 up to @xmath615 . [ zad](zador , see @xcite ) @xmath154 let @xmath619 , @xmath620 , @xmath621 , such that @xmath622 . then @xmath623 @xmath159 non asymptotic estimate ( see @xmath0 @xcite ) : let @xmath620 , @xmath624 . there exists a real constant @xmath625 and an integer @xmath626 such that or any @xmath52-valued random vector @xmath496 , for @xmath627 , @xmath628 now we are in position to apply the above results to provide an error bound for the pricing of swing options by optimal quantization : assume there is a real exponent @xmath596 such that the ( @xmath629-dimensional ) markov structure process @xmath560 satisfies @xmath630 at each time @xmath59 , we implement a ( quadratic ) optimal quantization grid @xmath631 of @xmath113 with constant size @xmath632 . then the general error bound result ( [ ratequantswing ] ) combined with theorem [ zad]@xmath159 says that , if @xmath633 , @xmath634 where as the complexity of the procedure is bounded by @xmath635 ( up to a constant ) . in fact this error bound turns out to be conservative and several numerical experiments , as those presented below , suggest that in fact the true rate ( for a fixed number @xmath71 of purchase instants ) behaves like @xmath636 . another approach could be to minimize the complexity of the procedure by considering ( optimal ) grids with variable sizes @xmath537 satisfying @xmath637 . we refer to @xcite for further results in that direction . however , numerical experiments were carried out with constant size grids for both programming convenience and memory saving . we considered a two factor continuous model for the price of future contracts which leads to the following dynamics for the spot price @xmath638,\ ] ] where @xmath639 and @xmath640 are two standard brownian motions with correlation coefficient @xmath641 and @xmath642 then , we consider a ( daily ) discretization of the gaussian process @xmath643 at times @xmath644 @xmath645 , @xmath646 ) . the sequence @xmath647 is clearly not markov . however , adding an appropriate auxiliary processes , one can build a higher dimensional ( homogenous ) markov process @xmath648 whom @xmath649 is a linear combination . this calls upon classical methods coming from time series analysis . then a fast quantization method has been developed to make makes possible a parallel implementation of the quantized probability transitions of @xmath650 . for further details about this model and the way it can be quantized , we refer to @xcite . in @xcite , the optimal quantization method described above is extensively tested from a numerical viewpoint ( rates of convergence , needed memory , swapping effect , etc ) . its performances are compared those of the longstaff - schwartz approach introduced in @xcite . this comparison emphasizes the accuracy and the velocity of our approach , even if only one contract is to be computed and the computation of the probability transitions is included in the computation time of the quantization method . furthermore , it seems that it needs significantly less memory capacity when implemented on our tested model . we simply reproduced below a complete graph of the function @xmath651 when @xmath101 runs over the whole set of admissible global constraints @xmath142 . the parameters were settled at the following values @xmath652 0 ( 2001 ) . a stochastic quantization method for non - linear problems , _ monte carlo methods and applications _ , * 7 * , n@xmath6541 - 2 , pp.21 - 34 . a quantization algorithm for solving multi - dimensional optimal stopping problems , _ bernoulli _ , * 6*(9 ) , 1 - 47 . error analysis of the quantization algorithm for obstacle problems , stochastic processes & their applications , * 106*(1 ) , 1 - 40 . a quantization tree method for pricing and hedging multi - dimensional american options , _ mathematical finance _ , * 15*(1 ) , 119 - 168 . ( 2006 ) pricing swing options using optimal quantization . ( 2006 ) numerical methods for the pricing of swing options : a stochastic control approach , _ methodology and computing in applied probability_. ( 2007 ) optimal switching with applications to energy tolling agreements , pre - print . ( 2005 ) gas storage and supply guarantees : an optimal switching approach , pre - print . ( 2003 ) optimal multiple - stopping of linear diffusions and swing options , pre - print . ( 2004 ) optimal multiple stopping and valuation of swing options , _ mathematical finance _ , to appear . ( 2002 ) risk analysis of swing contracts , _ energy and power risk management_. ( 2005 ) _ commodities and commodity derivatives - modeling and pricing for agriculturals , metals and energy _ , wiley finance . , special issue on quantization , _ ieee trans . on inf . theory _ , * 29 * , @xmath655&@xmath46 , 1982 . ( 1992 ) . _ vector quantization and signal compression _ , kluwer , boston . , _ foundations of quantization for probability distributions_. lect . notes in math . 1730 , springer , berlin . , distortion mismatch in the quantization of probability measures , to appear in _ esaim p & s _ , prep - print lpma-@xmath656 . valuation of commodity - based swing options , _ management science _ , * 50 * , 909 - 921 . ( 2004 ) pricing of electricity swing options , _ journal of derivatives _ , * 11 * , 26 - 43 . ( 2001 ) a discrete valuation of swing options , _ canadian applied mathematics quarterly _ , * 9*(1 ) , 35 - 74 . ( 2001 ) valuing american options by simulation : a simple least squares approach , _ the review of financial studies _ , * 14 * , 113 - 147 . , ( 2006 ) . functional quantization rate and mean regularity of processes with an application to lvy processes , pre - pub . ( 2004 ) monte carlo methods for the valuation of multiple exercise options , _ mathematical finance _ , * 14 * , @xmath657 . ( 1998 ) a space vector quantization method for numerical integration , _ j. computational and applied mathematics _ , 89 , 1 - 38 . an optimal markovian quantization algorithm for multidimensional stochastic control problems , _ stochastics and dynamics _ , 4 , 501 - 545 . optimal quantization methods and applications to numerical problems in finance , _ handbook of numerical methods in finance _ , ed . s. rachev , birkhauser , 253 - 298 . optimal quadratic quantization for numerics : the gaussian case , _ monte carlo methods and applications _ , * 9 * , n@xmath658 , pp.135 - 166 . ( 2006 ) structure of swing contract value and optimal strategy , _ the 2006 stochastic modeling symposium and investment seminar , canadian institute of actuaries _ ( 1995 ) valuation of path - dependent contingent claims with multiple exercise decisions over time : the case of take or pay . _ journal of financial and quantitative analysis _ , * 30 * , 271 - 293 . ( 2006 ) finite element valuation of swing options , pre - print . asymptotic quantization error of continuous signals and the quantization dimension , _ ieee trans . , * 28 * , special issue on quantization , a. gersho & r.m . grey eds , 139 - 149 .
forced detachment of a single polymer chain , strongly - adsorbed on a solid substrate , is investigated by two complementary methods : a coarse - grained analytical dynamical model , based on the onsager stochastic equation , and molecular dynamics ( md ) simulations with langevin thermostat . the suggested approach makes it possible to go beyond the limitations of the conventional bell - evans model . we observe a series of characteristic force spikes when the pulling force is measured against the cantilever displacement during detachment at constant velocity ( displacement control mode ) and find that the average magnitude of this force increases as grows . the probability distributions of the pulling force and the end - monomer distance from the surface at the moment of final detachment are investigated for different adsorption energy and pulling velocity . our extensive md - simulations validate and support the main theoretical findings . moreover , the simulation reveals a novel behavior : for a strong - friction and massive cantilever the force spikes pattern is smeared out at large . as a challenging task for experimental bio - polymers sequencing in future we suggest the fabrication of stiff , super - light , nanometer - sized afm probe .
1310.3876
in recent years single - molecule pulling techniques based on the use of laser optical tweezers ( lot ) or atomic force microscope ( afm ) have gained prominence as a versatile tool in the studies of non - covalent bonds and self - associating bio - molecular systems @xcite . the latter could be exemplified by the base - pair binding in dna as well as by ligand - receptor interactions in proteins and has been studied recently by means of brownian dynamic simulations and the master equation approach @xcite . the lot and afm methods are commonly used to manipulate and exert mechanical forces on individual molecules . in lot experiments , a micron - sized polystyrene or silica bead is trapped in the focus of the laser beam by exerting forces in the range @xmath2 . typically , afm ( which covers forces interval in @xmath3 range ) is ideal for investigations of relatively strong inter- or intramolecular interactions which are involved in pulling experiments in biopolymers such as polysaccharides , proteins and nucleic acids . on the other hand , due to the relatively small signal - to - noise ratio , the afm experiments have limitations with regard to the mechanochemistry of weak interactions in the lower piconewton regime . the method of dynamic force spectroscopy ( dfs ) is used to probe the force - extension relationship , rupture force distribution , and the force vs loading rate dependence for single - molecule bonds or for more complicated multiply - bonded attachments . historically , the first theoretical interpretation of dfs has been suggested in the context of single cell adhesion by bell @xcite and developed by evans @xcite . the consideration has been based on the semi - phenomenological arrhenius relation which describes surface detachment under time - dependent pulling force , @xmath4 , with @xmath5 being the loading rate . it was also assumed that the effective activation energy , @xmath6 , may be approximated by a linear function of the force , i.e. , @xmath7 . here @xmath8 is the distance between the bonded state and the transition state where the activation barrier is located . the resulting bell - evans ( be ) equation then gives the mean detachment force as a function of temperature @xmath9 and loading rate @xmath5 , i.e. , @xmath10 , where @xmath11 is the desorption rate in the absence of applied pulling force . as one can see from this be - equation , the simple surmounting of be - activation barrier results in a linear dependence of detachment force on the logarithm of loading rate , provided one uses the applied force as a governing parameter in the detachment process ( i.e. , working in an _ isotensional _ ensemble when @xmath12 is controlled and the distance @xmath13 from the substrate to the clamped end - monomer of the polymer chain fluctuates ) . for multiply - bonded attachments the interpretation problem based on this equation becomes more complicated since a non - linear @xmath14 relationship is observed @xcite . in this case chain detachment involves passages over a cascade of activation barriers . for example , merkel et al . @xcite suggested that the net rate of detachments can be approximated by a reciprocal sum of characteristic times , corresponding to jumps over the single barriers . in particular , regarding the detachment of biotin - streptavidin single bonds , it was suggested that two consecutive barriers might be responsible for the desorption process . a simple example of multiply - bonded bio - assembly is presented by a singe - stranded dna ( ssdna ) macromolecule , strongly adsorbed on graphite substrate . the forced - induced desorption ( or peeling ) of this biopolymer has been studied analytically and by means of brownian dynamics ( bd ) simulation by jagota et al . @xcite . in ref . @xcite the equilibrium statistical thermodynamics of ssdna forced - induced desorption under force control ( fc ) and displacement control ( dc ) has been investigated . in the latter case one works in an _ isometric _ ensemble where @xmath13 is controlled and @xmath12 fluctuates . it has been demonstrated that the force response under dc exhibits a series of spikes which carry information about the underlying base sequence of ssdna . the brownian dynamics ( bd ) simulations @xcite confirmed the existence of such force spikes in the force - displacement curves under dc . the nonequilibrium theory of forced desorption has been developed by kreuzer et al . @xcite on the basis of master equation approach for the cases of constant velocity and force - ramp modes in an afm - experiment . the authors assumed that individual monomers detachments represent a fast process as compared to the removal of all monomers . this justifies a two - state model where all monomers either remain on the substrate or leave it abruptly . the corresponding transition rates ( which constitute a necessary input in the master equation approach ) must satisfy detailed balance . as a result of the master equation solution , the authors obtained a probability distribution of detachment heights ( i.e. , distances between the cantilever tip and the substrate ) as well as an average detachment height as a function of the pulling velocity . irrespective of all these efforts , a detailed theoretical interpretation of the dynamic force spectroscopy experiments is still missing . for example , in terms of kramers reaction - rate theory @xcite the arrhenius - like be - model holds only when the effective activation energy @xmath15 . on the other hand , it is clear that for large forces ( which we experience in afm ) , the case when @xmath16 occurs fairly often . in this common case the general approach , based on the be - model , becomes questionable . besides , it can be shown @xcite , that the activation energy vs force dependence , @xmath17 , is itself a _ nonlinear _ function , so that the conventional be - model , based on the linear approximation , @xmath18 , should be limited to small forces . moreover , the arrhenius - like relationship for the detachment rate , which was used in the be - model , is a consequence of a saddle - point approximation for the stationary solution of fokker- planck equation @xcite . this contradicts the typical loading regimes , used in experiments , where applied force or distance grow linearly with time . the present paper is devoted to the theoretical investigation of a single molecule desorption dynamics and aimed at interpretation of afm - or lot - based dynamic force spectroscopy in the dc constant - velocity mode . the organization of the paper is as follows : in sec . [ second_section ] we give the equilibrium theory of detachment for the case of strong polymer adsorption . the mean force ( measured at the cantilever tip ) versus displacement diagram is discussed in detail . in particular , the characteristic force-``spikes '' structure ( which was first discussed in ref . @xcite ) can be clearly seen . in sec . [ sec_dynamics ] we give a dynamical version of the detachment process . our approach rests on construction of general free energy functions , depending on coarse - grained variables , which govern the non - linear response and structural bonding changes in presence of external forces . the corresponding free - energy - based stochastic equations ( known as onsager equations @xcite ) are derived and solved numerically . this solution makes it possible to provide not only force - displacement diagrams and the ensuing dependence on cantilever displacement velocity @xmath0 but also the detachment force probability distribution function ( pdf ) . in sec [ sec_md ] the main theoretical results are then checked against extensive molecular dynamics ( md ) simulation . a brief discussion of results is offered in sec . [ sec_results ] . recently we suggested a theory of the force - induced polymer desorption ( for relatively weak adsorption energy ) in the isotensional @xcite and isometric @xcite equilibrium ensembles supported by extensive monte carlo ( mc ) simulations . in the former case , the fraction of adsorbed monomers changes abruptly ( undergoes a jump ) when one varies the adsorption energy or the external pulling force . in the second case , the order parameter varies steadily with changing height of the afm - tip , even though the phase transition is still of first order . the total phase diagram in terms of adsorption energy - pulling force , or , adsorption energy - end - monomer height , has been discussed theoretically and in terms of mc - simulations . on the other hand , the afm experiments deal with relatively strong forces ( @xmath19 @xcite ) so that in the case of a single molecule desorption experiment only a really strong adsorption energy is essential . this limit has been discussed in the recent papers by jagota et al . @xcite and kreuzer ey al . @xcite . here we consider this problem in a slightly more general form . in so doing we distinguish between two different models : with frictionless- and strong - friction substrates , as indicated in fig . [ geometry ] . this case has been considered in refs . @xcite and is based on the assumption that the force resisting sliding is sufficiently small , i.e. , the cantilever tip and the contact point @xmath20 are both placed along the same @xmath21-axis ( see fig . [ geometry ] ( left panel ) ) . the total partition function for a fixed cantilever distance @xmath13 , i.e. , @xmath22 , is a product of partition functions of the adsorbed part , @xmath23 , of the desorbed portion ( a stretched polymer portion ) , @xmath24 , and of the cantilever itself , @xmath25 , where @xmath26 is the number of _ desorbed _ polymer segments , and @xmath27 denotes the distance between the clamped end of this desorbed portion and the substrate . as a result , @xmath28 where the integration interval , @xmath29 , and the step - function , @xmath30 , imply that restrictions , @xmath31 and @xmath32 , should be applied simultaneously . in this representation @xmath13 is the control variable ( which is monitored by the corresponding afm operating mode ) whereas @xmath26 and @xmath27 are coarse - grained dynamic variables which should be integrated ( in our case , an integral over @xmath27 , and summation over @xmath26 ) out . moreover , if we introduce the function @xmath33 then eq . ( [ combination ] ) can be rewritten as @xmath34 in the strong adsorption regime , @xmath23 attains a simple form @xmath35 , \label{adsorption_z } \end{aligned}\ ] ] where the dimensionless adsorption energy @xmath36 . the cantilever manifests itself as a harmonic spring with a spring constant @xmath37 , i.e. , the corresponding partition function reads @xmath38 \label{cantilever}\end{aligned}\ ] ] finally , we derive the partition function of the desorbed part of the polymer as function of the dynamic variables @xmath26 and @xmath27 , based of the _ freely jointed bond vector _ ( fjbv ) model @xcite . the corresponding gibbs free energy ( i.e. , the free energy in the isotensional - ensemble ) is @xmath39 , \label{gibbs}\end{aligned}\ ] ] where the dimensionless force @xmath40 . the corresponding distance @xmath41 . @xmath42 where the so called langevin function @xmath43 has been used . in the isometric - ensemble , the proper thermodynamic potential is the helmholtz free energy , @xmath44 , which is related to @xmath45 by legendre transformation , @xmath46 where @xmath47 . taking the gibbs free energy , eq . ( [ gibbs ] ) , into account and the relation eq . ( [ langevin ] ) for the helmholtz free energy , we have @xmath48 + { \widetilde f } \ : k_bt \ : n \ : { \cal l } ( { \widetilde f})\nonumber\\ & = & - n \ : k_bt \ ; \left\lbrace \ln \left [ \dfrac{\sinh { \widetilde f}}{{\widetilde f}}\right ] + 1 - { \widetilde f } \ : \coth ( { \widetilde f } ) \right\rbrace \equiv - n \ : k_bt \ ; { \cal g } ( { \widetilde f } ) , \label{helmholtz}\end{aligned}\ ] ] where the function @xmath49 + 1 - x \coth ( x)$ ] . as a result , ( [ helmholtz ] ) along with eq . ( [ langevin ] ) parametrically define @xmath50 , and the corresponding partition function @xmath51 \label{polymer_z}\end{aligned}\ ] ] as function of @xmath26 and @xmath27 . by making use of eqs . ( [ adsorption_z ] ) , ( [ cantilever ] ) , ( [ polymer_z ] ) , the total partition function given by eq . ( [ combination ] ) reads @xmath52 \ : \exp \left [ n \ : { \cal g } ( { \widetilde f } ) \right ] \ : \exp \left [ - \dfrac{k_c}{2 k_bt } ( d - r)^2 \right ] . \label{total_z_final}\end{aligned}\ ] ] in eq . ( [ total_z_final ] ) the force @xmath53 should be expressed in terms of @xmath54 as follows : @xmath55 , where @xmath56 denotes the inverse langevin function . the corresponding effective free energy function in terms of @xmath26 and @xmath27 reads @xmath57 in the limit of a very stiff cantilever , @xmath58 , the cantilever partition function approaches a @xmath59-function @xcite : @xmath60 \rightarrow ( 2\pi k_bt / k_c ) ^{1/2 } \ : \delta ( d - r ) , \label{delta}\end{aligned}\ ] ] and eq . ( [ total_z_final ] ) takes the form @xmath61 \ : \exp \left [ n \ : { \cal g } ( { \widetilde f } ) \right ] \ : \theta ( n b - d ) , \label{total_z_simple}\end{aligned}\ ] ] where @xmath62 and the step - function @xmath63 ensures that the condition @xmath64 holds . it is this very stiff cantilever limit that was considered in ref . @xcite . for the isometric ensemble , i.e. , in the @xmath13-ensemble , the average force @xmath65 , measured by afm - experiment , is given by @xmath66 \ : \int\limits_{0}^{\min ( b n , d ) } \ : d r \ ; ( d - r ) \ : \exp \left [ n \ : { \cal g } ( { \widetilde f } ) \right]\ : \exp \left [ - \dfrac{k_c}{2 k_bt } ( d - r)^2 \right ] , \label{average_force_1}\end{aligned}\ ] ] where @xmath22 is given by eq . ( [ total_z_final ] ) . the numerical results , which follow from eq . ( [ average_force_1 ] ) , are shown in fig . [ equilibrium ] . one can immediately see the " sawtooth- , or force - spikes structure on the force - displacement diagram as it was also found by jagota et al . @xcite in the limit of very stiff cantilever . physically , spikes correspond to the reversible transitions @xmath67 , during which the release of polymer stretching energy is balanced by the adsorption energy . the corresponding thermodynamic condition reads @xmath68 . this condition also leads to the spikes amplitude law @xmath69 @xcite , i.e. the spikes amplitude gradually decreases in the process of chain detachment ( i.e. , with growing @xmath26 ) . this structure is more pronounced at larger adsorption energy @xmath1 and cantilever spring constant @xmath37 . thus , while the force oscillates , its mean value remains nearly constant in a broad interval of distances @xmath13 , exhibiting a kind of plateau . complementary information ( for fixed @xmath37 at different values of @xmath1 ) is given on fig . [ equilibrium_1 ] . one can verify that the plateau height is mainly determined by @xmath1 whereas the spikes amplitude is dictated by the cantilever spring constant @xmath37 . in this limit one has to take into account the specific geometry of an afm experiment , shown in fig . [ geometry ] ( right panel ) . for simplicity , an infinite friction of the polymer at the surface is assumed . the adsorbed polymer portion may be considered as a two - dimensional self - avoiding chain comprising @xmath70 segments . the last contact point ( marked as @xmath20 in fig . [ geometry ] ) can move due to adsorption or desorption elementary events . in ref . @xcite this was classified as the _ sticky _ case . in fig . [ geometry ] , @xmath13 is the distance from the cantilever base to to the substrate , @xmath71 is the height of the cantilever tip above the substrate , and @xmath27 is the distance between the cantilever tip and the contact point @xmath20 . eventually , @xmath72 is the lateral distance between cantilever base and the contact point @xmath20 . one may assume that initially the desorbed portion of @xmath26 segments has occupied a distance of @xmath72 which , due to self - avoiding @xmath73-configurations of an adsorbed chain , equals @xmath74 ( where @xmath75 ) . the specific geometry of the afm experiment in the case of strong polymer - substrate friction ( shown in fig . [ geometry ] ) brings about changes only in the cantilever partition function , i.e. , instead of eq . ( [ cantilever ] ) , one has @xmath76 . \nonumber\\ & = & \exp \left [ - \dfrac{k_c}{2 k_bt } \left(d - \sqrt{r^2 - b^2 n^{2\nu}}\right)^2 \right ] \label{cantilever_1}\end{aligned}\ ] ] as a result , the total partition function in this case is given by @xmath61 \ : \int\limits_{b n^{\nu}}^{b n } \ : d r \ ; \exp \left [ n \ : { \cal g } ( { \widetilde f } ) \right ] \ : \exp \left [ - \dfrac{k_c}{2 k_bt } \left(d - \sqrt{r^2 - b^2 n^{2 \nu } } \right)^2 \right ] \theta ( d - \sqrt{r^2 - b^2 n^{2 \nu } } ) \label{total_partition_function}\end{aligned}\ ] ] where again the variable @xmath53 should be excluded in favor of @xmath54 by means of the relation @xmath55 . in eq . ( [ total_partition_function ] ) the following constraints @xmath77 have been taken into account . the corresponding free energy functional in terms of dynamical variables @xmath26 and @xmath27 has the following form @xmath78 the average force , which is measured in afm - experiments , is given by @xmath79 \nonumber\\ & & \times \theta \left ( d - \sqrt{r^2 - b^2 n^{2 \nu}}\right ) . \label{average_force_2}\end{aligned}\ ] ] in our recent paper @xcite we have studied a single polymer force - induced desorption kinetics by making use of the notion of tensile blobs as well as by means of monte carlo and molecular dynamics simulations . it was clearly demonstrated that the total desorption time @xmath80 scales with polymer length @xmath81 as @xmath82 . in order to treat a realistic afm experiment in which the cantilever - substrate distance changes with constant velocity @xmath0 , i.e. , @xmath83 , one has to consider the afm tip dynamics . with this in mind , we will develop a coarse - grained stochastic model based on the free - energy functional eq . ( [ free_energy_frictionless ] ) . before proceeding any further , we need to define the adsorption - desorption potential profile @xmath84 . this plays the role of the potential of mean force ( pmf ) which depends on @xmath26 . in the helmholtz free - energy functional @xmath85 , given by eq . ( [ free_energy_frictionless ] ) and eq . ( [ free_energy ] ) , the free energy of the adsorbed portion is given by a simple contact potential , @xmath86 , where @xmath26 is an integer number in the range @xmath87 . considering desorption dynamics ( see below ) , we will treat @xmath26 as a continuous variable with a corresponding adsorption - desorption energy profile satisfying the following conditions : 1 . for integer @xmath26-values the energy profile has minima whereby we use the contact potential @xmath88 . 2 . for half - integer values of @xmath26 the adsorption potential goes over maxima . 3 . the activation barrier for monomer desorption , @xmath89 , and the corresponding adsorption activation barrier , @xmath90 , are proportional to the adsorption strength @xmath1 of the substrate whereby @xmath91 . the adsorption - desorption energy profile satisfies the boundary conditions : @xmath92 ( a fully adsorbed chain ) , and @xmath93 ( an entirely detached chain ) . one may show that the following energy profile , given as @xmath94 + n \right\rbrace - k_bt \epsilon n , \label{adsorption_profile}\end{aligned}\ ] ] meets the conditions ( 1 ) - ( 4 ) . the minima and maxima of eq . ( [ adsorption_profile ] ) are located in the points defined by @xmath95 = 1/2\pi$ ] with @xmath96 denoting the _ continuous _ index of a monomer . as a result , @xmath97 in eq . ( [ max_min ] ) the first term , @xmath98 , is very small and could be neglected . thus , the minima and maxima are located at the integer and half integer points respectively ( see fig . [ pealing_potential ] ) in order to calculate the activation barriers , we determine first @xmath99 at the half - integer points , i.e. , @xmath100 as well as at the integer points @xmath101 therefore , the activation barriers for the detachment , @xmath102 , and adsorption , @xmath103 , are given by @xmath104 i.e. , @xmath91 . finally , one may readily see that @xmath105 and @xmath93 which is in line with condition ( iv ) . the total helmholtz free energy for the frictionless substrate model is given by @xmath106 + n \right\rbrace - k_bt \epsilon n - k_bt n { \cal g } ( { \widetilde f } ) + \dfrac{k_c}{2 } \left(d - r \right)^2 , \label{free_energy_frictionless_dynamics}\end{aligned}\ ] ] whereas for the strong polymer - substrate friction model we have @xmath107 + n \right\rbrace - k_bt \epsilon n - k_bt n { \cal g } ( { \widetilde f } ) + \dfrac{k_c}{2 } \left(d - \sqrt{r^2 - b^2 n^{2 \nu } } \right)^2 . \label{free_energy_dynamics}\end{aligned}\ ] ] these helmholtz free energy functions govern the dissipative process which is described by the stochastic ( langevin ) differential equations @xmath108 where @xmath109 and @xmath110 are the onsager coefficients . the random forces @xmath111 and @xmath112 describe gaussian noise with means and correlators given by @xmath113 equations ( [ onsager ] ) are usually referred to as the onsager equations @xcite . the set of stochastic differential equations , eq.([onsager ] ) can be treated by a time integration scheme . each realization @xmath114 of the solution provides a time evolution of @xmath115 and @xmath116 . in order to get mean values of the observables , these trajectories should be averaged over many independent runs @xmath117 . for example , in order to obtain the average force , eq . ( [ average_force_2 ] ) , one should average over the runs @xmath118 ^ 2 - b^2 [ n^{(l)}(t)]^{2 \nu}}\right)\end{aligned}\ ] ] the thermodynamic forces which arise in eq . ( [ onsager ] ) , i.e. , @xmath119 and @xmath120 , could be calculated explicitly . for example , for the free energy function , eq . ( [ free_energy_frictionless_dynamics ] ) , one has @xmath121 \right\rbrace + k_bt \ : { \cal g } ( { \widetilde f } ) + k_bt \ : n \ : { \cal g}^{\prime } ( { \widetilde f } ) \ : \left ( \dfrac{\partial { \widetilde f } } { \partial n}\right)_{r } \nonumber\\ & = & - k_bt \epsilon \left\lbrace 1 - 2 \pi \sin [ ( 2 n + 1 ) \pi ] \right\rbrace + k_bt \ : { \cal g } ( { \widetilde f } ) - \dfrac{k_bt \ : r}{b\ : n } \ : \dfrac{{\cal g}^{\prime } ( { \widetilde f})}{{\cal l}^{\prime } ( { \widetilde f } ) } , \label{f_n_1}\end{aligned}\ ] ] where we have used ( recall that @xmath122 ) @xmath123 on the other hand , a direct calculation shows that @xmath124 ^ 2 } \ : - \ : \dfrac{1}{x}\nonumber\\ { \cal l}^{\prime } ( x ) & = & \dfrac{1}{x^2 } \ : - \ : \dfrac{1}{[\sinh ( x)]^2}\end{aligned}\ ] ] so that @xmath125 thus , for the force @xmath126 , given by eq . ( [ f_n_1 ] ) , one has @xmath127 \right\rbrace + k_bt \dfrac{r \ : { \widetilde f}}{b\ : n } + k_bt \ : { \cal g } ( { \widetilde f } ) . \label{force_frictionless}\end{aligned}\ ] ] in the strong friction case , eq . ( [ free_energy_dynamics ] ) leads to a more complicated expression for the thermodynamic force : @xmath128 \right\rbrace + k_bt \dfrac{r \ : { \widetilde f}}{b\ : n } + k_bt \ : { \cal g } ( { \widetilde f } ) + \ : \nu \ ; b^2 \ : k_c \ ; \left ( \dfrac{d}{\sqrt{r^2 - b^2 n^{2 \nu } } } - 1 \right ) . \label{force_strong_friction}\end{aligned}\ ] ] for @xmath129 one obtains @xmath130 where we have used @xmath131 taking into account eq . ( [ relationship ] ) , one finally derives @xmath132 for the model , given by eq . ( [ free_energy_dynamics ] ) , the corresponding force reads @xmath133 finally , the variable @xmath53 should be expressed in terms of @xmath54 by making use of the relationship @xmath55 , where @xmath56 is the inverse langevin function . a very good approximation for the inverse langevin function , published in ref . @xcite , is given by @xmath134 \label{interpolation}\end{aligned}\ ] ] it could be shown that for a strongly stretched desorbed portion of the polymer chain , the @xmath27 variable rapidly relaxes to its * quasi - stationary * value ( see appendix [ app ] ) . in other words , @xmath27 can quickly adjust to the slow evolution of @xmath26 ( governed by the kramers process ) . in this quasi - stationary approximation @xmath135 , and from eq . ( [ f_r_1 ] ) one has @xmath136 , so that the following nonlinear equation for @xmath27 emerges @xmath137 = \dfrac{k_bt}{b } \ : { \cal l}^{-1 } \left(\frac{r}{nb}\right ) \label{l^{-1}}\end{aligned}\ ] ] this could be represented as @xmath138}{k_bt } \right ) = 0 , \label{g}\end{aligned}\ ] ] i.e. , the height @xmath27 is instantaneously coupled to the number of desorbed beads , @xmath26 . inserting eq . ( [ interpolation ] ) into eq . ( [ l^-1 ] ) , one obtains @xmath139}{k_bt } - \left ( \dfrac{r}{b n}\right ) \ : \left [ \dfrac{3 - \left(\dfrac{r}{b n}\right)^2}{1 - \left(\dfrac{r}{b n}\right)^2 } \right ] = 0 . \label{p}\end{aligned}\ ] ] the onsager equation for the slow variable @xmath26 is given as @xmath140 \right\rbrace + k_bt \dfrac{r \ : { \widetilde f}}{b\ : n } + k_bt \ : { \cal g } ( { \widetilde f } ) ] + \xi_n ( t ) \label{kramers_process}\end{aligned}\ ] ] where @xmath53 is determined by eq . ( [ interpolation ] ) . eventually , we get a system of so - called semi - explicit differential - algebraic equations ( dae ) @xcite @xmath141 in this particular form of dae one can distinguish between the differential variable @xmath142 and the algebraic variable @xmath143 . ( [ dae ] ) can be solved numerically by making use of an appropriate runge - kutta ( rk ) algorithm , as shown in the appendix [ r - k ] . we have solved numerically our stochastic model , given by eq . ( [ kramers_process ] ) and eq . ( [ g ] ) , for the case of frictionless substrate . to this end we used the second order runge - kutta ( rk ) algorithm for stochastic differential - algebraic equations ( see appendix [ r - k ] for more details ) . the advantage of the stochastic differential equations approach as compared to the master equation method @xcite is that the former one gives a more detailed ( not averaged ) dynamic information corresponding to each individual force - displacement trajectory ( as is often in an experiment ) . the result of averaging over @xmath144 runs is shown in fig . [ averaging ] ( left ) . [ averaging ] ( right panel ) shows the resulting force - displacement diagram for @xmath145 and different detachment velocities . it it worth noting that the sawtooth pattern can be seen for all investigated detachment velocities ranging between @xmath146 and @xmath147 . for larger velocities the plateau height of the force grows substantially . in other words , the mean detachment force increases as the afm - tip velocity gets higher and the bonds stretching between successive monomers becomes stronger . we have also studied the detachment force behavior as well as that of the cantilever tip distance from the substrate at the moment of a full detachment , ( i.e. when @xmath148 ) , by repeating the detachment procedure @xmath149 times and plotting the probability distribution functions ( pdf ) for different adsorption energies @xmath1 and detachment velocities @xmath0 - fig . [ pdf_force ] . as one can see from fig . [ pdf_force]a , b , both the average and the dispersion of detachment force grow with @xmath0 which agrees with findings for reversible ( i.e. , when a broken bond can rebind ) bond - breaking dynamics @xcite . in contrast , the mean cantilever tip distance @xmath27 variance decreases and its average value increases with growing @xmath0 ( cf . fig . [ pdf_force]c , d ) . the average detachment force dependence on cantilever velocity @xmath0 is a widely covered subject in the literature in the context of biopolymers unfolding @xcite or forced separation of two adhesive surfaces @xcite . figure [ force_average]a , which shows the result of our calculations , has the characteristic features discussed also in ref . one observes a well expressed crossover from a shallow - slope for relatively small detachment rates to a steep - slope region as detachment speed increases . one remarkable feature is that this crossover practically does not depend on the adsorption energy @xmath1 : the curve is merely shifted upwards upon increasing of @xmath1 . therefore , the crossover is not related to a competition between the kramers rate and the cantilever velocity but rather accounts for the highly nonlinear chain stretching as the velocity @xmath0 increases . the corresponding detachment distance of the cantilever tip @xmath27 ( detachment height ) , fig . [ force_average]b , reveals a specific sigmoidal shape in agreement with the results based on the master equation @xcite . at low velocities of pulling , @xmath0 , when the chain still largely succeeds in relaxing back to equilibrium during detachment , an interesting entropy effect is manifested in fig . [ force_average]b : the ( effectively ) stiffer coil at @xmath150 leaves the substrate at lower values of @xmath27 than in the case of the colder system , @xmath151 . as the pulling velocity grows , however , this entropic effect vanishes and the departure from the substrate is largely governed by the stretching of the bonds rather than of the coil itself whereby the difference in behavior between @xmath150 and @xmath151 disappears . eventually , as it can be seen from fig . [ detachment_time ] , the total detachment ( peel ) time @xmath152 vs. velocity @xmath0 relationship has a well - defined power - law behavior , @xmath153 , with the power @xmath154 , in line with previous theoretical findings @xcite . in our md - simulations we use a coarse - grained model of a polymer chain of @xmath81 beads connected by finitely extendable elastic bonds . the bonded interactions in the chain is described by the frequently used kremer - grest potential , @xmath155 . the fene ( finitely extensible nonlinear elastic ) potential is given by @xmath156 } \label{fene}\ ] ] with @xmath157 and @xmath158 . in order to allow properly for excluded volume interactions between bonded monomers , the repulsion term is taken as weeks - chandler - anderson ( wca ) potential ( i.e. , the shifted and truncated repulsive branch of the lennard - jones potential , ) given by @xmath159\theta(2^{1/6}\sigma - r ) \label{wca}\ ] ] with @xmath160 or @xmath161 for @xmath162 , or @xmath163 , and @xmath164 , @xmath165 . the overall potential @xmath166 has a minimum at bond length @xmath167 . the nonbonded interaction between monomers are taken into account by means of the wca potential , eq . thus , the interactions in our model correspond to good solvent conditions . the substrate in the present investigation is considered simply as a structureless adsorbing plane , with a lennard - jones potential acting with strength @xmath168 in the perpendicular @xmath21direction , @xmath169 $ ] . in our simulations we consider as a rule the case of strong adsorption @xmath170 , where @xmath171 is a temperature of langevin thermal bath described below . the dynamics of the chain is obtain by solving the langevin equations of motion for the position @xmath172 $ ] of each bead in the chain , @xmath173 which describes the brownian motion of a set of bonded particles . the influence of solvent is split into slowly evolving viscous force and rapidly fluctuating stochastic force . the random gaussian force @xmath174 is related to friction coefficient @xmath175 by the fluctuation - dissipation theorem . the integration step is @xmath176 and time in measured in units of @xmath177 , where @xmath178 denotes the mass of the polymer beads , @xmath179 . in all our simulations the velocity - verlet algorithm was used to integrate equations of motion ( [ langevin ] ) . monomers during detachment from a substrate at @xmath180 ( left ) and @xmath181 ( right ) . here the interaction range of the adsorption potential is shaded ( transparent ) green . the cantilever tip is shown schematically in blue . one may clearly see that the polymer chain is more relaxed ( less stretched ) at @xmath180 , and the adsorbed monomers do not stick tightly to the surface but partially exit the range of surface adsorption . , title="fig:",width=8 ] monomers during detachment from a substrate at @xmath180 ( left ) and @xmath181 ( right ) . here the interaction range of the adsorption potential is shaded ( transparent ) green . the cantilever tip is shown schematically in blue . one may clearly see that the polymer chain is more relaxed ( less stretched ) at @xmath180 , and the adsorbed monomers do not stick tightly to the surface but partially exit the range of surface adsorption . , title="fig:",width=8 ] the molecule is pulled by a cantilever at constant velocity @xmath183 $ ] . the cantilever is imitated by two beads connected by harmonic spring and attached to one of the ends of the chain . the mass of beads @xmath184 , forming the cantilever , was set either to @xmath185 or to @xmath186 . the equilibrium size of this harmonic spring was set to @xmath187 and the spring constant was varied in the range @xmath188 . the hydrodynamics radius @xmath189 of beads composing the cantilever was varied by changing the friction coefficient @xmath190 , taking into account the stokes law , @xmath191 , where @xmath192 is the solvent viscosity . taking the value of the thermal energy @xmath193 j at @xmath194 k , the typical kuhn length of @xmath195 nm and the mass of coarse - grained monomer as @xmath196 kg setups the unit of time in our simulations which is given in @xmath197 ps . the velocities used in simulations are in units of @xmath198 . spring constants of our cantilever in real units are : @xmath199 . two typical snapshots of a polymer chain during slow detachment from an adsorbing substrate with different strengths of adsorption , @xmath200 and @xmath201 are shown in fig . [ fig_snapshots ] . evidently , the chain is much more stretched for the strongly - attractive substrate where all adsorbed monomers stick firmly to the surface . at the cantilever was calculated by monitoring extensions @xmath202 of a harmonic spring , i.e. @xmath203 . here @xmath204 , @xmath205 , @xmath206 and @xmath207 . @xmath185 and @xmath208 . ] as we have already seen in sec . [ res ] , the averaging of the force profile over many runs reveals the inherent sawtooth - structure of the force vs distance dependence ( see fig . [ averaging ] ) which is otherwise overshaded by thermal noise . our md - simulation result , depicted in fig . [ average_1 ] , show the same tendency against the noisy background of a single detachment event . therefore , for better clarity and physical insight , all our graphic results that are given below result from such averaging procedure . figure [ force_vs_distance_md]a shows how adsorption energy @xmath1 affects the force @xmath12 vs distance @xmath13 relationship . apparently , with increasing @xmath1 the mean force ( plateau height ) is found to grow in agreement with our equilibrium theory results , given in fig . [ equilibrium_1 ] . as suggested by our recent theory @xcite , the plateau height goes up as @xmath209 , or as @xmath210 , for relatively small or large @xmath1 values , respectively . the amplitude of spikes increases with growing @xmath1 too , in line with the equilibrium findings ( see fig . [ equilibrium_1 ] ) . moreover , as found by jagota _ @xcite , the amplitude of spikes follows an exponential law , @xmath211 , where @xmath26 is the number of desorbed polymer segments . on the other hand , the comparison of fig . [ force_vs_distance_md]b and fig.[equilibrium ] suggests that the stiffness of the cantilever spring constant @xmath37 affects mainly the spike amplitude especially at large @xmath1 . at the cantilever as a function of a distance from the substrate @xmath13 during chain detachment . here @xmath204 , @xmath206 , @xmath185 and @xmath208 . presented results are for different adsorption strengths @xmath1 , as indicated in the legend , and @xmath212 ( a ) . figure ( b ) presents results for @xmath213 and different spring constants @xmath37 . each curve originates from @xmath214 independent simulations.,title="fig : " ] at the cantilever as a function of a distance from the substrate @xmath13 during chain detachment . here @xmath204 , @xmath206 , @xmath185 and @xmath208 . presented results are for different adsorption strengths @xmath1 , as indicated in the legend , and @xmath212 ( a ) . figure ( b ) presents results for @xmath213 and different spring constants @xmath37 . each curve originates from @xmath214 independent simulations.,title="fig : " ] vs distance above the substrate @xmath13 displayed for different pulling velocities @xmath0 . here @xmath204 , @xmath205 and @xmath207 and ( a ) @xmath215 , ( b ) @xmath216 , ( c ) @xmath217 , ( d ) @xmath218.,title="fig : " ] vs distance above the substrate @xmath13 displayed for different pulling velocities @xmath0 . here @xmath204 , @xmath205 and @xmath207 and ( a ) @xmath215 , ( b ) @xmath216 , ( c ) @xmath217 , ( d ) @xmath218.,title="fig : " ] vs distance above the substrate @xmath13 displayed for different pulling velocities @xmath0 . here @xmath204 , @xmath205 and @xmath207 and ( a ) @xmath215 , ( b ) @xmath216 , ( c ) @xmath217 , ( d ) @xmath218.,title="fig : " ] vs distance above the substrate @xmath13 displayed for different pulling velocities @xmath0 . here @xmath204 , @xmath205 and @xmath207 and ( a ) @xmath215 , ( b ) @xmath216 , ( c ) @xmath217 , ( d ) @xmath218.,title="fig : " ] eventually , we demonstrate the impact of cantilever velocity , @xmath0 , as well as of its mass , @xmath184 , and friction coefficient , @xmath219 , on the force - distance profile . apparently , these parameters affect differently strong the observed force - distance relationship . similar to the results , obtained for our coarse - grained model in sec . [ second_section ] , in the md - simulation data the plateau height grows less than twice upon velocity increase of three orders of magnitude ( see fig . [ velocity_impact ] ) ! only for a very massive , ( @xmath220 ) , and strong - friction , ( @xmath221 ) , cantilever , the plateau height grows significantly and gains a slight positive slope ( see fig.[velocity_impact]d ) whereby oscillations vanish . this occurs for the fastest detachment @xmath222 . evidently , this effect is related to the combined role of the friction force in the case of rapid detachment along with the much larger inertial force ( @xmath223 ) whereby the substrate - induced oscillations are overshadowed by the increased effort of pulling . in contrast , neither fig . [ velocity_impact]b , nor fig . [ velocity_impact]c indicate any major qualitative changes in the @xmath12-vs-@xmath13-behavior when medium - friction , or mass cantilever alone are drastically changed . the pdf of the detachment force and its velocity @xmath0 dependence are shown in fig . [ pdf_force_md ] . similarly as in sec . [ second_section ] , the average value and dispersion grow with increasing speed of pulling and this is weakly sensitive with regard to the adsorption strength of the substrate @xmath1 . remarkably , the mean detachment force @xmath224 shows a similar nonlinear dependence on @xmath225 ( cf . [ force_average]a ) . the crossover position does not change practically as the adhesion strength is varied , and the variation of the other parameters ( @xmath226 ) towards a massive and strong - friction cantilever render this crossover considerably more pronounced . the complementary pdf for the detachment height @xmath27 is given in fig . [ pdf_r_md]a together with the corresponding average @xmath227 vs @xmath0 relationship . as predicted by our analytic model , cf . section [ second_section ] , the height of final detachment of the chain from the substrate becomes larger for faster peeling @xmath0 and stronger adhesion @xmath1 , which is consistent with the md data . one can see again the typical sigmoidal - shape in the @xmath227 vs @xmath0 dependence . the two panels for different temperature , shown in fig . [ pdf_r_md]b , indicate a smaller increase in @xmath228 at the higher temperature , provided the pulling velocity @xmath0 is sufficiently small too . this can be readily understood in terms entropic ( rubber ) elasticity of polymers and represents a case of delicate interplay between entropy and energy - dominated behavior . it is well known that a polymer coil becomes less elastic ( i.e. , it contracts ) upon a temperature increase , cf . the lowest ( grey ) curve in fig . [ pdf_r_md]b , ( left panel ) at @xmath150 , so that @xmath27 is smaller than in the corresponding lowest curve for @xmath151 in the right panel of fig . [ pdf_r_md]b . this occurs at low values of @xmath0 . on the other hand , the softer chain ( at @xmath151 ) stretches more easily and , therefore , @xmath27 goes up to @xmath229 for the highest speed @xmath230 instead of @xmath231 for @xmath232 . this entropic effect is well expressed at weak attraction to the surface , @xmath233 , which does not induce strong stretching of the bonds along the chain backbone . in contrast , at high @xmath234 , the bonds extend so strongly that the chain turns almost into a string and entropy effects become negligible . the energy cost of stretching then dominates and leads to higher values of @xmath27 at the higher temperature ( cf . upper most green symbols in fig . [ pdf_r_md]b ) since it is now the elasticity of the bonds between neighboring segments which governs the physics of detachment . in this case the elastic constant of the bonds effectively decreases with an increase of @xmath9 so that the distance of detachment @xmath27 in the left panel of fig . [ pdf_r_md]b for @xmath150 is higher than that for @xmath151 in the right panel . at the cantilever measured in the moment of detachment ( when the last monomer leaves the substrate ) . here @xmath204 , @xmath205 , @xmath207 , @xmath185 and @xmath208 . ( b ) averaged force @xmath224 plotted versus pulling velocity @xmath0 in the semilogaritmic scale . the values of parameters are the same . , title="fig : " ] at the cantilever measured in the moment of detachment ( when the last monomer leaves the substrate ) . here @xmath204 , @xmath205 , @xmath207 , @xmath185 and @xmath208 . ( b ) averaged force @xmath224 plotted versus pulling velocity @xmath0 in the semilogaritmic scale . the values of parameters are the same . , title="fig : " ] . upper panel present results for @xmath185 and @xmath208 while lower panel contains data for @xmath223 and @xmath235 . solid line represent results for @xmath207 while dashed line for @xmath236 . here @xmath204 , @xmath205 . ( b ) averaged detachment height @xmath228 as a function of pulling velocity @xmath0 in the semi - logarithmic scale plotted for different adsorption strengths as indicated : ( left panel ) @xmath237 , ( right panel ) @xmath238 . on both panels symbols plotted with same color correspond to the same ratio of @xmath239 while the corresponding values of @xmath168 and @xmath171 differ . here @xmath240 , @xmath205 , @xmath185 and @xmath208.,title="fig : " ] . upper panel present results for @xmath185 and @xmath208 while lower panel contains data for @xmath223 and @xmath235 . solid line represent results for @xmath207 while dashed line for @xmath236 . here @xmath204 , @xmath205 . ( b ) averaged detachment height @xmath228 as a function of pulling velocity @xmath0 in the semi - logarithmic scale plotted for different adsorption strengths as indicated : ( left panel ) @xmath237 , ( right panel ) @xmath238 . on both panels symbols plotted with same color correspond to the same ratio of @xmath239 while the corresponding values of @xmath168 and @xmath171 differ . here @xmath240 , @xmath205 , @xmath185 and @xmath208.,title="fig : " ] we have demonstrated in this paper that a simple theory , based on the onsager stochastic equations , yields an adequate description of a typical afm- experiment within the displacement - control mode . this approach makes it possible to relax most of the restrictions inherent in the be - model . for example , this approach also holds for small desorption activation barriers ( i.e. , for @xmath241 ) , and also for nonlinear barrier vs. force dependence . it naturally takes into account the reversible desorption - adsorption events @xcite which are neglected in be - model . moreover , it does not rest on the stationary approximation ( which is customary in the standard kramers rate calculation @xcite ) and is , therefore , ideally suited for description of driven force- ( fc ) or displacement - control ( dc ) regimes . one of the principal results in this analytic treatment is the predicted existence of characteristic spikes the mean force vs distance profile , observed in the dc - regime . these spikes depend on the adsorption energy @xmath1 , cantilever spring constant @xmath37 as well as on the cantilever velocity @xmath0 . in equilibrium , this has been found earlier by jagota and coworkers @xcite . the pdf of detachment forces and detachment distances are been thoroughly investigated . the relevant mean detachment force is found to be a strongly nonlinear function of @xmath0 which is mainly governed by the nonlinear chain stretching upon increasing @xmath0 . the average full detachment ( peeling ) time scales @xmath242 which is supported by early theoretical findings @xcite . some of these predictions were checked by means of md - simulation and found in a qualitative agreement with the results , gained by the analytic method . most notably , this applies to properties like the characteristic force oscillations pattern and the mean force vs cantilever velocity @xmath0 dependence . on the other hand , our md - simulation reveals a very strong increase in the magnitude of the force plateau for a strong - friction ( @xmath221 ) and massive ( @xmath243 ) cantilever . interestingly , in this case the spikes pattern is almost totally smeared out . this might be the reason why the force spikes pattern is not seen in laboratory detachment experiments . we recall that in a recent brownian dynamic simulation ( which totally ignores inertia forces ) @xcite , the friction coefficient of the cantilever was @xmath244 times larger than the friction coefficient of the chain segments . it was shown that for this high - friction cantilever and large velocity of pulling , the force spikes pattern was significantly attenuated @xcite so that information on the base sequence was hardly assessable . therefore , fabrication of a stiff and super - light , nanometer - sized afm probe would be a challenging task for future developments of biopolymer sequencing . as an outlook , our coarse - grained onsager stochastic model could be generalized to encompass investigations of forced unfolding of a multi - domain , self - associating biopolymers @xcite . this work was supported by grant sfb 625 from german research foundation ( dfg ) . computational time on pl - grid infrastructure is acknowledged . a.m. thanks the max planck institute for polymer research in mainz , germany , and cecam - mpi for hospitality during his visit at the institute . due to strong adsorption , the desorbed portion of polymer chain is expected to be strongly stretched . one could simplify the force , @xmath245/[1 - ( r / bn)^2 ] \approx 1/[1 - ( r / bn)]$ ] , where eq . ( [ interpolation ] ) has been used and the contribution of the cantilever has been neglected . therefore , the simplified equation which governs @xmath27 reads @xmath246 } \label{simplified}\end{aligned}\ ] ] this equation can be easily solved and the corresponding solution has the form @xmath247 where the relaxation time @xmath248 . this result suggests that for a strongly stretched chain , i.e. , for @xmath249 , the relaxation time @xmath250 is very small @xcite . for example , in the case that @xmath251 we have @xmath252 this relaxation time should be compared to the characteristic time , @xmath253 , of the slow variable @xmath142 which is governed by the kramers process . according to the semi - phenomenological bell model @xcite , the characteristic time of unbonding ( that is , desorption in our case ) is given by @xmath254 $ ] where @xmath255 is the segmental time , @xmath256 is the activation energy for single monomer desorption , @xmath257 stands for the width of adsorption potential , and @xmath258 is the plateau height . the free energies in the desorbed , @xmath259 , and in the adsorbed , @xmath260 , states are given by @xmath261 and @xmath262 where @xmath263 and @xmath264 are the so - called connective constants in two- and three dimensions respectively @xcite . as mentioned in sec . [ mm ] , for large adsorption energies @xmath1 the dimensionless plateau height @xmath265 . taking this into account , one could represent @xmath266 in the following form : @xmath267 , \label{kk}\end{aligned}\ ] ] where @xmath268 . therefore , in the case when @xmath269 , the distance @xmath27 could be treated as the fast variable . with eqs . ( [ tau_r ] ) and ( [ kk ] ) and taking into account that the onsager coefficient @xmath270 , this condition means that @xmath271 \gg 1 . \label{gg}\end{aligned}\ ] ] this condition holds for all typical values of the relevant parameters . in order to solve the dae ( [ dae ] ) numerically , one may employ the second order runge - kutta ( rk ) algorithm . to this end the first equation in eq . ( [ dae ] ) may be rewritten as an integral equation which relates the @xmath272-th and @xmath273 grid points ( using discrete time points @xmath274 ) @xmath275 where @xmath276 is the time step and @xmath277 . moreover , @xmath278 describes a wiener process with zero mean and with variance : @xmath279 the integral over the deterministic force in eq . ( [ integral_equation_2 ] ) within this @xmath280 order approximation reads @xmath281 + { \cal o } ( h^3)\end{aligned}\ ] ] this is so - called trapezoidal rule for approximation of the integral . in order to calculate @xmath282 , one should first take the initial value @xmath283 , and find @xmath284 through the solution of the nonlinear equation @xmath285 . for the calculation of @xmath286 , one can use the forward euler method of order @xmath161 , i.e. , @xmath287 and @xmath288 are obtained as solution of the equation @xmath289 . here the random variable @xmath290 is gaussian with zero mean value and with variance @xmath291 1 . for a given initial value of @xmath283 , go to eq . ( [ g ] ) or eq . ( [ p ] ) and solve this nonlinear equation ( e.g. @xmath292 ) with respect to @xmath284 . 2 . compute @xmath293 . 3 . compute @xmath294 and @xmath295 within the euler approximation , i.e. , calculate first @xmath296 and then solve @xmath289 with respect to @xmath288 . 4 . compute @xmath297 . 5 . compute the corrected @xmath294 , i.e. @xmath298 6 . finally , with the value of @xmath294 , go to item @xmath161 and solve the nonlinear equation @xmath299 with respect to @xmath295 . 99 f. ritort , j. phys . : condens . matter , 2006 , * 18 * , r531 - r583 . i. franco , m.a . ratner , g.c . schatz , single - molecule pulling : phenomenology and interpretation , in nano and cell mechanics : fundamentals and frontiers , edited by h.d . esinosa and g. bao ( wiley , microsystem and nanotechnology series , 2013 ) chap . 14 , p. 359 - 388 . balzer , m. gallei , m.v . hauf , m. stallhofer , l. wiegleb , a. holleiner , m. rehahn , t. hugel , angew . int . ed . , 2013 , * 52 * , 6541 - 6544 . butt , b. cappella , m. kappl , surf . , 2005 , * 59 * , 1 - 152 . y. gao , g. sirinakis , y. zhang , j. am . soc . , 2011 , * 133 * , 12749 - 12757 . a. janshof , m. neitzert , y. oberdorfer , h. fuchs , angew . , int . ed . , 2000 , * 39 * , 3212 - 3237 . m. carrion - vazquez , a. f. oberhauser , s. fowler , p. e. marszalek , s.e . broedel , j. clarke , j. m. fernandez , proc . natl , acad . , 1999 , * 96 * , 3694 - 3699 . t. e. fisher , a. f. oberhauser , m. carrion - vazquez , p. e. marszalek , j. m. fernandez , trends biochem . , 1999 , * 24 * , 379 - 384 . h. clausen - schaumann , m. seitz , r. krautbauer , h. e. gaub , curr . biol . , 2000 , * 4 * , 524 - 530 sing , a. alexander - katz , macromolecules , 2011 , * 44 * , 6962 - 6971 . sing , a. alexander - katz , macromolecules , 2012 , * 45 * , 6704 - 6718 . g. i. bell , science , 1978 , * 200 * 618 - 627 . e. evans , k. ritchie , biophys . j. , 1997 , * 72 * , 1541 - 1555 . e. evans , annu . struct . , 2001 , * 30 * , 105 - 128 . e. evans , d. a. calderwood , science , 2007 , * 316 * 1148 - 1153 . r. merkel , p. nassoy , k. ritchi , e. evans , nature , 1999 , * 397 * , 50 - 53 . s. manohar , a. jagota , phys . e , 2010 , * 81 * , 021805 . s. manohar , a. r. manz , k. e. bancroft , ch - y . hui , a. jagota , d. v. vezenov , nano lett . , 2008 , * 8 * , 4365 - 4372 . s. iliafar , d. vezenov , a. jagota , langmuir , 2013 , * 29 * , 1435 - 1444 . s. iliafar , k. wagner , s. manohar , a. jagota , d. vezenov , j. phys . c , 2012 , * 116 * , 13896 - 13903 . h. j. kreuzer , s.h . payne , l. livadaru , biophys . j. , 2001 , * 80 * , 2505 - 2514 . d. b. staple , f. hanke , h. j. kreuzer , phys . e , 2008 , * 77 * , 021801 . d. b. staple , m. geisler , t. hugel , l. kreplak , h.j . kreuzer , new j. phys . , 2011 , * 13 * , 013025 . p. hnggi , p. talkner , m. borcovec , rev . phys . , 1990 , * 62 * , 251 - 341 . f. hanke , h. j. kreuzer , phys . e , 2006 , * 74 * , 031909 . s. dattagupta , s. puri , dissipative phenomena in condensed matter , springer - verlag , berlin , 2004 . s. bhattacharya , v. g. rostiashvili , a. milchev , t.a . vilgis , macromolecules , 2009 , * 42 * , 2236 - 2250 . s. bhattacharya , v. g. rostiashvili , a. milchev , t.a . vilgis , phys . e , 2009 , * 79 * , 030802(r ) . s. bhattacharya , a. milchev , v. g. rostiashvili , t.a.vilgis , eur . j. e , 2009 , * 29 * , 285 - 297 . sheng , p .- y . lai , phys . e , 1997 , * 56 * , 1900 - 1909 . j. u. schurr , s.b . smith , biopolymers , 1990 , * 29 * , 1161 - 1165 . a. serr , r.r . netz , europhys . lett . , 2006 , * 73 * , 292 - 298 . j. paturej , a. milchev , v.g . rostiashvili and t.a . vilgis , macromolecules , 2012 , * 45 * , 4371 - 4380 . a. cohen , rheol . acta , 1991 , * 30 * , 270 - 273 . u. m. ascher , l. r. petzold , computer methods for ordinary differential equations and differential - algebraic equations , siam , philadelphia , 1998 . g. diezemann , a. janshoff , j. chem . phys . , 2008 , * 129 * , 084904 . d. b. staple , s. h. payne , a. l. c. reddin , h. j. kreuzer , phys . biol . , 2009 , * 6 * , 025005 . k. eom , d. e. makarov , g. j. rodin , phys . e , 2005 , * 71 * , 021904 . g. yoon , s. na , k. eom , j. chem . phys . , 2012 , * 137 * , 025102 . f. li , d. leckband , j. chem . phys . , 2006 , * 125 * , 194702 . u. seifert , phys . lett . , 2000 , * 84 * , 2750 - 2753 . m. febbo , a. milchev , v. rostiashvili , d. dimitrov , t.a . vilgis , j. chem . phys . , 2008 , * 129 * , 154908 . c. vanderzande , _ lattice models of polymers _ cambridge university press , cambridge , 1998 .
the problem of expressing a specific polynomial as the determinant of a square matrix of affine - linear forms arises from algebraic geometry , optimisation , complexity theory , and scientific computing . motivated by recent developments in this last area , we introduce the notion of a _ uniform determinantal representation _ , not of a single polynomial but rather of all polynomials in a given number of variables and of a given maximal degree . we derive a lower bound on the size of the matrix , and present a construction achieving that lower bound up to a constant factor as the number of variables is fixed and the degree grows . this construction marks an improvement upon a recent construction due to plestenjak hochstenbach , and we investigate the performance of new representations in their root - finding technique for bivariate systems . furthermore , we relate uniform determinantal representations to vector spaces of singular matrices , and we conclude with a number of future research directions . determinantal representation , system of polynomial equations , multiparameter matrix eigenvalue problem , space of singular matrices . 13p15 , 65h04 , 65f15 , 65f50 .
1607.04873
consider an @xmath0-variate polynomial of degree at most @xmath1 : @xmath2 where @xmath3 , @xmath4 , @xmath5 , @xmath6 , and where each coefficient @xmath7 is taken from a ground field @xmath8 . a _ determinantal representation _ of @xmath9 is an @xmath10-matrix @xmath11 of the form @xmath12 where each @xmath13 with @xmath14 . we call @xmath15 the _ size _ of the determinantal representation . clearly , since the entries of @xmath11 are affine - linear forms in @xmath16 , @xmath15 must be at least the degree of @xmath9 . determinantal representations of polynomials play a fundamental role in several mathematical areas : from _ algebraic geometry _ it is known that each plane curve ( @xmath17 ) of degree @xmath1 over an algebraically closed field @xmath8 admits a determinantal representation of size @xmath1 @xcite . over non - algebraically closed fields , and especially when restricting to symmetric determinantal representations , the situation is much more subtle @xcite . for larger @xmath0 , only certain hypersurfaces have a determinantal representation of size equal to their degree @xcite . in _ optimisation _ , and notably in the theory of _ hyperbolic polynomials _ @xcite , one is particularly interested in the case where @xmath18 , @xmath19 is symmetric positive definite , and the @xmath20 are symmetric . in this case , the restriction of @xmath9 to any line through 0 has only real roots . for @xmath17 the converse also holds @xcite ; for counterexamples to this converse holding for higher @xmath0 , see @xcite . in _ complexity theory _ a central role is played by valiant s conjecture that the permanent of an @xmath21-matrix does not admit a determinantal representation of size polynomial in @xmath22 @xcite . via the _ geometric complexity theory _ programme @xcite this leads to the study of polynomials in the boundary of the orbit of the @xmath10-determinant under the action of the group @xmath23 permuting matrix entries . recent developments in this field include the study of this boundary for @xmath24 @xcite and the exciting negative result in @xcite that valiant s conjecture can _ not _ be proved using occurrence obstructions proposed earlier in @xcite . our motivation comes from _ scientific computing _ , where determinantal representations of polynomials have recently been proposed for efficiently solving systems of equations @xcite . for this application , it is crucial to have determinantal representations not of a _ single _ polynomial @xmath9 , but rather of all @xmath0-variate polynomials of degree at most @xmath1 . moreover , the representation should be easily computable from the coefficients of @xmath9 . specifically , in @xcite determinantal representations are constructed for the bivariate case ( @xmath17 ) in which the entries of the matrices @xmath25 themselves _ depend affine - linearly _ on the coefficients @xmath7 . this is what we call a _ uniform determinantal representation _ of the generic polynomial @xmath9 of degree @xmath1 in @xmath0 variables ; see section [ sec : problem ] for a precise definition . [ ex : binaryquadric ] the identity @xmath26 exhibits the matrix on the right as a uniform determinantal representation of the generic bivariate quadric . @xmath27 in applications , the matrix @xmath11 is used as input to algorithms in numerical linear algebra that scale unfavourably with @xmath15 , such as a complexity of @xmath28 . consequently , we are led to consider the following fundamental question . what is the minimal size @xmath29 of any uniform determinantal representation of the generic polynomial of degree @xmath1 in @xmath0 variables ? a construction from @xcite shows that for fixed @xmath17 and @xmath30 we have @xmath31 ; this construction is reviewed in section [ sec : first ] . we improve the construction from @xcite by giving a particularly elegant uniform determinantal representations of bivariate polynomials of size @xmath32 in example [ ex:2d ] , and of size @xmath33 in example [ ex:2d1 ] . in view of the obvious lower bound of @xmath1 this is clearly sharp up to a constant factor for @xmath30 , although we do not know where in the interval @xmath34 $ ] the true answer lies . we show in section [ sec : numerics ] how to use these small determinantal representations of bivariate polynomials for solving systems of equations . before that , we focus on the asymptotic behaviour of @xmath29 for fixed @xmath0 and @xmath30 . in this setting , we derive the following result . [ thm : main ] for fixed @xmath35 there exist positive constants @xmath36 ( depending on @xmath0 ) such that for each @xmath37 the smallest size @xmath29 of a uniform determinantal representation of the generic polynomial of degree @xmath1 in @xmath0 variables satisfies @xmath38 . we will also compare our results with previous constructions , most notably with those by quarez ( * ? ? ? 4.4 ) , who proves the existence of a symmetric representation of size @xmath39 . for fixed @xmath0 and @xmath30 , @xcite therefore has the asymptotic rate @xmath40 , meaning that the results of this paper represent a clear improvement . for fixed @xmath1 and @xmath41 , @xcite leads to the asymptotic behavior @xmath42 , which is similar to our bounds ; we will discuss more details in section [ sec : outlook ] . in section [ sec : problem ] we formalise the notion of uniform determinantal representations , study their symmetries , and derive some simple properties . in particular , we relate uniform determinantal representations to spaces of singular @xmath10-matrices . in section [ sec : singular ] we briefly review some of the existing literature on these singular spaces , and we prove that for @xmath43 there are infinitely many equivalence classes of such objects ; this poses an obstruction to a `` brute - force '' approach towards finding lower bounds on @xmath29 . in section [ sec : first ] we present a first construction , of which however the size is of the order of @xmath44 , rather than @xmath45 , for @xmath30 . in section [ sec : second ] we give a more efficient construction and prove theorem [ thm : main ] . in section [ sec : small ] we give upper bounds on @xmath29 for small @xmath0 and @xmath1 and determine @xmath46 and @xmath47 exactly . we extend representations from scalar to matrix polynomials in section [ sec : matpol ] . in section [ sec : numerics ] we give some numerical results that show that for @xmath17 and small @xmath1 we get a competitive method for computing zeros of polynomials systems . finally , in section [ sec : outlook ] we summarise our main conclusions and collect some questions that arise naturally from our work . in this section we give a formal definition of uniform determinantal representations , and introduce a group that acts on such representations . we also show that a uniform determinantal representation gives rise to a vector space consisting entirely of singular matrices ; such spaces are the topic of next section . let @xmath8 be a field and fix @xmath48 . let @xmath49 denote the polynomials of degree at most @xmath1 in the polynomial ring @xmath50 $ ] . furthermore , let @xmath51 be the generic polynomial of that degree , i.e. , @xmath52 where @xmath3 , @xmath4 , @xmath5 , @xmath6 , and where we consider @xmath7 as a variable for each @xmath53 . for @xmath54 , a _ uniform determinantal representation _ of @xmath51 is an @xmath10-matrix @xmath11 with entries from @xmath55 $ ] , of degree at most 1 in each of these two sets of variables , such that @xmath56 . the number @xmath15 is called the _ size _ of the determinantal representation . to be explicit , we require each entry of @xmath11 to be a @xmath8-linear combination of the monomials @xmath57 . this means that we can decompose @xmath11 as @xmath58 , where @xmath59 contains all terms in @xmath11 that do not contain any @xmath7 , and where @xmath60 contains all terms in @xmath11 that do . we will use the notation @xmath61 throughout the paper . when @xmath0 and @xmath1 are fixed in the context , we will also speak of a uniform determinantal representation without reference to @xmath51 . our ultimate aim is to determine the following quantity . for @xmath54 , @xmath62 is the minimum among all sizes of uniform determinantal representations of @xmath51 . this minimal size could potentially depend on the ground field @xmath8 , but the bounds that we will prove do not . note that in the definition of @xmath29 we do not allow terms in @xmath11 of degree strictly larger than one in the @xmath7 . relaxing this condition to polynomial dependence on the @xmath7 might affect the exact value of @xmath63 , but it will not affect our bounds see remark [ re : poly ] . given a uniform determinantal representation @xmath11 of size @xmath15 , and given matrices @xmath64 in @xmath65 , the group of determinant - one matrices with entries in @xmath8 , the matrix @xmath66 is another uniform determinantal representation of @xmath51 . in this manner , the group @xmath67 acts on the set of uniform determinantal representations of @xmath51 . moreover , there exist further symmetries , arising from affine transformations of the @xmath0-space . recall that these transformations form the group @xmath68 generated by invertible linear transformations and translations . [ lm : affine ] the group @xmath69 acts on uniform determinantal representations of @xmath51 . the statement of this lemma is empty without making the action explicit , as we do in the proof . let @xmath70 be an affine transformation of @xmath71 , and expand @xmath72 where the @xmath73 are linear combinations of the @xmath7 . more precisely , the vector @xmath74 can be written as @xmath75 , where @xmath76 is the representation of @xmath69 on polynomials of degree at most @xmath1 regarded as a matrix representation relative to the monomial basis . now let @xmath77 be a uniform determinantal representation of @xmath51 . then @xmath78 i.e. , @xmath79 is another uniform determinantal representation of @xmath51 . the action of @xmath80 is given by @xmath81 . take @xmath82 and the affine transformation @xmath83 with inverse @xmath84 . we have @xmath85 we find @xmath86 if we make the substitutions @xmath87 in the uniform determinantal representation of example [ ex : binaryquadric ] , then we arrive at the matrix @xmath88 whose determinant also equals @xmath89 . @xmath27 the action of the affine group will be used in section [ sec : small ] to determine the exact value of @xmath90 for @xmath17 and @xmath91 . we now turn our attention to the component @xmath59 of a uniform determinantal representation @xmath11 . [ lm : singspace ] for any uniform determinantal representation @xmath61 of size @xmath15 , the determinant of @xmath59 is the zero polynomial in @xmath50 $ ] . moreover , at every point @xmath92 , the rank of the specialisation @xmath93 is exactly @xmath94 . the first statement follows from the fact that @xmath95 is the part of the polynomial @xmath96 which is homogeneous of degree zero in the @xmath7 ; hence zero . by specialising the vector @xmath97 of variables to a point @xmath92 , the rank of @xmath59 can only drop , so the rank of @xmath98 is at most @xmath94 . however , if it were at most @xmath99 , then after column operations on @xmath11 by means of determinant - one matrices with entries in @xmath8 we may assume that @xmath98 has its last two columns equal to 0 . this means that all entries of @xmath100 in these columns are linear in the @xmath7 . this in turn implies that any term in the polynomial @xmath101 is at least quadratic in the @xmath7 . but on the other hand @xmath102 equals @xmath103 , which is a non - zero _ linear _ polynomial in the @xmath7 ( nonzero since not every polynomial of degree at most @xmath1 vanishes at @xmath104 ) . this contradiction implies that the rank of @xmath98 is @xmath94 . [ lm : dets ] if @xmath61 is a uniform determinantal representation of size @xmath15 , then @xmath105 spanned by the @xmath106-subdeterminants of @xmath59 satisfies @xmath107 . here , as in the rest of this paper , by the product of two spaces of polynomials we mean the @xmath8-linear span of all the products . let @xmath108 be the determinant of the submatrix of @xmath59 obtained by deleting the @xmath109th row and the @xmath110th column . on the one hand , @xmath56 is linear in the @xmath7 by assumption , and on the other hand , by expanding @xmath96 we see that the part that is homogeneous of degree one in the @xmath7 is @xmath111 this therefore equals @xmath51 . hence any element @xmath112 of @xmath49 is obtained from the expression above by specialising the variables @xmath7 to the coefficients of @xmath112 . since each @xmath113 is then specialised to an element of @xmath114 , we find @xmath115 . [ re : poly ] note that the proof above still applies if we allow determinantal representations of the form @xmath116 where @xmath117 is homogeneous of degree @xmath118 in the @xmath7 and affine - linear in the @xmath119 . since our upper bound in section [ sec : second ] builds directly on this lemma , the upper bound holds in this more general setting , as well . let @xmath61 be a uniform determinantal representation of @xmath51 . writing @xmath120 , lemma [ lm : singspace ] implies that the linear span @xmath121 consists entirely of singular matrices ( and indeed that this remains true when extending scalars from @xmath8 to an extension field ) . there is an extensive literature on such _ singular matrix spaces _ ; see , e.g. , @xcite and the references therein . the easiest examples are the following . a subspace @xmath122 is called a _ compression space _ if there exists a subspace @xmath123 with @xmath124 . we call the space @xmath125 a _ witness _ for the singularity of @xmath126 . given any two subspaces @xmath127 with @xmath128 , the space of all matrices which map @xmath125 into @xmath129 ( acting on row vectors ) is a compression space with witness @xmath125 . it is easy to see that these spaces are inclusion - wise maximal among all singular spaces . if @xmath126 is a singular matrix space , then so is @xmath130 for any pair @xmath131 . we call the latter space _ conjugate _ to the former . [ ex : smallsingspace ] for @xmath132 , every singular matrix space is a compression space , hence conjugate to a subspace of one of the two spaces @xmath133 where the @xmath134s indicate entries that can be filled arbitrarily . a witness for the first space is the span @xmath135 of the second standard basis vector , and a witness for the second space is @xmath136 . for @xmath24 , there are four conjugacy classes of inclusion - maximal singular matrix spaces , represented by the three maximal compression spaces @xmath137 and the space of skew - symmetric @xmath138-matrices @xcite ; the latter is not a compression space . for @xmath139 , there are still finitely many ( namely , @xmath140 ) conjugacy classes of inclusion - maximal singular matrix spaces @xcite , but this is not true for @xmath141 , as theorem [ thm : sing5 ] below shows . this theorem is presumably folklore ; we include a proof since we have not been able to find a literature reference for it . [ prop : max ] assume that @xmath8 is algebraically closed . for any @xmath22 and @xmath142 the locus @xmath143 in the grassmannian @xmath144 of @xmath22-dimensional subspaces of @xmath145 consisting of all _ singular _ subspaces is closed in the zariski topology . moreover , the locus @xmath146 in @xmath143 consisting of all _ inclusion - wise maximal _ singular subspaces is open inside @xmath143 . the first statement is standard . for the second statement , consider the incidence variety @xmath147 which is a closed subvariety of @xmath148 . the projection of @xmath149 into @xmath143 is the complement of @xmath146 , and it is closed because @xmath150 is a projective variety . [ thm : sing5 ] assume that @xmath8 is infinite and of characteristic unequal to two . for @xmath141 there are infinitely many conjugacy classes of inclusion - wise maximal singular @xmath10-matrix spaces . take @xmath141 . for sufficiently general skew - symmetric matrices @xmath151 set @xmath152 and define the space @xmath153 each matrix in this space is singular , since for @xmath154 we have @xmath155 in @xcite it is proved that , for a specific choice of the tuple @xmath156 , the space @xmath157 is maximal among the singular subspaces of @xmath158 . by proposition [ prop : max ] , @xmath157 is maximal for sufficiently general @xmath156 , as well ( note that we may first extend @xmath8 to its algebraic closure to apply the proposition ) . in the notation of that proposition , we have a rational map @xmath159 where @xmath160 denotes the subspace of skew - symmetric matrices ; the dashed arrow indicates that the map is defined only in an open dense subset of @xmath161 . for any nonzero scalar @xmath162 , @xmath163 . we claim that , in fact , the general fibre of @xmath164 is indeed one - dimensional . as the fibre dimension is semicontinuous , it suffices to verify this at a particular point where @xmath164 is defined . we take @xmath165 for @xmath166 and @xmath167 general ; here @xmath168 is the matrix with zeros everywhere except for a 1 at position @xmath169 . let @xmath170 ; if @xmath171 , then there exists an invertible matrix @xmath172 such that @xmath173 for all @xmath97 , so that @xmath174 . using skew - symmetry of @xmath20 and @xmath175 , we find that @xmath176 . substituting our choice of @xmath20 for @xmath177 yields @xmath178 for all @xmath110 with @xmath179 , @xmath180 , so @xmath181 since @xmath182 , and @xmath183 . hence @xmath80 is a scalar multiple of the identity . it follows that the fibre of @xmath164 through @xmath156 is one - dimensional as claimed . since @xmath184 , we have thus constructed an @xmath185-dimensional family inside @xmath186 . given any point @xmath126 in @xmath186 , its orbit under @xmath187 has dimension at most @xmath188 ( scalars act trivially ) . now for @xmath189 we have @xmath190 so that we have found ( at least ) a one - parameter family of conjugacy classes of singular spaces . for @xmath191 the difference between @xmath192 and @xmath188 is even larger . for large @xmath15 it seems impossible to classify maximal singular matrix spaces . the construction above already gives an infinite number of conjugacy classes , but there are many other sources of examples . for instance , for infinitely many @xmath15 there exists a maximal singular matrix space in @xmath145 of constant dimension @xmath193 , at least if we assume that @xmath8 has characteristic 0 @xcite . on the other hand , if the singular matrix space @xmath126 has dimension at least @xmath194 , then it is a compression space with either a one - dimensional witness or all of @xmath195 as witness @xcite ( and hence of dimension exactly @xmath194 ) . a sharpening of this result is proved in @xcite ( see also @xcite ) . it should be noted that in many cases not even the dimension of such singular matrix spaces is known , for fixed values of the size and rank of the matrices . there is a considerable body of work devoted to giving lower and upper bounds for such dimensions , both in the case of bounded and constant rank , but these bounds are rarely sharp , see , among many other references , @xcite and the more recent works on skew - symmetric matrices of constant rank @xcite . hence the fact that @xmath59 represents a singular matrix space of dimension ( at most ) @xmath196 does not much narrow down our search for good uniform determinantal representations , except in small cases discussed in section [ sec : small ] . however , for our constructions in sections [ sec : first ] and [ sec : second ] we will only use compression spaces where the witness has dimension 1 or about @xmath197 , respectively ; and our lower bounds on @xmath29 are independent of the literature on singular matrix spaces . in this section we restrict our attention to determinantal representations @xmath61 where @xmath59 represents a compression space with a one - dimensional witness ( or , dually by transposition , with a full - dimensional witness ) . under this assumption we will prove quite tight bounds on the minimal size of a uniform matrix representation . the following fundamental notion will be used throughout below . we say that a subspace @xmath198 $ ] is _ connected to @xmath199 _ if it is nonzero and its intersections @xmath200 satisfy @xmath201 for each @xmath202 . note that this implies that @xmath203 . we borrow the terminology from the theory of border bases @xcite , where a set @xmath204 of monomials is called connected to @xmath199 if @xmath205 and each nonconstant monomial in @xmath204 can be divided by some variable to obtain another monomial in @xmath204 . the linear span of @xmath204 is then connected to @xmath199 in our sense . translating monomials to their exponent vectors , we will call a subset @xmath204 of @xmath206 _ connected to @xmath207 _ if it contains 0 and for each @xmath208 there exists an @xmath109 such that @xmath209 , where @xmath210 is the @xmath109-th standard basis vector . let @xmath129 be a finite - dimensional subspace of @xmath50 $ ] connected to 1 . choose a @xmath8-basis @xmath211 of @xmath129 whose total degrees increase weakly . for each @xmath212 write @xmath213 for suitable elements @xmath214 . let @xmath215 be the @xmath216-matrix whose @xmath109th row equals @xmath217 note that @xmath215 depends on the choice of basis , but we suppress this dependence in the notation , since the property of @xmath215 in the next lemma does not depend on the choice of basis . [ lm : block ] the @xmath8-linear subspace of @xmath50 $ ] spanned by the @xmath218-subdeterminants of @xmath215 equals @xmath129 . by construction , @xmath215 has rank @xmath219 over the field @xmath220 and satisfies @xmath221 . by ( a version of ) cramer s rule , the kernel of @xmath215 is also spanned by @xmath222 where @xmath223 is the determinant of the submatrix of @xmath215 obtained by removing the @xmath110th column . so these two vectors differ by a factor in @xmath220 . since @xmath224 we find that they are , in fact , equal . hence @xmath225 as claimed . we can now formulate our first general construction . this generalises a construction from @xcite to the multivariate case . [ prop : cons1 ] let @xmath129 be an @xmath22-dimensional space connected to @xmath199 and suppose that @xmath107 . then there exists a uniform determinantal representation of size @xmath22 for the generic polynomial of degree @xmath1 in @xmath0 variables . let @xmath226 $ ] be the @xmath227-subdeterminants of @xmath215 . by @xmath228 and lemma [ lm : block ] we can find , for each monomial @xmath229 of degree at most @xmath1 , affine - linear forms @xmath230 such that @xmath231 . define @xmath232 a row vector of bi - affine linear forms in the @xmath119 and the @xmath233 . then , by laplace expansion along the first row , we find that the determinant of @xmath234 is the generic polynomial of degree @xmath1 in @xmath16 . [ ex : lin1 ] for @xmath17 the following picture gives a space @xmath129 , connected to @xmath199 and spanned by the monomials marked with black vertices , such that @xmath235 : this is the construction of @xcite , which shows that there exists a uniform determinantal representation of the generic bivariate polynomial of degree @xmath1 of size @xmath236 as @xmath30 . @xmath27 the bivariate case generalises as follows . [ thm : cons1 ] for fixed @xmath0 , there exists a determinantal representation @xmath61 of the generic @xmath0-variate polynomial of degree @xmath1 of size @xmath237 such that , moreover , the singular matrix space represented by @xmath59 is a compression space with a one - dimensional witness . moreover , under this latter additional condition on @xmath59 , the bound is sharp . note that @xmath238 . hence by proposition [ prop : cons1 ] it suffices to show the existence of a subspace @xmath239 connected to @xmath199 and such that @xmath240 , where @xmath241 . we will , in fact , show that @xmath129 can be chosen to be spanned by monomials . first , recall that there exists a lattice @xmath242 in @xmath243 such that @xmath243 is the disjoint union of @xmath242 and its cosets @xmath244 for @xmath245 , namely , the root lattice of type @xmath246 generated by the rows of the @xmath247-cartan matrix @xmath248 , \ ] ] where the empty positions represent zeros ( * ? ? ? * planche 1 ) . in particular , the index of @xmath242 in @xmath243 equals @xmath0 . for example , if @xmath249 , here is the root lattice @xmath242 ( in black ) and its two cosets ( in gray and white ) : now let @xmath250 be the simplex in @xmath251 with vertices @xmath252 , for @xmath253 let @xmath254 be the set of lattice points in @xmath250 that have @xmath109th coordinate zero , and set @xmath255 . define @xmath256 a subset of the lattice points in @xmath250 . we claim that @xmath204 is connected to 0 . indeed , for each @xmath253 the set @xmath254 is connected to @xmath207 , and from each point @xmath53 in @xmath257 one can walk within @xmath257 to @xmath258 by subtracting @xmath259 times an @xmath260 . next , we claim that for each @xmath261 there exists a @xmath262 with @xmath263 . indeed , there is a ( unique ) @xmath264 with this property in @xmath265 . if this @xmath264 has nonnegative entries , then set @xmath266 . otherwise , @xmath53 itself has a zero entry , say on the @xmath109th position , and we set @xmath267 . furthermore , for @xmath253 the set @xmath254 contains @xmath268 vertices , and @xmath257 contains @xmath269 vertices . this concludes the construction note that in the construction of proposition [ prop : cons1 ] the matrix @xmath59 has a zero row , so that it represents a compression space with a one - dimensional witness . for sharpness , assume that @xmath61 is a uniform determinantal representation of size @xmath15 such that the singular matrix space represented by @xmath59 is a compression space with a one - dimensional witness . after a choice of basis of @xmath71 , we may assume that the first row of @xmath59 is identically zero ; write @xmath270^t$ ] accordingly . let @xmath271 be the first row of @xmath60 and write @xmath272^t$ ] . then we have @xmath273.\ ] ] let @xmath274 denote the @xmath275 subdeterminants of @xmath276 . by lemma [ lm : dets ] , the space @xmath129 spanned by these satisfies @xmath277 . this already gives a lower bound of @xmath129 equal to @xmath278 . to improve the @xmath196 in the denominator into an @xmath0 , we observe that by cramer s rule , the map @xmath279,\ ( \ell_1,\ldots,\ell_n ) \mapsto \sum_i ( -1)^i \ell_i d_i\ ] ] has every column of @xmath276 in its kernel . these columns are linearly independent over @xmath8 ( indeed over @xmath220 ; see lemma [ lm : singspace ] ) . we conclude that @xmath280 so that @xmath281 as desired . in the next section we derive a second general construction of uniform determinantal representations , which we use to prove theorem [ thm : main ] . for a while , we believed that the uniform determinantal representations of theorem [ thm : cons1 ] were optimal . but then we realised that if one relaxes the condition that @xmath59 represent a compression space with one - dimensional witness to the condition that @xmath59 represent _ some _ compression space , smaller - size representations are possible . the basic example is the following . [ ex:2d ] let @xmath282 be the generic polynomial of degree @xmath283 in @xmath17 variables . it has the following uniform determinantal representation : @xmath284 where the empty positions denote zeros . let @xmath61 be the matrix on the right - hand side . in this case , @xmath59 represents a compression space with witness @xmath285 , which is mapped into @xmath286 . to verify the identity above without too many calculations , note that the 5 maximal subdeterminants of the @xmath287-block with @xmath97 s are , consecutively , @xmath288 , and similarly for @xmath289 . the matrix obtained from @xmath11 by deleting the column corresponding to @xmath290 and the row corresponding to @xmath291 has determinant @xmath292 . this example extends to a uniform determinantal representation of size @xmath32 for the generic bivariate polynomial @xmath9 of degree @xmath1 . we get @xmath293 , where @xmath294 @xmath295 and @xmath296 are @xmath297 matrices with 1 on the first upper diagonal and @xmath298 and @xmath299 respectively on the main diagonal , while @xmath300 is a @xmath301 triangular matrix such that @xmath302 for @xmath303 and 0 otherwise . note that we will slightly improve on the size @xmath32 in example [ ex:2d1 ] . @xmath27 example [ ex:2d ] generalises as follows . [ prop : cons2 ] let @xmath304 $ ] be subspaces connected to @xmath199 such that @xmath305 . then there exists a uniform determinantal representation of the generic @xmath0-variate polynomial of degree @xmath1 of size @xmath306 set @xmath307 and @xmath308 . consider the matrix @xmath309 with @xmath215 and @xmath310 the matrices of sizes @xmath311 and @xmath312 from lemma [ lm : block ] , and where @xmath313 is an @xmath314-matrix to be determined . note that the determinant of @xmath11 is linear in the entries of @xmath300 . indeed , setting @xmath315 yields the singular matrix @xmath59 , so @xmath96 contains no terms of degree 0 in the entries of @xmath300 . furthermore , deleting from @xmath11 two or more of the first @xmath316 columns from @xmath215 , we end up with a matrix that is singular since , when acting on rows , it maps the span of @xmath317 into a space of dimension at most @xmath318 , so @xmath96 does not contain terms that are of degree @xmath319 in the entries of @xmath300 . hence the determinant equals @xmath320 where the @xmath223 are the maximal subdeterminants of @xmath215 and the @xmath321 are the maximal subdeterminants of @xmath310 . by lemma [ lm : block ] we have @xmath322 and @xmath323 . hence the assumption that @xmath305 ensures that we can choose the @xmath214 in such a manner that the determinant of @xmath11 equals the generic polynomial @xmath9 . [ ex:2d1 ] example [ ex:2d ] can be slightly improved to a representation of size @xmath33 by taking @xmath324 and @xmath325 ; note that , indeed , @xmath326 . a representation of size @xmath33 for the polynomial @xmath9 from is @xmath327 we do not know whether the factor 2 can be improved . @xmath27 a representation of the form can also be obtained from the linearisations based on dual basis from @xcite . there , linearisations of a univariate polynomial are presented that use the basis of the form @xmath328 , where @xmath329 and @xmath330 are polynomials . if we use the same approach for a bivariate polynomial with the standard basis @xmath331 and @xmath332 , we get a representation of the form up to permutations of rows and columns . we will now prove our main theorem . for the lower bound on @xmath29 , let @xmath61 be a size-@xmath15 uniform determinantal representation of the generic @xmath0-variate polynomial of degree @xmath1 . let @xmath108 be the @xmath275-determinant of the submatrix of @xmath59 obtained by deleting the entry at position @xmath169 . then the image of the linear map @xmath333,\ ( \ell_{ij})_{i , j } \mapsto \sum_{i , j } ( -1)^{i+j } \ell_{ij } d_{ij}\ ] ] contains @xmath49 ( see lemma [ lm : dets ] ) . we claim that @xmath164 has a kernel of dimension at least @xmath334 . indeed , fix any row index @xmath335 . if the @xmath336 are all zero , then we obtain an @xmath337-dimensional subspace of @xmath338 by setting all @xmath339 with @xmath340 equal to zero and choosing the @xmath341 arbitrarily . if they are not all zero , then the @xmath94 rows with indices @xmath342 are linearly independent over @xmath220 and hence _ a fortiori _ over @xmath8 . for each such @xmath343 define @xmath344 by @xmath345 then , by cramer s rule , we have @xmath346 , and these @xmath94 vectors are linearly independent . hence , for each @xmath335 we find a subspace of @xmath338 of dimension at least @xmath94 , and these subspaces are linearly independent . thus we find that @xmath347 from which the existence of @xmath348 follows . in the proof of the lower bound we have been a bit more careful than strictly needed : without the discussion of the kernel it follows that @xmath349 . but one derives a better constant ( for @xmath350 ) by using the kernel . for the upper bound , we first give a simple construction for even @xmath0 . for odd @xmath0 , a trickier analysis is needed ( which also applies in the even case ) ; see below . assume that @xmath351 with @xmath352 . let @xmath129 be the space of polynomials in @xmath353 of degree at most @xmath1 , and let @xmath354 be the space of polynomials in @xmath355 of degree at most @xmath1 . then @xmath129 and @xmath354 are connected to @xmath199 and we have @xmath356 , so that by proposition [ prop : cons2 ] we have @xmath357 . now compute @xmath358 this implies the existence of @xmath359 for even @xmath0 . we now give a construction that works for all @xmath360 , for which we thank aart blokhuis . for @xmath361 let @xmath362 denote the set of nonnegative integers that can be expressed as @xmath363 with @xmath364 , i.e. , whose binary expansions have ones only at even positions ( for @xmath365 , counting the least significant bit as zeroeth position ) or only at odd positions ( for @xmath366 ) . observe that @xmath367 and that both @xmath368 and @xmath369 contain roughly @xmath370 of the first @xmath1 nonnegative integers for every @xmath1they have `` dimension @xmath371 '' . now set @xmath372 for @xmath361 , so that @xmath373 and the number of elements of @xmath20 intersected with a large box @xmath374^n$ ] is roughly @xmath45 . unfortunately , @xmath19 and @xmath375 are not connected to @xmath207 . however , we can connect them to 0 as follows . for a lattice point @xmath376 let @xmath377 be the minimum among the 2-adic valuations of its entries , attained , say , by @xmath378 . then set @xmath379 . setting the remaining coordinates of @xmath380 equal to those of @xmath53 we have @xmath381 and @xmath382 we propose to add to @xmath20 the sequence @xmath383 to connect @xmath53 to @xmath380 . we need to verify , however , that in doing this , @xmath20 retains dimension @xmath384 . the fraction of @xmath53 in ( a large box intersected with ) @xmath20 for which the minimal valuation is at least @xmath377 equals roughly @xmath385after all , the condition is that for each @xmath386 , @xmath378 has zeros on the first @xmath387 positions where it is allowed to have ones . write @xmath388 . thus by adding the sequences above , the total increase of @xmath20 is by a factor of at most @xmath389 this is a convergent series as @xmath360 , and hence @xmath20 retains dimension @xmath384 . we have thus constructed subsets @xmath390 that satisfy @xmath373 , are connected to @xmath199 , and that contain roughly a constant times @xmath45 points in each box @xmath374^n$ ] . for @xmath361 let @xmath391 be the space spanned by the monomials whose exponent vectors lie in @xmath392^n$ ] . then @xmath393 and @xmath394 are connected to @xmath199 , so by proposition [ prop : cons2 ] there exists a uniform determinantal representation for @xmath51 of size @xmath395 , as desired . the construction in the proof is by no means tight . for example , one could also replace @xmath378 by @xmath396 , which yields a shorter sequence to be added ; and for large @xmath377 we have also counted additional , shorter sequences since they also have valuation larger than numbers smaller than @xmath377 . we think that for even @xmath0 the previous construction , subdividing the variables into two sets of equal size , may lead to a better constant , but we have not verified this . carrying out the construction in the proof for @xmath19 with @xmath249 , always choosing for @xmath110 the smallest index of a coordinate of @xmath53 with minimal valuation , we arrive at the following fractal - like structure ( the circles indicate the points of @xmath19 , the black edges show that @xmath19 is connected to @xmath207 ) : in this section we give several uniform representations of to our knowledge the smallest possible size for cases where @xmath0 and @xmath1 are small . we start with the two cases where we can compute @xmath29 exactly . @xmath397 . taking @xmath398 in proposition [ prop : cons1 ] we see that @xmath399 ; this is the representation of example [ ex : binaryquadric ] . suppose that a uniform determinantal representation @xmath61 of size @xmath132 exists . then , by example [ ex : smallsingspace ] , after acting with @xmath400 and transposing if necessary , we may assume that the singular space represented by @xmath59 is a compression space with a one - dimensional witness . but then reads @xmath401 a contradiction . hence @xmath397 . @xmath402 . taking @xmath403 in proposition [ prop : cons1 ] we see that @xmath404 . suppose that a uniform representation of size @xmath24 exists . up to transposition , there are three possibilities for the singular space @xmath126 represented by @xmath59 ; see example [ ex : smallsingspace ] ( where the third compression space is conjugate to the transpose of the first ) : 1 . assume that @xmath126 is a compression space with a one - dimensional witness , so that after acting with @xmath405 we have @xmath406 let @xmath223 denote the determinant of the minor of @xmath59 obtained by deleting the first row and @xmath110th column . then the linear map @xmath407,\ ( l_1,l_2,l_3 ) \mapsto l_1d_1-l_2d_2+l_3d_3\ ] ] has @xmath408 . now inequality reads @xmath409 which holds with equality . this means that , in fact , @xmath410 must _ equal _ @xmath411 . in particular , @xmath412 must all be of degree one ( or else @xmath410 would contain cubic polynomials ) . the image of @xmath413 depends only on the span @xmath414 . if @xmath415 , then there exists an affine transformation in @xmath416 that maps @xmath129 into a subspace of @xmath417 . then @xmath418 , a contradiction . if @xmath419 , then after an affine transformation we find @xmath420 . in that case , @xmath421 , another contradiction . 2 . assume that @xmath126 is a compression space with a two - dimensional witness , so that after row and column operations we have @xmath422 where @xmath423 . write @xmath424 . using that @xmath96 is assumed to be linear in the @xmath7s , we find that @xmath425 consequently , setting @xmath426 and @xmath427 , we have @xmath428 . if @xmath429 , then by acting with a suitable element of @xmath416 we achieve that @xmath430 . but then @xmath431 . the same applies when @xmath432 . on the other hand , if @xmath433 , then by an element in @xmath416 we achieve that @xmath434 . in that case , 3 . finally , assume that @xmath126 is conjugate to a space of skew - symmetric matrices , so that after conjugation @xmath436\ ] ] where @xmath437 . set @xmath438 . then the space spanned by the @xmath439-determinants of @xmath59 is @xmath440 , of dimension at most @xmath441 . moreover , we have @xmath442 . if @xmath415 , then by acting with @xmath416 we achieve that @xmath443 , and hence @xmath444 . if , on the other hand , @xmath419 , then we achieve that @xmath420 , and @xmath445 . in each of these cases we arrive at a contradiction . consequently , @xmath402 as claimed . the proofs above use the classification of spaces of small singular matrices in an essential manner , as well as the action of @xmath69 on uniform determinantal representations . we conjecture that @xmath446 , and that this can still be proved in the same manner , using the classification of @xmath447-singular matrix spaces from @xcite . but as theorem [ thm : sing5 ] shows , fundamentally new ideas will be needed to prove lower bounds in larger situations . for some pairs of small @xmath0 and @xmath1 we now give the smallest uniform representations that we have been able to find . for the constructions we use proposition [ prop : cons2 ] with subspaces @xmath448 $ ] spanned by the monomials and connected to @xmath199 . first , we give in table [ tab : minnd ] the minimal sizes known to us of uniform determinantal representations for some small values of @xmath0 and @xmath1 . c|ccccccccc ' '' '' @xmath0 & @xmath449 & @xmath450 & @xmath283 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 + ' '' '' 2 & 3 & 15 & 17 & 19 & 11 & 13 & 15 & 17 + 3 & 4 & 17 & 10 & 14 & 18 & 22 & 27 & 34 + 4 & 5 & 19 & 14 & 19 & 26 & 34 & 44 + 5 & 6 & 11 & 18 & 26 & + 6 & 7 & 13 & 22 & 33 & + 7 & 8 & 15 & 27 & 39 & + 8 & 9 & 17 & 32 & + the corresponding representations for the entries in table [ tab : minnd ] for @xmath17 , which are of size @xmath33 , are given in example [ ex:2d1 ] . for @xmath449 we take @xmath456 and @xmath457 , therefore @xmath458 , while for @xmath450 we can take @xmath459 , and hence @xmath460 . in table [ tab : wv ] we give sets @xmath129 and @xmath354 for the remaining nonzero entries in table [ tab : minnd ] . the subspaces @xmath129 and @xmath354 have the form @xmath461 and @xmath462 , where @xmath463 w_0=\langle1,x_1,\ldots , x_n,\ldots , x_1^f,\ldots , x_n^f\rangle \end{matrix}\ ] ] for @xmath464 and @xmath465 , which yields @xmath466 . for clarity and brevity , the variables @xmath467 in table [ tab : wv ] stand for @xmath468 , respectively . ll|ll ' '' '' @xmath0 & @xmath1 & @xmath469 & @xmath470 + ' '' '' 3 & 4 & @xmath471 & @xmath471 + 3 & 5 & @xmath471 & @xmath472 + 3 & 6 & @xmath471 & @xmath473 + 3 & 7 & @xmath471 & @xmath474 + 3 & 8 & @xmath471 & @xmath475 + 3 & 9 & @xmath476 & @xmath477 + 4 & 4 & @xmath471 & @xmath472 + 4 & 5 & @xmath471 & @xmath478 + 4 & 6 & @xmath479 & @xmath480 + 4 & 7 & @xmath481 & @xmath482 + 4 & 8 & @xmath483 & @xmath484 + 5 & 4 & @xmath471 & @xmath485 + 5 & 5 & @xmath486 & @xmath487 + 6 & 4 & @xmath471 & @xmath488 + 6 & 5 & @xmath489 & @xmath490 + 7 & 4 & @xmath471 & @xmath491 + 7 & 5 & @xmath492 & @xmath493 + 8 & 4 & @xmath471 & @xmath494 + [ ex : turan ] to show how things get complicated , let us consider the construction for @xmath283 . we take @xmath495 and @xmath496 where @xmath497 and @xmath22 is as small as possible . if we take all possible pairs @xmath498 , then clearly @xmath499 , while on the other hand , when @xmath500 , @xmath501 does not contain any monomials of the form @xmath502 for @xmath503 . we need a minimal set of @xmath504 to cover all possible monomials , which is related to the following covering problem . given positive integers @xmath505 , we say that a system @xmath204 of @xmath118-subsets ( called blocks ) of @xmath506 is called a turn @xmath507-system if every @xmath508-subset of @xmath506 contains at least one block from @xmath204 @xcite . the minimum size of @xmath204 is called the turn number @xmath509 . in our case , additional terms @xmath510 form a turn @xmath511-system . while for most cases only upper and lower bounds for @xmath509 are known , turn proved that @xmath512 where @xmath513 . to obtain the minimal set one has to divide @xmath514 into three nearly equal groups ( their sizes do not differ for more than one ) and then take all pairs @xmath504 such that @xmath53 and @xmath515 belong to the same group . as a result , such construction gives a uniform representation of size @xmath15 , where @xmath516 , which therefore implies @xmath517 . @xmath27 suppose that we have a uniform representation @xmath11 of @xmath51 as in , and write @xmath518 where each @xmath519 . now consider the matrix polynomial ( cf . ) @xmath520 where @xmath521 is a @xmath522 matrix . we will show that under certain assumptions we can construct from @xmath11 a matrix @xmath523 that represents @xmath524 in the sense that @xmath525 . we obtain @xmath526 from @xmath11 in the following way . each element of the form @xmath527 is replaced by the @xmath522 matrix @xmath528 , where @xmath529 is the @xmath530 identity , and each @xmath7 is replaced by the matrix @xmath521 . [ thm : matpollin ] let be a uniform representation of the generic polynomial of degree @xmath1 in @xmath0 variables and assume that there exist matrices @xmath531 and @xmath149 , whose elements are polynomials in @xmath16 , such that @xmath532 , and @xmath533 is a triangular matrix with one diagonal element equal to @xmath51 and all other diagonal elements equal to 1 . then @xmath534 is a representation for the matrix polynomial @xmath524 , i.e. , @xmath525 . it is easy to see that @xmath535 is a block triangular matrix with one diagonal block @xmath524 while all other diagonal blocks are equal to @xmath529 . since @xmath536 , it follows that @xmath525 . theorem [ thm : matpollin ] applies to the uniform representation . indeed , take @xmath537 then @xmath538 it is easy to see that there exist permutation matrices @xmath539 and @xmath540 such that @xmath541 is triangular and has the diagonal which satisfies theorem [ thm : matpollin ] . therefore , we can apply for matrix polynomials by using block matrices . this can be generalised to a uniform representation of size @xmath32 of the form . in a similar way we can show that this also holds for representations of the form of size @xmath33 . @xmath27 unfortunately , not all uniform determinantal representations induce a determinantal representation of a general matrix polynomial in this manner . as a counterexample , let @xmath11 be such a uniform determinantal representation of the polynomial @xmath51 , @xmath542 , and construct a representation of larger size @xmath543 . \ ] ] then @xmath544 , but @xmath545 is not a representation for the matrix polynomial @xmath524 as the coefficient matrices @xmath546 and @xmath547 do not commute in general . this motivates the following definition . a uniform determinantal representation @xmath11 is _ minimal _ if there do not exist constant matrices @xmath548 and @xmath149 such that @xmath549 and @xmath550 we speculate that each minimal uniform representation gives rise to a representation for a matrix polynomial . recently , a new numerical approach for computing roots of systems of bivariate polynomials was proposed in @xcite . the main idea is to treat the system as a two - parameter eigenvalue problem using determinantal representations . suppose that we are looking for roots of a system of bivariate polynomials @xmath551 q=\sum_{i+j \le d_2 } \beta_{ij } x^i y^j=0,\end{matrix}\ ] ] where @xmath9 and @xmath112 are polynomials of degree @xmath552 and @xmath553 over @xmath554 . let @xmath555 and @xmath556 , where @xmath557 and @xmath558 , with @xmath559 and @xmath560 , be determinantal representations of @xmath9 and @xmath112 , respectively . then a root @xmath561 of is an eigenvalue of the two - parameter eigenvalue problem @xmath562 ( b_0+x b_1+y b_2 ) \ , v=0 , \end{matrix}\ ] ] where @xmath563 and @xmath564 are nonzero vectors . the standard way to solve is to consider a joint pair of generalized eigenvalue problems @xcite @xmath565 ( \delta_2-y \delta_0 ) \ , w=0 , \end{matrix}\ ] ] where @xmath566 and @xmath567 . in this particular application we can expect that the pencils in are singular , i.e. , @xmath568 and @xmath569 . namely , by bzout s theorem a generic system has @xmath570 solutions , while a generic problem has @xmath571 eigenvalues . unless @xmath572 , both pencils in are singular . in this case we first have to apply the staircase algorithm from @xcite to extract the finite regular eigenvalues . the method returns smaller matrices @xmath573 , @xmath574 , and @xmath575 ( of size @xmath576 for a generic ) such that @xmath573 is nonsingular and @xmath577 and @xmath578 commute . from @xmath579 ( \widetilde\delta_2-y \widetilde\delta_0 ) \ , \widetilde w=0 , \end{matrix}\ ] ] we compute the eigenvalues @xmath561 using a variant of the qz algorithm @xcite and thus obtain the roots of . the above approach is implemented in the matlab package biroots @xcite together with the two determinantal representations from @xcite . the first one , to which we refer as lin1 , is a uniform one from example [ ex : lin1 ] of size @xmath580 for a polynomial of degree @xmath1 . the second one , which we refer to as lin2 , is not uniform and involves some computation to obtain a smaller size @xmath581 . although the construction of lin2 is more time consuming , this pays off later , when the staircase algorithm is applied to . table [ tbl : compare1 ] shows the sizes of determinantal representations for polynomials of small degree . as expected , the new uniform determinantal representation of size @xmath33 , to which we refer as minunif , returns smaller matrices , which reflects later in faster computational times . it is also important that lin1 and minunif return real matrices for polynomials with real coefficients , which is not true for lin2 . l|cccccccccc ' '' '' method & @xmath450 & @xmath283 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 & @xmath582 & @xmath583 & @xmath584 + ' '' '' lin1 & 5 & 8 & 11 & 15 & 19 & 24 & 29 & 35 & 41 & 48 + lin2 & 3 & 5 & 18 & 10 & 13 & 17 & 20 & 24 & 29 & 34 + minunif & 5 & 7 & 19 & 11 & 13 & 15 & 17 & 19 & 21 & 23 + it was reported in @xcite that the determinantal representation approach for solving systems of bivariate polynomials is competitive for polynomials of degree 9 or less . as we show below , the new uniform representation minunif extends this to degree 15 and , in addition , performs better than the existing representations for polynomials of degree 6 or more . in @xcite the approach was compared numerically to the following state - of - the art numerical methods for polynomial systems : nsolve in mathematica 9 @xcite , bertinilab 1.4 @xcite running bertini 1.5 @xcite , naclab 3.0 @xcite , and phclab 1.04 @xcite running phcpack 2.3.84 , which turned out as the fastest of these methods . to show the improved performance of the new determinantal representation , we compare minunif to lin1 , lin2 , and phclab in table [ tbl : two ] . for each @xmath1 we run the methods on the same set of 50 real and 50 complex random polynomial systems of degree @xmath1 and measure the average time . for lin1 and minunif , where determinantal representations have real matrices for real polynomials , we report separate results for polynomials with real and complex coefficients . the timings for lin1 and lin2 are given only for @xmath585 as for larger @xmath0 these two linearisations are no longer competitive . r|rrrrrr ' '' '' @xmath1 & lin1 ( @xmath586 ) & lin1 ( @xmath554 ) & lin2 & phclab & minunif ( @xmath586 ) & minunif ( @xmath554 ) + ' '' '' 3 & 1116 & 1118 & 114 & 1116 & 116 & 117 + 4 & 1119 & 1111 & 116 & 1130 & 112 & 1113 + 5 & 1120 & 1126 & 113 & 1151 & 118 & 1120 + 6 & 1139 & 1171 & 128 & 1174 & 127 & 1127 + 7 & 1196 & 1160 & 151 & 1217 & 136 & 1144 + 8 & 1205 & 1395 & 118 & 1264 & 159 & 1174 + 9 & 1467 & 1124 & 279 & 1329 & 195 & 1125 + 10 & 1424 & 3412 & 600 & 1414 & 147 & 1221 + 11 & & & & 1538 & 248 & 1354 + 12 & & & & 1650 & 361 & 1530 + 13 & & & & 1911 & 592 & 1740 + 14 & & & & 1142 & 842 & 1148 + 15 & & & & 1531 & 1237 & 1835 + of course , the computational time is not the only important factor , we also have to consider the accuracy and reliability . in each step of the staircase algorithm a rank of a matrix has to be estimated numerically , which is a delicate task . after several steps it may happen that the gap between the important singular values and the meaningful ones that should be zero in exact computation , virtually disappears . in such case the algorithm fails and does not return any roots . as the number of steps in the staircase algorithm increases with degree of the polynomials , such problems occur more often for polynomials of large degree . a heuristic that usually helps in such cases is to apply the algorithm on a transformed system @xmath587 for random @xmath588 and @xmath589 such that @xmath590 . as this transformation does not change the conditioning of the roots , we can conclude that the difficulties with the staircase algorithm are not directly related to the conditioning . the trick does not work every time , and it seems that for some systems the only way to make the determinantal representation approach to work is to increase the machine precision . we can apply the same approach to systems of polynomials in more than two variables . however , since the size of the corresponding @xmath591 matrices is the product of sizes of all representations , this is competitive only for @xmath249 and @xmath592 . for a comparison , if we have a system of three polynomials in three variables of degree @xmath91 , then the size of the @xmath591 matrices is @xmath593 . for degree the size increases to @xmath595 and phcpack is faster . finally , for @xmath596 and the smallest nontrivial @xmath449 we already get @xmath591 matrices of size @xmath597 and the method is not efficient . we have introduced _ uniform determinantal representations _ , which rather than representing a single polynomial as the determinant of a matrix of affine - linear forms , represent all polynomials of degree at most @xmath1 in @xmath0 variables as such a determinant . we have seen that in the bivariate case , these determinantal representations are useful for numerically solving bivariate systems of equations ; and in the general multivariate case we have determined , up to constants , the asymptotic behaviour of @xmath29 , the minimal size of such a representation , for @xmath0 fixed and @xmath30 . * for fixed @xmath0 and @xmath30 , @xmath598 , see theorem [ thm : main ] . this is a noticeable improvement on @xcite , where an asymptotic rate of @xmath599 is shown , with the remark that the representation in @xcite are symmetric . however , symmetry currently can not be exploited by methods that compute roots of multivariate polynomial systems . * for fixed odd @xmath1 and @xmath41 , @xmath600 , which is the same rate as quarez @xcite , who also manages to get symmetric representations . for fixed even @xmath1 and @xmath41 , @xmath601 , which again is the same rate as in @xcite . however , we have a slightly smaller lower bound for the asymptotic rate of @xmath602 . * tables [ tab : minnd ] and [ tab : wv ] give constructions for the smallest representations that we have been able to find for some small values of @xmath0 and @xmath1 . * @xmath603 ; cf . table [ tab : minnd ] . * @xmath604 ; cf . table [ tab : minnd ] . * @xmath605 ; see example [ ex : turan ] . * @xmath606 ; cf . table [ tab : minnd ] . note that this result satisfies dixon @xcite up to an asymptotic factor 2 whereby no computations are necessary for the determinantal representation . in particular , it is a major improvement on the @xmath607 of @xcite . * due to the smaller sizes of the representations , the numerical approach for bivariate polynomials ( @xmath17 ) is competitive to ( say ) mathematica for degree @xmath1 up to @xmath608 ( see section [ sec : numerics ] ) ; this in contrast to @xmath609 as obtained in @xcite . * under some conditions , the results carry over to the case of matrix coefficients ( see section [ sec : matpol ] ) . there are still many interesting open questions , both of intrinsic mathematical interest and of relevance to polynomial system solving . first of all , in a situation where the degree @xmath1 is fixed and the number @xmath0 of variables grows , what is the asymptotic behaviour of @xmath29 ? although we expect that the above inequalities are equalities , whether this holds is still open . for general fixed @xmath1 , the proof of theorem [ thm : main ] yields a lower bound which is a constant ( depending on @xmath1 ) times @xmath602 . for odd @xmath1 we obtain a matching upper bound ( with a different constant ) by using proposition [ prop : cons2 ] with @xmath610 . however , for even @xmath1 we only know how to obtain @xmath611 . we remark that this latter is ( up to a constant ) the same bound as obtained in ( * ? ? ? 4.4 ) for _ symmetric _ uniform representations . second , in the case of fixed @xmath0 and varying @xmath1 studied in this paper , what are the best constants in theorem [ thm : main ] ? more specifically , for fixed @xmath0 , does @xmath612 exist , and if so , what is its value ? third , how can our techniques for upper bounds and lower bounds be further sharpened ? can singular matrix spaces _ other _ than compression spaces be used to obtain tighter upper bounds ( constructions ) on @xmath29 ? can the action of the affine group be used more systematically to find lower bounds on @xmath29 ? finally , we have restricted our attention to matrices that , apart from being affine - linear in @xmath613 , are also affine - linear in the coefficients @xmath7 . our proofs give the same asymptotic behaviour ( with different constants ) if we require , in addition , that no quadratic terms @xmath614 may occur in @xmath11 . if , instead , we relax the condition that @xmath11 be affine - linear in the @xmath7 to a _ polynomial _ dependence on the @xmath7 , then the same bounds still apply ; see remark [ re : poly ] . but what if we relax this to _ rational _ dependence of @xmath11 on the @xmath7 ? given that @xmath51 is only linear in the @xmath7 it seems unlikely that allowing @xmath11 to be rational in the @xmath7 we would gain anything , but we currently do not know how to formalise this intuition . on the other hand , in cases where a ( non - uniform ) determinantal representation of size @xmath1 is known to exist for every ( or sufficiently general ) polynomials of degree @xmath1 in @xmath1 variables ( e.g. , in the case of plane curves ) , it follows that this representation can be chosen to have entries _ algebraic _ in the @xmath7 . this observation rules out approaches aimed at proving lower bounds in a too general setting . , phclab : a matlab / octave interface to phcpack . in : m. stillman , j. verschelde , and n. takayama ( eds ) , _ software for algebraic geometry _ , volume 148 of the i m a volumes in mathematics and its applications , springer , new york ( 2008 ) , 1532 .
the subject of graph pebbling has seen dramatic growth recently , both in the number of publications and in the breadth of variations and applications . here we update the reader on the many developments that have occurred since the original _ survey of graph pebbling _ in 1999 . c * v * p s # 1#2
math0509339
since the publication of the graph pebbling survey @xcite there has been a great deal of activity in the subject . there are now over 50 papers by roughly 80 authors in the field ; the web page @xcite maintains a current list of these papers . many researchers have asked for an updated survey thanks to the new york academy of sciences for providing the opportunity . the following is based on the talk , `` _ everything you always wanted to know about graph pebbling but were afraid to ask _ '' , an obvious ripoff / homage to one of new york s favorite directors . we begin by introducing relevant terminology and background on the subject . here , the term _ graph _ refers to a simple graph without loops or multiple edges . for the definitions of other graph theoretical terms see any standard graph theory text such as @xcite . the pebbling number of a disconnected graph will be seen to be undefined . henceforth we will assume all graphs to be connected . a configuration @xmath0 of pebbles on a graph @xmath1 can be thought of as a function @xmath2 . the value @xmath3 equals the number of pebbles placed at vertex @xmath4 , and the _ size _ of the configuration is the number @xmath5 of pebbles placed in total on @xmath6 . a pebbling step along an edge from @xmath7 to @xmath4 reduces by 2 the number of pebbles at @xmath7 and increases by 1 the number of pebbles at @xmath4 . we say that a vertex @xmath8 can be _ reached _ by @xmath0 if one can repeatedly apply pebbling steps so that , in the resulting configuration @xmath9 , we have @xmath10 ( and @xmath11 for all @xmath4 ) . the _ pebbling number _ , @xmath12 , is defined to be the smallest integer @xmath13 so that any specified _ root _ vertex @xmath14 of @xmath6 can be reached by every configuration @xmath0 of size @xmath13 . a configuration that reaches every vertex is called _ solvable _ , and _ unsolvable _ otherwise . the origins of graph pebbling reside in combinatorial number theory and group theory . a sequence of elements of a finite group @xmath15 is called a _ zero - sum sequence _ if it sums to the identity of @xmath15 . a simple pigeonhole argument ( on the sequence of partial sums ) proves the following theorem . [ origin ] any sequence of @xmath16 elements of a finite group @xmath15 contains a zero - sum subsequence . in fact , a subsequence of consecutive terms can be guaranteed by the pigeonhole argument . furthermore , one can instead stipulate that the zero - sum subsequence has at most @xmath17 terms , where @xmath18 is the exponent of @xmath15 ( i.e. the maximum order of an element of @xmath15 ) , and this is best possible . initiated in 1956 by erds @xcite , the study of zero - sum sequences has a long history with many important applications in number theory and group theory . in 1961 erds et al . @xcite proved that every sequence of @xmath19 elements of a cyclic group @xmath15 contains a zero - sum subsequence of length exactly @xmath16 . in 1969 van emde boas and kruyswijk @xcite proved that any sequence of @xmath20 elements of a finite abelian group contains a zero - sum sequence . in 1994 alford et al . @xcite used this result and modified erds s arguments to prove that there are infinitely many carmichael numbers . much of the recent study has involved finding davenport s constant @xmath21 , defined to be the smallest @xmath22 such that every sequence of @xmath22 elements contains a zero - sum subsequence @xcite . there are a wealth of results on this problem @xcite and its variations @xcite , as well as applications to factorization theory @xcite and to graph theory @xcite . in 1989 kleitman and lemke @xcite , and independently chung @xcite , proved the following theorem ( originally stated number - theoretically ) , strengthening theorem [ origin ] . let @xmath23 denote the cyclic group of order @xmath24 , and let @xmath25 denote the order of an element @xmath26 in the group to which it belongs . [ cyclic ] for every sequence @xmath27 of @xmath24 elements from @xmath23 there is a zero - sum subsequence @xmath28 such that @xmath29 . for a sequence @xmath30 the sum @xmath31 is known as the _ cross number _ of @xmath30 and is an important invariant in factorization theory . guaranteeing cross number at most 1 strengthens the extension of theorem [ origin ] that @xmath32 , and shows that equality holds if and only if every @xmath33 . the concept of pebbling in graphs arose from an attempt by lagarias and saks to give an alternative ( and more natural and structural ) proof than that of kleitman and lemke ; it was chung who carried out their idea . see also @xcite for another extension of this result . kleitman and lemke then conjectured that theorem [ cyclic ] holds for all finite groups . for a subgroup @xmath34 of @xmath15 , call a sequence of elements of @xmath15 an @xmath34-_sum sequence _ if its elements sum to an element of @xmath34 . in @xcite is proved the following theorem ( the methods of @xcite use graph pebbling ) . [ subgroupconj ] let @xmath34 be a subgroup of a finite abelian group @xmath15 with @xmath35 . for every sequence @xmath27 of @xmath24 elements from @xmath15 there is an @xmath34-sum subsequence @xmath28 such that @xmath36 . the case @xmath37 here gives theorem [ cyclic ] for finite abelian groups , strengthening the van emde boas and kruyswijk result @xcite . kleitman and lemke also conjectured that theorem [ subgroupconj ] holds for all finite groups , and verified their conjecture for all dihedral groups ( see @xcite ) . for other nonabelian groups , it has been shown recently to hold for the nonabelian solvable group of order 21 ( see @xcite ) . it would be interesting to see whether graph pebbling methods can shed light on the davenport constant for finite abelian groups . in this regard , one of the most pressing questions is as follows . write @xmath38 , where @xmath39 . then @xmath40 is the _ rank _ of @xmath15 . it is natural to guess that @xmath41 from pigeonhole intuition . this was conjectured in @xcite and is true by theorem [ cyclic ] for rank 1 groups . moreover , it was proven in @xcite for rank 2 groups and @xmath42-groups as well . however , it was proven in @xcite that the conjecture is false for some groups each rank at least 4 . what remains open is the instance of rank 3 . [ davenport ] if @xmath15 is a finite abelian group of rank @xmath43 then its davenport constant satisfies @xmath44 . there are many known results regarding @xmath12 . if one pebble is placed at each vertex other than the root vertex , @xmath14 , then no pebble can be moved to @xmath14 . also , if @xmath8 is at distance @xmath45 from @xmath14 , and @xmath46 pebbles are placed at @xmath8 , then no pebble can be moved to @xmath14 . thus we have that @xmath47 . graphs @xmath6 that satisfy @xmath48 are known as _ class 0 _ graphs , which include the complete graph @xmath49 , the @xmath45-dimensional cube @xmath50 @xcite , complete bipartite graphs @xmath51 @xcite , and many others . we will say more about such graphs in section [ dcc ] . any graph @xmath6 with a cut vertex @xmath52 has @xmath53 . ( indeed , let @xmath54 and @xmath55 , where @xmath56 and @xmath57 are two components of @xmath58 . define the configuration @xmath0 by @xmath59 , @xmath60 and @xmath61 for every other vertex @xmath8 . then @xmath62 and @xmath0 can not reach @xmath4 . ) the path @xmath63 , the cube @xmath50 @xcite , the petersen graph @xmath64 @xcite , the even cycle @xmath65 @xcite , and the line graph @xmath66 of the complete graph @xmath49 @xcite are examples of graphs @xmath6 that satisfy @xmath67 , while the odd cycle @xmath68 @xcite is an example of a graph not satisfying either lower bound . another standard result is the pebbling number of a tree , which is worked out in @xcite . regarding upper bounds , it follows immediately from the pigeonhole principle that a graph @xmath6 on @xmath24 vertices with diameter @xmath45 has pebbling number @xmath69 . it would be interesting to find better general bounds on @xmath12 , especially not involving @xmath24 . the independence number seems not to be useful . for example , there is no function @xmath26 such that every graph @xmath6 of independence number @xmath70 and diameter @xmath45 has pebbling number @xmath71 . indeed , we define a family of graphs @xmath72 which satisfy @xmath73 and @xmath74 , but which have pebbling number @xmath75 as @xmath76 . define @xmath77 , where @xmath78 and the edge set @xmath79 . since @xmath52 is a cut vertex we know by the above comment that @xmath80 . however , the domination number ( minimum size of a dominating set ) can be useful . chan and godbole @xcite made the following improvements on the general upper bound . [ betterupper ] let @xmath81 denote the domination number of @xmath6 . then 1 . @xmath82 , 2 . @xmath83 , and 3 . @xmath84 . the inequalities in parts 1 and 2 are sharp , and the coefficient of @xmath85 in part 3 can be reduced to @xmath86 in the case of perfect domination . for any two graphs @xmath56 and @xmath57 , we define the _ cartesian product _ @xmath87 to be the graph with vertex set @xmath88 and edge set @xmath89 and @xmath90 or @xmath91 and @xmath92}. the cube @xmath50 can be built recursively from the cartesian product , and chung s result @xcite that @xmath93 ( and more generally theorem [ pathprod ] below ) would follow easily from graham s conjecture , which has generated a great deal of interest . ( graham)[graham ] @xmath94 . it is worth mentioning that there are some results which verify graham s conjecture . among these , the conjecture holds for a tree by a tree @xcite , a cycle by a cycle @xcite , and a clique by a graph with the @xmath85-pebbling property @xcite ( see below ) . recently , feng and kim verified the conjecture for complete bipartite graphs @xcite and for wheels or fans @xcite . it is also proven in @xcite that the conjecture holds when each @xmath95 is a path . let @xmath63 be a path with @xmath24 vertices and for @xmath96 let @xmath97 denote the graph @xmath98 . [ pathprod ] for nonnegative integers @xmath99 , @xmath100 . the conjecture was also verified recently @xcite for graphs of high minimum degree , using theorem [ conn ] . [ mindeg ] if @xmath56 and @xmath57 are connected graphs on @xmath24 vertices that satisfy @xmath101 and @xmath102 , then @xmath94 . in particular , there is a constant @xmath103 so that if @xmath104 then @xmath105 is class 0 . we will present probabilistic versions of conjecture [ graham ] in sections [ graphthresh ] and [ pebblingthresh ] . a graph @xmath6 is said to have the _ 2-pebbling property _ if two pebbles can be moved to any specified vertex when the initial configuration @xmath0 has size @xmath106 , where @xmath107 is the number of vertices @xmath4 with @xmath108 . this property is crucial in the proof of theorem [ pathprod ] . until recently , only one graph @xmath109 ( see figure [ lemke ] ) , due to lemke , was known not to have the 2-pebbling property , although a family of related graphs were conjectured in @xcite not to have the property either . although that conjecture is still unresolved , wang @xcite proved that a slight modification of snevily s graphs yield an infinite family of graphs , none of which have the 2-pebbling property . also found in @xcite is the conjecture that all bipartite graphs have the 2-pebbling property . it is possible that the square of the lemke graph might be a counterexample to graham s conjecture . find @xmath110 . in order to describe the generalization employed by chung to prove the kleitman - lemke theorem , we need to introduce generalized pebbling . a _ @xmath111-pebbling step _ in @xmath6 consists of removing @xmath111 pebbles from a vertex @xmath7 , and placing one pebble on a neighbor @xmath4 of @xmath7 . we say that a pebbling step from @xmath7 to @xmath4 is _ greedy _ ( _ semigreedy _ ) if @xmath112 ( @xmath113 ) , where @xmath14 is the root vertex , and that a graph @xmath6 is _ ( semi ) greedy _ if for any configuration of @xmath12 pebbles on the vertices of @xmath6 we can move a pebble to any specified root vertex @xmath14 , in such a way that each pebbling step is ( semi ) greedy . let @xmath114 be a product of paths , where @xmath115 . then each vertex @xmath116 can be represented by a vector @xmath117 , with @xmath118 for each @xmath119 . let @xmath120 , be the @xmath121 standard basis vector . denote the vector @xmath122 by * 0*. then two vertices @xmath7 and @xmath4 are adjacent in @xmath97 if and only if @xmath123 for some integer @xmath124 . if @xmath125 , then we may define * q*-_pebbling _ in @xmath97 to be such that each pebbling step from * u * to * v * is a @xmath126-pebbling step whenever @xmath123 . we denote the * q*-pebbling number of @xmath97 by @xmath127 . chungs s proof of theorem [ pathprod ] uses the following theorem @xcite . for integers @xmath128 , @xmath124 , we use @xmath129 as shorthand for the product @xmath130 . [ genpathprod ] suppose that @xmath129 pebbles are assigned to the vertices of @xmath97 and that the root @xmath131 . then it is possible to move one pebble to @xmath14 via greedy * q*-pebbling . in addition , it was shown in @xcite that @xmath132 , and moreover , that @xmath97 is greedy . also in @xcite the following generalization of graham s conjecture to * q*-pebbling was made . [ gengraham ] @xmath133 . can conjecture [ graham ] be proved in the case when both graphs share certain extra properties such as greediness and tree - solvability ? we say that a graph @xmath6 is _ tree - solvable _ if , whenever @xmath0 is a configuration on @xmath6 of size @xmath12 , it is possible to solve @xmath0 in such a way that the edges traversed by pebbling steps form an acyclic graph . there are graphs which are neither greedy nor tree - solvable . for example , let @xmath134 be the vertices ( in order ) of the cycle @xmath135 , and form @xmath6 from @xmath135 by adjoining new vertices @xmath26 to @xmath136 and @xmath103 , and @xmath137 to @xmath136 and @xmath138 . it is not difficult to show that @xmath139 . with root @xmath140 and @xmath141 , @xmath0 has no greedy solution ( nor does it have a semi - greedy solution ) . with @xmath142 and @xmath143 , @xmath0 has no tree - solution . ( it is worth noting that @xmath6 is bipartite . ) another illustrative graph is @xmath144 , which is also bipartite . ( @xmath145 is the _ star _ with @xmath24 vertices , also denoted @xmath146 ) . easily , @xmath147 and @xmath148 ( these numbers also follow from our classification of diameter two graphs below ) . although each pebbling number is only one more than the number of vertices , @xmath149 is far greater than @xmath150 ; it would be interesting to discover how much greater this gap can be for other graphs . also , notice that @xmath151 , a strict inequality . more importantly , as observed by moews @xcite , @xmath6 is not semi - greedy . indeed , think of @xmath6 as three pages of a book , let @xmath14 be the corner vertex of one of the pages , @xmath52 the farthest corner vertex of another page , @xmath7 , @xmath4 and @xmath8 the three vertices of the third page , and let @xmath152 . this shows that even semi - greediness is not preserved by the product of two greedy graphs . probably , the following holds . [ greedyconj ] almost every graph is greedy and tree - solvable . because of the lower bound of @xmath153 on the pebbling number @xmath12 , it is natural to try to classify class 0 graphs , or at least give conditions which either guarantee or prohibit class 0 . this is of course extremely difficult , although some preliminary results have proved quite interesting . as argued above , graphs of connectivity 1 are not class 0 . in @xcite we find the following theorem . [ diam2 ] if @xmath154 then @xmath48 or @xmath155 . class 0 graphs of diameter 2 are classified in @xcite . a particularly crucial graph in the characterization is the graph @xmath6 , built from the bipartite graph @xmath135 by connecting all the vertices of one of the parts of the bipartition to each other . @xmath6 has connectivity 2 and diameter 2 but @xmath156 , as witnessed by the configuration @xmath157 , where @xmath136 , @xmath158 and the root @xmath14 are independent . the following corollary to the characterization appears in @xcite . denote the connectivity of a graph @xmath6 by @xmath159 . [ 3conn ] if @xmath154 , and @xmath160 then @xmath6 is of class 0 . from this it follows that almost all graphs ( in the probabilistic sense ) are of class 0 , since almost every graph is 3-connected with diameter 2 . the following result , conjectured in @xcite , was proved in @xcite . this result was used to prove a number of other theorems , including theorems [ mindeg ] , [ girth ] and [ class ] . [ conn ] there is a function @xmath161 such that if @xmath6 is a graph with @xmath73 and @xmath162 then @xmath6 is of class 0 . moreover , @xmath163 . an upper bound for diameter 3 graphs was recently obtained in @xcite . [ diam3 ] if @xmath164 then @xmath165 , which is best possible . another guarantee for class 0 membership may reside in the following , as yet unexplored , question . [ greedy ] is it true that every greedy graph is of class 0 ? a nice family of graphs in relation to theorem [ conn ] is the following . for @xmath166 , the _ kneser graph _ , @xmath167 , is the graph with vertices @xmath168 \choose t}$ ] and edges @xmath169 whenever @xmath170 . the case @xmath171 yields the complete graph @xmath172 and the case @xmath173 and @xmath174 yields the petersen graph @xmath64 , both of which are class 0 . when @xmath175 and @xmath176 we have @xmath177 . also , it is not difficult to show that @xmath178 in this range , implying that @xmath167 is class 0 by theorem [ 3conn ] . furthermore , chen and lih @xcite have shown that @xmath167 is connected , edge transitive , and regular of degree @xmath179 . a theorem of lovsz @xcite states that such a graph has connectivity equal to its degree , and thus @xmath180 . therefore , using theorem [ conn ] , it is not difficult to prove the following . @xcite[kneser ] for any constant @xmath181 , there is an integer @xmath182 such that , for @xmath183 , @xmath184 and @xmath185 , we have that @xmath167 is class 0 . in the context of graph pebbling , the family of kneser graphs is interesting precisely because the graphs become more sparse as @xmath13 decreases toward @xmath186 , so the diameter ( as well as the girth ) increases and yet the connectivity decreases . [ classkneser ] for @xmath187 , is @xmath188 class 0 ? because pachter et al . @xcite also proved that diameter two graphs have the 2-pebbling property , it is interesting as well to ask whether @xmath167 has the 2-pebbling property when @xmath189 ( i.e. , when its diameter is at least 3 ) . in addition , one should ask whether graphs of large enough connectivity ( in terms of the diameter ) have the 2-pebbling property . this would follow from theorem [ conn ] if all class 0 graphs have the 2-pebbling property . regarding conditions which prohibit class 0 membership , one can easily show that if @xmath190 then @xmath53 . the following question was asked in @xcite : is there a constant @xmath26 so that @xmath191 implies @xmath53 ? this question was answered in the negative in @xcite by using theorem [ conn ] , along with a probabilistic ( deletion ) method analogous to erds s construction @xcite of graphs of arbitrarily high girth and chromatic number . [ girth ] let @xmath192 denote the maximum number @xmath26 such that there exists a class 0 graph @xmath6 on at most @xmath24 vertices with finite @xmath193 . the for all @xmath194 we have @xmath195 finally , denote by @xmath196 the family of all connected , bipartite graphs on @xmath197 vertices with minimum degree at least @xmath198 . what can be said of the pebbling numbers of graphs in @xmath196 ? in particular , the only graph in @xmath199 is @xmath200 , having pebbling number @xmath201 . the only graph in @xmath202 is @xmath51 , having pebbling number @xmath197 , as first shown in @xcite . also shown in @xcite is the result that if @xmath203 then @xmath6 is of class 0 . in @xcite we find that if @xmath204 ( and @xmath205 ) then every @xmath206 is of class 0 . define @xmath207 to be the smallest @xmath198 such that @xmath6 is of class 0 for every @xmath206 . recently in @xcite was proved the following result . [ regularbipartite ] for all @xmath208 we have @xmath209 . for all @xmath210 we have @xmath211 . if @xmath6 has some pseudo - random properties then the bound for @xmath212 seems to be significantly smaller . as far as sparse graphs are concerned , a related problem is to find the transitions for @xmath198 at which the pebbling numbers of the graphs in @xmath196 change from exponential to subexponential in @xmath24 , and from superlinear to linear in @xmath24 . in connection with theorem [ regularbipartite ] , it is natural to consider the family @xmath213 of connected , minimum degree @xmath198 graphs instead of bipartite graphs . in @xcite is proved an analogous result . let @xmath214 be the minimum @xmath198 such that every graph in @xmath213 is class 0 . [ normal ] for all @xmath215 we have @xmath216 . for all @xmath217 we have @xmath218 . in terms of possible applications it may be relevant to consider having to move a pebble to every vertex simultaneously , rather than to just one root . more generally , given a weight function @xmath8 on the vertices of a graph @xmath6 , we define the _ weighted cover pebbling number _ @xmath219 to be the minimum number of pebbles needed to place , after a sequence of pebbling steps , @xmath220 pebbles on vertex @xmath4 , for all @xmath4 , regardless of the initial configuration . the _ cover pebbling number _ @xmath221 regards the case @xmath222 for all @xmath4 . we say that @xmath8 is _ positive _ if @xmath223 for every vertex @xmath4 . in @xcite , for positive weight functions the weighted cover pebbling number for cliques and trees is found , and it is shown that the ratio between this parameter and the pebbling number can be arbtrarily large , even within the class of trees . given @xmath8 define @xmath224 and @xmath225 . the results are as follows . [ wcovpebbk ] for every positive weight function @xmath8 we have @xmath226 . the _ fuse _ @xmath227 is the graph on @xmath228 vertices composed of a path ( or _ wick _ ) on @xmath229 vertices with @xmath230 independent vertices ( _ sparks _ ) incident to one of its endpoints . [ wcovrat ] for every @xmath24 and @xmath229 we have @xmath231 . thus , for @xmath232 , we have @xmath233 . given a graph @xmath6 define @xmath234 , where @xmath235 denotes the distance between @xmath7 and @xmath4 . [ wcovpebbt ] for every positive weight function @xmath8 we have @xmath236 for every tree @xmath237 . it would be interesting to find the weighted pebbling numbers of other graphs such as cycles , cubes , complete @xmath14-partite graphs and random graphs , among others . the main technique in proving theorem [ wcovpebbt ] involves showing that the largest @xmath8-unsolvable configuration is _ simple _ ; that is , concentrated on a single vertex . the same can not be said in general for weight functions that are not positive . an important question to investigate along these lines is whether , for arbitrary graphs and positive weight functions , the largest @xmath8-unsolvable configuration is always simple . a positive answer verifies the so - called stacking conjecture ( now theorem ) , below . [ stack ] for every graph @xmath6 and positive weight function @xmath8 we have @xmath238 . this theorem originally was verified in @xcite for the case of covering @xmath45-dimensional cubes ( @xmath239 ) . other graphs were considered in @xcite . theorem [ stack ] was recently proved in general independently in @xcite . one should also search for classes of graphs for which there is some constant @xmath103 so that every graph @xmath6 in the class satisfies @xmath240 . for example , the classes of cliques and paths both have @xmath241 . finally , we define the weighted pebbling number @xmath242 to be the minimum number @xmath212 so that , for every weight function @xmath8 with @xmath243 , every configuration of @xmath212 pebbles is @xmath8-solvable . if one is concerned only with positive @xmath8 , then such an evaluation on any tree @xmath237 is trivial . one would compute @xmath244 . however , nonpositive @xmath8 must be considered , and thus finding weighted cover pebbling numbers for such @xmath8 is critical . instead of looking for the size of the largest unsolvable configuration in a graph , which is essentially the task in finding the pebbling number of the graph ( minus one ) , one could look for the size of the smallest solvable configuration , which is the task in finding what is called the _ optimal _ pebbling number , @xmath245 . the first results in this direction showed that @xmath246 @xcite . in @xcite it is proved that @xmath247 for all @xmath6 ( equality also holds for cycles ) , and the lower bound @xmath248 is found in @xcite . caterpillars , cycles and other graphs have been considered in @xcite , and the following interesting analog of graham s conjecture was recently proven in @xcite . [ optgraham ] for all graphs @xmath6 and @xmath249 we have @xmath250 . what is most surprising is that , while @xmath251 is known exactly , only @xmath252 is known at present @xcite . clearly , much work needs to be done in this area . some of the newest results consider graphs of high minimum degree . the following result of czygrinow appears in @xcite . [ optdeg ] if @xmath6 is a connected graph with @xmath24 vertices and @xmath253 , then @xmath254 . the result is not known to be sharp , as given by the following result in @xcite . [ optdeglow ] for all @xmath255 , @xmath256 and @xmath257 , there is a graph @xmath6 with @xmath253 and @xmath258 . the graphs discovered for this theorem are clever modifications of a blow - up of the vertices of a cycle into cliques . it is noted that the @xmath259 term disappears when @xmath24 is a multiple of @xmath260 . it would be interesting to discover the correct asymptotic coefficient of @xmath261 , somewhere between 2.4 and 4 . the authors also ask if the general upper bound of @xmath262 can be improved because of high minimum degree . [ nover2 ] is it true that @xmath263 for all graphs @xmath6 with @xmath264 ? such a bound would be sharp for @xmath265 . if girth is also considered then one can say more . let @xmath266 and @xmath267 . the following theorem of @xcite displays an asymptotic bound of @xmath268 . [ deggir ] let @xmath269 , @xmath175 and @xmath270 . then every @xmath24-vertex graph @xmath6 with @xmath253 and @xmath271 satisfies @xmath272 . the optimal pebbling numbers of linear ( @xmath273 ) , circular ( @xmath274 ) and m " obius ( circular with a twist ) are also determined in @xcite : @xmath275 unless @xmath276 , with @xmath13 as a lower bound always . in terms of a taxi service , army unit , or other operational outfit being able to strategically place their limited resources so as to reach any desired location , this line of research may be most fruitful and significant . many combinatorial optimization parameters have fractional counterparts . graph parameters like matching number , chromatic number @xcite are well known examples , as is the dimension of posets @xcite , and the pebbling number is no exception . let @xmath277 be the minimum @xmath212 such that , for every choice of root @xmath14 and for every configuration of @xmath212 pebbles on @xmath6 , it is possible to move @xmath198 pebbles to @xmath14 . then the fractional pebbling number @xmath278 . for example , @xmath279 . also , using moews s tree theorem @xcite , it is easy to compute @xmath280 for every tree @xmath237 . the following general result was recently proven in @xcite . [ fracpebbthm ] for every graph @xmath6 we have that @xmath281 . typically , the parameter in question can be formulated in terms of an integer linear program , and the relaxation of the program gives rise to the fractional version of the parameter . in the case of graph pebbling , while it is routine to verify in polynomial time that a particular configuration reaches a particular root , no simple matrix condition exists to capture whether every configuration of a fixed size @xmath212 is solvable for a particular root . thus minimizing @xmath212 as an integer program remains elusive , and so it is difficult to say at present whether or not a parameter exists that is dual to @xmath282 . recently , lourdusamy and somasundaram @xcite found a proof that verifies graham s conjecture for the case @xmath283 . their proof uses linear programming and lends credence to the above ideas while suggesting that further explorations into the connections between linear programming duality and pebbling numbers may be fruitful and worthwhile indeed . because it is easy to verify a pebbling solution , the question of deciding whether a given configuration @xmath0 reaches a given root @xmath14 is in np . in fact it has been shown @xcite that this question is at least has hard as np - complete ( see also @xcite ) . [ hard ] for every @xmath284-regular hypergraph @xmath249 , there is a graph @xmath6 , a configuration @xmath0 , and a root @xmath14 , so that deciding if @xmath0 reaches @xmath14 in @xmath6 is at least as hard as deciding if @xmath249 has a perfect matching . a complexity upper bound is found in @xcite , where it shown that deciding whether @xmath285 is a @xmath286-complete problem . this means that it is complete for the class of languages computable in polynomial time by conp machines equipped with an oracle for an np - complete language . they also proved that deciding whether @xmath287 is np - complete , while the same result is proved in @xcite for deciding whether @xmath288 . instead of computing @xmath12 exactly , one could try to approximate it . in other words the following question is of interest . [ comp2 ] does there exist a ( constant / linear / polynomial ) function @xmath289 and an algorithm which , given a graph @xmath6 on @xmath24 vertices , can find a number in polynomial time @xmath290 such that @xmath291 just like in the case of other `` hard '' problems one can design a `` good '' approximation algorithm for special cases of graphs . in particular , is it possible to design an approximation algorithm for the case when @xmath6 is a dense graph ? perhaps an easier and more natural question is to ask for an algorithm which , given a configuration and a root , will pebble to the root efficiently . [ comp3 ] does there exist an algorithm which , given a graph @xmath6 , a root @xmath14 , and an @xmath14-solvable configuration @xmath0 , can pebble to @xmath14 in time polynomial in @xmath24 ? of course , theorem [ hard ] shows that this is in general impossible unless @xmath292 , although the question for appropriately restricted classes of graphs , such as cubes , remains interesting . note that for some graphs ( like trees ) the greedy algorithm will pebble to @xmath14 efficiently . however , as noted in section [ pebnum ] , there are graphs which are neither tree - solvable , nor greedy ( nor semi - greedy ) , and so the question of how to pebble is far from obvious . even in the case of greedy graphs like @xmath50 , there are solvable configurations that require decision - making ; that is , some greedy approaches fail while others succeed . the notion that graphs with very few edges tend to have large pebbling number and graphs with very many edges tend to have small pebbling number can be made precise as follows . let @xmath293 be the random graph model in which each of the @xmath294 possible edges of a random graph having @xmath24 vertices appears independently with probability @xmath42 . for functions @xmath295 and @xmath26 on the natural numbers we write that @xmath296 ( or @xmath297 ) when @xmath298 as @xmath299 . let @xmath300 and define @xmath301 ( resp . , @xmath302 ) to be the set of functions @xmath295 for which there are constants @xmath303 such that @xmath304 ( resp . , @xmath305 ) whenever @xmath306 . finally , let @xmath307 . let @xmath308 be a property of graphs and consider the probability @xmath309 that the random graph @xmath293 has @xmath308 . for large @xmath42 it may be that @xmath310 as @xmath299 , and for small @xmath42 it may be that @xmath311 as @xmath299 . more precisely , define the _ threshold _ of @xmath308 , @xmath312 , to be the set of functions @xmath212 for which @xmath313 implies that @xmath310 as @xmath299 , and @xmath314 implies that @xmath311 as @xmath299 . it is not clear that such thresholds exist for arbitrary @xmath308 . however , we observe that class 0 is a monotone property ( adding edges to a class 0 graph maintains the property ) , and a theorem of bollobs and thomason @xcite states that @xmath312 is nonempty for every monotone @xmath308 . it is well known @xcite that @xmath315 , and since connectedness is required for class 0 , we see that @xmath316 . in @xcite it is noted that @xmath317 . in @xcite theorem [ conn ] was used to prove the following result . [ class ] for all @xmath318 , @xmath319 . as a probabilistic version of theorem [ regularbipartite ] , we might replace the probability space of @xmath293 by @xmath320 , the random graph model in which each of the @xmath321 possible edges of a random bipartite graph having @xmath13 vertices in each bipartition appear independently with probability @xmath42 . one might be interested in the following problem . [ randbip ] find the class 0 threshold @xmath322 of the random bipartite graph @xmath320 . we now reconsider graham s conjecture [ graham ] from a probabilistic point of view . we would like to say at the very least that it holds almost always . more precisely , we offer the following . [ randgraham ] let @xmath323 for any @xmath324 . then @xmath325\rar 1 $ ] as @xmath299 . if it turns out that the threshold for class 0 is bigger than that for connectivity , then it would be a very interesting to investigate the pebbling numbers of graphs in @xmath293 for @xmath42 in that range . another important problem , especially in light of the previous conjecture , is the following . [ 2ppthresh ] find the threshold @xmath326 for the 2-pebbling property of the random graph @xmath293 . for this section we will fix notation as follows . the vertex set for any graph on @xmath17 vertices will be taken to be @xmath327\}$ ] , where @xmath328=\{0,\ldots , n-1\}$ ] . that way , any configuration @xmath329 is independent of @xmath330 . ( here we make the distinction that @xmath24 is the index of the graph @xmath330 in a sequence @xmath331 , whereas @xmath332 denotes its number of vertices . ) let @xmath333 denote the sequence of complete graphs , @xmath334 the sequence of paths , and @xmath335 the sequence of @xmath24-dimensional cubes . let @xmath336\rar \n$ ] denote a configuration on @xmath332 vertices . let @xmath337 and for fixed @xmath17 consider the probability space @xmath338 of all configurations @xmath339 of size @xmath340 , we denote by @xmath341 the probability that @xmath342 is @xmath330-solvable and let @xmath343 . we say that @xmath212 is a _ pebbling threshold _ for @xmath344 , and write @xmath345 , if @xmath346 whenever @xmath347 and @xmath348 whenever @xmath349 . the existence of such thresholds was recently established in @xcite . [ existthresh ] every graph sequence @xmath344 has nonempty @xmath350 . the first threshold result is found in @xcite . the result is merely an unlabeled version of the so - called `` birthday problem '' , in which one finds the probability that 2 of @xmath212 people share the same birthday , assuming @xmath17 days in a year . [ cliquethresh ] @xmath351 . the same threshold applies to the sequence of stars ( @xmath352 ) . it was discovered in @xcite that every graph sequence @xmath344 satisfies @xmath353 , where @xmath354 is meant to signify that @xmath355 for every @xmath356 . the authors also discovered that @xmath357 when @xmath344 is a sequence of graphs of bounded diameter , that @xmath358 , and that @xmath359 . surprisingly , the threshold for the sequence of paths has not been determined . the lower bound found in @xcite was improved in @xcite to @xmath360 for every @xmath361 , while the upper bound of @xmath362 found in @xcite was improved in @xcite to @xmath363 for every @xmath364 . finally the lower bound was tightened recently in @xcite to nearly match the upper bound . [ lower ] for any constant @xmath365 , we have @xmath360 . frustratingly , this still leaves room for a wide range of possible threshold functions . it is interesting that even within the family of trees , the pebbling thresholds can vary so dramatically , as in the case for paths and stars . diameter seems to be a critical parameter . it is quite natural to guess that families of graphs with higher pebbling numbers have a higher threshold , but this kind of monotonicity result remains unproven . [ monothresh ] if @xmath366 for all @xmath24 then @xmath367 . in fact , this conjecture remains unproven even in the case that @xmath344 and @xmath368 are sequences of trees . moreover , there is some reason to believe the conjecture may be false , since the pebbling number is a worst - case scenario , while the threshold is an average - case scenario . consider the following . for a positive integer @xmath212 and a graph @xmath6 denote by @xmath369 the probability that a randomly chosen configuration @xmath0 of size @xmath212 on @xmath6 solvable . then conjecture [ monothresh ] would follow from the statement that , if @xmath366 then for all @xmath212 we have @xmath370 . unfortunately , although seemingly intuitive , this implication is false . using the class 0 pebbling characterization theorem of @xcite , in @xcite is found a family of pairs of graphs @xmath371 , one pair for each @xmath372 , for which the implication fails . however , the implication may yet hold when @xmath330 and @xmath373 are trees . [ trees ] let @xmath330 and @xmath373 both be trees on @xmath24 vertices so that @xmath366 . is it true that , for all @xmath212 ( or those @xmath212 `` near '' one of the thresholds ) , we have @xmath370 ? at the heart of such investigations into monotonicity is the following , most natural conjecture . [ spec ] let @xmath374 be any functions satisfying @xmath375 . then there is some graph sequence @xmath344 such that @xmath376 . this conjecture was proven in @xcite in the case that @xmath377 is replaced by @xmath378 . in fact , the family of fuses ( defined in section [ coverpebbopt ] ) covers this whole range . what behavior lives above @xmath378 remains unknown . it is interesting to consider a pebbling threshold version of graham s conjecture . given graph sequences @xmath379 and @xmath344 , define the sequence @xmath380 . suppose that @xmath381 , @xmath382 , and @xmath383 , where @xmath384 , @xmath385 , and @xmath386 . [ prodthresh ] is it true that , for @xmath379 , @xmath344 , and @xmath368 as defined above , we have @xmath387 ? in particular , one can define the sequence of graphs @xmath388 in the obvious way . in @xcite one finds tight enough bounds on @xmath389 to show that the answer to this question is yes for @xmath390 and @xmath391 . another important instance is @xmath392 . boyle @xcite proved that @xmath393 . this was improved in @xcite , answering question [ prodthresh ] affirmatively for squares of complete graphs . [ squarecliques ] for @xmath394 we have @xmath395 . this result is interesting because , by squaring , the graphs become fairly sparse , and yet their structure maintains the low pebbling threshold . the proof of the result tied the behavior of pebbling in @xmath396 to the existence of large components in various models of random complete bipartite graphs . another interesting related sequence to consider is @xmath397 . when @xmath398 we have @xmath399 , and the best result to date is the following theorem of @xcite ( obtained independently in @xcite ) . [ cubes ] for the sequence of cubes we have @xmath400 for all @xmath401 . let @xmath402 and @xmath403 and denote by @xmath404 the graph sequence @xmath405 , where @xmath406 . most likely , fixed @xmath229 yields similar behavior to theorem [ cubes ] . [ grids ] for fixed @xmath229 we have @xmath407 . in contrast , the results of @xcite show that @xmath408 for fixed @xmath45 . thus it is reasonable to believe that there should be some relationship between the two functions @xmath402 and @xmath403 , both of which tend to infinity , for which the sequence @xmath404 has threshold on the order of @xmath17 . [ gridbalance ] find a function @xmath403 for which @xmath409 . in particular , how does @xmath45 compare to @xmath24 ? finally one might consider the behavior of graphs of high minimum degree . define @xmath410 to be the set of all connected graphs on @xmath24 vertices having minimum degree at least @xmath411 . let @xmath412 denote any sequence of graphs with each @xmath413 . in @xcite is proven the following . [ dense ] for every function @xmath414 , @xmath415 . in particular , if in addition @xmath416 then @xmath417 . a very recent incarnation of graph pebbling is introduced in @xcite , in which the _ critical pebbling number _ of a graph @xmath6 is defined . a configuration @xmath0 of pebbles on @xmath6 is _ minimally solvable _ if it is solvable but the removal of any pebble leaves it unsolvable . while the optimal pebbling number measures the size of the smallest minimally solvable configuration on @xmath6 , the critical pebbling number measures the size of the largest minimally solvable configuration . another variation is developed in @xcite , combining ideas from sections [ pebnum ] and [ coverpebbopt ] . the _ support _ of a configuration @xmath0 is the set of vertices @xmath4 having @xmath418 . the _ domination cover pebbling number _ @xmath419 is the minimum number @xmath212 so that from every configuration of @xmath212 pebbles one can reach a configuration whose support is a dominating set of @xmath6 . their motivation stems from transporting devices from initial positions to eventual positions that allow them to monitor the entire graph . because of the superficial similarity of graph pebbling to other positional games on graphs , like `` cops - and - robbers '' @xcite and `` chip - firing '' @xcite for instance , and the high degree of applicability of many of these games to structural graph theory @xcite and theoretical computer science @xcite , one should nt neglect the possibility that graph pebbling will have similar impact . for example , one can think of the loss of a pebble during a pebbling step as a toll or as a loss of information , fuel or electrical charge . @xmath111-pebbling is one generalization of this rate of loss ; another is simply to choose any fixed rate @xmath70 of loss . in any case , instead of restricting the initial configuration to integer values , let @xmath0 range among all nonnegative reals . a pebbling step removes weight @xmath52 from one vertex and places weight @xmath420 at an adjacent vertex , for some fixed @xmath421 . still the objective is to place weight 1 at any prescribed root @xmath14 so that there is enough money , fuel , information , or energy at that location in the network . of course , all of the questions raised herein may be asked about this more general @xmath70-_pebbling_. it is conceivable that chip - firing may even come into play as a useful model . for a given graph @xmath6 , one might be able to build an auxiliary graph @xmath249 , so that chip - firing results on @xmath249 can be brought to bear on @xmath12 . this opens up the theory to questions of the eigenvalues of the laplacian of @xmath6 , and so on . 999 w. r. alford , a. granville and c. pomerance , there are infinitely many carmichael numbers , _ annals math . ser . 2 _ * 139 * , no . 3 ( 1994 ) , 703722 . n. alon , _ personal communication _ ( 2003 ) . n. alon , s. friedland and g. kalai , regular subgraphs of almost regular graphs , _ j. combin . theory ser . b _ * 37 * ( 1984 ) , 7991 . r. anderson , l. lovsz , p. shor , j. spencer , e. tardos and s. winograd , disks , balls , and walls : analysis of a combinatorial game , _ amer . monthly _ , * 96 * ( 1989 ) , 481493 . s. arnborg , d. g. corneil and a. proskurowski , complexity of finding embeddings in a @xmath198-tree , _ siam j. algebraic disc . methods _ * 8 * ( 1987 ) , 277284 . a. bekmetjev , g. brightwell , a. czygrinow and g. hurlbert , thresholds for families of multisets , with an application to graph pebbling , _ discrete math . _ * 269 * ( 2003 ) , 2134 . a. bekmetjev and g. hurlbert , the pebbling threshold of square cliques , _ preprint _ b. bollobs and a. thomason , threshold functions , _ combinatorica _ * 7 * ( 1987 ) , 3538 . j. boyle , thresholds for random distributions on graph sequences with applications to pebbling , _ discrete math . _ * 259 * ( 2002 ) , 5969 . g. brightwell and e. scheinerman , fractional dimension of partial orders , _ order _ * 9 * ( 1992 ) , 139158 . b. bukh , maximum pebbling number of graphs of diameter three , _ preprint _ d. bunde , e. chambers , d. cranston , k. milans and d. west , pebbling and optimally pebbling in graphs , _ preprint _ ( 2005 ) . y. caro , zero - sum problems a survey , _ discrete math . _ * 152 * ( 1996 ) , 93113 . m. chan and a. godbole , improved pebbling bounds , _ preprint _ s. chapman , on the davenport constant , the cross number , and their application in factorization theory , in : _ zero - dimensional commutative rings , lecture notes in pure appl . _ * 171 * , marcel dekker , new york , 1995 , 167190 . chen and k.w . lih , hamiltonian uniform subset graphs , _ j. combin . theory ser . b _ * 42 * ( 1987 ) , 257263 . f. r. k. chung , pebbling in hypercubes , _ siam j. disc . _ * 2 * ( 1989 ) , 467472 . f. r. k. chung and r. ellis , a chip - firing game and dirichlet eigenvalues , _ discrete math . _ * 257 * ( 2002 ) , 341355 . t. clarke , pebbling on graphs , _ master s thesis _ , arizona st . univ . t. clarke , r. hochberg and g. hurlbert , pebbling in diameter two graphs and products of paths , _ j. graph th . _ * 25 * ( 1997 ) , 119128 . b. crull , t. cundif , p. feltman , g. hurlbert , l. pudwell , z. szaniszlo and z. tuza , the cover pebbling number of graphs , _ discrete math . _ * 296 * ( 2005 ) , 1523 . a. czygrinow , n. eaton , g. hurlbert and p. m. kayll , on pebbling threshold functions for graph sequences , _ discrete math . _ * 247 * ( 2002 ) , 93105 . a. czygrinow and g. hurlbert , pebbling in dense graphs , _ austral . j. combin . _ , * 29 * ( 2003 ) , 201208 . a. czygrinow and g. hurlbert , girth , pebbling and grid thresholds , _ siam j. discrete math . _ , to appear . a. czygrinow and g. hurlbert , on the pebbling threshold of paths and the pebbling threshold spectrum , _ preprint _ a. czygrinow , g. hurlbert , h. kierstead and w. t. trotter , a note on graph pebbling , _ graphs and combin . _ * 18 * ( 2002 ) , 219225 . a. czygrinow and m. wagner , on the pebbling threshold of the hypercube , _ preprint _ t. denley , on a result of lemke and kleitman , _ combin . , prob . and comput . _ * 6 * ( 1997 ) , 3943 . s. elledge and g. hurlbert , an application of graph pebbling to zero - sum sequences in abelian groups , _ integers : elec . j. number theory _ * 5(1 ) * ( 2005 ) , # a17 . p. van emde boas and d. kruyswijk , a combinatorial problem on finite abelian groups iii , _ report zw-1969 - 007 _ , math . centre , amsterdam . p. erds , on pseudoprimes and carmichael numbers , _ publ . debrecen _ * 4 * ( 1956 ) , 201206 . p. erds , applications of probabilistic methods to graph theory , in _ a seminar on graph theory _ , holt , rinehart and winston , new york ( 1967 ) , 6061 . p. erds , a. ginzburg and a. ziv , a theorem in additive number theory , _ bull . council israel _ * 10f * ( 1961 ) , 4143 . p. erds and a. rnyi , on random graphs i , _ publ . debrecen _ * 6 * ( 1959 ) , 290297 . r. feng and j. y. kim , graham s pebbling conjecture on product of complete bipartite graphs , _ sci . china ser . a _ * 44 * ( 2001 ) , 817822 . r. feng and j. y. kim , pebbling numbers of some graphs , _ sci . china ser . a _ * 45 * ( 2002 ) , 470478 . j. foster and h. snevily , the 2-pebbling property and a conjecture of graham s , _ graphs and combin . _ * 16 * ( 2000 ) , 231344 . t. friedman and c. wyels , optimal pebbling of paths and cycles , _ preprint _ h. l. fu and c. l. shiue , the optimal pebbling number of the complete @xmath13-ary tree , _ discrete math . _ * 222 * ( 2000 ) , 89100 . h. l. fu and c. l. shiue , the optimal pebbling number of the caterpillar , _ elec . notes in disc . _ * 11 * ( 2002 ) . w. gao , on davenport s constant of finite abelian groups with rank three , _ discrete math . _ * 222 * ( 2000 ) , 111124 . w. gao and a. geroldinger , zero - sum problems and coverings by proper cosets , _ euro . j. combin . _ * 24 * ( 2003 ) , 531549 . w. gao and x. jin , weighted sums in finite cyclic groups , _ discrete math . _ * 283 * ( 2004 ) , 243247 . w. gao and r. thangadurai , on the structure of sequences with forbidden zero - sum subsequences , _ colloq . _ * 98 * ( 2003 ) , 213222 . j. gardner , a. godbole , a. teguia , a. vuong , n. watson and c. yerger , domination cover pebbling : graph families , _ preprint _ a. geroldinger , on a conjecture of kleitman and lemke , _ j. number theory _ * 44 * ( 1993 ) , 6065 . a. geroldinger and r. schneider , on davenport s constant , _ j. combin . theory ser . a _ * 61 * ( 1992 ) , 147152 . c. gibbons , j. laison and e. paul , critical pebbling numbers of graphs , _ preprint _ a. godbole , m. jablonski , j. salzman and a. wierman , an improved upper bound for the pebbling threshold of the @xmath24-path , _ discrete math . _ * 275 * ( 2004 ) , 367373 . g. gunda and a. higgins , pebbling on directed graphs , _ elec . day _ * 1 * ( 2003 ) , 113 . d. herscovici , graham s pebbling conjecture on products of cycles , _ j. graph theory _ * 42 * ( 2003 ) , 141154 . d. herscovici and a. higgins , the pebbling number of @xmath283 , _ discrete math . _ * 187 * ( 1998 ) , 123135 . g. hurlbert , two pebbling theorems , _ congr . * 135 * ( 1998 ) , 5563 . g. hurlbert , a survey of graph pebbling , _ congr . _ * 139 * ( 1999 ) , 4164 . g. hurlbert , the graph pebbling page , g. hurlbert and h. kierstead , on the complexity of graph pebbling , _ preprint _ g. hurlbert and b. munyan , cover pebbling hypercubes , _ bull . _ , to appear . p. lemke and d. j. kleitman , an addition theorem on the integers modulo @xmath24 , _ j. number th . _ * 31 * ( 1989 ) , 335345 . a. lourdusamy and s. somasundaram , pebbling @xmath422 using linear programming , _ j. discrete math . sci . cryptography _ * 4 * ( 2001 ) , 115 . l. lovsz , _ combinatorial problems and exercises _ , north holland pub . , amsterdam , new york , oxford ( 1979 ) . k. milans and b. clark , the complexity of graph pebbling , _ preprint _ ( 2005 ) . d. moews , pebbling graphs , _ j. combin ( b ) _ * 55 * ( 1992 ) , 244252 . d. moews , _ personal communication _ ( 1997 ) . d. moews , optimally pebbling hypercubes and powers . _ discrete math . _ * 190 * ( 1998 ) , 271276 . m. nathanson , _ additive number theory : inverse problems and the geometry of sumsets _ ( graduate texts in mathematics ; 165 ) , springer - verlag , new york , 1996 , 4851 . r. nowakowski and p. winkler , vertex - to - vertex pursuit in a graph , _ discrete math . _ * 43 * ( 1983 ) , 235239 . j. olson , a combinatorial problem on finite abelian groups i , _ j. number theory _ * 1 * ( 1969 ) , 810 . l. pachter , h. s. snevily , and b. voxman , on pebbling graphs , _ congr . _ * 107 * ( 1995 ) , 6580 . n. robertson and p. d. seymour , graph minors _ iii _ : planar tree - width , _ j. combin . theory ser . b _ * 36 * ( 1984 ) , 4964 . e. scheinerman and d. ullman , _ fractional graph theory : a rational approach to the theory of graphs _ , wiley , new york ( 1997 ) . p. d. seymour and r. thomas , graph searching and a min - max theorem for tree - width , _ j. combin . theory ser . b _ * 58 * ( 1993 ) , 2233 . j. sjostrand , the cover pebbling theorem , _ preprint _ ( 2004 ) . z. sun , unification of zero - sum problems , subset sums and covers of @xmath423 , _ elec . soc . _ * 9 * ( 2003 ) , 5160 . m. tomova and c. wyels , cover pebbling cycles and certain graph products , _ preprint _ ( 2005 ) . a. vuong and i. wyckoff , conditions for weighted cover pebbling of graphs , _ preprint _ ( 2004 ) . m. wagner and m. wokasch , graph pebbling numbers of line graphs , _ preprint _ s. wang , pebbling and graham s conjecture , _ discrete math . _ * 226 * ( 2001 ) , 431438 . n. watson , the complexity of pebbling and cover pebbling , _ preprint _ n. watson and c. yerger , cover pebbling numbers and bounds for certain families of graphs , _ preprint _ ( 2005 ) . d. b. west , _ introduction to graph theory _ , prentice - hall , upper saddle river , nj ( 1996 ) .
in black hole x - ray binaries , a misalignment between the spin axis of the black hole and the orbital angular momentum can occur during the supernova explosion that forms the compact object . in this letter we present population synthesis models of galactic black hole x - ray binaries , and study the probability density function of the misalignment angle , and its dependence on our model parameters . in our modeling , we also take into account the evolution of misalignment angle due to accretion of material onto the black hole during the x - ray binary phase . the major factor that sets the misalignment angle for x - ray binaries is the natal kick that the black hole may receive at its formation . however , large kicks tend to disrupt binaries , while small kicks allow the formation of xrbs and naturally select systems with small misalignment angles . our calculations predict that the majority ( ) of galactic field bh xrbs have rather small ( ) misalignment angles , while some systems may reach misalignment angles as high as and even higher . this results is robust among all population synthesis models . the assumption of small small misalignment angles is extensively used to observationally estimate black hole spin magnitudes , and for the first time we are able to confirm this assumption using detailed population synthesis calculations .
1001.1107
the formation of roche lobe overflowing x - ray binaries ( xrbs ) with black hole ( bh ) accretors in galactic fields , involves the evolution of a primordial isolated binary of which at least one of the component stars is massive enough to form a bh . right before the supernova ( sn ) explosion that will create the bh , the binary orbit is assumed to be circular , due to tidal interactions and possible preceding mass - transfer ( mt ) episodes , and the spins of the two stars are assumed to be aligned with the orbital angular momentum at the formation of the binary . the subsequent sn explosion will change the orbital characteristics of the binary . if the sn explosion is symmetric , the mass lost from the binary will cause the orbit to expand and induce some eccentricity , however the bh spin axis will remain aligned with the orbital angular momentum . in contrast , if an asymmetric kick is imparted on the bh during its formation , the orientation of the orbit will also change and this in turn will cause a misalignment between the spin axis of the bh and the orbital angular momentum . through studies of the position and spatial velocity of radio pulsars in the galaxy , it has been established in the literature that neutron stars ( nss ) can receive large asymmetric sn kicks ( @xmath3 ) @xcite . on the other hand , the possibility of an asymmetric kick imparted to bhs remains still an open question . stellar bhs need to be in xrbs in order to be observable , and only for a handful of galactic bh xrbs there are accurate measurements of their position and spatial velocity . however , indirect bh kick observations indicate that low mass bh receive high asymmetric kicks , while high mass bhs low or no kicks @xcite . on the other hand , detailed theoretical modeling of the evolutionary history of two galactic bh xrbs ( groj1655 - 40 and xtej1118 + 480 ) showed that for the case of groj1655 - 40 a moderate asymmetric kick is possible , while in the case of xtej1118 + 480 an asymmetric kick larger than @xmath4 is necessary in order to explain the current properties of the system @xcite . unless the direction of this asymmetric kick during the formation of the bh is finely tuned to be on the orbital plane , it will induce a misalignment of the bh spin and the orbital angular momentum . in fact , two microquasar type galactic bh xrbs ( groj1655 - 40 and v4641 sgr ) have been observed with relativistic radio jets that are misaligned with respect to their orbital plane . this misalignment angle is estimated to be @xmath5 for groj1655 - 40 @xcite , and @xmath6 for v4641 sgr @xcite . assuming that the radio jet has the same direction as the spin of the bh , this misalignment angle can be translated to a tilt of the bh spin axis with respect to the orbital plane . in this letter we present population synthesis ( ps ) models of roche - lobe overflowing , galactic field bh xrbs , and study their bh spin - orbit misalignment and its their dependence on model parameters . we performed the simulations presented here with _ startrack _ @xcite , a current binary ps code that has been tested and calibrated using detailed binary star calculations and incorporates all the important physical processes of binary evolution ( see * ? ? ? * for a detailed description of all code elements ) . we constructed a library of ps models appropriate for the milky - way . we performed a parameter study varying some of the important parameters involved in the formation and evolution of xrbs , such as the imf , the stellar wind strength for the late evolutionary stages of massive stars , the common envelope ( ce ) efficiency , the distribution of magnitude and direction of asymmetric kicks , and the initial spin magnitude of bhs . in all our models we adopted a solar metallicity for the stellar population , and a constant star formation rate over the age of the galaxy , which is assumed to be @xmath7 . for the full list of the model parameter combinations we considered , and the naming convention we used , see table 1 . ccccc 1 & 0.5 & -2.35 & 1.0 & 265 + 2 & 0.3 & -2.35 & 1.0 & 265 + 3 & 0.7 & -2.35 & 1.0 & 265 + 4 & 0.5 & -2.35 & 1.0 & 50 + 5 & 0.5 & -2.35 & 1.0 & 150 + 6 & 0.5 & -2.35 & 1.0 & 400 + 7 & 0.5 & -2.7 & 1.0 & 265 + 8 & 0.5 & -2.7 & 1.0 & 50 + 9 & 0.5 & -2.7 & 1.0 & 150 + 10 & 0.5 & -2.7 & 1.0 & 400 + 11 & 0.5 & -2.35 & 0.25 & 265 + 12 & 0.5 & -2.35 & 0.25 & 50 + 13 & 0.5 & -2.35 & 0.25 & 150 + 14 & 0.5 & -2.35 & 0.25 & 400 + 15 & 0.5 & -2.7 & 0.25 & 265 + 16 & 0.5 & -2.7 & 0.25 & 50 + 17 & 0.5 & -2.7 & 0.25 & 150 + 18 & 0.5 & -2.7 & 0.25 & 400 + 19 & 0.5 & -2.35 & 1.0 & 265 + at the time of birth , bhs and nss may receive an instantaneous asymmetric kick , due to asymmetries involved in the sn explosion mechanisms . the distribution of kick velocities for nss has been estimated through studies of the position and spatial velocity of radio pulsars in the galaxy . @xcite studied the proper motion of 233 galactic pulsars and found that their inferred kick velocities are consistent with a single maxwellian distribution with @xmath8 . in the sn explosions of more massive stars that can form bhs , the effects of material fall - back ( ejected initially in the sn explosion ) during the star s final collapse are included . for each core - collapse event we calculate the fall - back factor ( @xmath9 ) , i.e. , the fraction of the stellar envelope that falls back . the value of @xmath9 is calculate based on the mass of the carbon / oxygen core ( @xmath10 ) of the pre - sn star , and is interpolated linearly between @xmath11 ( @xmath12 ) and @xmath13 ( @xmath14 ) . the regimes of no fall - back , partial fall - back , and direct collapse are estimated from earlier studies by @xcite . in the formation of bhs , where partial fall - back or direct collapse takes place , kicks are lowered proportionally to the amount of fall - back associated with core - collapse event , according to the relation : @xmath15 where v is the kick magnitude drawn form the single maxwellian . furthermore , we consider the extreme case where the magnitude of the bh kick is drawn directly for the maxwellian distribution and not scale to fall - back mass ( model 19 ) . finally , the standard assumption is that the direction of the kick is isotropically distributed , but we also consider the case of polar kicks , where the kick direction is aligned with the spin axis of the bh progenitor . here we assume that any asymmetry in the sn explosion starts at , or very close to , the center of the progenitor star . if this natal kick has a component perpendicular to the plane of the orbit , then the inclination of the system will change while the spin of the bh will remain virtually unaffected , causing the bh spin to no longer be aligned with the orbit . _ startrack _ treats the orbital dynamics of the sn explosion by taking into account mass / angular momentum losses as well as asymmetries through natal kicks . the only simplifying assumption in these calculations is that the sn explosion is instantaneous ( for details see * ? ? ? * ; * ? ? ? knowing the exact geometry of the orbit right before and after the sn , and assuming that the direction of the bh spin axis remains unchanged relative to the spin axis of the bh progenitor , we are able to calculate the post - sn bh spin - orbit misalignment angle . during the xrb phase , the material that is being accreted onto the bh can be a significant fraction of the bh s initial mass , and the effects of the accretion on the bh spin can not be considered negligible . we therefore want to investigate whether this mass accretion can alter significantly the post - sn spin - orbit misalignment . in this section we describe how we model these effects . let us assume a bh of spin @xmath16 and mass @xmath17 surrounded by an accretion disk which has an inclination @xmath18 with respect to the bh s spin ( the accetion disk is assumed to be parallel to the orbital plane ) , and that mass @xmath19 is gradually being accreted onto the bh . the mass accreted by the bh , carries into the bh not only its mass , but also its angular momentum @xmath20 and as a result the orbital inclination is altered . it is reasonable to assume that @xmath20 is going to be the angular momentum right at the innermost stable circular orbit ( isco ) of the bh , since inside the isco the infalling mass is following a radial plunge . the position of the isco depends on the bh mass and spin and on the infalling particle s orbital inclination . therefore , during an accretion episode we need to find the position of the isco , calculate the angular momentum of the accreted mass at the isco , and then calculate the bh s new spin and mass and finally update the value of the inclination . the equation that describes the radial geodesic motion of a mass particle in the kerr spacetime in terms of the boyer - lindquist coordinates @xmath21 , is given by @xcite @xmath22 ^ 2-\delta[r^2+(l_z - ae)^2+q]\equiv r\ ] ] @xmath23 + al_z\left(1 - { r^2+a^2\over\delta}\right)\ ] ] where @xmath24 and @xmath25 . the quantities @xmath26 , @xmath27 and @xmath28 ( the three constants of motion ) are the specific energy , angular momentum and carter constant , with the carter constant being defined as : @xmath29 . notice that all equations make use of geometrized units ( @xmath30 ) . a mass particle orbiting at the isco should fulfill three conditions : @xmath31 ( for the orbit to be circular ) , @xmath32 ( for the orbit to remain circular , i.e. zero radial acceleration ) and finally @xmath33 ( for the orbit to be stable ) . by solving this set of equations ( @xmath34 ) we get the position of the isco @xmath35 , as well as the angular momentum ( @xmath36 ) and energy ( @xmath37 ) of the accreted mass . in our calculations the accretion of mass @xmath19 is treated as a series of accretion episodes of infinitesimal mass parcels @xmath38 ( @xmath39 ) . after every accretion episode , the mass and angular momentum of the bh is updated and so is the inclination . we proceed to the next accretion episode using these updated values until @xmath19 has been accreted . we considered a large library of ps models ( over 200 ) , where we varied a number of ps model parameters ( see table 1 ) . for each of the models , we examined the pdf of post - sn bh spin - orbit misalignment angles in bh xrbs and its dependence on the masses of the bh and the donor star . we note here that in all the results we present here , we considered only the transient bh xrb population . this selection was made in order to facilitate a direct comparison with the currently observed sample of galactic bh xrbs . the dynamical measurement of the bh mass , which is required in order to identify the compact object accretor of an xrb as bh , can only be done when the xrb is in quiescence . the same limitation holds for the two currently available observational technics ( continuum - fitting and fe k methods ) for the measurement of the bh spin magnitude . , second column : @xmath40 , and third column : @xmath41 ) . the upper six panels correspond to our _ `` standard '' _ model @xmath42 , while the the lower six panels correspond to model @xmath43 which has the same ps parameters as the _ `` standard '' _ model , except that the direction of the sn kicks is always aligned with the spin of the bh progenitor and not isotropic . the dotted vertical line shows the 67% confidence level ( i.e. 67% of the systems have misalignment angle less than that indicate by the position of the line ) , while the dashed and dash - dotted indicate the 95.4% and 99.7% confidence level correspondingly . ] in figure 1 we present the pdf of the post - sn bh spin - orbit misalignment angle and the two - dimensional pdf of the post - sn misalignment angle versus bh mass , split into three donor mass ranges , for our _ `` standard '' _ model @xmath42 , and for model @xmath43 . model @xmath43 has the same ps parameters as the _ `` standard '' _ model , except that the direction of the sn kicks is always perpendicular to the orbital plane and not isotropic . one general characteristic of the pdfs is that they are heavily skewed towards small misalignment angles . this characteristic is the result of the combination of two factors . bhs are generally believed to receive asymmetric kicks of smaller magnitude compared to that of ns . on the other hand , large kick magnitudes , that would result in a large misalignment angle , usually lead to the disruption of the binary , and thus these systems are effectively filtered out of the population . the latter effect becomes more prominent in the case of polar kicks ( model @xmath43 ) . one would initially expect that kicks always directed perpendicular to the orbital plane would result in larger misalignment angles . however , a large kick perpendicular to the orbital plane is also more efficient at disrupting the binary . figure 1 shows that the binary disruption effect is dominant , and that only bhs that received a small kick manage to remain bound in a binary , and thus polar kicks lead to overall smaller misalignment angles . the two - dimensional pdfs of the post - sn misalignment angle versus bh mass in figure 1 show that massive bhs tend to have smaller misalignment kicks . this trend is common in all our models , and is a consequence of our prescription for the asymmetric kicks imparted to bhs . as described in section 2.3 , massive bhs receive on average smaller kicks , which in turn lead to smaller misalignment angles . moreover , the higher the total mass of the binary is , the less the orbit will change from a given sn kick . the same physical mechanism is also associated with our finding that xrbs with massive donor stars have smaller bh spin tilts . , where the initial spin magnitude is drawn from a maxwellian distribution with a maximum at 0.2 and truncated at 1 . the lower six panels correspond to model @xmath44 which has the same ps parameters as model @xmath45 , but the maximum of the maxwellian initial bh spin magnitude distribution is at 0.9 . ] figure 2 shows the pdf of the current bh spin - orbit misalignment angle and the two - dimensional pdf of the current misalignment angle versus bh mass , split into three donor mass ranges , for models @xmath45 and @xmath44 . in this case , we take into account the evolution of the bh spin s direction and magnitude due to accretion during the xrb phase . the accretion of matter onto the bh always results on a gradual alignment of its spin with the orbital angular momentum . the efficiency of this process depends on the initial spin magnitude of the bh - the higher the initial bh spin is , the more difficult it is to change its direction via accretion . in figure 2 we show two models with different distributions of initial bh spin magnitute , one with a maxwellian distribution with a maximum at 0.2 and truncated at 1 , and one where the maximum of the maxwellian is at 0.9 . comparing these pdfs with the upper six panels of figure 1 , which show the pdfs of the post - sn misalignment angle of the same model , we see that for both models ( @xmath45 and @xmath44 ) the pdfs of figure 2 are shifted toward lower misalignment angles , and that indeed this effect is more prominent for smaller initial bh spin magnitudes ( model @xmath45 ) . lcccccccccccc @xmath42 & & 8.8 & 48.8 & 133.0 & & 10.1 & 43.2 & 100.7 & & 2.2 & 55.3 & 134.5 + @xmath43 & & 3.9 & 17.9 & 27.8 & & 10.7 & 24.2 & 29.7 & & 3.7 & 19.0 & 26.3 + @xmath46 & & 11.5 & 65.1 & 133.7 & & 7.9 & 37.7 & 131.3 & & 4.1 & 46.6 & 113.8 + @xmath47 & & 11.4 & 61.7 & 135.8 & & 10.1 & 45.3 & 117.8 & & 3.2 & 62.5 & 128.3 + @xmath48 & & 1.9 & 17.4 & 94.4 & & 1.8 & 6.2 & 111.2 & & 1.4 & 16.2 & 101.6 + @xmath49 & & 5.7 & 31.8 & 128.6 & & 5.4 & 22.0 & 75.5 & & 1.8 & 46.6 & 127.0 + @xmath50 & & 11.5 & 61.0 & 132.3 & & 13.6 & 60.5 & 136.4 & & 3.9 & 64.3 & 160.0 + @xmath51 & & 10.9 & 57.7 & 126.5 & & 10.1 & 43.8 & 97.7 & & 2.2 & 58.8 & 118.4 + @xmath52 & & 1.8 & 29.1 & 87.7 & & 2.2 & 25.4 & 81.6 & & 1.7 & 8.0 & 86.5 + @xmath53 & & 3.2 & 19.0 & 90.0 & & 2.3 & 9.1 & 69.8 & & 1.7 & 28.8 & 123.6 + @xmath54 & & 4.5 & 26.6 & 92.0 & & 4.2 & 15.9 & 86.1 & & 2.1 & 36.6 & 128.4 + @xmath55 & & 5.7 & 32.7 & 92.6 & & 6.0 & 21.7 & 89.3 & & 2.1 & 39.5 & 133.7 + @xmath45 & & 3.2 & 20.1 & 90.0 & & 2.4 & 10.7 & 60.9 & & 1.6 & 32.5 & 123.0 + @xmath56 & & 4.3 & 26.4 & 92.5 & & 4.3 & 17.3 & 86.1 & & 2.0 & 36.7 & 127.5 + @xmath44 & & 4.9 & 28.1 & 92.5 & & 4.8 & 18.0 & 89.2 & & 2.1 & 37.4 & 130.7 + @xmath57 & & 15.3 & 52.2 & 118.4 & & 11.6 & 50.8 & 93.4 & & 12.8 & 41.2 & 118.0 + table 2 contains the 67% , 95.4% , and 99.7% confidence levels of bh spin - orbit misalignment angle pdfs for a selected representative list of models ( for the complete list of ps models see the online supplemental material ) . after accounting for the partial alignment due to accretion onto the bh , we see that the majority of bh xrbs ( at a 67% confidence level ) have bh spin - orbit misalignment angles below - and some times well below - @xmath58 . this is a robust conclusion as it is true for _ all _ our models . at the same time , we find that in most of our models there is a small part of the xrb population ( @xmath59 ) with misalignment angles above @xmath60 . the main effect of varying the imf or the stellar wind strength is on the bh mass spectrum . a flatter imf or a weaker stellar wind lead to more massive bhs this in turn , as we explained earlier , lead to smaller sn kicks and overall more massive binaries , and thus smaller bh spin - orbit misalignment angles . on the other hand , a variation of the ce efficiency does not seem to affect spin misalignment of our bh xrb population . variations of the @xmath61 parameter of the maxwellian distribution for the sn kicks affect , as expected , the bh tilts , with higher kicks resulting to higher misalignment angles . however this effect saturates for very high kicks , as they lead to disruption rather than to very high misalignment angles . finally , we examine the extreme case where bh receive the same kicks as ns , directly drawn from a maxwellian distribution and not scaled to the fall - back mass ( model 19 ) . although this model predicts unrealistically large bh kicks , it serves as an absolute upper limit on the possible bh spin - orbit misalignment . we find the even in this extreme case , the majority of the bh xrbs have misalignment angles below @xmath62 . we performed ps simulations of a galactic field population of bh xrbs and studied the distribution of bh spin - orbit misalignment angles of transient , roche - lobe overflowing bh xrbs . in the determination of the current misalignment angle we took into account the evolution of the bh spin magnitude and direction due to accretion of matter during the xrb phase . we examined over 200 ps models , varying ps parameters such as imf , stellar wind strength , ce efficiency , sn kick s magnitude and direction distributions , and bh s initial spin magnitude distribution . we found that for the majority of bh xrbs in our model populations , the misalignment angle is below @xmath58 , while at the same time there is a small part of the population ( @xmath59 ) that has misalignment angles higher than @xmath60 which in some cases can even exceed @xmath63 . however this sub - population of highly misaligned bh xrbs account for less than @xmath64 of the total population . these results are robust among all our models , and all combinations of ps parameters . one caveat of our study is the assumption that the accretion disk around the bh is parallel to the orbital plane all the way down to the isco . @xcite showed that the inner parts of an accretion disk around a spinning bh are forced to align with the spin of the bh by the bardeen - petterson effect , and that the same torque that aligns the inner disk with the bh tends also to align the spin of the bh with the outer accretion disk . effectively this means that angular momentum is transfered to the bh from the outer parts of the disk . although this mechanism depends on the details of the accretion disk model adopted ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the overall effect is to accelerate the alignment of the bh spin compared to our calculations . in this sense , our calculation of bh spin evolution due to accretion pose an upper limit on the current bh spin - orbit misalignment in galactic bh xrbs . in recent years , several research groups have attempted to measure the bh spin magnitude in bh xrbs . both currently available methods for the measurement of bh spins depend on the bh spin - orbit misalignment angle @xcite . the x - ray continuum spectral fitting method @xcite assumes a full alignment between the bh spin and the orbital angular momentum . the fe k spectral line method @xcite does not make any explicit assumption , but its accuracy could be improved from additional constraints on the bh spin tilts . this letter is the first study that provides theoretical constraints on the bh spin - orbit misalignment in bh xrbs . the authors would like to thank prof . kalogera , dr . mcclintock , and dr . mandel for their comments on the manuscript and other helpful discussions that greatly improved this work . tf acknowledges support from the northwestern presidential fellowship . lcccccccccccc @xmath42 & & 8.8 & 48.8 & 133.0 & & 10.1 & 43.2 & 100.7 & & 2.2 & 55.3 & 134.5 + @xmath53 & & 3.2 & 19.0 & 90.0 & & 2.3 & 9.1 & 69.8 & & 1.7 & 28.8 & 123.6 + @xmath54 & & 4.5 & 26.6 & 92.0 & & 4.2 & 15.9 & 86.1 & & 2.1 & 36.6 & 128.4 + @xmath55 & & 5.7 & 32.7 & 92.6 & & 6.0 & 21.7 & 89.3 & & 2.1 & 39.5 & 133.7 + @xmath45 & & 3.2 & 20.1 & 90.0 & & 2.4 & 10.7 & 60.9 & & 1.6 & 32.5 & 123.0 + @xmath56 & & 4.3 & 26.4 & 92.5 & & 4.3 & 17.3 & 86.1 & & 2.0 & 36.7 & 127.5 + @xmath44 & & 4.9 & 28.1 & 92.5 & & 4.8 & 18.0 & 89.2 & & 2.1 & 37.4 & 130.7 + @xmath43 & & 3.9 & 17.9 & 27.8 & & 10.7 & 24.2 & 29.7 & & 3.7 & 19.0 & 26.3 + @xmath65 & & 2.1 & 11.3 & 20.6 & & 4.1 & 13.5 & 22.8 & & 1.4 & 8.0 & 13.7 + @xmath66 & & 2.7 & 13.9 & 23.4 & & 6.4 & 16.9 & 24.9 & & 2.7 & 12.5 & 20.5 + @xmath67 & & 3.1 & 15.5 & 24.9 & & 8.2 & 19.5 & 26.6 & & 4.3 & 15.5 & 22.9 + @xmath68 & & 2.1 & 11.4 & 21.6 & & 4.1 & 13.6 & 22.7 & & 1.6 & 8.4 & 15.5 + @xmath69 & & 2.6 & 13.9 & 24.3 & & 6.2 & 17.3 & 25.4 & & 2.8 & 12.8 & 20.2 + @xmath70 & & 2.8 & 14.6 & 24.7 & & 7.1 & 18.1 & 25.2 & & 2.9 & 13.6 & 22.8 + @xmath46 & & 11.5 & 65.1 & 133.7 & & 7.9 & 37.7 & 131.3 & & 4.1 & 46.6 & 113.8 + @xmath71 & & 4.4 & 33.7 & 119.7 & & 2.0 & 8.5 & 103.7 & & 1.9 & 33.5 & 115.1 + @xmath72 & & 7.2 & 52.7 & 126.6 & & 3.5 & 14.2 & 118.3 & & 3.5 & 45.8 & 118.6 + @xmath73 & & 8.9 & 58.6 & 128.8 & & 4.7 & 20.5 & 123.7 & & 3.9 & 48.6 & 120.1 + @xmath74 & & 4.4 & 33.7 & 121.8 & & 2.0 & 10.2 & 111.2 & & 2.1 & 34.5 & 100.3 + @xmath75 & & 6.9 & 50.3 & 128.8 & & 3.4 & 15.1 & 119.6 & & 3.2 & 45.6 & 118.5 + @xmath76 & & 8.1 & 57.7 & 115.3 & & 4.0 & 16.7 & 119.2 & & 3.5 & 48.0 & 118.7 + @xmath47 & & 11.4 & 61.7 & 135.8 & & 10.1 & 45.3 & 117.8 & & 3.2 & 62.5 & 128.3 + @xmath77 & & 3.9 & 27.9 & 103.6 & & 2.7 & 11.1 & 90.7 & & 1.9 & 89.7 & 124.7 + @xmath78 & & 6.1 & 38.8 & 111.8 & & 4.9 & 17.5 & 91.7 & & 2.2 & 89.8 & 125.2 + @xmath79 & & 7.8 & 44.6 & 119.1 & & 7.1 & 24.3 & 94.5 & & 3.5 & 89.8 & 125.2 + @xmath80 & & 4.0 & 28.0 & 107.1 & & 2.7 & 11.2 & 90.3 & & 1.9 & 87.5 & 124.2 + @xmath81 & & 6.0 & 35.9 & 111.6 & & 4.8 & 18.3 & 91.0 & & 2.0 & 89.8 & 125.2 + @xmath82 & & 6.7 & 41.0 & 112.9 & & 5.6 & 21.1 & 93.3 & & 3.1 & 89.8 & 125.2 + @xmath48 & & 1.9 & 17.4 & 94.4 & & 1.8 & 6.2 & 111.2 & & 1.4 & 16.2 & 101.6 + @xmath83 & & 1.7 & 8.7 & 51.1 & & 1.5 & 2.8 & 90.5 & & 1.3 & 19.9 & 94.8 + @xmath84 & & 1.8 & 12.4 & 53.2 & & 1.6 & 4.0 & 90.5 & & 1.4 & 24.0 & 98.7 + @xmath85 & & 1.9 & 14.6 & 55.7 & & 1.7 & 4.5 & 93.0 & & 1.4 & 25.6 & 99.3 + @xmath86 & & 1.7 & 9.0 & 51.6 & & 1.5 & 3.1 & 90.5 & & 1.3 & 19.6 & 96.6 + @xmath87 & & 1.8 & 12.8 & 53.9 & & 1.6 & 4.1 & 91.2 & & 1.3 & 21.8 & 97.6 + @xmath88 & & 1.8 & 13.6 & 55.3 & & 1.6 & 4.3 & 90.7 & & 1.3 & 25.5 & 99.3 + @xmath89 & & 2.8 & 11.3 & 24.6 & & 3.6 & 8.9 & 13.5 & & 1.1 & 3.5 & 6.7 + @xmath90 & & 1.8 & 6.5 & 14.5 & & 1.6 & 4.0 & 8.0 & & 1.1 & 2.2 & 4.4 + @xmath91 & & 2.1 & 8.3 & 18.8 & & 2.1 & 5.3 & 9.2 & & 1.1 & 2.2 & 6.6 + @xmath92 & & 2.4 & 9.5 & 21.8 & & 2.7 & 6.5 & 10.9 & & 1.1 & 2.9 & 6.7 + @xmath93 & & 1.7 & 6.6 & 14.9 & & 1.7 & 4.1 & 8.2 & & 1.1 & 2.2 & 4.4 + @xmath94 & & 2.0 & 8.4 & 18.1 & & 2.1 & 5.8 & 10.0 & & 1.1 & 2.6 & 4.9 + @xmath95 & & 2.2 & 8.7 & 21.8 & & 2.2 & 5.9 & 10.6 & & 1.1 & 2.2 & 6.7 + @xmath49 & & 5.7 & 31.8 & 128.6 & & 5.4 & 22.0 & 75.5 & & 1.8 & 46.6 & 127.0 + @xmath96 & & 2.5 & 16.0 & 91.9 & & 1.8 & 6.7 & 17.4 & & 1.3 & 33.2 & 113.7 + @xmath97 & & 3.8 & 21.8 & 97.1 & & 2.5 & 10.5 & 22.5 & & 1.4 & 40.2 & 121.4 + @xmath98 & & 4.4 & 25.6 & 102.6 & & 3.6 & 13.8 & 33.4 & & 1.6 & 43.0 & 123.9 + @xmath99 & & 2.5 & 16.8 & 91.2 & & 1.9 & 7.0 & 17.9 & & 1.3 & 32.8 & 111.4 + @xmath100 & & 3.7 & 21.8 & 97.3 & & 2.5 & 10.2 & 24.9 & & 1.4 & 37.8 & 122.9 + @xmath101 & & 4.0 & 23.7 & 103.7 & & 3.0 & 12.3 & 30.8 & & 1.4 & 40.6 & 123.1 + @xmath102 & & 4.3 & 15.3 & 29.7 & & 8.2 & 19.5 & 26.7 & & 2.1 & 10.7 & 29.0 + @xmath103 & & 2.4 & 9.7 & 23.6 & & 2.9 & 8.8 & 16.4 & & 1.0 & 4.6 & 22.7 + @xmath104 & & 3.3 & 11.4 & 26.9 & & 4.7 & 11.8 & 18.9 & & 1.2 & 6.2 & 27.1 + @xmath105 & & 3.9 & 12.9 & 28.8 & & 6.2 & 14.0 & 20.6 & & 1.3 & 6.7 & 28.9 + @xmath106 & & 2.4 & 9.9 & 23.8 & & 2.7 & 9.1 & 17.1 & & 0.9 & 5.5 & 23.8 + @xmath107 & & 3.3 & 11.7 & 26.8 & & 4.5 & 12.4 & 19.2 & & 1.3 & 5.6 & 27.7 + @xmath108 & & 3.5 & 11.9 & 27.4 & & 5.4 & 13.0 & 20.2 & & 1.3 & 6.2 & 29.0 + @xmath50 & & 11.5 & 61.0 & 132.3 & & 13.6 & 60.5 & 136.4 & & 3.9 & 64.3 & 160.0 + @xmath109 & & 4.5 & 31.8 & 102.7 & & 3.5 & 17.7 & 91.3 & & 1.9 & 21.5 & 110.1 + @xmath110 & & 7.2 & 47.1 & 109.7 & & 6.1 & 28.5 & 92.4 & & 1.9 & 29.6 & 117.1 + @xmath111 & & 9.0 & 52.7 & 113.8 & & 8.5 & 38.4 & 95.4 & & 3.0 & 34.1 & 117.1 + @xmath112 & & 4.5 & 36.4 & 101.1 & & 3.6 & 18.7 & 91.3 & & 1.9 & 21.9 & 113.3 + @xmath113 & & 6.7 & 45.9 & 109.7 & & 5.8 & 31.0 & 91.4 & & 1.9 & 24.7 & 117.1 + @xmath114 & & 7.9 & 51.7 & 112.8 & & 6.8 & 34.6 & 92.6 & & 2.1 & 33.9 & 117.1 + @xmath115 & & 4.0 & 20.5 & 30.3 & & 10.4 & 26.3 & 32.0 & & 4.0 & 22.9 & 35.4 + @xmath116 & & 2.2 & 13.4 & 26.0 & & 4.4 & 14.7 & 25.7 & & 1.5 & 10.8 & 22.6 + @xmath117 & & 2.8 & 15.8 & 28.8 & & 6.7 & 18.9 & 27.3 & & 2.2 & 19.0 & 22.8 + @xmath118 & & 3.1 & 17.5 & 30.3 & & 8.1 & 21.7 & 28.9 & & 2.4 & 21.1 & 27.2 + @xmath119 & & 3.6 & 12.8 & 23.1 & & 4.3 & 15.1 & 25.9 & & 1.5 & 9.5 & 22.6 + @xmath120 & & 5.5 & 15.8 & 25.8 & & 6.1 & 19.2 & 27.6 & & 2.2 & 18.6 & 23.5 + @xmath121 & & 7.1 & 17.3 & 26.0 & & 7.1 & 20.5 & 28.4 & & 2.0 & 20.6 & 22.9 + @xmath51 & & 10.9 & 57.7 & 126.5 & & 10.1 & 43.8 & 97.7 & & 2.2 & 58.8 & 118.4 + @xmath122 & & 3.9 & 29.6 & 97.6 & & 2.3 & 10.8 & 91.2 & & 1.4 & 45.2 & 105.7 + @xmath123 & & 6.0 & 42.4 & 108.5 & & 4.1 & 19.0 & 91.3 & & 1.6 & 53.2 & 107.5 + @xmath124 & & 7.7 & 51.0 & 128.9 & & 5.7 & 25.2 & 92.2 & & 1.9 & 55.6 & 108.4 + @xmath125 & & 3.9 & 28.6 & 98.9 & & 2.2 & 12.9 & 91.1 & & 1.6 & 46.4 & 106.6 + @xmath126 & & 6.0 & 42.4 & 107.9 & & 4.1 & 17.2 & 91.3 & & 1.7 & 52.5 & 107.7 + @xmath127 & & 6.9 & 48.1 & 122.2 & & 4.5 & 21.3 & 92.4 & & 1.9 & 55.2 & 108.4 + @xmath128 & & 5.7 & 19.6 & 29.4 & & 11.5 & 24.0 & 30.0 & & 7.5 & 20.4 & 26.7 + @xmath129 & & 3.3 & 13.6 & 24.4 & & 4.2 & 12.8 & 22.8 & & 2.1 & 9.8 & 18.3 + @xmath130 & & 4.3 & 16.3 & 26.8 & & 6.9 & 16.8 & 24.9 & & 3.8 & 12.1 & 24.2 + @xmath131 & & 4.8 & 17.8 & 27.6 & & 9.0 & 20.0 & 26.4 & & 4.0 & 15.2 & 26.4 + @xmath132 & & 3.2 & 13.9 & 24.3 & & 4.3 & 14.5 & 24.1 & & 1.9 & 9.8 & 20.1 + @xmath133 & & 4.2 & 15.9 & 26.1 & & 6.7 & 17.5 & 25.5 & & 3.4 & 15.2 & 22.4 + @xmath134 & & 4.5 & 17.5 & 27.3 & & 7.6 & 18.0 & 25.1 & & 4.0 & 14.5 & 21.6 + @xmath135 & & 2.2 & 25.2 & 78.6 & & 1.9 & 6.9 & 83.1 & & 1.7 & 26.0 & 112.9 + @xmath136 & & 1.8 & 16.4 & 34.1 & & 1.5 & 3.3 & 28.8 & & 1.5 & 16.1 & 112.4 + @xmath137 & & 2.0 & 24.0 & 42.0 & & 1.7 & 4.3 & 68.5 & & 1.5 & 18.2 & 113.1 + @xmath138 & & 2.1 & 25.0 & 67.0 & & 1.8 & 5.3 & 78.3 & & 1.5 & 20.5 & 113.1 + @xmath139 & & 1.8 & 18.4 & 34.4 & & 1.5 & 3.5 & 50.0 & & 1.5 & 16.6 & 112.4 + @xmath140 & & 2.0 & 23.1 & 38.4 & & 1.7 & 4.4 & 61.5 & & 1.5 & 18.8 & 113.1 + @xmath141 & & 2.0 & 24.2 & 68.0 & & 1.7 & 4.5 & 78.6 & & 1.5 & 19.9 & 113.1 + @xmath142 & & 3.3 & 13.6 & 25.1 & & 3.9 & 9.5 & 15.5 & & 1.4 & 6.0 & 13.7 + @xmath143 & & 1.8 & 7.5 & 14.9 & & 1.7 & 4.4 & 8.4 & & 0.9 & 2.1 & 9.4 + @xmath144 & & 2.2 & 9.9 & 17.9 & & 2.2 & 6.2 & 10.1 & & 1.0 & 2.2 & 13.6 + @xmath145 & & 2.6 & 11.5 & 20.1 & & 3.1 & 7.2 & 11.6 & & 1.0 & 3.0 & 15.9 + @xmath146 & & 1.9 & 7.4 & 15.4 & & 1.7 & 4.7 & 8.7 & & 0.9 & 2.1 & 10.2 + @xmath147 & & 2.1 & 10.2 & 19.0 & & 2.2 & 6.3 & 9.8 & & 1.0 & 2.2 & 12.8 + @xmath148 & & 2.2 & 10.9 & 20.1 & & 2.7 & 6.5 & 10.6 & & 1.0 & 3.0 & 15.9 + @xmath149 & & 7.3 & 29.7 & 103.7 & & 6.3 & 23.4 & 102.3 & & 2.2 & 37.0 & 144.5 + @xmath150 & & 2.7 & 16.0 & 94.7 & & 2.0 & 7.6 & 23.2 & & 2.1 & 28.1 & 123.6 + @xmath151 & & 4.2 & 20.9 & 99.3 & & 3.2 & 12.2 & 89.6 & & 3.2 & 31.7 & 137.3 + @xmath152 & & 5.2 & 24.4 & 101.6 & & 4.3 & 17.0 & 89.6 & & 3.6 & 34.3 & 140.4 + @xmath153 & & 2.8 & 16.7 & 96.8 & & 2.1 & 8.5 & 89.6 & & 1.9 & 29.2 & 125.9 + @xmath154 & & 4.0 & 21.7 & 101.6 & & 3.2 & 11.8 & 89.6 & & 2.9 & 29.7 & 139.6 + @xmath155 & & 4.5 & 22.8 & 101.6 & & 3.7 & 13.9 & 89.6 & & 3.3 & 33.3 & 138.3 + @xmath156 & & 5.1 & 17.0 & 34.1 & & 9.0 & 19.5 & 26.3 & & 2.7 & 13.0 & 20.5 + @xmath157 & & 3.1 & 10.6 & 25.6 & & 3.4 & 9.0 & 15.8 & & 0.3 & 4.4 & 15.0 + @xmath158 & & 4.2 & 12.9 & 30.6 & & 5.4 & 12.1 & 17.9 & & 0.5 & 4.5 & 23.5 + @xmath159 & & 5.0 & 14.2 & 33.3 & & 6.9 & 14.3 & 20.0 & & 0.7 & 6.8 & 27.2 + @xmath160 & & 3.1 & 10.8 & 26.3 & & 3.5 & 9.6 & 16.1 & & 0.3 & 4.4 & 12.2 + @xmath161 & & 4.2 & 13.0 & 30.6 & & 5.4 & 12.3 & 18.2 & & 0.5 & 4.5 & 23.1 + @xmath162 & & 4.5 & 13.6 & 32.3 & & 6.0 & 13.4 & 20.2 & & 0.7 & 6.7 & 26.1 + @xmath163 & & 13.5 & 71.6 & 139.4 & & 13.3 & 64.0 & 138.2 & & 3.5 & 39.1 & 133.3 + @xmath164 & & 5.5 & 38.8 & 98.3 & & 3.7 & 20.6 & 104.5 & & 1.9 & 27.8 & 104.2 + @xmath165 & & 8.7 & 50.4 & 108.3 & & 6.7 & 30.8 & 121.9 & & 2.3 & 34.0 & 112.3 + @xmath166 & & 11.0 & 55.5 & 114.9 & & 9.5 & 43.2 & 129.7 & & 2.5 & 37.5 & 117.9 + @xmath167 & & 5.5 & 38.5 & 97.3 & & 3.8 & 21.5 & 116.4 & & 1.9 & 31.5 & 106.2 + @xmath168 & & 8.4 & 50.8 & 105.8 & & 6.4 & 31.1 & 124.9 & & 2.3 & 35.1 & 108.3 + @xmath169 & & 9.6 & 54.2 & 108.7 & & 7.6 & 34.4 & 119.9 & & 3.2 & 36.6 & 111.3 + @xmath170 & & 4.5 & 20.0 & 33.1 & & 12.3 & 26.4 & 33.0 & & 8.2 & 26.5 & 29.7 + @xmath171 & & 3.7 & 14.3 & 25.2 & & 5.4 & 15.6 & 23.8 & & 2.8 & 11.9 & 16.1 + @xmath172 & & 4.8 & 17.1 & 28.3 & & 8.7 & 19.1 & 25.2 & & 5.2 & 16.5 & 25.2 + @xmath173 & & 5.5 & 18.8 & 29.5 & & 11.0 & 21.8 & 27.4 & & 7.4 & 17.8 & 27.7 + @xmath174 & & 3.6 & 14.5 & 25.2 & & 4.8 & 15.0 & 22.9 & & 3.3 & 14.2 & 16.0 + @xmath175 & & 4.6 & 17.0 & 29.0 & & 8.4 & 19.4 & 25.8 & & 3.7 & 17.2 & 25.4 + @xmath176 & & 5.2 & 18.0 & 29.3 & & 9.4 & 20.7 & 26.9 & & 6.0 & 17.8 & 27.5 + @xmath52 & & 1.8 & 29.1 & 87.7 & & 2.2 & 25.4 & 81.6 & & 1.7 & 8.0 & 86.5 + @xmath177 & & 1.3 & 13.3 & 83.3 & & 1.1 & 6.4 & 56.5 & & 1.5 & 2.1 & 46.2 + @xmath178 & & 1.6 & 18.5 & 85.8 & & 1.5 & 10.5 & 76.0 & & 1.5 & 2.7 & 54.4 + @xmath179 & & 1.7 & 22.1 & 87.0 & & 1.8 & 14.2 & 78.1 & & 1.5 & 4.0 & 55.9 + @xmath180 & & 1.3 & 13.4 & 82.1 & & 1.2 & 6.7 & 58.5 & & 1.5 & 2.2 & 48.0 + @xmath181 & & 1.6 & 18.5 & 85.4 & & 1.4 & 11.0 & 60.7 & & 1.5 & 2.7 & 49.2 + @xmath182 & & 1.6 & 19.8 & 86.8 & & 1.6 & 12.3 & 58.6 & & 1.5 & 3.5 & 55.6 + @xmath183 & & 0.4 & 14.1 & 25.0 & & 1.1 & 19.6 & 28.6 & & 0.0 & 8.6 & 27.8 + @xmath184 & & 0.2 & 8.7 & 20.8 & & 0.6 & 8.1 & 17.4 & & 0.0 & 2.2 & 6.9 + @xmath185 & & 0.2 & 10.6 & 23.0 & & 0.9 & 11.4 & 20.0 & & 0.0 & 4.2 & 13.1 + @xmath186 & & 0.2 & 11.9 & 24.3 & & 1.1 & 14.2 & 22.2 & & 0.0 & 6.4 & 17.5 + @xmath187 & & 0.2 & 8.8 & 21.6 & & 0.6 & 8.4 & 17.5 & & 0.0 & 2.2 & 10.2 + @xmath188 & & 0.2 & 10.7 & 23.2 & & 0.8 & 11.5 & 20.3 & & 0.0 & 4.4 & 15.2 + @xmath189 & & 0.2 & 11.0 & 23.4 & & 0.9 & 12.9 & 21.6 & & 0.0 & 5.5 & 19.5 + @xmath190 & & 0.7 & 5.6 & 53.0 & & 0.6 & 3.9 & 9.1 & & 1.7 & 1.9 & 35.3 + @xmath191 & & 0.6 & 2.8 & 24.5 & & 0.6 & 2.1 & 4.5 & & 0.0 & 2.0 & 26.8 + @xmath192 & & 0.7 & 3.9 & 38.0 & & 0.6 & 2.3 & 6.4 & & 0.0 & 2.0 & 31.7 + @xmath193 & & 0.7 & 4.4 & 43.6 & & 0.6 & 3.0 & 7.9 & & 0.0 & 2.0 & 34.6 + @xmath194 & & 0.6 & 2.9 & 21.9 & & 0.6 & 2.1 & 5.1 & & 1.5 & 2.0 & 27.8 + @xmath195 & & 0.7 & 3.9 & 35.7 & & 0.6 & 2.3 & 6.8 & & 1.5 & 2.0 & 31.2 + @xmath196 & & 0.7 & 4.1 & 41.4 & & 0.6 & 2.4 & 7.0 & & 1.5 & 2.0 & 32.8 + @xmath197 & & 0.9 & 7.6 & 23.5 & & 1.1 & 6.8 & 12.8 & & 0.0 & 1.9 & 4.6 + @xmath198 & & 0.7 & 4.2 & 12.8 & & 0.6 & 2.8 & 6.5 & & 0.0 & 1.7 & 2.3 + @xmath199 & & 0.8 & 5.3 & 17.6 & & 0.7 & 4.1 & 7.9 & & 0.0 & 1.7 & 3.0 + @xmath200 & & 0.9 & 6.2 & 20.7 & & 0.9 & 5.1 & 9.0 & & 0.0 & 1.9 & 4.3 + @xmath201 & & 0.7 & 4.2 & 13.6 & & 0.6 & 3.0 & 6.6 & & 0.0 & 1.7 & 2.3 + @xmath202 & & 0.8 & 5.4 & 16.6 & & 0.7 & 4.2 & 8.2 & & 0.0 & 1.8 & 4.0 + @xmath203 & & 0.8 & 5.7 & 19.5 & & 0.8 & 4.5 & 8.6 & & 0.0 & 1.9 & 4.2 + @xmath204 & & 1.5 & 18.3 & 69.1 & & 1.6 & 14.9 & 40.5 & & 1.7 & 4.5 & 62.2 + @xmath205 & & 1.2 & 8.3 & 43.3 & & 0.9 & 4.2 & 15.2 & & 1.5 & 2.0 & 43.3 + @xmath206 & & 1.4 & 11.7 & 56.9 & & 1.1 & 6.8 & 21.8 & & 1.5 & 2.2 & 50.5 + @xmath207 & & 1.5 & 13.9 & 62.7 & & 1.3 & 9.3 & 28.0 & & 1.5 & 3.0 & 53.4 + @xmath208 & & 1.2 & 8.7 & 45.7 & & 0.9 & 4.4 & 15.8 & & 1.5 & 2.1 & 40.6 + @xmath209 & & 1.4 & 12.0 & 60.8 & & 1.1 & 6.8 & 22.2 & & 1.5 & 2.2 & 51.8 + @xmath210 & & 1.4 & 12.8 & 58.3 & & 1.1 & 8.2 & 25.2 & & 1.5 & 2.3 & 51.4 + @xmath211 & & 0.8 & 11.0 & 21.6 & & 1.5 & 15.8 & 24.9 & & 0.0 & 5.8 & 20.9 + @xmath212 & & 0.6 & 6.7 & 14.7 & & 0.7 & 6.3 & 14.5 & & 0.0 & 2.0 & 10.7 + @xmath213 & & 0.7 & 8.4 & 17.0 & & 1.1 & 9.0 & 17.1 & & 0.0 & 3.0 & 14.0 + @xmath214 & & 0.8 & 9.4 & 18.1 & & 1.3 & 11.3 & 18.6 & & 0.0 & 4.3 & 16.4 + @xmath215 & & 0.6 & 6.7 & 15.4 & & 0.7 & 6.5 & 14.8 & & 0.0 & 1.9 & 10.7 + @xmath216 & & 0.7 & 8.4 & 16.7 & & 1.0 & 9.3 & 17.2 & & 0.0 & 3.1 & 16.0 + @xmath217 & & 0.8 & 8.8 & 17.8 & & 1.2 & 10.2 & 18.0 & & 0.0 & 3.5 & 16.2 + @xmath218 & & 1.1 & 34.0 & 112.6 & & 2.6 & 36.0 & 100.1 & & 1.7 & 9.4 & 102.1 + @xmath219 & & 0.6 & 12.5 & 92.8 & & 1.2 & 8.9 & 90.1 & & 0.0 & 4.0 & 43.2 + @xmath220 & & 0.8 & 19.9 & 100.6 & & 1.6 & 15.8 & 90.5 & & 0.0 & 6.2 & 55.4 + @xmath221 & & 0.9 & 25.7 & 106.0 & & 1.9 & 21.8 & 90.9 & & 0.0 & 8.4 & 59.6 + @xmath222 & & 0.6 & 13.7 & 93.2 & & 1.2 & 9.7 & 89.9 & & 1.5 & 3.8 & 48.3 + @xmath223 & & 0.8 & 19.9 & 104.3 & & 1.5 & 16.2 & 90.5 & & 1.5 & 6.3 & 59.6 + @xmath224 & & 0.8 & 22.1 & 104.3 & & 1.7 & 19.2 & 90.6 & & 1.5 & 9.5 & 59.6 + @xmath225 & & 0.4 & 15.9 & 28.5 & & 1.1 & 21.9 & 31.5 & & 0.0 & 13.3 & 33.1 + @xmath226 & & 0.2 & 10.2 & 23.9 & & 1.1 & 10.2 & 23.5 & & 0.0 & 3.9 & 9.8 + @xmath227 & & 0.2 & 12.3 & 26.3 & & 1.1 & 14.2 & 26.6 & & 0.0 & 6.7 & 18.8 + @xmath228 & & 0.2 & 13.7 & 27.4 & & 1.1 & 17.1 & 28.9 & & 0.0 & 10.5 & 23.6 + @xmath229 & & 0.2 & 10.3 & 24.2 & & 0.9 & 10.1 & 24.0 & & 0.0 & 4.0 & 11.0 + @xmath230 & & 0.2 & 12.4 & 26.5 & & 0.9 & 14.8 & 27.7 & & 0.0 & 6.7 & 16.8 + @xmath231 & & 0.2 & 12.9 & 26.9 & & 0.9 & 15.6 & 27.8 & & 0.0 & 9.8 & 23.2 + @xmath232 & & 2.9 & 35.2 & 115.2 & & 2.9 & 29.4 & 96.9 & & 1.7 & 8.2 & 83.9 + @xmath233 & & 1.6 & 14.0 & 89.6 & & 1.3 & 7.1 & 79.9 & & 0.0 & 3.3 & 89.8 + @xmath234 & & 2.0 & 20.4 & 90.9 & & 1.8 & 12.6 & 89.9 & & 0.0 & 5.4 & 92.1 + @xmath235 & & 2.2 & 25.0 & 94.3 & & 2.1 & 18.0 & 90.1 & & 0.0 & 7.5 & 92.6 + @xmath236 & & 1.6 & 14.5 & 89.5 & & 1.4 & 7.8 & 86.4 & & 1.5 & 3.1 & 90.2 + @xmath237 & & 1.9 & 20.2 & 90.3 & & 1.7 & 12.2 & 89.9 & & 1.5 & 6.0 & 91.8 + @xmath238 & & 2.1 & 23.2 & 92.1 & & 1.9 & 15.0 & 89.9 & & 1.5 & 5.4 & 92.9 + @xmath239 & & 0.9 & 14.1 & 26.0 & & 1.6 & 20.4 & 28.8 & & 0.0 & 5.9 & 19.8 + @xmath240 & & 0.7 & 9.4 & 18.6 & & 0.7 & 8.6 & 17.4 & & 0.0 & 2.0 & 8.3 + @xmath241 & & 0.8 & 11.4 & 20.1 & & 1.1 & 11.9 & 20.3 & & 0.0 & 3.4 & 13.1 + @xmath242 & & 0.9 & 12.7 & 20.7 & & 1.3 & 14.3 & 22.5 & & 0.0 & 3.7 & 16.4 + @xmath243 & & 0.7 & 9.5 & 18.7 & & 0.7 & 8.8 & 17.6 & & 0.0 & 2.0 & 11.9 + @xmath244 & & 0.8 & 11.4 & 19.9 & & 1.0 & 12.1 & 20.5 & & 0.0 & 2.8 & 12.5 + @xmath245 & & 0.9 & 12.1 & 20.4 & & 1.1 & 13.1 & 22.0 & & 0.0 & 3.4 & 15.5 + @xmath246 & & 0.9 & 6.4 & 56.8 & & 0.9 & 4.3 & 15.7 & & 1.7 & 2.0 & 45.1 + @xmath247 & & 0.7 & 3.3 & 39.2 & & 0.6 & 2.2 & 6.8 & & 0.0 & 1.9 & 37.7 + @xmath248 & & 0.8 & 4.4 & 50.6 & & 0.6 & 2.3 & 10.6 & & 0.0 & 1.9 & 40.3 + @xmath249 & & 0.8 & 5.1 & 53.9 & & 0.6 & 3.0 & 12.8 & & 0.0 & 1.9 & 42.7 + @xmath250 & & 0.7 & 3.5 & 41.2 & & 0.6 & 2.2 & 7.9 & & 1.5 & 1.9 & 37.9 + @xmath251 & & 0.8 & 4.3 & 42.0 & & 0.6 & 2.3 & 9.9 & & 1.5 & 1.9 & 41.4 + @xmath252 & & 0.8 & 4.6 & 53.2 & & 0.6 & 2.5 & 12.3 & & 1.5 & 1.9 & 40.1 + @xmath253 & & 1.4 & 8.4 & 20.7 & & 1.4 & 7.2 & 13.1 & & 0.0 & 2.1 & 4.4 + @xmath254 & & 1.0 & 4.4 & 13.8 & & 0.7 & 2.8 & 6.5 & & 0.0 & 1.7 & 2.3 + @xmath255 & & 1.1 & 5.9 & 15.9 & & 0.8 & 4.2 & 8.4 & & 0.0 & 1.7 & 2.3 + @xmath256 & & 1.2 & 6.5 & 18.0 & & 1.0 & 5.1 & 10.2 & & 0.0 & 1.8 & 3.2 + @xmath257 & & 1.0 & 4.5 & 13.9 & & 0.7 & 3.0 & 6.7 & & 0.0 & 1.7 & 2.3 + @xmath258 & & 1.1 & 5.9 & 16.7 & & 0.8 & 4.2 & 8.6 & & 0.0 & 1.7 & 2.3 + @xmath259 & & 1.2 & 6.2 & 16.9 & & 0.9 & 4.5 & 9.1 & & 0.0 & 1.8 & 3.2 + @xmath260 & & 1.8 & 18.3 & 81.3 & & 1.8 & 15.5 & 53.1 & & 1.7 & 5.0 & 110.4 + @xmath261 & & 1.3 & 9.0 & 34.1 & & 1.0 & 4.2 & 14.8 & & 0.0 & 2.2 & 99.7 + @xmath262 & & 1.5 & 12.6 & 41.6 & & 1.2 & 6.9 & 21.8 & & 0.0 & 3.0 & 106.5 + @xmath263 & & 1.6 & 15.0 & 45.3 & & 1.4 & 9.7 & 26.6 & & 0.0 & 4.4 & 108.8 + @xmath264 & & 1.3 & 9.0 & 32.6 & & 1.1 & 4.4 & 14.6 & & 1.5 & 2.1 & 104.2 + @xmath265 & & 1.5 & 13.2 & 43.7 & & 1.2 & 6.8 & 22.5 & & 1.5 & 3.6 & 102.0 + @xmath266 & & 1.5 & 14.0 & 42.6 & & 1.3 & 8.3 & 24.8 & & 1.5 & 4.2 & 108.8 + @xmath267 & & 1.1 & 11.4 & 22.0 & & 2.1 & 15.9 & 24.7 & & 0.0 & 5.7 & 18.6 + @xmath268 & & 0.7 & 6.9 & 16.1 & & 1.0 & 6.0 & 13.7 & & 0.0 & 2.0 & 6.0 + @xmath269 & & 0.9 & 8.8 & 18.2 & & 1.4 & 8.6 & 16.3 & & 0.0 & 3.0 & 10.6 + @xmath270 & & 1.0 & 10.1 & 19.4 & & 1.7 & 11.0 & 18.1 & & 0.0 & 3.6 & 15.1 + @xmath271 & & 0.7 & 6.9 & 16.1 & & 1.0 & 6.2 & 13.7 & & 0.0 & 2.0 & 8.2 + @xmath272 & & 0.9 & 8.9 & 18.4 & & 1.4 & 8.8 & 17.2 & & 0.0 & 3.3 & 12.7 + @xmath273 & & 1.0 & 9.4 & 18.8 & & 1.5 & 9.6 & 17.5 & & 0.0 & 3.3 & 15.0 + @xmath274 & & 2.2 & 39.7 & 124.6 & & 3.9 & 39.6 & 117.2 & & 1.7 & 8.8 & 62.2 + @xmath275 & & 1.3 & 13.9 & 92.7 & & 1.4 & 9.0 & 90.8 & & 0.0 & 3.3 & 49.0 + @xmath276 & & 1.7 & 21.4 & 95.6 & & 1.9 & 15.5 & 91.1 & & 0.0 & 5.3 & 53.4 + @xmath277 & & 1.9 & 26.2 & 96.8 & & 2.2 & 21.8 & 92.0 & & 0.0 & 7.5 & 55.2 + @xmath278 & & 1.3 & 14.8 & 92.5 & & 1.3 & 9.4 & 90.5 & & 1.5 & 3.1 & 46.1 + @xmath279 & & 1.6 & 21.2 & 96.6 & & 1.8 & 17.1 & 91.1 & & 1.5 & 5.5 & 55.2 + @xmath280 & & 1.8 & 23.0 & 95.8 & & 2.0 & 18.6 & 91.3 & & 1.5 & 7.0 & 53.7 + @xmath281 & & 0.9 & 18.0 & 27.2 & & 0.1 & 20.7 & 32.7 & & 0.0 & 13.8 & 30.1 + @xmath282 & & 0.9 & 12.3 & 22.4 & & 1.3 & 9.9 & 22.9 & & 0.0 & 4.0 & 8.7 + @xmath283 & & 0.9 & 15.3 & 24.6 & & 1.3 & 13.4 & 26.8 & & 0.0 & 6.8 & 13.4 + @xmath284 & & 0.9 & 16.9 & 26.0 & & 1.3 & 15.6 & 29.7 & & 0.0 & 8.9 & 19.8 + @xmath285 & & 0.9 & 12.1 & 22.5 & & 1.1 & 10.2 & 24.0 & & 0.0 & 4.4 & 10.1 + @xmath286 & & 0.9 & 14.9 & 24.8 & & 1.1 & 13.4 & 27.9 & & 0.0 & 6.5 & 17.5 + @xmath287 & & 0.9 & 15.8 & 25.1 & & 1.1 & 14.9 & 29.4 & & 0.0 & 8.6 & 15.2 + @xmath288 & & 25.1 & 73.0 & 140.3 & & 26.7 & 78.0 & 123.5 & & 26.5 & 115.3 & 119.1 + @xmath289 & & 10.2 & 43.2 & 104.9 & & 6.3 & 34.0 & 91.3 & & 5.9 & 25.0 & 113.2 + @xmath290 & & 15.2 & 53.2 & 116.1 & & 11.6 & 50.6 & 92.6 & & 11.3 & 36.6 & 117.1 + @xmath291 & & 18.4 & 58.4 & 136.6 & & 16.0 & 58.9 & 95.8 & & 16.9 & 41.2 & 118.0 + @xmath292 & & 11.0 & 43.5 & 105.0 & & 6.4 & 35.2 & 91.3 & & 8.2 & 31.9 & 112.6 + @xmath57 & & 15.3 & 52.2 & 118.4 & & 11.6 & 50.8 & 93.4 & & 12.8 & 41.2 & 118.0 + @xmath293 & & 16.7 & 55.7 & 134.0 & & 13.3 & 55.3 & 94.3 & & 13.1 & 41.2 & 115.4 +
we present deep spectropolarimetric observations obtained with the w. m. keck telescope of the very high redshift (=3.79786.0024 ) radio galaxy c. we find that the bright , spatially extended rest - frame uv continuum emission from this galaxy , which is aligned with the radio axis , is unpolarized ( ) . this implies that scattered agn light , which is generally the dominant contributor to the rest - frame uv emission in radio galaxies , is unlikely to be a major component of the uv flux from c. the resulting total light spectrum shows absorption lines and p - cygni like features that are similar to those detected in the spectra of the recently discovered population of star forming galaxies at slightly lower ( ) redshifts . it may be possible for a galaxian outflow to contribute partially to the absorption line profiles of the low ionization species ; however , it is unlikely that the high velocity wings of the high ionization lines are dominated by a galaxian wind since the outflow mass implied by the absorption line strengths is very large . the detection of the stellar photospheric absorption line , the shape of the blue wing of the profile , the unpolarized continuum emission , the inability of any agn - related processes to account for the uv continuum flux , and the overall similarity of the uv continuum spectra of c and the nearby star - forming region ngc 1741b1 strongly suggest that the uv light from c is dominated by young , hot stars . if all of the uv emission is due to starlight from a young population , the implied star - formation rate is roughly 140 1100 /yr . the deep spectroscopy presented here combined with the morphology of the system at radio and optical wavelengths and the possibly comparable ages for the radio source structure and the uv stellar population suggest that star formation in c was triggered by the expansion of the radio source into the ambient medium . our current observations are consistent with the hypothesis that c is undergoing its major epoch of star formation at , and that by it will have evolved to have spectral and morphological properties similar to those observed in known powerful radio galaxies . -0.4 in 0.0 in 0.0 in 9.0 in 6.5 in c4c 41.17 # 1 # 1 # 1to 0pt#1 3 2 3mpch^-3mpc^3 0 0 # 1 # 110^#1 # 1to 0pt#1 a&a # 1;#2;#3;#4 # 1 , # 2 , # 3 , # 4 # 1#2#3#4 , # 3 , # 4 # 1#2#3 # 1 accepted for publication in the december 1 , 1997 apj
astro-ph9707166
until recently , radio source samples provided astronomers with the only known galaxy like objects at cosmological distances . currently , three of the four brightest known visible galaxies at redshifts @xmath13 are radio galaxies . one of the most intriguing properties of powerful radio galaxies at high - redshift is that their spatially extended uv morphology is preferentially aligned with the major axis of their radio emission ( mccarthy 1987 , chambers 1987 ) . several possible explanations have been proposed for this ` alignment effect ' ( see mccarthy 1993 for a recent review ) ; the two most compelling of these are ( 1 ) that the uv morphology is dominated by dust- and electron - scattered light from an anisotropically radiating nuclear source ( e.g. , tadhunter 1988 ) , and ( 2 ) that the uv morphology is the result of star - formation triggered by the radio source as it expands into the dense ambient medium ( e.g. , de young 1981 , 1989 ; rees 1989 ; begelman & cioffi 1989 ) . although there is circumstantial evidence at low redshift for radio sources triggering star - formation ( e.g. , in minkowski s object , van breugel 1985 ; in 3c 285 , van breugel & dey 1993 ; and in the abell 1795 cd galaxy , mcnamara 1996 ) , the majority of powerful @xmath3 radio galaxies investigated exhibit strong polarization ( @xmath14 ) of the continuum emission and broad emission lines , a result which confirms that much of the spatially extended rest - frame uv continuum emission in these systems is scattered light from a hidden agn ( di serego alighieri 1989 , tadhunter 1988 , jannuzi & elston 1991 , jannuzi 1995 , dey 1996 , cimatti 1996 , 1997 ) . the fractional polarization in most of the @xmath15 powerful radio galaxies investigated thus far show either no significant variation of polarization with wavelength ( , @xmath16 ) or , in a few cases , polarization increasing towards shorter wavelengths . in addition , @xmath3 radio galaxies of lower radio power ( , in which the agn makes an insignificant contribution to the uv light ) show underlying stellar populations that are both dynamically and spectrally old ( , spinrad 1997 , dunlop 1996 , rigler 1992 ) , providing further evidence that triggered star - formation is not the dominant process responsible for the alignment effect in @xmath3 radio galaxies . despite the success of the scattering hypothesis in explaining the alignment effect in most @xmath3 radio galaxies , the processes responsible for its origin in the highest redshift systems remain open to debate . the @xmath17 radio galaxies are faint , and until now have remained beyond the reach of most modern polarimeters . in this paper , we present spectropolarimetric observations of the @xmath18 radio galaxy c. the broad - band spectral energy distribution of c was first investigated by chambers ( 1990 ) , who suggested that starlight from a young population produced the bulk of the observed optical and near - ir continuum emission , and that the alignment of the uv morphology with the radio source axis is most likely due to star - formation triggered by the radio source . although there is close spatial coincidence between the compact uv and radio continuum emitting components ( miley 1992 ) , the observed uv emission is vastly in excess of that which can be produced by synchrotron emission from relativistic electrons responsible for the radio emission , or inverse compton scattering of the radio or cosmic microwave background photons off these electrons ( carilli 1994 ) . our data demonstrate that the uv light in this system is not dominated by scattered light from an active nucleus , but instead is very likely to be dominated by starlight from a young stellar population . we discuss our observations of this high - redshift galaxy in the context of a scenario where the @xmath7 powerful radio galaxies are the progenitors of the @xmath3 population . throughout this paper we assume that = 50 and = 0.1 . the scale at @xmath18 is then 10.8@xmath19 kpc/ and the lookback time is nearly 14.3@xmath19 gyr , or 86% of the age of the universe . in comparison , for = 0.5 the angular scale is 6.6@xmath19 kpc/ and the lookback time is 11.8@xmath19 gyr , or 90% of the age of the universe . we observed c on the nights of u.t . 1996 december 9 and 10 using the spectropolarimetric mode of the low resolution imaging spectrometer ( oke 1995 , goodrich 1995 ) at the cassegrain focus of the keck ii telescope . the observations were obtained using the 400 line / mm grating ( @xmath20 ; dispersion = 1.86 / pixel ) to sample the wavelength range @xmath2155009280 ( corresponding to @xmath2111461933 in the rest frame of c ) , and a 1 wide slit ( effective resolution fwhm in the observed frame of @xmath228 ) . the lris detector is a tek 2048@xmath23 ccd with 24@xmath24 m pixels with a pixel scale of 0214 pix@xmath25 . the ccd read noise during our run was 6.2@xmath26 and the gain was @xmath27/adu . over the two nights , we obtained a total of seven sets of observations ( three on dec . 9 and four on dec . 10 ) with the spectrograph slit oriented along the major axis of the rest - frame uv continuum emission from the galaxy ( pa=76.5@xmath28 ) . each set is comprised of observations made in four waveplate positions ( @xmath29 , @xmath30 , @xmath31 , @xmath32 ) which sample the electric vector in four position angles on the sky ( @xmath33 , @xmath34 , @xmath35 , @xmath36 respectively ) . the exposure time per waveplate position was 20 minutes for each set , resulting in a total exposure time of 140 minutes per waveplate position , and 9.33 hours for the total spectrum . after two sets of observations , we reacquired the galaxy using an offset from a nearby star . our first night was affected by minor telescope focus problems and poor seeing , but the seeing during our second night was good ( @xmath37 0.8 1 ) and conditions were photometric . during our observations the parallactic angle rotated during our observations from pa@xmath38 to @xmath3975 ( hour angle @xmath40 to @xmath41 ) , and the maximum airmass was 1.45 . in order to calibrate the instrumental polarization position angle ( @xmath42 ) correction and the polarization efficiency , we observed the star bd+28@xmath434211 through uv and ir polaroid filters in four waveplate positions . we find the @xmath42 correction to be very stable ( consistent with that measured on previous runs with the polarimeter ) and the polarization measurement efficiency to be @xmath44 ; hence , no correction for the latter was made to the data . we also observed the zero polarization standard gd 319 as a check on any residual polarization effects : our measurement yields @xmath45% ( 2@xmath46 ) . the electric vector position angle zero point offset was calibrated using observations of the polarization standard stars hd204827 and hd245310 ( schmidt , elston , & lupie 1992 ) ; our measurements are in very good agreement with the published values . the data were corrected for overscan bias and flat fielded using internal lamps taken immediately preceding and following the observations . the flux calibration was performed using observations of feige 110 and feige 34 ( massey 1988 , massey & gronwall 1990 ) obtained both with and without an order sorting og570 filter in order to correct for the second order light contamination in the spectral region @xmath47 . all of the spectroscopic reductions were performed using the noao iraf package . the spectropolarimetric analysis was carried out using our own software , and is based on the methods described in miller , robinson , & goodrich ( 1988 ) . our polarimetric results are presented in table [ poldata ] . since c is very faint ( peak continuum @xmath48 surface brightness @xmath49 mag/ ) , the continuum polarization was determined in large spectral bins ( cf . , dey 1996 ) . for each wavelength bin , we tabulate the normalized linear stokes parameters , @xmath50 and @xmath51 , where @xmath52 is the effective intensity of light polarized with its electric vector at position angle @xmath53 . since the polarization is low , we also tabulate an unbiased estimate of the percentage polarization @xmath54 , where @xmath55 is defined to be negative if @xmath56 ( wardle & kronberg 1974 , simmons & stewart 1985 ) . the continuum emission , sampled in these bins over the 342@xmath5710 extracted aperture ( , sampling the brightest continuum emitting region in the galaxy ) , is found to be unpolarized , with formal 2@xmath46 upper limits of @xmath58 in the wavelength range @xmath59 ( i.e. , between ly@xmath60 and ) , and @xmath61 in the region @xmath62 ( , between and ] ) . within our measurement errors , the narrow emission lines appear to be unpolarized . the galactic latitude of c is 17.5 and the interstellar percentage polarization is estimated to be low ( _ e(b v)[email protected] , which implies @xmath63 ; burstein & heiles 1982 ; appenzeller 1968 , mathewson & ford 1970 ) . a slightly better constraint on the interstellar polarization results from the formal 2@xmath46 upper limit of @xmath64 for the polarization of the narrow component of the ly@xmath60 emission line . the fractional polarization of the continuum emission in c is therefore much lower than that measured in the lower redshift ( @xmath65 ) radio galaxies , which are generally found to have rest - frame uv continuum percentage polarizations of @xmath66 ( di serego alighieri 1989 , cimatti 1996 , 1997 , dey 1996 ) . this difference in polarization is not simply the result of the different rest - frame wavelength ranges sampled at @xmath67 and @xmath15 . spectropolarimetric observations of the @xmath3 radio galaxies show either no significant wavelength dependence of the continuum polarization in the near uv ( after accounting for dilution by starlight ) or , in a few cases , some evidence for increasing @xmath68 with decreasing wavelength . moreover , the spectropolarimetry of c may be directly compared with existing observations of 3c 256 , a @xmath69 powerful radio galaxy for which spectropolarimetric observations have been made down to 1400 in the rest frame ( dey 1996 ) . the continuum polarization of 3c 256 at wavelengths shortward of is high ( @xmath70 ) and @xmath71 is roughly wavelength independent between 1400 and 2700 . these results suggest that , in comparison , c has intrinsically low polarization . the summed spectrum of c was corrected for galactic absorption using the extinction curve of cardelli ( 1989 ) and an @xmath72=0.15 . the parameters of the emission and absorption lines listed in tables [ emdata ] and [ absdata ] were derived from the dereddened spectrum ( using specfit in iraf ; kriss 1994 ) extracted in a @xmath73 aperture centered on the brightest region of the galaxy ( figure [ 4cf702 ] ) . the redshift of the galaxy determined from the @xmath51640 emission line is [email protected] ; as expected , the resonance lines ( ly@xmath60 , , and ) result in a slightly higher redshift , since the blue side of the lines are modified by associated absorption . the ] @xmath51909 emission line may be slightly blue shifted ( by @xmath74 ) relative to the systemic velocity determined from emission line , but since it is also much broader ( see below ) than any of the other emission lines , we consider it likely that it arises in a different region than the bulk of the narrow line emitting gas . the redshift is consistent with ( but possibly slightly redshifted by @xmath75 relative to ) the average redshift determined from the low ionization absorption lines . since most of the absorption lines may be contaminated by ( stronger ) emission and have asymmetric profiles , in this paper we adopt the redshift as the systemic redshift of the galaxy . the ratios of the narrow emission line strengths of ] , and can be well matched by a simple photoionization model incorporating a power law ionizing source with @xmath76 ( @xmath77 ) , a high ionization parameter @xmath78 and solar metallicity clouds ( e.g. , villar - martin , tadhunter & clark 1997 ) . the large / ] ratio also favors a geometry where the illuminated sides of the line emitting clouds are viewed directly ( villar - martin 1996 ) ; if the photoionizing source is the central agn , this implies that the bulk of the line emitting region ( the ne part of the galaxy ) lies on the _ far _ side of the agn nucleus . this geometry is also supported by the radio rotation measure ( carilli 1994 ) , which implies a larger faraday optical depth to the eastern part of the source than to the western lobe . the profiles of both ly@xmath60 and ] @xmath51909 reveal relatively broad components . on fitting these lines with a two - gaussian model , we find that the broad components have fwhms of @xmath221100 1400 ( e.g. , figure [ ciii ] ) . although large , these widths may be simply a high - velocity component of the emission line gas resulting from the interaction of the radio source with the ambient medium , or due to entrainment of the gas in the radio jet . the fact that the emission lines are unpolarized supports this suggestion . alternatively , the broad components may arise in the nuclear broad - line region , and may imply that there is a small contribution to the uv continuum and line flux from the active nucleus . although the measured fwhm of the emission lines are smaller than those typically measured in the broad line regions of steep spectrum radio - loud quasars ( @xmath79 , with almost all having @xmath80 ; e.g. , brotherton 1994 , corbin 1991 ) , the rest frame equivalent width of the broad component of the ] emission line ( @xmath81 ) is roughly comparable to that observed in these quasars . it is therefore possible that there may be some contribution from agn light to the uv flux , at least at wavelengths @xmath82 . we note that although we do not detect any significant polarization of the ] line , the errors are large in this part of the spectrum ; the 2@xmath46 upper limit from these data is large ( @xmath83 ) , and therefore does not provide a useful constraint . one of the most remarkable results from our observations of c is the detection of strong absorption features in its total light uv spectrum . figure [ spectrum ] shows the total light spectrum of the brightest @xmath73 region of the galaxy ( smoothed using a 5-pixel boxcar ) , which clearly shows strong absorption lines of , , , , and ly@xmath60 . some of these features have ` p - cygni like ' profiles , , an emission component juxtaposed with a blue - shifted absorption component . p - cygni like profiles may arise either in stellar winds or be the result of a different mechanism ( e.g. , galaxian outflows ) ; we discuss the possible physical origin of these profiles in c in 4.2 . the spectrum of c in this regard is similar to that of actively star forming regions in nearby galaxies such as 30 doradus ( vacca 1995 ) , ngc 4214 ( leitherer 1996 ) and the b1 star - forming knot in the nearby starburst galaxy ngc 1741 ( hereafter referred to as ngc 1741b1 , conti , vacca & leitherer 1996 ; see also figure [ spectrum ] ) , and to the @xmath84 population of star forming galaxies recently discovered by virtue of their lyman limit absorption ( e.g. , steidel 1996 , giavalisco 1996 , lowenthal 1997 ) . the parameters of the strongest detected absorption lines are listed in table [ absdata ] along with limits on the strengths of known stellar absorption features . the absorption lines that are largely uncontaminated by strong emission have rest frame equivalent widths of @xmath85 ( table [ absdata ] ) , and are therefore slightly smaller than in most starbursts systems ( e.g. , ngc 1741b1 has equivalent widths of @xmath86 for these lines ; conti 1996 ) . many of the absorption features have asymmetric profiles with blue wings extending to more than 2500 from the line centroid , and some show p - cygni like profiles . if we define the systemic velocity to be that measured from the emission line , then the absorption troughs associated with ly@xmath60 and are blue shifted relative to the systemic velocity by @xmath87 , whereas the absorption in the low - ionization species of , and are at the systemic velocity , or perhaps slightly blue shifted by @xmath88 . the blue shifts observed in ly@xmath60 and are somewhat difficult to determine accurately since these absorption features are contaminated by very strong line emission . in addition to the absorption lines that are clearly associated with p - cygni like emission , there are a few features for which we have only tentative identifications . these features are likely to be either very high - velocity blue - shifted absorption arising in a outflowing gas , or due to foreground ( intervening ) metal - line systems . the strong absorption feature at 6555 appears to be ( at the present resolution ) a composite of three absorption lines at 6544.3 , 6560.0 and 6573.6 ( figure [ siiv ] . the ratios of the first two features suggest that these may be @xmath212796.35,2803.53 in an intervening system at @xmath89 . we tentatively identify the 6573.5 line with the @xmath51371 wind feature in c. this line is strong in the spectra of o3 giant and supergiant stars ( walborn 1995a ) . an unambiguous understanding of the true origin of this complex must await higher resolution and higher signal - to - noise ratio data . table [ absdata ] also lists absorption lines at observed wavelengths of 6351.5 , 6661.8 , and 7256.5 . these features may correspond to the @xmath51335 , @xmath51400 , and @xmath51527 features blue shifted by @xmath222500 from the systemic velocity . alternatively , the 6661.8 feature is more likely to simply be the blue wing of the absorption profile , and the 6351.5 line may correspond to a similar feature at @xmath90 observed in ngc 1741b1 ( figure [ spectrum ] ) . in summary , absorption line systems physically associated with c are observed at approximately the systemic velocity defined by the line emitting gas , with the high ionization p - cygni like lines extending to velocities of @xmath91 , and possibly also at @xmath222500 blue shifted relative to the line - emitting gas . the majority of radio galaxies at redshifts @xmath3 exhibit uv continuum polarization , with the position angle of the electric vector generally perpendicular to the major axis of the uv extent ( e.g. , tadhunter 1988 , di serego alighieri 1989 , jannuzi & elston 1991 , jannuzi 1995 , dey 1996 , cimatti 1996 , 1997 ) . in almost all known cases , the percentage polarization is large ( @xmath66 ) at rest - frame uv wavelengths and , in a few cases , monotonically decreases with increasing wavelength . in a few well studied cases , broad emission lines have been detected both in the total light and polarized flux spectra . these results have led to the conclusion that the spatially extended , aligned uv morphologies of the @xmath3 powerful radio galaxies are largely a result of scattered light from the agn . c is therefore an exception to the rule the low measured percentage polarization does not support the hypothesis that the uv emission is dominated by light from the agn . however , the lack of polarization does not necessarily _ prove _ that the uv continuum emission from c does not have a scattered agn contribution . the polarization could be diluted by geometrical effects resulting from averaging over large areas ( our spectral extraction covers an aperture of 32@xmath5710 ) which have different polarization angles . in order to investigate this possibility , we measured the polarization in the uv continuum between ly@xmath60 and civ as a function of spatial position along the slit from our best seeing data . within the errors , there is no convincing detection of polarization in the extended continuum . it is also possible to dilute the fractional polarization by multiple scattering . this would require that the scattering be due to dust particles rather than electrons , which would in turn result in large reddening and extinction of the spectrum ; no significant extinction or reddening is observed . the upper limit on the estimated extinction ( @xmath92 ) corresponds to a dust optical depth at a rest wavelength of 1500 of @xmath93 , which should not result in significant dilution of @xmath68 by multiple scattering . finally , some theoretical models suggest that the dust scattering efficiency may drop sharply at wavelengths less than @xmath94 ( a. laor , personal communication ) . although such a feature has never been observed in the scattered light uv spectra of galactic reflection nebulae ( e.g. , calzetti 1995 ) , if this were indeed the case , our limits may be consistent with dust scattering contributing significantly to the longer wavelength radiation @xmath95 , but very little to the shorter wavelength radiation . we therefore conclude that the uv spectrum , at least at wavelengths @xmath96 , is largely uncontaminated by scattered agn light . is it possible that other agn - related processes dominate the uv continuum emission from c ? due to the strong ly@xmath60 emission observed in c , it has been suggested by dickson ( 1995 ) that the uv continuum emission in c may be dominated by thermal continuum emission . the blue ( @xmath97 ) uv continuum spectrum of c indicates that it is unlikely that the recombination continuum emission from hydrogen dominates the continuum emission in the region between ly@xmath60 and . the flux in the ly@xmath60 emission line can provide a fairly good estimate of the contribution of the nebular emission to the uv spectrum in our 21@xmath5710 aperture . however , this resonance line is easily attenuated by dust , and instead a more reliable estimate may be derived from the @xmath51640 emission line . for a solar metallicity gas ionized by a power - law continuum ( ionization parameter @xmath98 to @xmath99 ) , the typical ratio of ly@xmath60/ is @xmath100 . hence , assuming that @xmath101 , and that the typical temperature of the line - emitting gas is @xmath102 k , the total contribution of recombination and bremsstrahlung emission from at @xmath103 is @xmath104 , or @xmath105% of the uv continuum emission ( e.g. , osterbrock 1989 , brown & mathews 1970 ) . hence , it is unlikely that any obvious agn - related process ( , scattered light , thermal bremsstrahlung and recombination emission ) contributes significantly to the uv continuum emission from c , at least at rest wavelengths @xmath106 . we are therefore left with the possibility that this uv continuum emission is dominated by starlight . it is possible , however , that the longer wavelength radiation has a larger contribution from agn emission . for wavelengths longward of the @xmath51640 line , the strong telluric oh emission precludes a high signal - to - noise measurement of the fractional polarization , and the upper limits from our measurements are less restrictive ( e.g. , @xmath107 for @xmath108 ) . in addition , the ] emission line is observed to have a relatively broad component with a rest frame equivalent width similar to that observed in quasars , and the redshift determined from the line is slightly different from that determined from the narrow line which suggests that this line arises in a different region ( at a different velocity ) from the bulk of the narrow line emitting gas . it is therefore possible that there is a significant agn contribution to the emission at the longest observed wavelengths . indeed , the observed continuum emission from the central body of the galaxy can be fairly well modelled by a double power law : the region between ly@xmath60 and has a @xmath109 dependence , whereas the spectral region between and ] is roughly wavelength independent . the hypothesis of an agn contribution rising to redder wavelengths would imply that the agn is intrinsically much redder ( @xmath110 , where @xmath111 ) than that observed in the luminous , steep spectrum radio loud quasars ( typically @xmath112 ) . in light of the submillimeter detection of warm dust emission from c ( dunlop 1994 ) , it is possible that its active nucleus is still enshrouded in dust and heavily reddened at least along our line of sight . if the polarization in the region @xmath113 is real , it is noteworthy that the electric vector position angle is parallel to the radio axis , rather than perpendicular as it is in most @xmath3 radio galaxies ; this result suggests a different scattering geometry or polarization mechanism . spectroscopic and polarimetric observations at near - ir wavelengths are necessary to determine the agn contribution , if any , to the observed emission at longer wavelengths . it is of great importance to understand the origin of the absorption lines in a galaxy observed at such an early epoch ( for = 50 , = 0.1 , @xmath114=0 , the universe is only 2.2 gyr old at @xmath0=3.8 ) . as discussed in the previous section , the uv continuum emission at @xmath106 can not be adequately explained by agn - related processes , and is therefore likely to be dominated by starlight . strong stellar uv continuum emission can be produced by very young populations , in which the integrated uv spectrum is dominated by the flux from hot , high - mass stars . hence , it is tempting to interpret the uv absorption lines observed in c as photospheric or dense wind features from hot , young stars . however , it is important to note that most of the strong resonance lines detected near the systemic velocity in the uv spectra of nearby star - forming galaxies are generally dominated by interstellar rather than stellar components ( , york et al . 1991 ) . in this section we investigate whether or not the absorption line spectrum of c contains a component due to starlight , and the possibility that the absorption spectrum can arise in a galaxian outflow . a clear test of whether or not the uv continuum emission is due to young stars is the detection of stellar photospheric lines which are uncontaminated by interstellar absorption in the integrated spectrum ( e.g. , @xmath51247 , @xmath51296 , @xmath51417 , @xmath51502 , @xmath51720 ; leitherer , robert & heckman 1995 , heckman & leitherer 1997 , heckman 1997 ; m. pettini personal communication ) . these lines arise from excited levels ( e.g. , the lower energy level of the @xmath51502 transition is the 3@xmath115(j=1 ) level which lies @xmath2215.8ev above ground ; bashkin & stoner 1975 ) . while these features can be easily formed in the dense , hot winds of early - type stars , they are unlikely to arise in the cooler , less dense interstellar medium . the integrated spectrum of c shows at least one unambiguous signature of starlight : we detect the weak @xmath51502 photospheric absorption line ( figure [ sv ] ) , with an equivalent width of @xmath116 in the rest frame . the strength of the line in c is comparable to that observed in nearby starbursts ( @xmath117 in ngc 1741b1 ) . this suggests that agn light does not contribute significantly to the spectrum at these wavelengths , although the signal - to - noise ratio and resolution of the present data do not provide a strong constraint on this issue . we also marginally detect weak absorption due to stellar photospheric lines of @xmath211294.6,1296.7 ( @xmath118 ) . the shape of the absorption profile also indicates that hot stars make a significant contribution to the uv spectrum . figure [ siiv ] shows that the feature has narrow components that appear slightly blue shifted from the systemic velocity as well as a blue wing that extends to more than 2500 from the systemic velocity . although the narrow components are almost certainly interstellar , the shape and extent of the blue wing closely resembles that observed in ngc 1741b1 ( figure [ siiv ] ; conti 1996 ) , where it is believed to arise in hot stellar winds ( e.g. , conti 1996 , heckman & leitherer 1997 ; in particular , see their figure 2 ) . finally , as reported in 3.3 , there is a marginal detection of the @xmath51371 wind feature ( figure [ siiv ] ) . if the existence of this feature can be verified with higher resolution data ( to separate the line from the foreground absorber ) , it will provide strong support for the presence of early spectral type o giants and supergiants ( walborn 1995a ) . we have also searched for other pure photospheric absorption lines ( e.g. , @xmath51247 and @xmath51720 ) , but these are undetected in the present data . unfortunately , the upper limits on the strengths of the undetected lines do not provide any useful constraints or rule out the presence of starlight in the spectrum ( see table [ absdata ] ) . is it possible for the p - cygni like absorption lines to be formed in a galaxian outflow rather than in a stellar wind ? some low - ionization lines such as appear to show asymmetric absorption profiles with blue wings in c. this is a property rarely seen in the spectra of starburst regions or individual hot stars ( with the possible exception of some b supergiants ) . since these absorption lines are likely to have significant interstellar components , it is certainly possible that these low - ionization lines arise in a galaxian - scale outflow . however , it is unlikely that the blue wing of the absorption line is also formed in a dense , hot wind from the galaxy driven by supernovae or the agn . supernovae driven ` super winds ' observed in some starburst galaxies generally exhibit liner like line ratios and rarely show strong high - ionization lines ( e.g. , lehnert & heckman 1996 ) and are therefore unlikely to be the origin of the features observed in c. absorption observed in broad absorption line ( bal ) qsos exhibit high - ionization lines , but these tend to show absorption extending over a very large velocity range ( @xmath1193000 ) and are not spatially extended sources . nuclear bal - like absorption seen reflected off dust and gas in the ambient medium can be ruled out by the lack of detectable polarization in c. nevertheless , it is worth investigating whether a massive galaxian outflow could result in the observed profile . a consideration of the energetics of such an outflow and the mass of outflowing material necessary to produce the observed high - ionization line profiles may provide a method by which to discriminate between the ` galaxian outflow ' and ` stellar wind ' hypotheses described above . although a detailed estimate of these quantities is beyond the scope of this paper , a very rough calculation can be used to show that the amount of mass implied by the absorption in the blue wing of @xmath51393 is large if this blue wing arises in an outflowing galaxian wind . the blue wing of the line has a rest frame equivalent width of @xmath120 . although this line may be saturated , the _ minimum _ column density required to produce the line is @xmath121 . the mass of the absorbing region ( , the region of the outflow with velocities between 700 and 2500 blueward of the systemic velocity ) is @xmath122 where @xmath123 is the abundance of si in c relative to the cosmic abundance , and @xmath124 is the depletion of si onto dust grains . we have assumed that the continuum emitting region is @xmath125 kpc in size , and used the measured rest equivalent width of 1.5 . for the purposes of this order of magnitude estimate , we have assumed that the ion fraction of is the maximum possible in thermal equilibrium ( , at @xmath126 k ; shull & van steenberg 1982 ) . the energy input required to drive this minimum mass is equivalent to that of @xmath127 supernovae . recent determinations of the abundances in @xmath7 damped ly@xmath60 absorbers have resulted in estimates of @xmath128 \approx -2}$ ] ( , @xmath129 ) in undepleted gas ( lu 1996 ) , suggesting that the minimum mass estimate derived above may be higher by at least two orders of magnitude . also , @xmath130 corresponds solely to the mass in the region in velocity space in which the blue wing of the feature is formed ; accounting for the probable line saturation , the range of ionization states and velocities that would be present in a real outflow , depletion of si onto dust grains , and the integration over all solid angles will increase this number further by at least an order of magnitude . this estimate for the mass of the outflowing material in front of the continuum emitting region of the galaxy is therefore potentially very large ( @xmath131 ) and is comparable to the _ total _ mass of material in the entire system ( e.g. , van ojik 1997 , chambers 1990 ) . although more detailed modelling of the ionization state , velocity structure and geometry of the outflow are required to investigate these constraints further , the order - of - magnitude estimate presented here suggests that it is unlikely that the blue wing of arises in outflowing material . on the other hand , photospheric emission from hot stars provides not only the one viable explanation for the uv continuum emission observed from c ( 4.1 ) , but also a natural origin for the blue wings of the high ionization absorption line profiles . one of the main differences between the spectrum of c and those of nearby star forming galaxies is that the spectrum of c is dominated by strong line emission , which is likely the result of photoionization by the hard spectrum of the active nucleus and possibly shock ionization resulting from the interaction of the radio source plasma with the ambient medium . as discussed in 3.2 the ] // ratios suggest a high ionization parameter with high energy photons , implying that the observed emission lines are not due to photoionization by starlight , but instead to photoionization by the hard agn spectrum ( e.g. , villar martin 1996 ) or fast shocks ( e.g. , dopita & sutherland 1996 ) . the agn - driven emission lines dilute the stellar and interstellar absorption lines weakening their observed strengths , and overwhelm the much weaker emission component that may be associated with stellar wind features ( , the ` true ' p - cygni profiles ) . although the emission lines are not powered by hot stars , it is intriguing that a fairly good fit to the overall spectrum of c can be produced by adding strong emission lines to the spectrum of ngc 1741b1 , suggesting that the uv _ continuum _ emission shortward of may indeed be starlight . in summary , although the present data are inadequate to unambiguously discriminate between stellar wind and galaxian outflow origins for the resonance absorption transitions , several lines of argument suggest that the uv continuum emission at wavelengths @xmath132 is dominated by light from a young , hot stellar population . the detection of , the marginal detection of , the blue wing of the profile , the unpolarized continuum emission , the inability of agn - related processes to account for the uv continuum flux and the overall similarity of the spectra of 4c 41.17 and ngc 1741b1 together provide a compelling argument for the existence of starlight in this @xmath133 galaxy . if the stellar populations in c and ngc 1741b1 have similar metallicities and initial mass funtion ( imf ) , the similarity of the shape and equivalent width of the blue wing of in c and ngc 1741b1 may reflect comparable numbers of hot stars per unit mass in the two systems . the high - ionization resonance lines may therefore have both stellar and interstellar components : our observations are consistent with the hypothesis that the high - velocity blue - wings of the and absorption are produced in fast winds from hot stars , whereas the absorption associated with the lower ionization states arises in the larger scale dense , galaxian wind . a test of this hypothesis can be provided by detections of the fainter stellar photospheric lines , but this must await higher signal - to - noise and higher resolution spectroscopy . the discussion in the remainder of this paper is based on our favoured interpretation that the uv continuum spectrum of c is dominated by the emission from a young stellar population . if the observed uv continuum emission is indeed due entirely to starlight , the slope of the continuum and the profiles of the stellar absorption lines can be used to constrain the global properties of the stellar population . since our spectral extraction samples a large volume of the galaxy ( our 21@xmath5710 aperture corresponds to [email protected] @xmath19 kpc for = 0.1 ) , it is very plausible that we are observing a mix of stellar populations corresponding to a range of ages rather than a burst of a single age . nevertheless , the absorption lines in the integrated spectrum can provide a few constraints on the luminosity - weighted mean age of the stellar population dominating the uv light . detailed spectral synthesis of the uv spectrum of c will be presented by vacca & dey ( 1997 ) ; in this section , we present some preliminary conclusions based on the general appearance of the spectrum . we may constrain the upper mass cutoff ( @xmath134 ) of the imf and the age of the stellar population by using the absorption feature . as discussed above , although the strong , narrow low - velocity components of the doublet are largely interstellar , the high - velocity blue wing of the profile ( figure [ siiv ] ) is probably formed in hot stellar winds . the line is a powerful age diagnostic , as it only appears in the integrated light of young populations ( ages @xmath1351 gyr ) , reaching maximum strength in b0/b1 stars ( leitherer et al . 1996 ) , with broad , blue - shifted wings of the profile being formed in supergiant winds . since this absorption line is observed only in massive stars , the very presence of a broad line implies an @xmath136 . in addition to the line , the @xmath51502 absorption line is weakly detected in the integrated spectrum . this line is only observed in the spectrum of o - type stars , is weaker in late o and early b subtypes , and no longer detectable in the spectra of mid b subtypes ( walborn 1985 , 1995b ) . hence , the uv light is dominated by a very young population , implying that the stars were formed in a burst less than a few million years before , or ( more likely ) that we are observing c during an epoch of ongoing star formation . an additional constraint is provided by the shape of the uv spectrum . the population synthesis models of leitherer and heckman ( 1995 ) demonstrate that the uv spectrum of a young star - forming population can never be bluer than @xmath137 , and should gradually redden with age . hence , the observed slope of the spectrum of c ( @xmath138 ) provides an upper limit on the age and the extinction . in the absence of any extinction , the constant star - formation solar metallicity salpeter imf models of leitherer & heckman ( 1995 ) imply maximum ages for the stellar population of @xmath139 600 myr . if there is a moderate amount of extinction then the intrinsic slope of the uv spectrum of c will be bluer than the observed slope and the age will be younger . the extinction for c is probably not larger than @xmath140 mag , because the intrinsic slope would then be bluer than the bluest model values . ( here we have assumed the extinction law given by kinney et al . the change in the slope as a function of reddening for this extinction law is given by leitherer et al . 1996 . ) for the instantaneous burst model , the uv continuum slope should become redder much faster than in a continuous star formation scenario . using the same arguments about the continuum slope as those given above , we can place an upper limit of about 16 myr for the age of the population . for an instantaneous burst model , the line profiles place much stronger constraints on the age and @xmath141 than in the continuous ( constant - rate ) star formation case , since high - velocity blue wings are only observed for ages less than 6 myr and @xmath142 . if the age is less than 6 myr , then the intrinsic spectral shape of the continuum emission must be between @xmath143 and @xmath144 , which implies that the observed continuum is moderately reddened ( _ e(b v)_=0.04 to 0.10 ) . again , this age estimate is based on the assumption that the emission component is from hot stars rather than from the agn . an additional constraint on the stellar content of c may be derived from existing near - infrared photometry . the near - infrared @xmath145 and @xmath146 bands are largely uncontaminated by strong line emission from c. in the 10@xmath5721 spectroscopic aperture discussed here , c has measured magnitudes of @xmath147 and @xmath148 ( adam stanford , personal communication ) . under the null hypothesis that all the observed near - infrared and optical emission is starlight , we find that unreddened , young continuous ( constant - rate ) star - forming models with ages less than @xmath149 myr and instantaneous burst models with ages less than @xmath150 myr that match the observed optical spectrum do not fit the observed near - infrared fluxes . in fact , these simple unreddened models do not provide satisfactory fits to both the @xmath145 and @xmath146 fluxes at any age , either overpredicting the @xmath146 flux or underpredicting the observed @xmath145 flux . in contrast , slightly reddened ( @xmath151 ) , models with ages 420 myr ( for instantaneous burst models ) or 10 300 myr ( for continuous star - forming models ) provide a satisfactory fit to both the overall optical - ir spectral energy distribution and ( at the young age end ) to the line profile . if the observed infrared emission contains a reddened agn component , the spectral energy distribution of the underlying population is bluer than observed , and therefore younger . if the population is indeed young ( @xmath1391 gyr ) , then a reddened agn can contribute no more than 60 - 65% of the total @xmath146 flux . in summary , if the galaxy is forming stars continuously , the age ( derived from its uv spectrum ) is less than 600 myr . the age estimates presented in this section are at present phenomenological and are uncertain as they depend upon several assumptions regarding the shape of the imf , the star - formation history , the metallicity of the star - forming population , and the reddening intrinsic and along the line - of - sight to c. nevertheless , it is of considerable interest that a simple model can provide an adequate fit to both the observed optical and near - infrared continuum fluxes of c , suggesting that stronger constraints may be eventually derived when high signal - to - noise moderate resolution optical and infrared spectra of the uv light and the 4000 or balmer - break region are available . in a companion paper , van breugel ( 1997 ) present hst _ imaging of c , and demonstrate that the brightest central region is composed of several compact ( @xmath152 kpc ) knots , which are distributed in an ` edge - brightened ' structure suggestively lying on the outer envelope of the central radio source structure ( also see figure [ 4cf702 ] ) . the deep spectroscopic data presented in this paper strongly suggest that the uv absorption line spectra of these knots are similar to that of star forming regions in nearby galaxies . if the compact uv knots in c are indeed star forming regions , then the spatial distribution of these knots relative to the radio source emission is suggestive of models where star - formation may be triggered by the radio source ( de young 1989 , begelman & cioffi 1989 ) . de young ( 1989 ) proposed that as radio jets propagate into dense protogalactic gas , star formation occurs in the compressed post shock gas as it cools . this results in a spatial distribution of hot stars that is both aligned with the radio source and edge brightened in the direction normal to the radio jet axis , on a scale of roughly @xmath153 kpc from the jet axis depending upon the jet parameters and age of the stellar population ( de young 1989 ) . in the scenario proposed by begelman & cioffi ( 1989 ) , star formation also occurs when intergalactic clouds are overtaken and compressed by the expanding over - pressured cocoon surrounding the radio source . this mechanism can result in a large star formation rate ( @xmath154/yr ; begelman & cioffi 1989 ) and also results in young stars in an edge brightened structure ( de young 1989 ) . the observed distribution of star forming clumps in c is edge brightened as predicted in these scenarios , and the spatial extent in the transverse direction is indeed comparable to that expected . the strongly asymmetric radio source morphology ( shorter eastern arm ) and brightness distribution ( the eastern knots are more luminous both in the radio and the optical ) also support a picture of stronger ongoing interactions between the radio source and the ambient medium on the eastern side of the galaxy . although the evidence presented here ( spectroscopy coupled with the van breugel 1997 _ hst _ imaging data ) is circumstantial , at present there is no clearer support for this picture than these data . our observations of c are consistent with the interpretation that the alignment effect in at least this @xmath18 radio galaxy is due to star formation induced by the radio source , and not due to scattering as is the case in the bulk of the @xmath3 radio galaxies . if the continuum emission is indeed due to starlight from a young population , we can use the observed uv continuum luminosity to estimate the star - formation rate in the system . the specific luminosity at @xmath155 is @xmath156 ( = 0.1 ) . for a salpeter imf with an upper mass cutoff of 80 , the stellar population synthesis models of leitherer ( 1993 ) suggest that a continuously star forming population has @xmath157 for ages between 1 and 9 myr after the onset of star - formation . hence , if _ all _ of the 1500 flux in c is due to young stars , the star formation rate is between 140 1100 @xmath6 /yr . if we account for screen extinction of @xmath72=0.1 , these estimates are increased by a factor of @xmath222.9 . although this range is large , at the lower end , this rate is observed for extreme starburst galaxies . for a high redshift galaxy , possibly in its early stages of formation , such a large rate may be expected . at this sustained rate of star formation , it will only take @xmath158 gyr to form a @xmath159 system of stars , a timescale comparable to the dynamical timescale . this rate of star formation is higher than that observed in the @xmath160 `` lyman limit '' galaxies discovered by steidel ( 1996 ) , which are typically observed to have star formation rates at @xmath161 of @xmath162 /yr ( = 0.1 ) . the lower rates observed in these systems may be due to the fact that these galaxies are being observed at a later stage in their history ( @xmath163 gyr between @xmath164 and @xmath165 for @xmath166 ) , or that these are lower mass systems than c , or that their star - formation rates have been under - estimated due to dust - extinction . the timescale derived above for the age of the starburst can be compared with the expansion timescale of the radio source . in the instantaneous burst model , the age of the stellar population is less than 16 myr . during this time , the radio source will expand by roughly 16.4 @xmath167 kpc , or @xmath168 from the location at which it triggered the star formation , where @xmath169 is the radio source expansion speed , @xmath170 is the age of the population , and @xmath53 is the angle between the direction of expansion and the line of sight . the lack of a bright core component to the radio source and the absence of any significant agn component to the uv emission suggest that the radio source is oriented close to the plane of the sky , with our direct view of the nuclear region obscured by dust . hence , since the uv continuum emission is observed to be roughly coincident with the radio emission in the central regions ( miley et al . 1992 , carilli 1994 ) , we may conclude that the triggering was recent , or that the radio source is expanding slowly , or both . the large star formation rate derived for c requires a large reservoir of cold gas . indeed , there is some evidence for an extended cold component from the observation of spatially extended ly@xmath60 absorption in c ( hippelein & meisenheimer 1993 , van ojik 1997 , dey 1997@xmath171 ) suggesting at least @xmath172 of cold gas in the central regions , and @xmath173 in a large extended halo ( = 0.1 ) . in addition , the detection of submillimeter emission from c implies that the galaxy contains a very large mass in dust ( @xmath174 for = 0.1 ; dunlop 1994 ) , and suggests the presence of such a reservoir . under the assumption that the dust is heated purely by starlight , dunlop derive star formation rates of @xmath175/yr ( = 0.1 ) ; although there may be a significant agn contribution to the dust heating , it is noteworthy that this estimate is of the same order of magnitude as the estimates derived above from the uv luminosity alone . the blue uv / optical color and lack of strong color gradients in c implies that the observed continuum emission is not very reddened , and is likely to be indicative of an asymmetric distribution of dust . we speculate that c may well be a young galaxy emerging from its dust cocoon , with only the very outer regions visible at present . in addition to its different polarization properties , c also exhibits an important morphological difference at rest - frame uv and optical wavelengths compared to lower redshift objects . the @xmath3 3cr radio galaxies tend to be clumpy and spatially extended in the rest - frame uv , whereas their rest - frame optical morphologies ( observed near - ir ) are generally more symmetric ; , rather than reflecting the peculiar uv structures , the optical structures look more like ` normal ' elliptical galaxies . c also has a clumpy uv morphology ( van breugel 1997 ) , but its rest - frame optical morphology is also clumpy and extended along the major axis of the radio emission ( chambers 1990 ) ; in fact , there is near spatial coincidence between the rest frame uv and optical emission ( graham 1995 ) . although the rest wavelengths sampled at these different redshifts are different , the @xmath3 radio galaxies exhibit a noticeable morphological change across the 4000 break ( i.e. , the approximate wavelength beyond which stars begin to dominate over the agn component ) , whereas c ( @xmath18 ) does not exhibit this property . the dissimilarity between the polarization and morphological properties of c and those of the lower redshift powerful radio galaxies raises the intriguing question of whether we are witnessing an evolutionary phenomenon . a tempting hypothesis is that radio galaxies at @xmath13 are in the process of formation : they are forming stars on the dense borders of the expanding cocoon inflated by the radio lobes ( e.g. , begelman & cioffi 1989 ) , and their uv and optical emission is dominated by the light from young stellar populations . the agn component , scattered or directly viewed , is likely to be diluted by the young starlight . in contrast , in the @xmath8 radio galaxies , the population has aged and dynamically relaxed , and no longer contributes any significant flux to the uv spectrum . the uv light in the @xmath8 population may instead be dominated by other blue components , such as the agn light ( viewed directly or scattered by dust and electrons in the ambient medium ) remnant star forming regions , or balmer continuum emission from the photo and shock ionized gas . in an = 50 , @xmath176=0.2 universe , the time between @xmath133 and @xmath177 is @xmath178 gyr , providing ample time for the population to age and ( possibly ) dynamically relax to explain the colors and morphology of the stellar component in @xmath3 radio galaxies . recent spectral synthesis modelling of the uv spectra of @xmath179 lower - power radio galaxies which have no discernible light contribution from the agn result in age estimates of @xmath180 3.5 gyr , consistent with this picture ( dunlop 1996 , spinrad 1997 , dey 1997@xmath181 ) . we have presented deep spectropolarimetric observations of one of the most distant known radio galaxies , c at @xmath18 . the spectrum is dominated by strong emission lines ( possibly excited by shocks and / or photoionization by the agn ) , a bright uv continuum emission and fairly strong resonance absorption lines of both low and high ionization species . the uv continuum emission from the brightest regions in this galaxy is unpolarized ( @xmath182 ) , suggesting that scattered agn light may not be a dominant contributor to the uv flux from the galaxy . many of the lines exhibit p - cygni like profiles , , emission components associated with blue - shifted absorption components . the uv absorption spectrum is similar to that of star forming regions in nearby starburst galaxies , and to the spectra of the recently - discovered high - redshift population of starburst systems . most of the strong low - velocity narrow absorption lines most likely arise in the interstellar medium of c. however , we report a detection of the @xmath51502 absorption line which arises in the photospheres of o - type stars and is uncontaminated by interstellar absorption , and also a possible detection of the stellar photospheric lines . we also find that the shape of the high - velocity blue wing of the doublet is similar to that observed in nearby starburst galaxies such as ngc 1741 , where it is interpreted as arising in an o star wind . it appears unlikely that the blue wing of the line arises in a galaxian outflow since the implied mass in such an outflow is very large . these considerations , combined with the lack of detectable polarization , the lack of strong agn - like broad - line emission , the inability of other agn - related processes to produce the bulk of the continuum emission , and the similarity of c s spectrum to that of the star - forming region ngc 1741b1 , together provide a compelling argument that the uv spectrum of c is dominated by starlight from a young stellar population . if the uv continuum emission is entirely dominated by starlight , instantaneous burst solar metallicity population synthesis models imply an age of less than 16 myr , and suggest that the spectrum is reddened by dust either intrinsic to c or along the line of sight . if the star - formation is continuous , then the age of the population is @xmath183 myr . under the assumption that _ all _ of the 1500 flux in c is due to young stars , the star formation rate is between 140 1100 @xmath6 /yr . a fairly young , slightly reddened population provides an adequate fit to the overall spectral energy distribution , but there potentially could be a red agn which dominates the near - infrared emission . c is one of the highest redshift aligned radio galaxies known . our data suggest that the uv properties of this galaxy are different from those of @xmath8 , aligned radio galaxies : the unpolarized uv continuum emission and the edge brightened , clumpy _ hst _ morphology combined with the spectroscopic evidence of a starburst - like spectrum may imply that the uv light is dominated by young stars forming on the dense edge of the pressurized cocoon formed by the radio source , as suggested by begelman & cioffi ( 1989 ) . if so , we are witnessing this galaxy in the process of formation . the young age derived for the starburst in c is consistent with the observation that the radio continuum emission is roughly spatially coincident with the uv emitting regions . the ages determined for the @xmath3 radio galaxies are consistent with the hypothesis that these objects form the bulk of their stars at @xmath7 in a low - density universe . we thank ron quick , randy campbell , tom bida , bob goodrich and mike brotherton for invaluable help during our keck run . we are grateful to claus leitherer for providing us with the spectrum of ngc 1741b1 , and to dave de young , mike brotherton , mike dopita , buell jannuzi , joan najita and adam stanford for useful discussions and comments on our manuscript . we thank adam stanford and daniel stern for sharing their near - infrared photometry of c in advance of publication . we thank the anonymous referee who provided useful suggestions . the w. m. keck observatory is a scientific partnership between the university of california and the california institute of technology , made possible by the generous gift of the w. m. keck foundation . r. a. acknowledges support from nsf grant ast 93 - 21441 . part of this research was performed at igpp / llnl under the auspices of the u. s. dept . of energy under contract # w-7405-eng-48 . leitherer , c. , vacca , w. d. , conti , p. s. , filippenko , a. v. , robert , c. & sargent , w. l. w. 1996 , , 465 , 717 lowenthal , j. d. , koo , d. c. , guzman , r. , gallego , j. , phillips , a. c. , faber , s. m. , illingworth , g. d. & gronwall , c. 1997 , , in press llrrcrcl 55005800 & 11461208 & [email protected] & @[email protected] & 3.48 & [email protected] & @[email protected] & 58005870 & 12081223 & @[email protected] & @[email protected] & 1.15 & [email protected] & [email protected] & ly@xmath60 58255845 & 12131218 & @[email protected] & @[email protected] & 0.32 & [email protected] & [email protected] & ly@xmath60 ( narrow ) 58707370 & 12231535 & [email protected] & [email protected] & 1.37 & [email protected] & [email protected] & 73707480 & 15351558 & [email protected] & [email protected] & 4.61 & [email protected] & [email protected] & 74807810 & 15581627 & @[email protected] & [email protected] & 2.27 & [email protected] & [email protected] & 78457900 & 16341646 & @[email protected] & @[email protected] & 5.84 & [email protected] & [email protected] & 78959050 & 16451885 & @[email protected] & [email protected] & 4.17 & [email protected] & [email protected] & 91009200 & 18961917 & [email protected] & @[email protected] & 6.06 & [email protected] & @[email protected] & ] [ poldata ] lccccc ly@xmath60(narrow ) & 1215.67 & [email protected] & 771@xmath137 & 613@xmath113 & 548 ly@xmath60(broad ) & 1215.67 & [email protected] & 689@xmath141 & 1373@xmath145 & 490 & 1238.81 & [email protected] & [email protected] & 628@xmath148 & 10.6 & 1242.80 & [email protected] & [email protected] & 628@xmath148 & 13.7 & 1264.89 & [email protected] & [email protected] & 546@xmath1106 & 4.0 & 1393.76 & [email protected] & [email protected] & 1134@xmath1172 & 6.3 & 1402.77 & [email protected] & [email protected] & 1134@xmath1172 & 8.4 & 1548.20 & [email protected] & [email protected] & 541@xmath114 & 58.8 & 1550.77 & [email protected] & [email protected] & 541@xmath114 & 44.8 & 1640.46 & [email protected] & [email protected] & 553@xmath128 & 51.2 ] ( narrow ) & 1908.73 & [email protected] & 19@xmath115 & 511@xmath1151 & 17.0 ] ( broad ) & 1908.73 & [email protected] & 72@xmath115 & 1120@xmath1135 & 64.6 [ emdata ] crcll [email protected] & [email protected] & 2022@xmath192 & ly@xmath60 & p - cygni like & @xmath1351.5 & & @xmath51247 & stellar [email protected] & [email protected] & 558@xmath1117 & @xmath51260.4 & [email protected] & [email protected] & 664@xmath1278 & @xmath211294.6,1296.7 ? & [email protected] & [email protected] & 639@xmath1162 & @[email protected] & 6351.5@xmath111 & [email protected] & 735@xmath1968 & @xmath51334.5 ? & [email protected] & [email protected] & 947@xmath1219 & @xmath51334.5 & [email protected] & [email protected] & 218@xmath163 & @xmath52796.4 at z=1.34 ? & foreground ? [email protected] & [email protected] & 654@xmath1152 & @xmath52803.5 at z=1.34 ? & foreground ? [email protected] & [email protected] & 263@xmath178 & @xmath51371 ? & stellar ? [email protected] & [email protected] & 931@xmath1456 & ? & [email protected] & [email protected] & 445@xmath169 & @xmath51393.8 & p - cygni like [email protected] & [email protected] & 362@xmath192 & @xmath51402.8 & p - cygni like & @xmath1351 & & @xmath51417 & stellar [email protected] & [email protected] & 416@xmath184 & @xmath51502 & stellar [email protected] & [email protected] & 754@xmath1285 & @xmath51526.7 ? & [email protected] & [email protected] & 739@xmath1310 & @xmath51526.7 & [email protected] & [email protected] & 748@xmath1210 & @xmath51548.2 & p - cygni like [email protected] & [email protected] & 748@xmath1210 & @xmath51550.8 & p - cygni like & @xmath1353 & & @xmath51720 & stellar [ absdata ]
the dust attenuation for a sample of local ( ) star forming galaxies is constrained as a function of their physical properties . we utilize aperture - matched multi - wavelength data available from the _ galaxy evolution explorer _ ( _ galex _ ) and the sloan digital sky survey ( sdss ) to ensure that regions of comparable size in each galaxy are being analyzed . we follow the method of and characterize the dust attenuation through the uv power - law index , , and the dust optical depth , which is quantified using the difference in balmer emission line optical depth , . the observed linear relationship between and is similar to the local starburst relation , but the large scatter ( ) suggests there is significant variation in the local universe . we derive a selective attenuation curve over the range and find that a single attenuation curve is effective for characterizing the majority of galaxies in our sample . this curve has a slightly lower selective attenuation in the uv compared to previously determined curves . we do not see evidence to suggest that a 2175 feature is significant in the average attenuation curve . significant positive correlations are seen between the amount of uv and optical reddening and galaxy metallicity , mass , star formation rate ( sfr ) , and sfr surface density . this provides a potential tool for gauging attenuation where the stellar population is unresolved , such as at high- .
1601.00208
the presence of dust in a galaxy causes its spectral energy distribution ( sed ) to experience reddening , a consequence of the highest attenuation occurring in the ultraviolet ( uv ) and decreasing towards longer wavelengths out to the infrared ( ir ) ( see review by * ? ? ? the nature of this reddening is dependent on both the dust properties and its geometry within the galaxy ( see review by * ? ? ? . taken together , these effects limit the interpretation of galaxy seds to determine fundamental quantities such as the stellar population age , stellar mass , and star formation rate ( sfr ) . in the milky way ( mw ; * ? ? ? * ; * ? ? ? * ) and magellanic clouds @xcite , where dust extinction can be directly measured using individual stars , these effects have been well studied and are routinely corrected using extinction curves . the size distribution and composition of dust grains in these galaxies is directly related to the observed extinction . when observing galaxies beyond the mw , the magellanic clouds , and other very nearby galaxies ( e.g. , m31 , * ? ? ? * ; * ? ? ? * ) , it is usually no longer feasible to utilize point sources to derive extinction curves . exceptions include using quasars ( e.g. , * ? ? ? * ) or gamma ray burst ( e.g. , * ? ? ? * ) . instead , we must rely on using unresolved stellar populations , which have more complicated seds as a result of being composed of many stars of varying spectral types and will depend on the star formation history and the initial mass function ( imf ) . in addition , using collections of stars , as opposed to single stars , introduces a wide range of possible geometries for the light sources with respect to the dust . the nature of this geometry determines the importance of light scattering into the line of sight @xcite , and is the reason that it is important to distinguish between extinction and attenuation . this additional component has the effect of flattening or `` graying '' the overall effective extinction , due to bluer light experiencing more efficient scattering . these added complications hinder a general prescription for correcting for dust attenuation in external galaxies . one exception to the aforementioned problem has been for starburst galaxies , for which a tight positive correlation exists between the slope of the uv flux density , @xmath2 , and the color excess of the nebular gas , @xmath8 , which are both related to the wavelength - dependent attenuation @xcite . the color excess can be inferred using the _ differential _ optical depth of the dust from the balmer decrement , @xmath4 , which we will term `` balmer optical depth '' for sake of brevity . tight positive correlations also exist between @xmath2 and the ratio of ir to uv luminosity , termed the `` infrared excess '' ( @xmath9 ) , which is a proxy for the total dust attenuation @xcite . the simple interpretation of the correlation between @xmath2 and @xmath10 is that dust attenuation increases with higher dust - to - gas ratios . unfortunately , the @xmath11 correlation has been shown to break down as one moves from starburst galaxies to more `` normal '' star forming galaxies ( sfgs ) @xcite . this breakdown has been attributed to effects of evolved stellar populations , different star formation histories , and variations in the dust / star geometry ( e.g. , * ? ? ? * ; * ? ? ? * ) , all of which can impact the @xmath10 and/or @xmath2 values and introduce large scatter . the influence of dust appears to increase at intermediate redshifts ( @xmath12 ) , as the sed of galaxies are more heavily attenuated by dust than in the local universe , corresponding to a larger fraction of the star formation within these galaxies being enshrouded by dust @xcite . these effects can result in imprecise values of sfrs , stellar mass , extinction corrections , and photometric redshifts of individual galaxies . uncertainties in the latter two quantities are among the biggest factors limiting current precision dark energy studies , affecting both weak - lensing and supernovae measurements . therefore , it is imperative to accurately characterize the dust attenuation in galaxies in order to reduce the uncertainties in the interpretation of data in future missions that seek to measure cosmological quantities with unprecedented precision . in addition , these quantities are crucial for studies of galaxy formation and evolution . understanding the dust attenuation in local sfgs as a function of their properties ( e.g. , metallicity , stellar mass , sfr ) creates a baseline with which to determine appropriate corrections for higher redshift galaxies . this will allow for accurate determination of galaxy properties when limited measurements are available , as is typically the case for higher redshift systems . for their sample of 39 local starburst galaxies , @xcite find a linear correlation between @xmath2 and @xmath4 . this result implies that the dust behaves as a foreground distribution to the ionized gas , because in this scenario the reddening of the stellar continuum linearly correlates with the reddening of nebular regions . they derive the selective attenuation for this sample by comparing the average seds of galaxies binned according to @xmath4 . virtually all studies of sfgs , both local and distant , make use of the attenuation curve derived from this sample @xcite to correct for effects of dust and determine properties of those galaxies . the appeal of this method stems from its simple approach ; determining rest - frame uv colors ( @xmath2 ) , which correspond to observer - frame optical colors for high-@xmath7 galaxies , allows for the dust - free luminosity to be recovered . however , since the @xcite attenuation curve was calibrated with a relatively small number of local starburst galaxies , it is not clear how accurate such generalizations are to more typical sfgs with lower specific star formation rates ( ssfrs ; @xmath13 ) . more importantly , the extent to which the relation holds true as a function of redshift has not been conclusively determined . recent results by @xcite suggest that the relationship between @xmath2 and @xmath4 is shallower for galaxies at @xmath14 and is dependent on the ssfr . they also find that the attenuation curve for these galaxies is lower by about 20% in the uv . a similar study of a large number of local sfgs will provide a strong foundation with which to compare and address why such differences exist . other studies have examined the nature of attenuation for large samples of local galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ) in order to address the degree to which it can vary . however , these studies have used different techniques than those described in @xcite , which can introduce different biases and make direct comparison unclear . more specifically , @xcite use average seds of galaxies separated according to @xmath10 and @xcite utilize a galaxy pair - matching technique ( matched in gas - phase metallicity , ssfr , axial ratio , and redshift ) to compare the seds of more dusty and less dusty galaxies as determined by @xmath4 . despite the different methods for constructing attenuation curves among these studies , a common picture has developed in which the dust content of galaxies appear to have two components ; one associated with short - lived dense clouds where massive stars form hii regions and another associated with the diffuse interstellar medium . for this study , we follow the methodology used in @xcite to determine the behavior of dust attenuation in a large sample of sfgs with a wide range of properties in order to test the extent to which the dust attenuation and geometry found in that work might hold for a more diverse sample of galaxies . understanding the geometry of the dust in these systems gives important information on where dust is located in galaxies . the large sample size will also allow us to examine sources of scatter in the attenuation properties of individual galaxies . furthermore , our results will allow for more detailed future analysis into the nature of the @xmath11 relationship breakdown for normal sfgs at @xmath15 . throughout this work we adopt a @xmath16-cdm concordance cosmological model , @xmath17 km / s / mpc , @xmath18 , @xmath19 . we also assume a kroupa imf @xcite when making comparisons with stellar population models . to avoid confusion , we make explicit distinction between the color excess of the stellar continuum @xmath20 , which traces the reddening of the bulk of the galaxy stellar population , and the color excess seen in the nebular gas emission @xmath8 , which traces the reddening of the ionized gas around massive stars located within hii regions . in principle these two parameters need not be related because @xmath20 and @xmath8 are a result of attenuation and extinction , respectively ( i.e. , they use different obscuration curves ; see [ method_tau ] ) , but they have been found to be correlated in starburst galaxies , with @xmath21 @xcite , and in star forming regions within local galaxies , with @xmath22 @xcite . our sample is constructed using the _ galaxy evolution explorer _ ( _ galex _ ; * ? ? ? * ; * ? ? ? * ) catalogs of @xcite . these catalogs represent unique sources in the _ galex _ data release 6/7 ( gr6/7 ) and are separated for the all - sky imaging survey ( ais ; depth @xmath23 mag in fuv / nuv ) , containing @xmath071 million sources , and the medium imaging survey ( mis ; depth @xmath24 mag ) , containing @xmath016.6 million sources @xcite . we only make use of sources that can be cross - matched to the sloan digital sky survey ( sdss ) data release 7 ( dr7 ; * ? ? ? * ) sources and have available sdss spectroscopy , as the latter is required to determine the balmer decrement . these cross - matched cases are determined using the mikulski archive for space telescopes ( mast ) casjobs website , with the requirement that the separation between objects be within @xmath25 to be a match . since our analysis requires detection in both the _ galex _ fuv ( @xmath26 ) and nuv ( @xmath27 ) bands , we have chosen to only consider galaxies within the mis catalog because spectroscopic sdss galaxies detected within both bands in the shallower ais are found to be highly biased towards very blue galaxies ( see [ sample_compare ] for details ) . within the mis catalog , this restricts the parent sample to 63691 galaxies . for the purpose of this study , we further constrain the sample to only star forming galaxies ( sfgs ) and exclude cases in which a significant fraction of the flux density is produced from an active galactic nucleus ( agn ) . the galaxy type is determined using the traditional optical emission line diagnostics @xmath28\lambda6583/h\alpha$ ] , a proxy for gas phase metallicity , and @xmath29\lambda5007/h\beta$ ] , a measure of the hardness of the radiation field . the optical spectroscopic measurements for these galaxies are from the max planck institute for astrophysics and johns hopkins university ( mpa / jhu ) group , which is based on the method presented in @xcite . to summarize , line fluxes are corrected for stellar absorption by fitting a non - negative combination of stellar population synthesis models from for the sdss dr4 and updated in dr7 using a newer version of these models ( unpublished ) . as recommended by the mpa / jhu group starting with dr4 , we increase the uncertainties associated with each emission line . we adopt the values listed in @xcite , which are updated for the dr7 dataset . it has been found by @xcite that the equivalent width of the h@xmath2 emission line of galaxies in the mpa / jhu catalog appear to be underestimated by @xmath0@xmath30 due to their method of correcting balmer absorption . however , this correction value was derived assuming these galaxies follow the @xcite attenuation curve and because we are seeking to determine if there are significant departures from that relation , we choose not to adopt it . we have examined the effect of including this correction and found that it causes a systematic decrease in the estimated balmer optical depth of @xmath31 . as was seen in @xcite , we find this shift to be uniform across the entire range of balmer emission line strengths ( i.e. , @xmath4 values ) . for this reason , we expect that this will not influence the attenuation curve derived later because only the difference in balmer optical depth is used and not the absolute value . we require that all emission lines have a signal - to - noise ratio ( @xmath32 ) greater than 5 for classification , with the additional constraint that the fuv and nuv measurements have @xmath33 and that the redshift of the galaxy be @xmath34 . the redshift restriction is required to ensure that the fuv passband ( @xmath351516 , fwhm=269 ) lies above the numerous stellar absorption features that occur below 1250 . using this selection criteria gives a final sample of 9813 sfgs . all photometry and spectroscopy has been corrected for foreground milky way extinction using the _ galex _ provided @xmath36 with the extinction curve of @xcite . for the _ galex _ bands , we adopt the values of @xmath37 is the total - to - selective extinction . ] and @xmath38 , which represent the average value of the mw extinction curve convolved with each filter on seds with uv slopes @xmath39 , the typical range for our sfg sample . all measurements of galaxy properties utilized in this work are those provided by the mpa / jhu group and correspond only to the 3 sdss fiber , which is typically centered on the nuclear region and represents only a fraction of the total galaxy . the stellar masses are based on fits to the photometric data following the methodology of @xcite and @xcite . the sfrs are based on the method presented in @xcite . the gas phase metallicities are estimated using models as outlined in @xcite . the fiber regions of the 9813 sfgs in this sample span the following range in properties : @xmath40<10.67 $ ] , @xmath41<1.60 $ ] , and @xmath42 . for comparison with @xcite , our sample consists of galaxies with lower ssfrs ( average h@xmath43 emission equivalent width , @xmath44 ; a proxy for ssfr ) relative to their starburst sample ( @xmath45 ; * ? ? ? * ; * ? ? ? in order to test the existence of any relation between the uv flux density measured by _ galex _ and the optical flux density measured by sdss it is crucial that the apertures be closely matched in order to ensure the they arise from regions of the galaxy that are comparable in size . this is also essential in order to utilize the seds of these galaxies to derive the underlying attenuation curve . the limiting factor in this respect is the @xmath25 diameter of the sdss spectroscopic fiber . this aperture is smaller than the point - spread function ( psf ) of _ galex _ at both fuv ( @xmath46 ) and nuv ( @xmath47 ) . using an aperture which is smaller than _ galex _ psf would add positional uncertainty that could introduce uv emission unassociated with the fiber location . however , using an aperture much larger than the psf would require large aperture corrections for our fiber measurements , which would also introduce uncertainty . in addition , the large psf of _ galex _ can lead to nearby objects contributing to the observed flux density within a given area and using a large aperture would increase the likelihood of this happening . as a compromise between these issues we choose to adopt a @xmath48 diameter aperture for our analysis . the uv photometry and associated uncertainties at this aperture were retrieved directly from the mast database using casjobs . we note that by adopting a fixed aperture , the physical sizes being probed will vary from sub - kpc at the lowest redshifts up to several kpc at the higher redshifts ( @xmath49 ) . we address the impact of this effect on our results in [ atten_vs_param ] & [ curve_vs_param ] when we separate the sample according to redshift . given that the chosen aperture is roughly the size of the psf for the _ galex _ bands , determining the appropriate aperture corrections for the uv is non - trivial as a significant amount of light from outside of the aperture region can be spread within it . to distinguish these effects we utilize the light profile models of sdss galaxies from the nyu value - added galaxy catalog ( nyu - vagc ; * ? ? ? these profiles are one - component srsic fits of each individual band of sdss , @xmath50 \,,\ ] ] where @xmath51 is an amplitude in nanomaggies / arcsec@xmath52 , @xmath53 is an effective radius in arcsec , and @xmath54 is the srsic index . a nanomaggie is a flux density unit defined such that 1 nanomaggie has a magnitude of 22.5 in any band , @xmath55 $ ] , which for sdss as a near ab magnitude system is @xmath56 @xmath57jy . if we assume that the uv colors follow the @xmath58-band light profile , which is the shortest wavelength sdss band , then it is possible to determine appropriate aperture corrections for each individual galaxy . this is done by measuring the amount of light within @xmath48 in the modeled @xmath58-band profile , @xmath59 , and comparing this to the same measurement after the light profile has been convolved with the _ galex _ psfs , @xmath60 . these psfs are obtained from the _ galex _ data analysis website . the aperture correction is taken as the ratio of these values , @xmath61 . the distribution of aperture corrections for the entire parent sample of galaxies is shown in figure [ fig : galex_aper_corr ] . roughly one - third of the sources in our sample are well approximated as point sources in _ galex _ ( corresponding to being near the peak in the distribution ) . the aperture correction in flux density for a point - source is a factor of 2.29 and 2.86 for fuv and nuv , respectively . it can be seen in figure [ fig : galex_aper_corr ] that there are some sources with aperture corrections that are larger than what is expected for a point source ( vertical dashed lines ) . to determine the cause of this interesting behavior we examined the relationship between the aperture corrections with the effective radius of the galaxy and its srsic index in the @xmath58-band . we choose to use the @xmath62 light radius from the nyu - vagc light profile , @xmath63 , instead of the variable @xmath53 in the srsic fit , as it is a better representation of the size of the galaxy . we find that the behavior of the aperture correction is directly related to @xmath63 and the srsic index , which we illustrate in figure [ fig : galex_aper_corr_demo ] and discuss below . we find that sources with radii of @xmath64 or with @xmath65 and large srsic indices ( i.e. , steep light profile ) are well described by a point - source correction ( see panel ( a ) in figure [ fig : galex_aper_corr_demo ] ) . for the regime of galaxies with @xmath66 and small srsic indices ( i.e. , shallow light profile ) , there is little galaxy flux density outside of the @xmath48 aperture to spread inside of it , due to the smearing effect of the psf , but the light within it is being spread out more than would be the case for a point - source . together these effects result in a larger aperture correction being necessary than would be the case for a point - source ( see panel ( b ) in figure [ fig : galex_aper_corr_demo ] ) . for galaxies with @xmath67 and small srsic indices , there is significant light outside of the @xmath48 aperture that can be spread into it and we find an aperture correction that is smaller than that for a point - source ( see panel ( c ) in figure [ fig : galex_aper_corr_demo ] ) . apertures based on the sdss @xmath58-band srsic light profile , @xmath59 , relative to the flux within the same aperture after convolving @xmath58-band light profile by the _ psfs , @xmath60 , for the 9813 sfgs in our sample . this ratio corresponds to the aperture correction , @xmath61 , under the assumption that the uv light follows a similar behavior to that of the @xmath58-band . roughly half of the sources are well approximated as point - sources , which corresponds to the peaks in the distributions ( vertical dashed lines ) . [ fig : galex_aper_corr],width=336 ] @xmath68{aper_corr_demo1.eps } \\ \includegraphics[width=3.5in]{aper_corr_demo2.eps } \\ \includegraphics[width=3.5in]{aper_corr_demo3.eps } \\ \end{array}$ ] since we have chosen to adopt a @xmath48 aperture for our galaxies , we also need to apply an aperture correction to the sdss spectroscopy ( @xmath25 ) . following a similar methodology to before , we determine the correction for the observed photometry using the light profile models from the nyu - vagc . the correction is taken to be the ratio of light within @xmath48 and @xmath69 in the modeled band profile . we then perform a chi - squared minimization to match the optical spectrum to the @xmath48 photometry . all nebular line diagnostics used in this study are ratios of emission lines ( @xmath28/h\alpha$ ] , @xmath29/h\beta$ ] , and h@xmath43/h@xmath2 ) , and the effects of the aperture corrections are very small . the reason for this is because the relative difference in the correction terms across the sdss bands is small ( @xmath70 , @xmath71 dispersion of @xmath72 ) . as a check , we estimated a correction for each the optical emission lines by taking a linear interpolation between the corrections of the two closest bands . the largest aperture effects for the line ratios will occur for lines which are separated the furthest in wavelength ; which in this study is the ratio of the h@xmath43 to h@xmath2 . we find that the distribution of the ratio of the aperture corrections for h@xmath43 to h@xmath2 has a mean of @xmath73 , with a @xmath71 dispersion of 0.03 . this implies these corrections will only change the ratio measurements at the level of a few percent and will be even smaller for the other line ratios ( @xmath74 ) . as these are minor changes we have chosen not to apply any aperture corrections to the emission lines . as a check on the accuracy of the aperture matching , we compare the corrected uv flux densities to the corrected optical spectra and inspected if the shape of the uv ( inferred from @xmath75 ) agrees with the shortest wavelength data available in the optical spectrum ( @xmath76 , for @xmath77 ) . in other words , we examine whether the region between @xmath78 , corresponding to the gap in our data , that would be extrapolated from the uv slope and the optical data shortward of the 4000 break feature are in agreement . if our aperture corrections are inaccurate , then we expect to see systematic offsets between the flux densities in the uv and optical . as will be shown with our average templates in [ attenuation curve ] , we find that on average the uv data agrees well with the optical data , with no significant offsets between them . the dust attenuation in a galaxy can be measured from the optical depth , @xmath79 . for the simple case of a uniform layer of dust between a source of intensity , @xmath80 , and the observer , the optical depth is defined as @xmath81 where @xmath82 is the observed intensity and all quantities are dependent on the wavelength . in the case of a point source such as a star with a well - characterized sed , the optical depth can easily be determined by comparing the observed and intrinsic seds . however , in the case of entire galaxies for which the underlying sed is strongly affected by many factors , including the underlying stellar population , star formation history , and imf , this becomes much more difficult . to mitigate these problems the flux ratio of h@xmath43 and h@xmath2 is often utilized , as the intrinsic flux ratio is set by quantum mechanics and is only affected by the electron temperature , @xmath83 , and density , @xmath84 , at the @xmath85 level @xcite . this ratio is also relatively insensitive to the underlying stellar population and imf @xcite . therefore , large variations from the intrinsic ratio can be directly attributed to the reddening of dust . for our work we will use the balmer decrement , @xmath86 , as a tracer of dust attenuation , with the assumption that this dust acts as a foreground screen for these lines . since the ionized gas from which these lines arise is primarily located in hii regions , which have small angular extent relative to the rest of the galaxy ( and presumably the dust ) , and are usually distributed within a short height of the galaxy s midplane , this is a reasonable assumption . following from equation ( [ eq : intensity ] ) and @xcite , we define the balmer optical depth as @xmath87 where @xmath88h@xmath89 and @xmath88h@xmath90 are the flux of the nebular emission lines located at 6562.8 and 4861.4 , respectively , and the value of 2.86 comes from the theoretical value expected for the unreddened ratio h@xmath91h@xmath2 undergoing case b recombination with @xmath92 k and @xmath93 @xmath94 @xcite . the superscript _ l _ is used to emphasize that this quantity is coming from emission lines and should be distinguished from optical depths associated with the stellar continuum . if one assumes knowledge of the total - to - selective extinction , @xmath95 , then @xmath4 can be directly related to the color excess of the nebular gas , @xmath8 , through @xmath96 where @xmath97 is the total extinction at a given wavelength . if the extinction in other galaxies at these wavelengths were to be identical to the mw , which is unlikely to always be the case , then we can use @xmath98 @xcite . since the hydrogen recombination lines are primarily produced within the hii regions of massive o and b stars , this implies that they are only sensitive to the attenuation of ionized gas around these massive stars . in general these same stars will also contribute greatly to the uv and optical light of the total stellar population , leading one to expect the reddening of the ionized gas to be related to the reddening seen in the stellar population ( traced by @xmath2 or @xmath20 ) . however , the distribution of these massive stars and the rest of the stellar population relative the distribution of dust , in addition to the age distribution of the stellar population ( i.e. , the relative contribution of older stars to the uv - optical flux density ) , will strongly influence how these quantities are related . for both starburst galaxies @xcite and star forming regions within local galaxies @xcite it appears that the stellar continuum suffers roughly one - half of the reddening of the ionized gas . this result suggests a scenario in which massive stars are more deeply embedded in molecular clouds than the long - lived stars ( e.g. , see descriptions in , @xcite , or @xcite ) . another concern is that the total amount of dust within a galaxy may only weakly relate to attenuation , as the regions of lowest optical depth ( @xmath99 ) are providing the majority of the flux density @xcite . large amounts of the dust might exist in the regions of high optical depth ( @xmath100 ) , which would provide negligible amounts of flux density and thus not be represented by our tracers . fortunately , the presence of dust attenuation beyond the level of what can be measured in the optical has not been shown to be an issue for local starburst galaxies @xcite , and therefore we do not expect this to be a major concern . for actively sfgs , where the uv is dominated by recent star formation ( i.e. , massive stars ) and there is little contamination from earlier generations of stars , the dust attenuation can also be measured from the uv flux density spectral slope , @xmath2 , which is defined as @xmath101 where @xmath102 is the flux density in the range @xmath103 . for reference , @xmath2 values cover the range between @xmath104 and @xmath105 for constant star formation ( i.e. , they have a fairly small intrinsic variation ; * ? ? ? this makes this parameter ideal for measuring the wavelength - dependence of attenuation in the uv . in appendix [ opt_slope ] , we examine the use of the optical slope as dust tracer when uv data is unavailable . for this study , we use the uv power - law index @xmath75 measured from observed _ galex _ fuv and nuv photometry , @xmath106}{\log[\lambda_{\mathrm{fuv}}/\lambda_{\mathrm{nuv}}]}\,,\ ] ] where the flux density is in erg s@xmath107 @xmath108 @xmath107 , @xmath109 , and @xmath110 . if one assumes a power - law fit to the region described above , there is no need to perform @xmath111-corrections on the flux density . we examine the possibility of a 2175 absorption feature biasing this uv slope measurement in [ influence_of_2175 ] . as is shown in @xcite , the quantities @xmath4 and @xmath2 can be used to _ derive _ a dust attenuation curve independent of any prior knowledge of its shape . such an analysis will be performed in [ attenuation curve ] . in order to characterize the dust attenuation in sfgs , we first examine how the reddening in the uv stellar continuum ( measured through @xmath75 ) is related to the optical reddening of the ionized gas ( measured through @xmath4 ) . in figure [ fig : beta_tau ] we show the @xmath75 and @xmath4 values for our sample of 9813 sfgs . there does appear to be a significant correlation present , but with a large degree of scatter . spearman and kendall nonparametric correlation tests give @xmath112 and @xmath113 , respectively , which indicate that this correlation is not particularly strong . given the large sample size we are dealing with , it is not possible to report the significance of these correlation coefficients because the probability of no correlation existing is found to be very close to zero . looking at figure [ fig : beta_tau ] , it can be seen that there are a number of cases for which @xmath114 . this corresponds to cases where the observed flux ratio of the balmer lines is below the assumed intrinsic value of 2.86 . this can result from the uncertainties in the measured values of these lines ( see representative error bar ) or from variations in the intrinsic line ratio ( e.g. , for @xmath115 k and/or @xmath116 @xmath94 the intrinsic ratio decreases ; * ? ? ? * ) . another interesting feature is the lack of data points at @xmath117 , which corresponds to galaxies experiencing the largest attenuation . it is worth determining if this is being driven by the selection criteria imposed for our sample , which requires the uv flux density and emission line fluxes to have @xmath33 , and if it could result in any biases . we postpone the analysis of a uv selection bias until [ sample_compare ] , where we will compare sources from _ galex _ surveys of different depths . to test for an emission line selection bias , we examine the effect that requiring a lower @xmath32 threshold for the weakest line in this study ( [ oiii]@xmath118 ) has on the sample being selected . imposing a threshold of @xmath119 for [ oiii]@xmath118 , while still requiring @xmath33 for the other lines , increases the sample size by 2160 galaxies . imposing lower thresholds on the other lines does not increase the sample size significantly . we do find that the weaker [ oiii ] systems are slightly more attenuated on average than the original sample ( i.e. , larger @xmath75 & @xmath4 ) , but the vast majority are still @xmath120 . including these galaxies has no effect on the relationship between @xmath75 and @xmath4 . therefore , we believe that we are not preferentially excluding objects located at @xmath117 . the lack of objects at these values may result from galaxies at this level of attenuation being relatively rare in the local universe , which is consistent with the results of @xcite . similar to the findings of @xcite , the observed relationship between @xmath75 and @xmath4 is linear , which indicates that the dust behaves as a foreground - like screen to ionized gas regions . we fit a linear relationship to the data in figure [ fig : beta_tau ] using the mpfitexy routine @xcite , which utilizes the mpfit package @xcite . this routine performs a linear least - squares fit with error in both variables while also including a term to account for intrinsic scatter in the data . a scatter within the data is expected for the y - axis given that @xmath75 is likely to be dependent on variations in the age of the stellar population , the star formation history , and/or the metallicity of each galaxy @xcite . a fit to the @xmath75 and @xmath4 values for all of the data gives @xmath121 with an intrinsic dispersion of @xmath122 . we denote the region of @xmath123 in our fit with a dashed line to indicate that there are limited data in this range . interestingly , the width of the scatter in @xmath75 does not appear to change with @xmath4 . as a result of this behavior , we do not expect the scatter to be strongly driven by possible variations in the dust geometry . this is because the scatter around the value of @xmath75 should decrease for `` low - dust '' systems ( @xmath124 ) , regardless of geometry , and approach the intrinsic value of @xmath75 for each galaxy ( e.g. , see * ? ? ? * ; * ? ? ? . we will examine this scatter in more detail in [ atten_vs_param ] . we also examine the data after binning it into 6 bins of @xmath4 , which is useful for comparison in our derivation of the attenuation curves in [ attenuation curve ] . the ranges of the bins along with their weighted - mean values , which account for uncertainties in both variables , are shown in table [ tab : tau_bin ] . the error bar shown for each bin correspond to the measurement uncertainty and the sample dispersion added in quadrature . it can be seen that these bins are in good agreement with the previous fit , given the uncertainties . we postpone a comparison of our @xmath125 relation to those in the literature until [ beta_glx_vs_z ] , as we will make use of our derived attenuation curve to determine an appropriate way to compare different techniques for measuring the uv slope . .values of @xmath4 bins [ tab : tau_bin ] [ cols="^,^,^,^,^",options="header " , ] * notes . * the functional form of these fits are @xmath126 . we also report the intrinsic dispersion , @xmath127 , which is taken as the standard deviation from the fitted relation . these fits are performed without accounting for the uncertainties in the variables and are only intended to illustrate general trends . the coefficients for the spearman , @xmath128 , and kendall , @xmath129 , nonparametric correlation tests are also given for each case . @xmath130{beta_oh_mis.eps } \includegraphics[width=3.in]{tau_oh_mis.eps } \\ \includegraphics[width=3.in]{beta_gm_mis.eps } \includegraphics[width=3.in]{tau_gm_mis.eps } \\ \includegraphics[width=3.in]{beta_sfr_mis.eps } \includegraphics[width=3.in]{tau_sfr_mis.eps } \\ \includegraphics[width=3.in]{beta_sfr_eff_mis.eps } \includegraphics[width=3.in]{tau_sfr_eff_mis.eps } \\ \end{array}$ ] given the large number of sources in our sample , we can also examine the variation in the behavior of the @xmath2-@xmath4 relation for galaxies with specific properties and identify key drivers of the large intrinsic scatter . recent findings by @xcite examining sfgs at @xmath14 , found that galaxies appear to show significant differences in the @xmath2-@xmath4 relation as a function of galaxy ssfr . we separate our sample into three subsamples according to @xmath131 and ssfr , as these may be expected to correlate to the intrinsic uv slope . we also divide the sample by @xmath7 in order to test whether a fixed aperture with redshift influences the measurements . the results are shown in figure [ fig : beta_tau_subsample ] . we chose the subsamples to consist of roughly one - third of the sample ( @xmath132 galaxies ) . we note regions with low sampling of galaxies with dashed lines . looking at figure [ fig : beta_tau_subsample ] , it can be seen that no significant differences are evident in the @xmath2 vs @xmath4 relation with @xmath131 , ssfr , or @xmath7 , which indicates that the intrinsic scatter is not driven primarily by these parameters . we do not observe large variations in the offset of @xmath2 with ssfr , as is seen in @xcite , but we note these studies probe different regimes , with the bin of highest ssfr ( @xmath133<-8.9 $ ] ) in our sample corresponding to the bin of lowest ssfr their sample ( @xmath133<-8.84 $ ] ) , which makes direct comparison difficult . we find that it is not particularly informative to divide the sample according to galaxy metallicity , mass , sfr , or @xmath134 because these show significant correlation with the attenuation parameters , which implies that these parameters segregate galaxies in @xmath2-@xmath4 parameter space . as a result , this makes it difficult to differentiate between variation in attenuation relations because low and high attenuation galaxies would be separated . as an example , we illustrate the sample separated into three equally spaced bins in metallicity , which is argueably more informative here instead of equal number bins , in figure [ fig : beta_tau_subsample ] . it can be seen that the higher metallicity galaxies ( @xmath135 ) , which correspond to the majority of our sample , are driving much of the observed trend between @xmath2 and @xmath4 . @xmath136{beta_corr_tau_mis_dn4000.eps } \\ \includegraphics[width=7.0in]{beta_corr_tau_mis_ssfr.eps } \\ \includegraphics[width=7.0in]{beta_corr_tau_mis_z.eps } \\ \includegraphics[width=7.0in]{beta_corr_tau_mis_oh.eps } \\ \end{array}$ ] in a similar manner to how the sample was divided to see the effect on the observed values of @xmath2 and @xmath4 , we can now examine how the attenuation curve changes with these properties . we utilize the same subsamples presented in [ atten_vs_param ] of @xmath131 , ssfr , and @xmath7 . for each parameter except @xmath131 , we add an additional constraint that @xmath137 in order to limit the stellar population age effects . we follow the same methodology presented in [ attenuation curve ] , dividing each sample into 6 bins of @xmath4 and constructing average flux templates , to derive the attenuation curve for these subsamples . we do not consider bins with less than 100 galaxies in determination of the effective attenuation curve . for nearly all cases the effective attenuation curve in the uv ( derived from @xmath75 ) and optical ( derived from sdss spectra ) appear to be in agreement and well approximated by a single third - order polynomial . all of the fits to these subsamples are presented in table [ tab : q_vs_param ] . we plot the subsample curves of @xmath138 alongside our average curve derived earlier and the @xcite curve ( left panels ) , as well as the curves normalized by @xmath139 and alongside the @xcite mw curve ( right panels ) , in figure [ fig : q_compare_subsample ] . slight differences in @xmath138 appear in the @xmath131 and ssfr subsamples , with the latter case being more significant . if we make the crude assumption that these galaxies have similar extinction curves , then this would indicate variation in @xmath140 . these differences suggest that galaxies with lower @xmath131 or higher ssfr have slightly higher ratios of @xmath140 ( lower @xmath139 ) . these changes are quite interesting because they indicate differences in the relative reddening of the ionized gas and the stellar continuum . one possible explanation for this is that in galaxies with elevated sfrs the uv reddening is more heavily weighted towards the same regions that dominate the balmer line emission , thus increasing @xmath140 . one can imagine that for the extreme scenario in which nearly all of the global flux density is exclusively produced in hii regions , that this ratio would approach unity . however , this behavior is dependent on the optical depth of these star forming regions . in their study of @xmath14 sfgs , @xcite find a lower ratio with increasing ssfr which they attribute to a larger fraction of the star formation in the galaxy becoming obscured in optically thick regions as the sfr increases . these obscured regions do not contribute significantly to the uv emission , but they continue to contribute to the balmer line emission . thus , the uv slope underestimates the dust attenuation relative to the balmer line inferred attenuation for these galaxies such that @xmath140 decreases toward larger sfrs . it is important to note that a lower ratio of @xmath140 is also possible if higher star formation activity gives rise to significant outflows which reduce the overall optical depth affecting the stellar continuum . it is also important to state again that the sample of @xcite probes a higher range of ssfr than this work and this may lead to differences in the underlying physical mechanisms at work between the two samples . given the complex dependence of this ratio , we can not make any clear statement as to the cause of the differences that we see in our sample . another notable result that can be seen in figure [ fig : q_compare_subsample ] is that there is virtually no dependence on the behavior of the attenuation curve as a function of the redshift spanned by our sample . we take this to indicate that the different physical aperture scales being probed as a result of our choice of fixed angular aperture does not seem to significantly alter the resulting curve . this also indicates that the average attenuation properties over smaller galaxy regions do not significantly deviate from the total values . despite the changes seen in the behavior of @xmath138 with proxies for stellar age , after normalizing the curves as @xmath141 the difference in the curves is significantly reduced in all cases ( see right panels of figure [ fig : q_compare_subsample ] ) . a similar result was found by @xcite for their @xmath14 sample of galaxies separated by ssfr , albeit with a different overall shape than we find for local galaxies . this remarkable result indicates that despite the differences in physical properties and sfhs that are spanned by sfgs in the local universe , on average they appear to suffer from a _ similar attenuation curve_. however , the large scatter in the @xmath2-@xmath4 relation and the flux density seds likely implies that there are variations in the attenuation on a case - by - case basis , part of which can stem from differences in the star - dust geometry . ccccccc @xmath142 & range & @xmath139 & @xmath143 & @xmath144 & @xmath145 & @xmath146 + @xmath131 & @xmath147 & 2.396@xmath148 & -2.488 & 1.803 & -2.609@xmath14910@xmath107 & 1.452@xmath14910@xmath150 + & @xmath151 & 2.283@xmath152 & -2.565 & 1.823 & -2.510@xmath14910@xmath107 & 1.332@xmath14910@xmath150 + & @xmath153 & 2.840@xmath154 & -2.139 & 1.551 & -2.320@xmath14910@xmath107 & 1.373@xmath14910@xmath150 + & @xmath155 & 2.840@xmath156 & -2.138 & 1.567 & -2.358@xmath14910@xmath107 & 1.364@xmath14910@xmath150 + + log[ssfr ( yr@xmath107 ) ] & @xmath157@xmath158 & 2.580@xmath159 & -2.423 & 1.817 & -2.862@xmath14910@xmath107 & 1.636@xmath14910@xmath150 + & @xmath160@xmath158 & 1.931@xmath161 & -3.065 & 2.225 & -3.168@xmath14910@xmath107 & 1.664@xmath14910@xmath150 + & @xmath162@xmath158 & 1.804@xmath163 & -3.185 & 2.275 & -3.063@xmath14910@xmath107 & 1.567@xmath14910@xmath150 + + @xmath7 & @xmath164@xmath158 & 2.302@xmath165 & -2.642 & 1.931 & -2.913@xmath14910@xmath107 & 1.717@xmath14910@xmath150 + & @xmath166@xmath158 & 2.296@xmath167 & -2.488 & 1.759 & -2.317@xmath14910@xmath107 & 1.169@xmath14910@xmath150 + & @xmath168@xmath158 & 2.260@xmath169 & -2.584 & 1.859 & -2.572@xmath14910@xmath107 & 1.321@xmath14910@xmath150 + * notes . * the uncertainty in @xmath139 denotes the maximum and minimum values from fits using individual @xmath170 for each subsample ( see [ attenuation curve ] ) . the functional form of these fits are @xmath171 . @xmath158these cases also have the constraint that @xmath172 . @xmath173{qn_compare_dn4000.eps } \includegraphics[width=3.3in]{fqn_compare_dn4000.eps } \\ \includegraphics[width=3.3in]{qn_compare_ssfr.eps } \includegraphics[width=3.3in]{fqn_compare_ssfr.eps } \\ \includegraphics[width=3.3in]{qn_compare_z.eps } \includegraphics[width=3.3in]{fqn_compare_z.eps } \\ \end{array}$ ] we use a sample of @xmath010000 local ( @xmath1 ) star forming galaxies to constrain the nature of dust attenuation in galaxies as a function of their physical properties . utilizing aperture - matched uv and optical data , we find a linear relationship between the uv power - law index , @xmath2 , and the balmer line optical depth , @xmath4 , which is similar to the local starburst relation . the large scatter ( @xmath5 ) of this relation suggests there is significant variation in the attenuation of individual galaxies local universe . using this large sample , we are able to quantify how the attenuation is influenced by varying galaxy parameters . we observe significant correlations between the amount of uv and ionized gas reddening with galaxy metallicity , @xmath174 , sfr , and @xmath175 . a weaker negative correlation is seen with the mean stellar age , traced by the 4000 break ( @xmath131 ) . these trends are consistent with a scenario in which the total dust content increases with star formation activity and also builds up slowly with age . these relationships can provide a way for determining attenuation in other studies if these parameters are available . however , we stress that the redshift evolution of some of these relationships ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) poses a problem in the application to higher redshift studies . using our sample we derive an attenuation curve over the wavelength range @xmath6 . we find a lower selective attenuation in the uv compared to previously determined attenuation curves and which is about 20% lower than the starburst curve from @xcite at 1250 . however , given that the normalization of our curve is still unknown , it is not clear whether this also corresponds to lower _ total _ attenuation . such an analysis will be the subject of a future study . we see no evidence to suggest that a significant 2175 feature is present in this curve , although this can not be conclusively determined without available uv spectroscopy . the relative reddening of the stellar continuum is roughly one - half of the amount suffered by the ionized gas , with @xmath176 ( assuming @xcite mw extinction for the ionized gas ) , in good agreement with previous studies @xcite . we emphasize that this is the average relation and that individual cases will vary from this ratio depending on their properties . when dividing the sample according to different galaxy properties we find that galaxies with larger ssfrs have smaller @xmath139 values ( larger ratios of @xmath140 ) . however , after normalizing the curves to remove the effects of differential reddening , the variation in the curves is significantly reduced in all cases . this result indicates that despite differences in physical properties and sfhs spanned by sfgs in the local universe , on average they appear to suffer from a _ similar underlying attenuation curve_. this single attenuation curve is well suited for application to large statistical studies of sfgs , but should be used with caution on a case - by - case basis . the authors thank the anonymous referee whose suggestions helped to clarify and improve the content of this work . ajb also thanks k. grasha for comments that improved the clarity of this paper . part of this work has been supported by nasa , via the jet propulsion laboratory euclid project office , as part of the `` science investigations as members of the euclid consortium and euclid science team '' program . this work is based on observations made with the nasa galaxy evolution explorer . _ galex _ is operated for nasa by the california institute of technology under nasa contract nas5 - 98034 . this work has made use of sdss data . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the us department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society and the higher education funding council for england . the sdss website is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . abazajian , k. n. , adelman - mccarthy , j. k. , ageros , marcel a. , et al . 2009 , apjs , 182 , 543 baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , pasp , 93 , 5 bianchi , l. , clayton , g. c. , bohlin , r. c. , hutchings , j. b. , & massey , p. 1996 , , 471 , 203 bianchi , l. , conti , a. , shiao , b. 2014 , adspr , 53 , 900 blanton , m. r. , eisenstein , d. , hogg , d. w. , schlegel , d. j. , & brinkmann , j. 2005 , , 629 , 143 boquien , m , calzetti , d. , kennicutt , r. , et al . 2009 , , 706 , 553 bouwens , r. j. , illingworth , g. d. , oesch , p. a. , et al . 2012 , , 754 , 83 brinchmann j. , charlot s. , white s. d. m. , tremonti c. , kauffmann g. , heckman t. , brinkmann j. , 2004 , mnras , 351 , 1151 bruzual g. , & charlot s. 2003 , mnras , 344 , 1000 buat , v. , iglesias - pramo , j. , seiber , m. et al . 2005 , , 619 , l51 buat , v. , giovannoli , e. , heinis , s. , et al . 2011 , a&a , 533 , a93 buat , v. , noll , s. , burgarella , d. , et al . 2012 , a&a , 545 , a141 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , , 429 , 582 calzetti , d. , & heckman , t. 1999 , , 519 , 27 calzetti , d. , armus , l. , bohlin , r. c. , et al . 2000 , , 533 , 682 calzetti , d. 2001 , pasp , 113 , 1449 calzetti , d. 2013 , secular evolution of galaxies ed . j. falcn - barroso & j. h. knapen ( cambridge : cambridge univ . press ) , 419 calzetti , d. , johnson , k. e. , adamo , a. 2015 , , 811 , 75 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 charlot , s. , & fall , s. m. 2000 , , 539 , 718 charlot , s. , & longhetti , m. 2001 , mnras , 323 , 887 clayton , g. c. , gordon , k. d. , bianchi l. c. , et al . 2015 , , in press conroy , c. , schiminovich , d. , & blanton , m. r. 2010 , , 718 , 184 cook , d. o. , dale , d. a. , johnson , b. d. , et al . 2014 , mnras , 445 , 899 domnguez , a. , siana , b. , henry , a. l. , et al . 2013 , , 763 , 145 draine , b.t . 2003 , araa , 41 , 241 draine , b. t. , dale , d. a. , bendo , g. , gordon , k. d. , et al . 2007 , , 663 , 866 dunlop , j. s. , mclure , r. j. , robertson , b. e. , et al . 2012 , mnras , 420 , 901 elbaz d. , dickinson , m. , hwang , h. s. , et al . 2011 , a&a , 533 , a119 fitzpatrick , e.l . 1999 , pasp , 111 , 63 galametz , m. , madden , s. c. , galliano , f. , hony , s. , et al . 2011 , a&a , 532 , a56 gallerani , s. , maiolino , r. , juarez , y. , et al . 2010 , a&a , 523 , a85 garn , t. , & best , p. n. 2010 , mnras , 409 , 421 gordon , k. d. , clayton , g. c. , misselt , k. a. , et al . 2003 , , 594 , 279 grasha , k. , calzetti , d. , andrews , j. e. , lee , j. c. , & dale , d. a. 2013 , , 773 , 174 groves , b. , brinchmann , j. , & walcher , c. j. 2012 , mnras , 419 , 1402 guo , q. , white , s. , li , c. , & boylan - kolchin , m. 2010 , mnras , 404 , 1111 hao , c .- , kennicutt , r. c. , johnson , b. d. , calzetti , d. et al . 2011 , , 741 , 124 hopkins , a. m. , connolly , a. j. , haarsma , d. b. , & cram , l. e. 2001 , aj , 122 , 288 iglesias - pramo , j. , vlchez , j. m. , galbany , l. , et al . 2013 , a&a , 553 , l7 juneau , s. , bournaud , f. , charlot , s. , et al . 2014 , , 788 , 88 johnson , b. d. , schiminovich , d. , seibert , m. , et al . 2007 , apjs , 173 , 392 kauffmann , g. , et al . 2003 , mnras , 341 , 33 kauffmann , g. , heckman , t. m. , white , s. d. m. , et al . 2003 , mnras , 341 , 54 kauffmann , g. , heckman , t. , tremonte , c. , et al . 2003 , mnras , 346 , 1055 kewley , l. , dopita , m. , sutherland , r. , heisler , c. , & trevena , j. 2001 , , 556 , 121 kewley , l. j. , jansen , r. a. , & geller , m. j. 2005 , pasp , 117 , 227 kewley , l. j. , & ellison , s. l. 2008 , , 681 , 1183 kong , x. , charlot , s. , brinchmann , j. , & fall , s. m. 2004 , mnras , 349 , 769 kreckel , k. , et al . 2013 , , 771 , 62 kriek , m. , & conroy , c. 2013 , apjl , 775 , l16 kroupa , p. 2001 , mnras , 322 , 231 le floch , e. , papovich , c. , dole , h. et al . 2005 , , 632 , 169 leitherer , c. , schaerer , d. , goldader , j. d. , et al . 1999 , apjs , 123 , 3 magnelli , b. , elbaz , d. , chary , r.r . 2009 , a&a , 496 , 57 markwardt , c. b. 2009 , astronomical data analysis software and systems xviii , 411 , 251 martin , d. c. , fanson , j. , schiminovich , d. , et al . 2005 , apjl , 619 , l1 mcquade , k. , calzetti , d. , & kinney , a. l. 1995 , apjs , 97 , 331 meurer , g. r. , heckman , t. m. , & calzetti , d. 1999 , , 521 , 64 morrissey , p. , conrow , t. , barlow , t. a. , et al . 2007 , apjs , 173 , 682 muoz - mateos , j. c. , gil de paz , a. , boissier , s. , et al . 2009 , apj , 701 , 1965 murphy , e.j . , chary , r .- r . , dickinson , m. et al . 2011 , , 732 , 126 natta , a. , & panagia , n. 1984 , , 287 , 228 nelson , e. j. , van dokkum , p. g. , momcheva , i. g. et al . 2015 , arxiv e - prints , arxiv:1511.04443 noll , s. , pierini , d. , cimatti , a. , et al . 2009 , a&a , 499 , 69 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley , ca : university science books ) osterbrock , d. e. , & ferland , g. j. 2006 , astrophysics of gaseous nebulae and active galactic nuclei ( sausalito , ca : university science books ) perley , d. a. , morgan , a. , updike , a. , et al . 2011 , aj , 141 , 36 pettini , m. , & pagel , b. e. j. 2004 , mnras , 348 , l59 reddy , n. a. , steidel , c. c. , fadda , d. , et al . 2006 , , 644 , 792 reddy , n. a. , erb , d. k. , pettini , m. , steidel , c. c. , & shapley , a. e. 2010 , , 712 , 1070 reddy , n.a . , dickinson , m. , elbaz , d. et al . 2012 , , 744 , 154 reddy , n. a. , kriek , m. , shapley , a. , et al . 2015 , , 806 , 259 salim s. , rich , r. m. , charlot , s. , et al . , 2007 , apjs , 173 , 267 scoville , n. , faisst , a. , capak , p. , et al . 2015 , , 800 , 108 sobral , d. , best , p. n. , matsuda , y. , et al . 2012 , mnras , 420 , 1926 storchi - bergmann , t. , kinney , a. l. , & challis , p. 1995 , apjs , 98 , 103 takeuchi , t. t. , yuan , f .- t . , ikeyama , a. , murata , k. l. , & inoue , a. k. 2012 , , 755 , 144 tremonti , c. a. , heckman , t. m. , kauffmann , g. , et al . 2004 , , 613 , 898 veilleux , s. & osterbrock , d. e. 1987 , apjs , 63 , 295 wang , b. , & heckman , t. m. 1996 , , 457 , 645 weingartner , j. , & draine , b. t. 2001 , , 548 , 296 wild , v. , charlot , s. , brinchmann , j. , et al . 2011 , mnras , 417 , 1760 williams m. j. , bureau m. , cappellari m. , 2010 , mnras , 409 , 1330 zahid , h. j. , yates , r. m. , kewley , l. j. , & kudritzki , r. p. 2013 , apj , 763 , 92 future large area ir surveys , such as those planned with _ euclid _ and the _ wide - field infrared survey telescope _ ( _ wfirst _ ) , will image vast numbers of galaxies . given the shortest wavelengths available to these missions , @xmath177 and @xmath178 for _ euclid _ and _ wfirst _ , respectively , the shortest rest - frame wavelengths available for galaxies with @xmath179 this will correspond to the optical portion of the spectrum . therefore , a proper utilization of the @xmath2-@xmath4 relation would require separate measurements from another facility to determine uv slope . here we investigate the possibility of using the optical slope , from the observed sdss @xmath58 ( @xmath180 ) and @xmath181 ( @xmath182 ) fiber photometry , as an indicator for reddening of the continuum instead of @xmath2 . this is calculated using the same method as for the uv slope , @xmath183}{\log(\lambda_{\mathrm{u}}/\lambda_{\mathrm{g}})}\,.\ ] ] we expect that the correlation between the optical slope and @xmath4 to be weaker than the uv slope , given that this region is less sensitive to the effects of dust and more sensitive to older stellar populations . in figure [ fig : ug_beta_tau ] we show the @xmath184 and @xmath4 values for our sample of 9813 sfgs . a linear fit to the data using the mpfitexy routine @xcite gives @xmath185 with an intrinsic dispersion of @xmath186 . as expected , this dispersion is larger than that found using the uv slope , but not by a very large amount . similar to earlier anaylsis , we divide the sample by @xmath131 to determine its role in the dispersion . we plot the sample divided into three ranges of @xmath131 in figure [ fig : ug_beta_tau_subsample ] and the fitted relationships are @xmath187 @xmath188 @xmath189 it is evident that the dispersion is reduced by @xmath190 . this indicates that @xmath131 acts as an indicator of the intrinsic optical slope ( i.e. , the normalization of the @xmath191 relation ) . from this we believe that the optical slope might be viable for determining appropriate corrections in a large statistical sample , especially if information on the 4000 break is available to correct for stellar age effects .
theoretical foundations of the problem of quantum spin tunneling in magnetic nanostructures are presented . several model problems are considered in detail , including recent new results on tunneling in antiferromagnetic nanoparticles and topologically nontrivial magnetic structures in systems with reduced dimension .
cond-mat9706292
it is well known that magnetic ordering is an essentially quantum phenomenon . according to the bohr van leeven theorem ( see , e.g. , @xcite ) , the magnetization of a thermodynamically equilibrium classical system of charged particles is zero even in presence of an external magnetic field . classical theories of magnetic properties were based on certain assumptions going beyond the limits of classical physics ( e.g. , the existence of stable micro - particles with nonzero magnetic moment assumed in langevin s theory of paramagnetism @xcite ) . the nature of magnetic ordering was revealed only after the discovery of modern quantum mechanics in the works of heisenberg , frenkel and dorfman . in 30s , many remarkable results were obtained within the microscopic quantum theory : bloch @xcite predicted the existence of magnons and low - temperature behavior of magnetization ; bethe @xcite was able to construct the complete set of excited states for a spin-@xmath0 chain , including nonlinear soliton - type excitations ( spin complexes ) . the ` undivided rule ' of the quantum theory of magnetism lasted only till 1935 , when in the well - known work landau and lifshitz @xcite formulated the equation describing the dynamics of macroscopic magnetization of a ferromagnet ( fm ) . when deriving the landau - lifshitz ( ll ) equation , a quantum picture of magnetic ordering was used , particularly , the exchange nature of spin interaction , but the ll equation itself has the form of a classical equation for the magnetization @xmath1 . later on the basis of the ll equation the macroscopic theory of magnetism was developed and enormous number of various phenomena were described @xcite ( an overview of modern phenomenological theory of magnetically ordered media can be also found in this book in the lecture by v.g.baryakhtar ) . this lecture presents an introduction to the foundations of a new , fast - developing topic in the physics of magnetism , macroscopic quantum tunnelling ( mqt ) . let us first address briefly the scope of problems belonging to this field . mqt problems can be roughly divided into two main types . first of all , there are phenomena connected with the underbarrier transition from a metastable state , corresponding to a local minimum of the magnet energy , to a stable one . such effects were observed in low - temperature remagnetization processes in small fm particles as well as in macroscopic samples ( due to the tunneling depinning of domain walls ) , see the recent review @xcite . such phenomena of `` quantum escape '' are typical not only for magnets , e.g. , quantum depinning of vortices contributes significantly to the energy losses in htsc materials @xcite . here we will concentrate on another type of phenomena , the so - called _ coherent mqt . _ to illustrate their main feature , let us consider a small fm particle with the easy axis along the @xmath2 direction . if the particle size is small enough ( much less than the domain wall thickness @xmath3 ) , the particle is in a single - domain state , because the exchange interaction makes the appearance of a state with magnetic inhomogeneities energetically unfavorable . then , from the point of view of classical physics , the ground state of the particle is twofold degenerate . those two states correspond to two local minima of the anisotropy energy and are macroscopically different since they have different values of macroscopic magnetization @xmath4 . the situation is the same as in the elementary mechanical problem of a particle in two - well potential @xmath5 having equivalent minima at @xmath6 , see fig . [ fig : two - well ] . in classical mechanics the minimum of energy corresponds to a particle located in one of the two local minima of the potential . however , from quantum mechanics textbooks it is well known that the actual situation is _ qualitatively _ different : the particle is `` spread '' over two wells , and the ground state is nondegenerate @xcite . one can expect that the same should be true for a fm particle : its correct ground state will be a superposition of `` up '' and `` down '' states , and the mean value of magnetization will be zero . such picture was first proposed by chudnovsky @xcite ; further calculations showed @xcite that such effects are possible for fm particles with rather large number of spins ( about @xmath7 ) . the tunneling effects , according to the theoretical estimates @xcite , should be even more important for small particles of antiferromagnet ( afm ) ; the effects of quantum coherence in afm particles were observed in ref . @xcite . thus , an important feature of quantum mechanics , a possibility of underbarrier transitions , can manifest itself in magnetic particles on a macroscopic ( strictly speaking , mesoscopic ) scale . maybe even more interesting is the manifestation of another characteristic feature of quantum physics , viz . the effects of quantum interference . such effects arise in the problem of mqt in magnetic nanostructures and can partially or completely suppress tunneling , restoring the initial degeneracy of the ground state @xcite . we wish to remark that understanding that motion of particles along very different classical trajectories can `` sum up '' in some sense and yield an interference picture was one of the crucial points in the development of quantum mechanics , and a considerable part of the well - known discussion between bohr and einstein was devoted to this problem . besides the importance of the tunneling phenomena in magnets from the fundamental point of view , they are potentially important for the future magnetic devices working on a nanoscale . in the present lecture we restrict ourselves to discussing the problems of coherent mqt in various mesoscopic magnetic structures . the paper is organized as follows : sect . [ sec : basics ] contains the elementary description of the instanton formalism , traditionally used in the theoretical treatment of mqt problems . since the instanton approach , though being the most straightforward one , is based on rather complicated mathematical formalism , we will discuss it in parallel with simple and widely known semiclassical approximation of quantum mechanics . the point is that those two approaches are equally adequate for treating the problem of mqt in small particles , and the `` standard '' semiclassical calculations , easily reproducible by anybody who learned foundations of quantum mechanics , may be helpful for understanding the structure of the results derived within the instanton technique . further , in sections [ sec : fm ] and [ sec : afm ] we discuss the problem of mqt in ferro- and antiferromagnetic small particles , with a special attention to the interference effects . for the description of afm we use simple but adequate approach based on the equations for the dynamics of the antiferromagnetism vector @xmath8 . this approach easily allows one to keep trace of the actual magnetic symmetry of the crystal ; the symmetry is lowered when external magnetic field is applied or when certain weak interactions , e.g. , the so - called dzyaloshinskii - moriya ( dm ) interaction , are taken into account , which leads to quite nontrivial interference phenomena . section [ sec : top ] is devoted to the analysis of coherent mqt in `` topological nanostructures , '' i.e. static inhomogeneous states of magnets with topologically nontrivial distribution of magnetization ; among the examples considered there are domain walls in one - dimensional ( 1d ) magnets @xcite , magnetic vortices @xcite and disclinations @xcite in 2d antiferromagnets , and antiferromagnetic rings with odd number of spins @xcite . for those problems , when the description of tunneling involves multidimensional ( space - time ) instantons , there is no alternative to the instanton approach and its use is decisive . finally , section [ sec : summary ] contains a brief summary and discussion of several problems which are either left out of our consideration or unsolved . for the sake of the presentation completeness , let us recall briefly the main concepts of the instanton technique , since we will extensively use them below . in quantum field theory , the propagator , i.e. , the amplitude of probability @xmath9 of the transition from any given state with the field configuration @xmath10 at @xmath11 to another state @xmath12 at @xmath13 is determined by the path integral @xmath14/\hbar \big\}\,,\ ] ] where @xmath15=\int_{0}^{t_{0 } } dt\int dx\ , { \cal l}[\varphi(x , t)]\ ] ] is the action functional . here @xmath16 is the lagrangian density , and the integration in ( [ 1to2 ] ) goes over all space - time field configurations @xmath17 satisfying the boundary conditions @xmath18 and @xmath19 . ( we leave out the problem of a consistent definition of the measure @xmath20 that arises for systems with infinitely many degrees of freedom , keeping in mind that we are going to talk about the application of field theory to the physics of spin systems on a discrete lattice , and thus all necessary regularizations are provided by the lattice in a natural way . ) instead of working with the propagator ( [ 1to2 ] ) in usual minkovsky s space - time , it is convenient to make the wick rotation @xmath21 ( essentially this procedure is an analytical continuation in @xmath22 ) , passing to the euclidean space - time . then one has the euclidean propagator @xmath23 the main contribution to the path integral comes from the global minimum of the euclidean action functional @xmath24 . this minimum corresponds to a trivial solution @xmath25 , where @xmath26 determines the minimal energy of the system . however , if several different values of @xmath26 are possible , it is often important to take into account the contribution from the _ local _ minima of the euclidean action as well . such a local minimum can correspond , e.g. , to a trajectory @xmath27 connecting two possible @xmath26 values ; it is clear that the probability @xmath28 will contain the factor @xmath29\}/\hbar$ ] . such a contribution can be calculated in a semiclassical approximation and describes effects which can not be accessed by means of the perturbation theory . we will illustrate the above arguments on the example of a simple quantum - mechanical problem . consider the motion of a particle of mass @xmath30 in a symmetric two - well potential @xmath5 of the type shown in fig.[fig : two - well ] , with two equivalent minima at @xmath6 . following the popular choice @xcite , we will assume this potential in the form @xmath31 where the parameters @xmath32 and @xmath33 determine the height and width of the barrier between two wells . this model is described by the lagrangian @xmath34 after passing to the imaginary time , the euclidean action is easily obtained in the form @xmath35 the classical ( global ) minimum of this functional is reached at @xmath36 or @xmath37 . equations of motion for the action ( [ fi4a ] ) @xmath38 correspond to the particle moving in the potential @xmath39 , so that @xmath6 are _ maxima _ of this effective potential , and there exist classical low - energy trajectories connecting them . such trajectories represent _ local _ minima of the euclidean action functional and are called _ instantons_. they can be easily found in implicit form , @xmath40 where @xmath41 is an arbitrary parameter determining the `` centre '' of instanton solution . for many potentials the integration can be performed explicitly , e.g. , in case of ( [ fi4 ] ) one obtains @xmath42\,,\ ] ] where @xmath43 is the frequency of linear oscillations around one of classical minima . euclidean action for the instanton trajectory can be written as @xmath44 for the model ( [ fi4 ] ) one has @xmath45 . thus , instantons are very much like solitons with the difference that they are localized in time . trajectories ( [ fi4inst ] ) begin at @xmath46 in one of the minima of @xmath47 and end at @xmath48 in the other one ; the contribution of those trajectories is responsible for the tunneling splitting of the lowest energy level in the two - well potential . indeed , the tunneling level splitting is proportional to the matrix element @xmath49 of the transition from one well to the other , and the probability amplitude of such a transition is given by the path integral from @xmath36 to @xmath37 . it is thus clear that the contribution of a single instanton to the transition amplitude is proportional to @xmath50 . the full calculation of this amplitude , however , is more complicated and should take into account not only the instanton trajectories but all trajectories close to them . further , the full variety of multiinstanton paths which bring the particle from one well to the other should be taken into account . if the problem is semiclassical , i.e. @xmath51 is large and the probability of tunneling is small , integration over `` close '' trajectories can be described as an effect of small fluctuations above the instanton solution . even this , usually elementary , problem of integrating over small ( linear ) fluctuations is nontrivial in case of instantons , because some of those fluctuations do not change the action . particularly , from ( [ fi4inst ] ) it is easy to see that changing the position of instanton centre @xmath41 has no effect on @xmath52 . such `` zero modes '' _ always _ arise in instanton problems and their contribution requires a special analysis . detailed description of this technique would take us out of the space limits , and we refer the interested reader to textbooks and review articles ( see , e.g. , @xcite ) . we will attempt to get the correct result for the probability amplitude @xmath28 by means of the `` traditional '' quantum mechanics ( without use of path integrals and instantons ) . first , let us note that , due to the symmetry of the potential @xmath53 , two lowest levels correspond to even and odd eigenfunctions @xmath54 and @xmath55 , with the energies @xmath56 and @xmath57 , respectively . multiplying the schrdinger equation for @xmath58 by @xmath59 and vice versa , then taking the difference of those two equations and finally integrating over @xmath60 from @xmath61 to @xmath62 , one obtains the relation @xmath63_{x=0}= \int_{0}^{\infty } \psi_{a}\psi_{s}\ , dx\,,\ ] ] which is _ exact _ and is nothing but a mere consequence of the symmetry properties . it is natural to try to use a semiclassical approximation . the semiclassical result is given , e.g. , in a popular textbook by landau and lifshitz ( @xcite , see the problem 3 after according to that result , @xmath64 , where @xmath65^{1/2}$ ] , @xmath66 and @xmath67^{1/2}$ ] , here @xmath68 is the turnover point of the classical trajectory with energy @xmath69 ( corresponding to a non - split level ) defined by the equation @xmath70 . however , this result is not adequate for our problem , and it does not coincide with the result of instanton calculation . the point is that , surprisingly , the problem of tunneling from one classical ground state to another _ is not semiclassical : _ semiclassical approximation can not be directly applied to the ground state wavefunction inside one well . therefore we will do as follows : let us represent the wavefunctions inside the barrier region as symmetric and antisymmetric combinations of the wkb exponents , @xmath71 where @xmath72}$ ] . those wavefunctions can be used inside the entire barrier region , except narrow intervals @xmath73 near the well minima , where @xmath74 is the amplitude of zero - point fluctuations . on the other hand , if the condition @xmath75 is satisfied , then for the description of the wavefunction inside the well any reasonable potential @xmath5 can be replaced by the parabolic one , @xmath76 . then in `` non - semiclassical '' regions one may use well - known expression for the ground state wavefunction of a harmonic oscillator , @xmath77\,.\ ] ] thus , in the regions @xmath78 both the expressions ( [ wkb ] ) and ( [ harm ] ) are valid . then , normalization factors @xmath79 can be determined from the condition of matching ( [ wkb ] ) and ( [ harm ] ) in the two above - mentioned regions , and after that the integration in ( [ rel1 ] ) can be performed explicitly . after some amount of algebra the tunneling level splitting can be represented in the form @xmath80 where the quantity @xmath81 coincides with the euclidean action for the instanton trajectory . one can see that the difference between the formula ( [ tunnel - qm ] ) and the usual semiclassical result consists in the pre - exponential factor containing the integral of the type @xmath82 . it is clear that the main contribution into this prefactor comes from the region @xmath83 , where the integral can be approximated as @xmath84 , so that it diverges logarithmically at @xmath85 . thus for any potential @xmath5 the prefactor can be represented in the form @xmath86 or , equivalently , @xmath87 . here @xmath88 is a numerical constant of the order of unity , it can be easily calculated for any given potential @xmath5 . so , finally we arrive at the following universal formula : @xmath89 for the model potentials @xmath90 and @xmath91 the value of @xmath88 is equal to @xmath92 and @xmath93 , respectively . the formulas ( [ tunnel - qm],[tunnel - inst ] ) give the desired result for any two - well potential with sufficiently large barrier . the main feature of this result is the presence of an exponentially small factor . the small parameter of the mqt problem is @xmath94 , which can be represented as a ratio of the zero - point fluctuations amplitude to the distance between wells , @xmath95 . the expression @xmath96 is non - analytical in the small parameter , and thus the mqt phenomenon can not be obtained in any order of the perturbation theory . we wish to emphasize that the correct result is roughly @xmath97 times greater than that following from `` naive '' semiclassical formula . this large additional factor appears due to the contribution from the regions close to the minima of the potential , where the motion is not semiclassic . let us try to understand this in the instanton language . as we mentioned before , the small exponential factor @xmath98 arises immediately in the instanton approach ; the main problem is to compute the pre - exponential factor , which is determined by the integration over all small deviations from the instanton solution . those deviations are of two types : real fluctuations of the instanton structure , which increase the euclidean action , and `` zero modes '' which correspond to moving the instanton centre . it is rather clear that `` nonzero '' modes have a characteristic energy of the order of @xmath99 , and that the quantity @xmath100 has nothing to do with the zero mode . thus , it is obvious that the factor @xmath99 arises from the integration over all `` nonzero '' modes , and the large factor @xmath101 arises due to the zero ( in our case translational ) mode . such a `` separation '' naturally arises in rigorous calculations @xcite . it is remarkable that the above result can be generalized to the case of much more complicated problems involving space - time instantons ( which , as we will see later , is important for the problem of mqt in topological nanostructures ) . for any instanton _ all _ nonzero modes yield a factor like @xmath99 , and _ each _ of the zero modes yields the factor @xmath102 @xcite , so that the final result can be reconstructed practically without calculations ( up to a numerical factor of the order of unity ) . to illustrate one more feature typical for tunneling problems , let us consider another model @xcite : a particle of mass @xmath30 which can move along the circle of radius @xmath103 , so that its coordinate is determined by a single angular variable @xmath104 , @xmath105 , in the two - well potential @xmath106 the model is described by the following lagrangian : @xmath107 the classical lagrangian can be modified by adding the arbitrary full derivative term , e.g. , @xmath108 which of course does not change the corresponding _ classical _ equations of motion . however , adding the full derivative ( [ poctop ] ) changes the definition of the canonical momentum conjugate to @xmath104 , which , as one can easily check , leads to a considerable change in the hamiltonian of the corresponding _ quantum - mechanical _ system after canonical quantization : for nonzero @xmath109 the correct hamiltonian would be @xmath110 thus , there is no one - to - one correspondence between classical and quantum - mechanical systems : several quantum systems can have the same classical system as a classical limit . for this model problem the instanton trajectories can be written down explicitly : @xmath111\,,\\ & & \omega=(4u_{0}/mr^{2})^{1/2}\ , . \nonumber \end{aligned}\ ] ] where @xmath112 is the arbitrary parameter determining the instanton position in the imaginary time axis and @xmath113 is the topological charge distinguishing instantons and antiinstantons ; the instanton action is finite and is given by @xmath114 . the importance of the full derivative term ( [ poctop ] ) can be most easily understood in terms of instantons . indeed , let us consider the tunneling amplitude @xmath9 from the @xmath115 well to @xmath116 one : it is clear that the contribution to this amplitude is made equally by instantons ( with @xmath104 changing from @xmath61 to @xmath117 ) and antiinstantons ( with @xmath104 changing from @xmath61 to @xmath118 ) . however , the term ( [ poctop ] ) becomes an _ part of the euclidean action and leads to the additional factor @xmath119 associated with the instanton contribution and a similar factor @xmath120 for antiinstanton paths . thus , the resulting transition amplitude for nonzero @xmath109 is modified as follows : @xmath121_{\gamma=0}\cos(\pi\gamma/\hbar)\,,\ ] ] where @xmath122_{\gamma=0 } \propto \omega \left({{\cal a}_{0}/\hbar}\right)^{1/2}e^{-{{\cal a}_{0}/\hbar}}$ ] , according to the general result described above . one can see that for half - integer @xmath123 the interference of instanton and antiinstanton paths is _ destructive , _ so that at @xmath124 the tunneling between two wells is _ completely suppressed . _ this effect is essentially _ topological _ because the topological charge appears in the answer : the contribution of configurations with different topological charge is different . the same result can be obtained directly by solving the schrdinger equation with the hamiltonian ( [ poch ] ) : for half - integer @xmath123 it can be mapped to the mathieu equation with _ antiperiodic _ boundary conditions , and the corresponding energy levels are known to be doubly degenerate @xcite , which also means absence of tunneling . in this section we consider the basic technique of field - theoretical description for spin systems on the simplest example , namely a nanoparticle of a ferromagnetic material . such an object may be viewed as a zero - dimensional magnetic system , because at very low temperature all spins in the particle can be considered as pointing in the same direction . it is worthwhile to consider first the dynamics of a single spin @xmath125 . in order to obtain the effective lagrangian describing the spin dynamics , it is convenient to use a coherent state path - integral approach ( see , e.g. , the excellent textbook by fradkin @xcite ) . let us introduce a set of generalized coherent states @xcite @xmath126 parameterized by the unit vector @xmath127 . here @xmath128 is a unit vector pointing along the @xmath129 axis , and @xmath130 denotes a spin-@xmath125 state with @xmath131 . they form a non - orthogonal ` overcomplete ' basis so that the following property , usually called a resolution of unity , holds : @xmath132 another useful property is that quantum average of @xmath133 on those coherent states is the same as of classical vector of length @xmath125 : @xmath134 in case of @xmath125=1/2 those coherent states have a very simple form and are general single - spin wavefunctions : @xmath135 we again start from the formula for propagator ( [ 1to2 ] ) which is essentially a definition of the effective lagrangian . slicing the time interval @xmath136 $ ] into infinitely small pieces @xmath137 , and successively using the identity ( [ complete ] ) , one can rewrite this propagator in @xmath138-representation as @xmath139 passing to the function @xmath140 of the continuum variable @xmath22 , one ends up with the coherent state path integral ( [ 1to2 ] ) where the action @xmath141 is determined by the effective lagrangian @xmath142 it can be shown that the dynamical part of this lagrangian has the form @xmath143 for arbitrary @xmath125 this calculation requires some algebra , but for the simplest case @xmath144 it is straightforward . the expression ( [ berry ] ) is nothing but the berry phase @xcite for adiabatic motion of a single spin . it should be remarked that the presence of the full derivative term @xmath145 is rather nontrivial and allows one to capture subtle differences between integer and half - integer spins , as we will see below . for example , consider a single spin @xmath125 in some crystal - field potential , with the effective hamiltonian @xmath146 where @xmath147 and the easy - plane anisotropy @xmath148 is much stronger than the in - plane anisotropy @xmath149 . the lagrangian is @xmath150 there are two equivalent classical minima of the potential at @xmath151 , @xmath115 and @xmath151 , @xmath116 . paths with @xmath152 make the main contribution into the tunneling amplitude , so that we can approximately set @xmath153 , @xmath154 , and expand in @xmath155 up to quadratic terms in the lagrangian ; in the term proportional to @xmath156 the @xmath149 contribution may be neglected as small comparing to the contribution of @xmath148 . after that , the `` slave '' variable @xmath155 can be excluded from the lagrangian ( `` integrated out '' of the path integral ) because the corresponding equation of motion @xmath157 allows to express @xmath155 through @xmath104 explicitly : @xmath158 substituting this solution into the original lagrangian ( [ singspinl ] ) , one obtains the effective lagrangian depending on @xmath104 only : @xmath159 we see that we end up with the lagrangian of a particle on a circle from the previous section , with the topological term @xmath160 . for each path where @xmath104 changes from @xmath61 to @xmath117 there is a corresponding antiinstanton path with @xmath104 changing from @xmath61 to @xmath118 , and those paths contribute to the tunneling amplitude with phase factors @xmath161 and @xmath162 . for half - integer @xmath125 those contributions precisely cancel each other , making the tunneling impossible . this is exactly in line with the well - known kramers theorem , which states that in absence of external magnetic field all energy levels of a system with half - integer total spin should be twofold degenerate . one can also straightforwardly check that for a single spin in magnetic field , i.e. , for @xmath163 , the correct energy levels can be obtained only with the full derivative term taken into account . now we are prepared enough , finally , to consider the problem of tunneling in a small ferromagnetic particle consisting of @xmath164 spin-@xmath125 spins . if we assume that ferromagnetic exchange interaction is so strong that we may consider all spins as having the same direction , then we come to the `` giant spin '' model where the entire particle is described as a quantum - mechanical ( `` zero - dimensional '' ) system with only two degrees of freedom @xmath165 and @xmath104 . in fact , we should postulate that in our path integral , when integrating over the coherent state configurations latexmath:[$\prod_{i=1}^{n } \otimes with all @xmath164 vectors @xmath167 replaced by the same vector @xmath127 , and we take into account only configurations from this subspace . assuming that the crystal - field anisotropy has the form ( [ singspinh ] ) , we come to essentially the same effective lagrangian ( [ singspinleff ] ) , and the only difference is that eq.([singspinleff ] ) should now be multiplied by the total number of spins @xmath164 . the tunnel splitting of the ground state level , according to eq.([ampltop ] ) , is given by @xmath168 where @xmath88 is a numerical constant of the order of @xmath169 . a remarkable property of the result ( [ splitfm ] ) is that presence of a large number @xmath164 in the exponent can be to some extent compensated by smallness of the ratio @xmath170 . however , when the in - plane anisotropy @xmath171 , the splitting vanishes ( this reflects the fact that in uniaxial case tunneling is impossible because of the conservation of the corresponding projection of the total spin ; the same is true for @xmath172 ) . another remarkable feature is that for half - integer @xmath125 the finite splitting can be observed only in particles with even number of spins @xmath164 ; since in any statistical ensemble @xmath164 fluctuates a bit , this roughly means that only one half of all particles gives nonzero contribution . statistical fluctuations of @xmath164 have another , more painful consequence : since @xmath164 stays in the exponent , even small fluctuations of the total number of spins in the particle lead to large fluctuations of the splitting . moreover , since @xmath164 scales as the third power of the linear size @xmath173 , small fluctuations of @xmath173 will be considerably enhanced in @xmath164 . this may be crucial if one tries to detect the splitting by means of some resonance technique : the initially weak signal would be even more weakened by the strong broadening of the resonance peak . actually , many factors can prevent one from observing the tunneling resonance , e.g. , relaxation , temperature effects , etc . here we will not at all touch the problem of relaxation because of its complexity ; instead of that we refer the interested reader to the review @xcite . taking into account the finite temperature effects is also nontrivial , particularly because it requires changing the procedure of taking averages in the path integral : statistical averages should be taken simultaneously with quantum - mechanical ones . roughly ( and without taking into account the temperature dependence of relaxation mechanisms ) the effects of finite temperature can be estimated with the help of the concept of a characteristic temperature @xmath174 below which the effects of quantum tunneling prevail over thermal transitions . rough estimate for @xmath174 is obtained from the comparison of the relative strength of two exponential factors : thermal exponent @xmath175 and tunneling exponent @xmath176 , where @xmath177 is the height of barrier separating two equivalent states and @xmath52 is the corresponding instanton action , then @xmath178 . it is easy to see that for the ferromagnetic particle problem considered above @xmath179 i.e. the temperature of crossover from classical to quantum transitions is in this case rather small since it is determined by weak ( relativistic ) anisotropy interaction constants ; for typical anisotropy values @xmath180 is about @xmath181 k. the problem of continuum field description of antiferromagnet ( afm ) is more complicated but also much more interesting than a similar problem for ferromagnet . antiferromagnet contains at least two different `` sublattices '' whose magnetizations compensate each other in the equilibrium state . thus , when choosing the coherent state wavefunction in the form @xmath182 as described above , one can not any more consider @xmath183 as a `` smooth '' function of the lattice site @xmath184 let us adopt the simplest two - sublattice model which , despite the fact that it may be inadequate for a specific material , still allows one to demonstrate the essential physics of antiferromagnetism . we assume that there are two equivalent sublattices with magnetizations @xmath185 and @xmath186 , @xmath187 . then , when passing to the continuum limit , one has to introduce smooth fields @xmath188 and @xmath189 describing net magnetization and sublattice magnetization , respectively . they satisfy the constraints @xmath190 , @xmath191 , and we further assume that @xmath192 . the energy of afm @xmath193 then can be expressed as a functional of @xmath194 and @xmath8 : @xmath195=m_{0}^{2 } \int dv\ , \left\ { { 1\over2}\delta \vec{m}^{2 } + { 1\over2}\alpha(\nabla\vec{l})^{2}+w_{a}(\vec{l } ) -{g\over m_{0}}(\vec{m}\cdot\vec{h})\right\ } \,.\ ] ] here the phenomenological constants @xmath196 and @xmath197 describe homogeneous and inhomogeneous exchange , respectively , @xmath198 is the external magnetic field , @xmath199 is the lande factor , the function @xmath200 describes the energy of magnetic anisotropy , and we use the notation @xmath201 . the magnitude of sublattice magnetization @xmath202 , where @xmath203 is the bohr magneton , @xmath125 is the spin of a magnetic ion , and @xmath204 is the volume of the magnetic elementary cell . as we learned from the previous section , the correct lagrangian , suitable for the quantum - mechanical treatment , has the form @xmath205\,,\ ] ] where the angular variables @xmath206 and @xmath207 determine the unit vectors describing the orientation of spins in first and second sublattice , respectively . note that we have kept intact the summation sign in the dynamical part of ( [ lafm ] ) : the reason is that the explicit expression for the berry phase in the continuum limit strongly depends on the details of the magnetic elementary cell structure ( which dictates the correct definition of @xmath194 and @xmath8 and the procedure of passing to the continuum limit ) . under the assumption that @xmath208 , the magnetization @xmath194 can be excluded from the lagrangian ( [ lafm ] ) , and one obtains the effective lagrangian depending only on @xmath8 ; after that step @xmath8 can be regarded as a unit vector , @xmath209 . for example , in the simplest case of an antiferromagnet with only two ( equivalent ) atoms in elementary magnetic cell the dynamic part of the lagrangian ( [ lafm ] ) can be written as @xmath210 and the density of the effective lagrangian takes the form @xmath211 where @xmath212 is the anisotropy energy renormalized by the magnetic field , @xmath213 @xmath214 is the gyromagnetic ratio , and @xmath215 is the limiting velocity of spin waves . using general phenomenological arguments , one can show @xcite that in case of arbitrary collinear antiferromagnet the lagrangian should have the form similar to ( [ leffafm ] ) . other , more complicated interactions can be present in eq . ( [ w ] ) . in some afm materials ( which are , strictly speaking , weak ferromagnets ) the so - called dzyaloshinskii - moriya ( dm ) interaction is possible . it can be described by including the term @xmath216 under the integration sign in into ( [ w ] ) , where @xmath217 is some tensor ( which is not necessarily symmetric or antisymmetric ) . the origin of the dm interaction is rather nontrivial , and there is a number of `` selection rules '' excluding the possibility of its existence , particularly the dm interaction can not exist ( i ) if there is an inversion center interchanging sublattices ; ( ii ) if there is a translation which interchanges sublattices , i.e. if the magnetic elementary cell is larger than the elementary cell of the original crystal lattice . it can be shown @xcite that presence of the dm interaction can be taken into account by the substitution @xmath218 in the lagrangian ( [ leffafm ] ) , where the components of vector @xmath219 are defined as @xmath220 . if there exists a sublattice - interchanging inversion center , another invariant may be present in ( [ w ] ) , namely @xmath221 ( here @xmath222 are certain exchange constants ) . it is very important for the physics of afm in one dimension , as we will see later . in case of a small particle one can consider @xmath194 and @xmath8 as being uniform throughout the particle , i.e. as not having any space dependence . then , the lagrangian ( [ leffafm ] ) takes the form @xmath223 \right\}- m_{0}^{2}\widetilde{w}_{a } \,,\nonumber \end{aligned}\ ] ] where @xmath164 is the total number of magnetic elementary cells in the particle , @xmath224 is the exchange field , the dot denotes differentiation with respect to time , and we used angular variables for the vector @xmath8 , @xmath225 there is another possible effect , typical only for antiferromagnetic particles : due to the boundary ( surface ) effects , the number of spins in two sublattices can differ from each other . in that case the lagrangian ( [ lpartafm ] ) will contain the additional term @xmath226 which is essentially the berry phase of @xmath227 non - compensated spins . such a sublattice decompensation in fact should be present in any ensemble of nanoparticles , so that @xmath227 has certain statistical variation . the full lagrangian ( [ lpartafm ] ) is rather complicated , and for the sake of clarity we will consider separately the effects of field and dm interaction . consider a small afm particle with easy - axis anisotropy @xmath228 in external magnetic field @xmath229 perpendicular to the easy axis . then the euclidean action takes the form @xmath230\bigg\ } \\ & + & i\hbar\nu s\int d\tau ( 1-\cos\theta){d\varphi\over d\tau } \ , \nonumber\end{aligned}\ ] ] where @xmath231 is the imaginary time , and @xmath232 is the characteristic magnon frequency ( @xmath99 is the magnon gap ) . there are two equivalent states @xmath233 and @xmath234 with opposite direction of @xmath8 along the easy axis @xmath235 , and obviously the most preferable instanton path is given by @xmath236 , @xmath237 . the instanton solution for @xmath104 is the same as in case of particle on a circle , and one - instanton action is @xmath238 where @xmath239 signs correspond to instantons and antiinstantons . thus , the tunneling amplitude @xmath28 is proportional to @xmath240 and the corresponding magnitude of tunneling level splitting ( proportional to @xmath241 ) oscillates with the period @xmath242 when changing the external field . this period @xmath243 may be rather small , for typical values of the exchange field @xmath244 oe and the number of spins in the particle @xmath245 one obtains @xmath246 oe . the effects of this type were studied in @xcite . the result ( [ ph ] ) illustrates also another remarkable feature : in any experiment probing the response of the ensemble of afm nanoparticles at each @xmath229 there must be only one possible value of splitting ( i.e. , only one peak in the low - frequency response ) when the spin of magnetic ions @xmath125 is integer ; but if @xmath125 is half - integer then , since in any ensemble @xmath227 arbitrarily takes even and odd values , for approximately one half of all particles the phase of cosine in ( [ ph ] ) is shifted by @xmath247 , and there should be _ two _ peaks at each @xmath229 . it is worthwhile to note that the real part of the one - instanton action , which enters the exponent in ( [ ph ] ) , is proportional to @xmath248 ( where @xmath249 and @xmath148 are the exchange and anisotropy constants ) while the corresponding quantity for ferromagnet , according to ( [ splitfm ] ) , does not contain the exchange constant and is determined by the rhombicity @xmath250 . one may conclude that tunneling in afm particles is more easy than in fm ; indeed , the characteristic crossover temperature below which quantum effects dominate over thermal ones , for antiferromagnets is @xmath251 which is much greater than for ferromagnets [ cf . eq.([tfm ] ) ] ; typically @xmath252 is about @xmath253 k. consider the same small afm particle from the previous subsection , but imagine that the dm interaction in its simplest form is present , with the energy given by @xmath254 then the dm interaction leads to the contribution into the lagrangian ( [ lpartafm ] ) of the form @xmath255 where @xmath256 is the so - called dzyaloshinskii field . this term will contribute to the imaginary part of the euclidean action ( [ ah ] ) , and as a result the cosine in ( [ ph ] ) will be modified as @xmath257 thus , presence of the dm interaction alone also leads to effective change of the berry phase and lifts the degeneracy for odd @xmath227 and half - integer @xmath125 . as we mentioned before , one of the most difficult experimental tasks when trying to detect the resonance on tunnel - splitted levels in small particles is to prepare the ensemble of particles with very sharp size distribution : even small fluctuations of size lead to large fluctuations of the tunneling probability since they contribute to the power of exponent . preparing such an ensemble requires high technologies and involves considerable difficulties . one may think about some other , `` natural '' type of magnetic nanostructures to observe spin tunneling phenomena in . one nice solution , which have actually been used in experiment , is to use biologically produced nanoparticles @xcite . another possible way , proposed in @xcite , is to use _ topologically nontrivial magnetic structures : _ kinks in quasi-1d materials , vortices and disclinations in 2d , etc . such objects have required mesoscopic scale ( e.g. , the thickness of a domain wall is usually about 100 lattice constants ) and , since their shape is determined by the material constants , they are identical to a high extent ( up to a possible inhomogeneity of the sample ) . here we consider several possible scenarios of tunneling in topological nanostructures and show that their use has a number of advantages . consider a one - dimensional two - sublattice antiferromagnet with rhombic anisotropy described by the hamiltonian @xmath258\,,\ ] ] where @xmath184 labels sites of the spin chain with the lattice constant @xmath33 , @xmath259 are the anisotropy constants ( so that @xmath2 is the difficult axis and @xmath235 is the easy axis ) , and @xmath249 is the exchange constant . for passing to the continuum field description one may introduce vectors @xmath194 and @xmath8 as @xmath260 and @xmath261 , where @xmath138 are the unit vectors describing the direction of spins ( the parameters of the corresponding coherent states , see the discussion in sect . [ sec : afmmodel ] ) above . these fields live on the lattice with the double spacing @xmath262 , and it is easy to see that the energy functional @xmath263 contains the term @xmath264 . using the equation @xmath265 , one may express @xmath194 through @xmath8 and its derivatives and exclude it from the lagrangian . the effective lagrangian takes the following form : @xmath266 which represents a ( 1 + 1)-dimensional nonlinear @xmath267-model with the so - called topological term @xmath268 it is easy to trace the origin of this term : because of presence of@xmath264 in the energy , the expression for @xmath194 contains @xmath269 which after the substitution into the berry phase ( [ berry ] ) yields the topological term . in agreement with general phenomenological result ( [ leffafm ] ) , the lagrangian ( [ lafm1d ] ) is lorentz - invariant , with the limiting velocity @xmath270 . a stable kink solution corresponds to rotation of vector @xmath8 in the easy plane @xmath271 : @xmath272 where @xmath273 is the characteristic kink thickness , and the quantities @xmath267 and @xmath274 may take the values @xmath275 . the topological charge of the kink @xmath274 is determined by the boundary conditions and can not change in any thermal or tunneling processes . the situation is different with the quantity @xmath267 which determines the sign of @xmath8 projection onto the `` intermediate '' axis @xmath276 . two states with @xmath277 are energetically equivalent ; change of @xmath267 is not forbidden by any conservation laws and describes the reorientation of the macroscopic number of spins @xmath278 `` inside '' a kink , typically @xmath279 . again , tunneling between the kink states with @xmath277 can be studied using the instanton formalism . in contrast to the case of a nanoparticle , here the tunneling between two _ inhomogeneous _ states takes place , so that nontrivial _ space - time _ instantons come into play . the instanton solution @xmath280 is now _ two - dimensional _ and has the following properties ( see fig . [ fig : inst1 ] ) : @xmath281 along any closed contour around the instanton center in the euclidean plane vector @xmath8 rotates through the angle @xmath282 in the easy plane @xmath271 , where @xmath283 . thus , the instanton configuration has the properties of a magnetic vortex and is characterized by two topological charges @xcite : vorticity @xmath227 and polarization @xmath284 . the instanton solution satisfies the equations @xmath285 = 0,\nonumber\\ & & \vec{\nabla}\cdot(\sin^2\theta\vec{\nabla}\varphi ) -(\rho/\delta^2)\sin^2\theta \sin\varphi\cos\varphi=0,\end{aligned}\ ] ] where we have introduced the angular variables @xmath286 , @xmath287 , @xmath288 is the rhombicity parameter , and @xmath289 is the euclidean gradient , @xmath290 . several important properties of the instanton can be obtained without using the explicit form of the solution . first of all , note that this instanton has _ two _ zero modes which correspond to shifting the position of its centre along the direction of @xmath231 and @xmath60 axes , respectively . the physical meaning of the first mode is the same as for 1d instanton , and the second mode corresponds to moving the kink center in real space ( the kink position in infinite 1d magnet is not fixed in our continuum model ) ; however , if the kink center is fixed due to some effects ( e.g. , because of pinning on the lattice , or by boundary conditions ) , so that the eigenfrequency of its oscillations is comparable with the characteristic magnon frequency , then only one zero - frequency mode is present . the euclidean action @xmath24 can be represented in the form @xmath291 \nonumber\\ & & q={1\over4\pi}\int d^{2}x\ , \varepsilon_{\alpha\beta}\sin\theta \partial_{\alpha}\theta \partial_{\beta}\varphi\,.\end{aligned}\ ] ] imaginary part of the euclidean action is in this case completely determined by the topological term @xmath292 . the word `` topological '' becomes now clear , because @xmath293 is the homotopical index of mapping of the @xmath294 plane onto the sphere @xmath209 ( the pontryagin index , or the winding number ) . for uniform boundary conditions at infinity in the @xmath294 plane @xmath293 can take only integer values , but in our case @xmath295 is half - integer , which is typical for vortices ( see , e.g. , @xcite ) . for a kink with given @xmath274 there are two instanton solutions with the same vorticity @xmath227 and different polarizations @xmath284 . thus , the tunneling amplitude is proportional to @xmath296 and vanishes when the spin @xmath125 of magnetic ions is half - integer . however , the degeneracy can be lifted in presence of external magnetic field or the dm interaction , as we will see below . we are not able to construct the exact solution of eqs . ( [ inst - eq ] ) , but the estimate of the tunneling amplitude in various limiting cases can be obtained from approximate arguments . for @xmath297 the characteristic space scale of @xmath104 variation @xmath298 is much greater than the kink thickness @xmath299 , and the problem can be mapped to one with a finite number of degrees of freedom ( one may introduce the variable @xmath300 having the meaning of the angle of deviation out of the easy plane `` inside a kink '' , so that the instanton solution can be seeked in the form @xmath301 ) , then it is easy to obtain @xcite @xmath302 in the opposite limiting case @xmath303 one again has two different length scales : the kink thickness @xmath299 and the `` core '' radius @xmath304 , @xmath305 . for @xmath306 all interactions except the exchange one can be neglected , and one may use the `` isotropic '' vortex solution @xmath307 where @xmath308 , @xmath309 are polar coordinates in the @xmath294 plane . for @xmath310 , i.e. , far outside the core , one can approximately assume that @xmath311 within a wide range of @xmath312 ( for @xmath313 ) the solutions ( [ iso ] ) and ( [ far ] ) can be regarded as coinciding , and the integrand in @xmath314 is proportional to @xmath315 . then , one may divide the integration domain into two parts : @xmath316 and @xmath317 , where @xmath103 is arbitrary in between @xmath318 and @xmath299 . for @xmath316 the solution ( [ iso ] ) may be used , yielding @xmath319 with @xmath320 @xcite . for @xmath317 , one can use a simple trial function approximately satisfying ( [ far ] ) , e.g. , @xmath321 which yields @xmath322 with @xmath323 . summing up the two contributions , we obtain @xmath324 the tunnel splitting of the `` ground state '' level of the kink @xmath325 where @xmath326 is the frequency of the out - of - plane magnon localized at the kink , @xmath327 is the factor determined by the imaginary part of the euclidean action [ in the simplest model @xmath328 , and @xmath329 is the number of zero modes which can be equal to @xmath169 or @xmath330 depending on whether the kink position is fixed , see above . it is easy to estimate the crossover temperature for the problem of tunneling in a kink , comparing the exponent in ( [ splitk ] ) with @xmath331 , where @xmath332 is the barrier height ; for @xmath333 ( i.e. , @xmath334 ) and @xmath335 one obtains @xmath336 which is only logarithmically smaller than the corresponding temperature for a particle ( [ tafm ] ) . let us discuss now the behavior of the imaginary part of the euclidean action in case of deviations from the simplest model ( [ hafm ] ) for which the tunneling is prohibited for half - integer @xmath125 . the most simple observation is that in a spin chain with alternated exchange interaction , when along the chain the strength of exchange constant alternates as @xmath337 , the topological term ( [ ltop ] ) acquires additional factor @xmath338 ( see , e.g. , @xcite ) , which leads to @xmath339 and allows tunneling for half - integer @xmath125 . another way to lift the degeneracy at half - integer @xmath125 is to `` switch on '' the dm interaction or external magnetic field . consider the same model ( [ hafm ] ) with the addition of a magnetic field @xmath198 applied in the easy plane @xmath271 . presence of the field leads to the additional contribution to the imaginary part of @xmath24 @xmath340 where @xmath341 . the mixed product in ( [ addh ] ) can be rewritten in angular variables as @xmath342 one may note that @xmath343 and @xmath165 significantly differ from zero only in the vortex core , and thus the isotropic vortex solution ( [ iso ] ) may be used for the calculation of @xmath344 . after integration we obtain @xmath345 where @xmath284 and @xmath227 , as earlier , denote the polarization and vorticity of the instanton solution , and @xmath346 are numerical constants ( recall that , according to ( [ iso ] ) , the isotropic solution @xmath347 may depend only on @xmath348 ) . after performing the summation in @xmath349 , and with the account taken of the contribution @xmath293 coming from the topological term , the factor @xmath327 in ( [ splitk ] ) will be modified as @xmath350 which means that for the given geometry only the field component perpendicular to the easy axis lifts the degeneracy existing for half - integer @xmath125 . similarly to the case of a small afm particle , the tunneling amplitude is an oscillating function of the external magnetic field @xmath229 , but here the situation is more complicated because the period of oscillations depends on the field orientation . another example of a magnetic nanostructure is a _ ring _ formed by magnetic atoms ; such rings may occur in a dislocation core of a 2d crystal as shown in fig . [ fig : ring ] , and the characteristic feature of this object is that the number of atoms in the ring is _ odd_. here we consider only _ antiferromagnetic _ rings . in terms of the vector @xmath8 such a ring is a spin disclination . let us assume that the magnetic anisotropy is of the easy - plane type , and all spins lie in the @xmath271 plane , @xmath351 where @xmath352 are the unit vectors along @xmath353 . then there are two energetically equivalent states of the ring , with @xmath354 and @xmath355 , where @xmath356 is the azimuthal coordinate of the @xmath184-th spin ( let us assume that the ring is a circle of radius @xmath103 ) . it is possible to construct the instanton solution which links the two states ; in terms of @xmath8 it can be written as @xmath357 calculation shows @xcite that the tunneling amplitude is proportional to @xmath358 i.e. , the probability of tunneling is sufficiently large if the radius of the ring is smaller than the characteristic thickness of the domain wall @xmath299 ( usually @xmath359 ) . again , the tunneling is suppressed for half - integer @xmath125 , and this can be changed with the help of external magnetic field . more detailed analysis @xcite shows that the field @xmath198 should be applied in the easy plane in order to lift the degeneracy , then the cosine in ( [ splitr ] ) will change into @xmath360 where @xmath33 is of the order of the lattice constant . for weak fields the above expression describes just the zeeman splitting of the ground state level of a ring ( recall that due to the odd number of spins the ring always has an uncompensated total spin if @xmath125 is half - integer ) . one more example of a magnetic topologically nontrivial structure is _ magnetic vortex _ in quasi-2d easy - plane antiferromagnet . consider the system described by the hamiltonian @xmath361 where @xmath362 is the anisotropy constant , and @xmath2 is the difficult axis . in terms of the angular variables for the antiferromagnetism vector , @xmath363 , @xmath364 , a vortex corresponds to the solution @xmath365 where @xmath347 satisfies the equation from the second line of eq.([far ] ) , @xmath366 , and the solutions @xmath367 and @xmath368 have the same vorticity @xmath227 but different polarizations @xmath369 . the vortex states with @xmath370 are energetically equivalent , and the transition between them corresponds to reorientation of a macroscopic number of spins @xmath371 , where @xmath372 is the characteristic radius of the vortex core and @xmath33 is the lattice constant . it is worthwhile to remark that such a transition would be forbidden in ferromagnet because of the conservation of the @xmath129-projection of the total spin @xmath373 . the instanton solution @xmath374 linking two vortex configurations @xmath375 with @xmath376 when the imaginary time @xmath231 changes from @xmath377 to @xmath378 is schematically shown in fig . [ fig : inst2 ] . in the 3d euclidean space @xmath379 it describes a topological configuration of the hedgehog type and has a singularity at the origin . such a singularity means that in a small space region around the origin ( roughly within the distance of about @xmath33 ) one has to take into account the change of magnitude of the sublattice magnetization : the length of the vector @xmath8 has to change so that @xmath380 . in this case there are four zero modes , three of them correspond to translations along @xmath60 , @xmath381 , @xmath231 , and the fourth one corresponds to changing the @xmath26 angle . if the position and structure of the vortex are fixed by some additional interactions , only one zero mode is left . the euclidean action derived from the lagrangian of the @xmath267-model has the following form @xmath382\right\ } \,,\ ] ] where @xmath383 describes the vortex solution ( [ vort ] ) and @xmath384 denotes the limiting velocity @xmath270 . away from the singularity ( for @xmath385 , @xmath386 ) the condition @xmath209 holds , and the equations for @xmath387 become @xmath388 = 0,\nonumber\\ & & \vec{\nabla}\cdot(\sin^2\theta\vec{\nabla}\varphi)=0\,.\end{aligned}\ ] ] in the region @xmath389 this system has an exact centrally symmetric solution of the hedgehog type : @xmath390 it can be shown that the contribution of the singularity itself is small and can be neglected . dividing the integration domain into two regions @xmath391 and @xmath392 , where @xmath393 , one can see that the contribution of the region of small distances @xmath391 to the euclidean action is given by @xmath394=4\pi(js^{2}/c)r\,.\ ] ] for estimating the contribution of the `` large '' distance region we use a variational procedure with the trial function of the form @xmath395\,,\ ] ] where @xmath396 is a `` smeared step function : '' @xmath397 as @xmath398 and the derivative of @xmath314 is nonzero in the region of the thickness @xmath399 around @xmath400 . a simple estimate shows that the resulting contribution of the region @xmath392 is described by @xmath401=(2\pi js^{2}/c)\big [ \xi_{1}\delta_{1}\ln(\delta / r ) + \xi_{2}\delta_{1}+\xi_{3}\delta^{2}/\delta_{1 } \big]\,,\ ] ] where @xmath402 are numerical constants of the order of unity . summing up ( [ lowrho ] ) and ( [ farrho ] ) and minimizing @xmath403 with respect to @xmath399 and @xmath103 , we find @xmath404 . thus , the total one - instanton euclidean action may be estimated as @xmath405 where @xmath406 . demanding that the tunneling exponent is not too large , e.g. , @xmath407 , we see that for @xmath408 this means @xmath409 , which is rather tight ; the continuum field approach we used here formally requires @xmath410 , but in practice it is still applicable for @xmath411 @xcite . the crossover temperature @xmath412 is not small since it is proportional to @xmath413 . let us mention briefly the problems which are closely related to the topic of this paper but were left out of discussion , and also those problems which are not clear at present , to our opinion . first of all we would like to remark that we did not touch at all _ microscopic _ essentially quantum effects in magnets , e.g. , predicted by haldane destruction of ( quasi)long - range order in 1d antiferromagnets with integer spin @xmath125 caused by quantum fluctuations . effects of quantum interference are also important for this phenomenon , and its existence is determined by the presence of topological term in the lagrangian of antiferromagnet ( see , e.g. , the reviews @xcite ) . for small @xmath125 and weak anisotropy the ground state of 1d antiferromagnetic system can differ drastically from its classical prototype ; e.g. , the ground state of a @xmath414 afm ring is not sensitive to whether the number of spins is odd or even and is always unique , and the ground state of a @xmath144 afm ring with odd number of spins is _ fourfold _ degenerate @xcite . we also did not consider the contribution of tunneling - generated internal soliton modes to the thermodynamics and response functions of 1d antiferromagnets , which can lead to interesting effects ( see @xcite ) . another problem which was ignored in our consideration is the role of relaxation and thermal fluctuations of different origin . even at low temperature the interaction of spins with other crystal subsystems ( lattice , nuclear spins , etc . ) may be very important , see @xcite . it is clear that stochastic influence on the dynamics of magnetization from thermal fluctuations leads to decoherence and suppresses coherent tunneling . description of this fundamental problem in any detail goes far beyond the scope of the present lecture , and we refer the reader to the review by caldeira and leggett @xcite . one more problem which is unclear from our point of view is a justification of considering all spins in a small particle as moving coherently ( a `` giant spin '' approximation usually used in treating the mqt problems and also adopted in the present paper ) . in fact , the only justification of this approximation is energetical : if the particle size is much smaller than the characteristic domain wall thickness , any inhomogeneous perturbation costs much energy . on the other hand , for the hamiltonian ( [ singspinh ] ) neither @xmath415 nor @xmath416 are good quantum numbers , which means presence of magnons ( deviations from collinear order ) in the ground state . our lecture was devoted first of all to the fundamental aspects of mqt considered as a beautiful physical phenomenon which is rather difficult to observe . but technological development can lead to the situation when this phenomenon will become practically important . the present tendency of increasing the density of recording in the development of information storage devices means decrease of the elementary magnetic scale corresponding to one bit of information , and one may expect that quantum effects will determine the `` natural limit '' of miniaturization in future . sections [ sec : dm ] and [ sec : rings ] were written together with v. kireev whom the authors wish to thank for cooperation . this work was partially supported by the grant 2.4/27 `` tunnel '' from the ukrainian ministry of science . v. g. baryakhtar , m. v. chetkin , b. a. ivanov , and s. n. gadetskii , _ dynamics of topological magnetic solitons . experiment and theory . _ springer tracts in modern physics , v. 129 , springer verlag , 1994 . d. d. awschalom , j. f. smyth , g. grinstein , d > p. divincenzo , and d. loss , phys . lett . * 68 * , 3092 ( 1992 ) ; see also the comment by a. garg , phys . * 70 * , 2198 ( 1993 ) and d. d. awschalom _ et al . _ reply , ibid . , p. 2199